141x PE Requires Six Load-Bearing Walls to Hold Simultaneously

ARM Holdings (NASDAQ: ARM) In-Depth Stock Research Report

Analysis Date: 2026-02-27 · Data as of: FY2026 Q3 (ending December 2025)

Chapter 1: The Essence of ARM's Business — The Semiconductor Industry's "Invisible Tax Collector"

1.1 Understanding ARM in One Sentence

If the global semiconductor industry were an economy, ARM would be its tax authority—it owns no factories, manufactures no chips, yet it collects an "IP tax" from the over 28 billion chips shipped annually. This tax rate is very low (about $0.065 per chip on average), but the tax base is extremely broad (covering over 99% of all smartphones, over 70% of IoT devices, and rapidly growing server and automotive markets).

This analogy isn't perfectly accurate—a tax authority has enforcement power, which ARM lacks. ARM's "tax base" is built on a thirty-five-year-old instruction set ecosystem: billions of lines of compiled code, the muscle memory of millions of developers, the design workflows of thousands of companies, and the implicit switching cost that "re-validating software for a new architecture would take years." This gives ARM's revenue a "tax-like" quality—low unit price, high certainty, and strong correlation with the volume of end-market chips.

Quantifying ARM's "Tax Base": As of 2025, the ARM ecosystem includes:

• Compiled Code: An estimated 10+ billion lines of ARM native compiled code running globally (including Android/iOS/server/embedded).
• Developers: Over 20 million registered developers using ARM toolchains (Source: ARM Developer Hub).
• Design Workflows: Over 500 semiconductor companies hold ARM licenses (from Apple to small Chinese MCU vendors).
• Shipment Base: Over 280 billion ARM-based chips shipped to date (cumulative, not annual).
• Educational Foundation: Most major universities worldwide use ARM as the teaching platform for computer architecture courses (though RISC-V has begun to penetrate in recent years).

The depth and breadth of this ecosystem constitute ARM's truest moat—and what is most difficult for RISC-V to replicate. RISC-V can be free at the instruction set level (open source), but it cannot replicate a 35-year-old software ecosystem. However, RISC-V doesn't need to fully replicate ARM's ecosystem—it only needs to establish sufficient software support in specific end markets (IoT, DC, Automotive). This is why ARM's moat is "erodible market by market" rather than "globally irreplaceable".

But unlike a real tax, ARM faces an open-source, royalty-free alternative architecture: RISC-V. It's as if a sovereign nation declared, "Any business that relocates here will be permanently tax-exempt." In the short term, relocation costs (software ecosystem migration) keep most businesses from leaving. But in the long term, every time ARM raises its "tax rate" (v9 royalties doubling, CSS doubling again), it increases the economic incentive to "relocate".

This is the core tension of the ARM investment thesis: Its current high growth (+26% YoY) and continuous price hikes (v9 = 2×v8) are both the engine for short-term financial performance and the catalyst for its own long-term disruption.

1.2 Business Model Deconstructed: The Dual Revenue Engines

ARM's revenue is driven by two complementary engines:

Engine 1: Licensing Revenue (FY2025: $1.839B, 46%)

• Customers pay a one-time or installment-based license fee to gain the right to design chips using ARM IP.
• Revenue Recognition: Upon contract signing or milestone achievement.
• Analogy: A software platform's "developer license fee"—charge an entry fee first, then collect transaction commissions.
• Characteristics: High-value, concentrated (top 5 customers account for 56%), and volatile (depends on the timing of major customer deals).
• FY2025: $1.839B (+28.7% YoY), growing faster than royalties.

Engine 2: Royalty Revenue (FY2025: $2.168B, 54%)

• For every chip shipped containing ARM IP, the customer pays ARM a royalty based on an agreed-upon rate.
• Revenue Recognition: When customers report their shipments (typically with a 1-2 quarter lag).
• Analogy: Visa/Mastercard's "per-transaction fee"—the more transactions, the higher the revenue.
• Characteristics: Stable, diversified (>500 licensees), and predictable (highly correlated with global chip shipments).
• FY2025: $2.168B (+20.2% YoY).

The Dual-Engine Flywheel: Licensing revenue is a "leading indicator" for royalties—licensing agreements signed today turn into royalty streams in 3-5 years. The acceleration in licensing revenue in FY2024-FY2025 (+28.5%/+28.7%) points to an acceleration in royalties in FY2028-FY2030. This flywheel effect leads analysts to give ARM a premium for its long-duration growth—they see not just the current $4B in revenue, but the seeds of a potential $11.2B revenue in 5 years already being sown.

But the flywheel has a dark side: Of the $505M in licensing revenue for Q3 FY2026, SoftBank (ARM's 90% controlling shareholder) contributed approximately $200M (~40%). When your largest customer is also your controlling shareholder, the commercial independence of "customer deals" must be questioned. One of the core reasons for BofA's downgrade of ARM was precisely that, after excluding SoftBank, licensing revenue growth was actually negative (~-5%).

Quarterly Revenue Trend (Last 8 Quarters):

Quarter	Total Revenue ($M)	YoY	Gross Profit ($M)	Gross Margin	Operating Profit ($M)	OPM
Q1 FY25	939	+39%	885	94.2%	182	19.4%
Q2 FY25	844	+5%	790	93.6%	64	7.6%
Q3 FY25	983	+19%	932	94.8%	175	17.8%
Q4 FY25	1,241	+34%	1,189	95.8%	410	33.1%
Q1 FY26	1,053	+12%	993	94.3%	107	10.2%
Q2 FY26	1,135	+34%	1,106	97.4%	163	14.4%
Q3 FY26	1,242	+26%	1,170	94.2%	191	15.4%
TTM	4,671	+24%	4,457	95.4%	871	18.7%

Key Observations: Operating profit margins fluctuate significantly between quarters (7.6%-33.1%), primarily due to the recognition of large licensing agreements in different quarters. The 33.1% OPM in Q4 FY25 is an exceptionally high point (possibly including a large SoftBank licensing deal), while the 7.6% in Q2 FY25 reflects a trough in license recognition. This quarterly volatility makes annual trends more meaningful than quarterly analysis.

1.3 Revenue Quality Prism: High Gross Margin ≠ High Quality

ARM's financial figures appear exceptionally strong on the surface:

However, upon closer inspection, at least three layers of distortions affect the authenticity of these figures:

Distortion 1: The SBC Black Hole
Quarterly tracking of SBC reveals its true scale. While some quarters in the FMP cash flow statement show SBC = 0 (a data parsing flaw), quarters with data provide reliable anchors:

Quarter	SBC (FMP, $M)	Notes
Q4 FY24	185	Benchmark
Q1 FY25	182	Stable
Q2 FY25	218	Beginning to climb
Q3 FY25	227	Continuing to rise
Q4 FY25	193	Cyclical fluctuation
Q1 FY26	241	New high
Q2 FY26	265	Accelerating
Q3 FY26	285	StockAnalysis confirmed (FMP missing)

Annualized SBC Estimate: FY2025 SBC = $820M. FY2026 first three quarters = $241+265+285 = $791M (9 months). Annualized approximately $1,050-1,140M — this indicates that SBC is accelerating (+28% YoY). It accounts for approximately 22-24% of revenue. The significant difference (25.6pp) between Non-GAAP operating profit margin of ~41% (Q3 FY2026) and GAAP operating profit margin of 15.4% is primarily caused by SBC. This means ARM's annual stock-based compensation expense is equivalent to nearly one-fifth of its revenue — while it doesn't impact current cash flow, it continuously dilutes shareholder equity.

Distortion 2: Abnormal Accounts Receivable
FY2025 accounts receivable surged from $1.117B to $1.749B (+57%), while revenue growth was only +24%. The growth rate of accounts receivable was 2.4 times that of revenue. FY2025 operating cash flow was only $397M (vs. net income of $792M), 50% of net income, with working capital consuming $1.465B.

Quarterly accounts receivable tracking:

Quarter	Accounts Receivable ($M)	QoQ Change	Revenue ($M) (Current Qtr)	DSO (Days)
Q4 FY24	1,117	—	928	109
Q1 FY25	1,113	-4	939	121
Q2 FY25	1,460	+347	844	174
Q3 FY25	1,451	-9	983	146
Q4 FY25	1,749	+298	1,241	138
Q1 FY26	1,315	-434	1,053	184
Q2 FY26	1,766	+451	1,135	157
Q3 FY26	1,832	+66	1,242	146

DSO consistently much higher than EDA peers (Synopsys ~70 days, Cadence ~80 days). ARM's average of ~150 days means each receivable takes nearly 5 months to collect. The 184-day DSO in Q1 FY26 is anomalous—it likely reflects the delayed settlement effect of large year-end contracts.

Distortion 3: Concentration of Related-Party Transactions
As a 90.6% controlling shareholder, SoftBank also contributes approximately 30% of ARM's licensing revenue. This creates a classic "paying yourself" scenario. Recently confirmed data clarifies this cycle further:

• Margin Loan Size: $20 billion facility (expanded from $13.5 billion in November 2025), $8.5 billion drawn, collateral held by 33 financial institutions
• This is the largest margin loan in tech industry history
• ARM accounts for 54.6% of SoftBank's Net Asset Value (NAV) ($122B out of $224B NAV)
• Margin Call Trigger: A **40%** drop in ARM's share price triggers a margin call
• SoftBank is also seeking an additional $5 billion loan (collateralized by ARM shares) specifically for OpenAI investment
• OpenAI Commitment: SoftBank must transfer $22.5 billion to OpenAI by December 31, 2025
• Stargate: A $500 billion AI data center infrastructure plan, co-invested by SoftBank and OpenAI
• Project Izanagi: A $100 billion initiative to transform ARM from an IP licensor into a vertically integrated AI chip champion

1.4 ARM vs Visa: A Dangerous Analogy

A popular narrative in the market is that "ARM is the Visa/Mastercard of the semiconductor industry"—transactional revenue, asset-light, high gross margins, network effects. Superficial data supports this analogy:

Dimension	ARM	Visa	Similarities	Differences
Gross Margin	95.4%	97.4%	Extremely high	Similar
Asset Model	IP Licensing (zero manufacturing)	Payment Network (zero lending)	Asset-light	Similar
Revenue Driver	Chip shipments	Transaction volume	Transaction volume driven	ARM has pricing leverage
Customer Switching Costs	Software ecosystem (high)	Merchant/cardholder inertia (high)	High lock-in	Different mechanisms
P/E (TTM)	140x	33x	—	4.2x Difference
Revenue Growth	+24%	+11%	—	ARM 2x faster
Open-source Alternative	RISC-V (yes)	None	—	Structural Difference
Regulatory Pressure	Low	High (rate regulation)	—	ARM advantage

However, the analogy breaks down at three critical points:

1. Visa has no open-source alternative. RISC-V is an alternative architecture with zero licensing fees and performance nearing parity. Payment networks have no equivalent—there is no "open-source Visa" for merchants to join for free. This is not a question of "ARM is better than Visa" or "ARM is worse than Visa," but rather two fundamentally different moat structures. Visa's moat is network effects + scale economies + regulatory access, while ARM's moat is ecosystem lock-in + intellectual property + performance leadership. The former implies "no alternative is possible" (the cost to replicate a two-sided network effect is astronomical), while the latter implies "no sufficiently good alternative yet" (but RISC-V is closing the gap).

2. Visa's "tax rate" is on a downward trend. Global payment processing fees are continuously decreasing due to regulatory pressure (EU cap of 0.3%, US Durbin Amendment). ARM, in contrast, is raising prices—v9 royalty rates are double those of v8. This is not a reflection of market pricing power, but opportunistic pricing during a window when RISC-V is not yet mature. Visa's "tax rate" is falling due to regulatory mandates; ARM's "tax rate" is rising due to a short-term lack of effective competition. Once RISC-V becomes "effective competition" (possibly by 2028-2030), ARM's pricing power will face Visa-like compression—but the driver will be market competition, not regulation.

3. Visa's two-sided network effect is self-reinforcing. More merchants → more cardholders → more merchants. ARM's network effect is one-sided—more developers → more software → but developers are also writing code for RISC-V concurrently (Google has integrated RISC-V into Android GKI). ARM's network effect is being hedged. More precisely, ARM faces a "parallel ecosystem development" risk—RISC-V does not need to "pull developers away from the ARM ecosystem," but rather "enable new developers to work on both platforms simultaneously." Software toolchains (LLVM/GCC) already support both ARM and RISC-V, meaning the switching costs at the compilation layer have been eliminated.

4. Valuation Implications: Visa's P/E (33x) already incorporates negative expectations regarding fee regulation. ARM's P/E (140x) not only fails to incorporate negative expectations from RISC-V competition but also implies optimistic assumptions of continuous price increases (v9 → CSS → v10). If ARM's P/E were to revert to a reasonable range for an "infrastructure company with alternatives" (40-60x P/E), the share price would fall by 57-71% to $55-$37. This is not a prediction; it is mathematics.

Conclusion: ARM is not Visa. ARM is more like a government collecting as much tax as possible before open-source alternatives mature—this strategy is very effective during the window of opportunity (v9 price increase + CSS bundling = +20% royalties in FY2025), but each price increase shortens that window.

1.5 Evolution of ARM's Global Chip "Tax Rate" and International Comparison

ARM's "tax rate" (royalties as a percentage of total chip revenue) has undergone significant changes over the past 10 years:

Evolution of ARM's "Tax Rate":

FY	Global ARM Chip Shipments (B)	ARM Royalties ($B)	Global Semiconductor Market ($B)	ARM "Tax Rate"	YoY
FY2019	22.8	1.34	412	0.33%	—
FY2020	25.3	1.38	440	0.31%	-2pp
FY2021	27.8	1.56	560	0.28%	-3pp
FY2022	30.6	1.68	574	0.29%	+1pp
FY2023	30.6	1.80	527	0.34%	+5pp
FY2024	31.1	1.80	542	0.33%	-1pp
FY2025	30.6	2.17	600E	0.36%	+3pp
FY2026E	31.0	2.95	630E	0.47%	+11pp
FY2030E	35.0	7.50	800E	0.94%	—

Key Observations:

1. Between FY2019-FY2022, ARM's "tax rate" actually declined (0.33%→0.28%)—because chip shipment growth (+34%) significantly outpaced royalty growth (+25%), with low-value IoT chips dragging down the average tax rate.
2. Starting FY2023, v9 price increases began to take effect, and the tax rate rebounded to 0.36%.
3. The consensus FY2030E tax rate of 0.94% requires nearly a 3x increase from the current 0.36%—this has no precedent in IP licensing history.
4. Even Qualcomm QTL (the closest analogy) has a "tax rate" of only ~3.3%, but QTL covers a much narrower market than ARM.

International Tax Rate Analogy: If ARM's royalties are analogous to "taxes," an effective tax rate of 0.36% is comparable to a very low-tax country (Singapore/Ireland level). The market-implied 0.94% (FY2030E) approaches a "medium tax rate"—the problem is that ARM lacks the enforcement power of a sovereign nation to maintain this tax rate. In a world with RISC-V "tax havens," a rise in ARM's tax rate will accelerate "capital flight" (customer migration).

1.6 ARM's "Second Category Growth Trap"

Understanding ARM requires distinguishing between two categories of growth:

First Category Growth: Royalty Volume-Driven — More chip shipments → More royalties. This is sustainable, low-risk growth, similar to Visa's transaction volume growth. ARM's royalty growth in mobile falls into this category (stable growth in global smartphone shipments).

Second Category Growth: Royalty Rate-Driven — Earning more money per chip. This has been ARM's primary growth engine in recent years: v8→v9 (royalty rate doubles), TLA→CSS (royalty rate doubles again by 2-3x). This is not sustainable growth, but a one-time pricing ladder—each step of which will be completed, and with each step completed, the economic incentive for RISC-V substitution increases.

The Trap: The market misprices Second Category Growth (royalty rate increases) as First Category Growth (sustainable volume growth). When v9 penetration reaches 80%+ and CSS penetration reaches 50%+ (estimated FY2028-FY2029), royalty rate-driven growth will decelerate significantly. At that point, the market will "discover" that ARM's growth is not the kind of growth they thought it was—this is typically a trigger for significant valuation correction.

Quantitative Measurement:

• Of the +20% royalty growth in FY2025, approximately 12pp comes from royalty rate increases (v9 penetration from 20%→30%), and only about 8pp from shipment volume growth.
• FY2026-FY2027: v9 penetration from 50%→70%, royalty rate contribution is still accelerating.
• FY2028+: v9 penetration approaches saturation, royalty rate contribution decelerates → Unless v10/CSS continues to double → but that would further accelerate RISC-V adoption.

Core Questions (CQ) Checklist

This report analyzes six core questions. Each CQ is progressively examined in subsequent chapters, culminating in a closed-loop evaluation in Chapter 32.

CQ-1: Royalty Economics End State (Weight 25%)

Question: What level can ARM royalties reach by FY2030? Can v10 price increases be sustained?
End State Judgment: Confidence level 57%. Royalty trajectory determines 60% of ARM's revenue, directly impacting fair value by ±$30-50B. v9→v10 price increase by 1.3-2.0x, but when the RISC-V pricing ceiling takes effect is the biggest uncertainty.
Key Uncertainties: v10 royalty rate unconfirmed; CSS mass production cancellation rate unknown; when RISC-V's pricing ceiling for royalties takes effect.
See Also: Chapter 2 (Royalty Economics), Chapter 21 (Royalty Reverse Engineering), Chapter 32 (CQ Closed Loop)

CQ-2: Data Center Journey (Weight 20%)

Question: How much can ARM's share in data centers grow from 12%? Can the CSS model be sustained?
End State Judgment: Confidence level 52%. Data centers are ARM's fastest-growing segment, with an estimated FY2030 share of 18-38% (scenario-dependent), impacting fair value by ±$20-40B. Out of 19 CSS licenses, only 5 are in mass production, raising doubts about success rate.
Key Uncertainties: When the AI CapEx cycle will peak; CSS mass production cancellation rate; RISC-V penetration speed in data centers.
See Also: Chapter 5 (Data Center Journey), Chapter 15 (Quantifying RISC-V CDS Pricing Power), Chapter 32 (CQ Closed Loop)

CQ-3: SoftBank's Double-Edged Sword (Weight 18%)

Question: Is SoftBank's 90% ownership a protection or a threat? When and how will it reduce its stake?
End State Judgment: Confidence level 58%. Net effect over 5 years approximately -$1B (neutral to slightly negative). SoftBank's actions could lead to a -$30~50B stock price impact (extreme scenario); margin call trigger point is approximately $85/share, and once initiated, the positive feedback loop is irreversible.
Key Uncertainties: Masayoshi Son's personal decisions are unpredictable; scale of Izanagi's funding needs; precise margin terms not disclosed.
See Also: Chapter 20 (SoftBank Related Party Transactions), Chapter 32 (CQ Closed Loop)

CQ-4: RISC-V Pricing Ceiling (Weight 15%)

Question: How significant is the long-term limitation of RISC-V's existence on ARM's pricing power?
End State Judgment: Confidence level 63% (highest among 6 CQs). Pricing power is the 'linchpin' among the six load-bearing walls—with a risk product of $23B, it is the highest, and the only single factor that can independently alter valuation by more than 20%. ARM can raise prices up to the 'annualized migration cost' but no higher, and this ceiling will significantly decrease by 2030.
Key Uncertainties: Timing of RISC-V software ecosystem's 'critical mass'; whether v10 features are sufficient to 'reset' ecosystem lock-in; the actual impetus from the dual engines of Google and China.
See Also: Chapter 14 (RISC-V Special Report), Chapter 15 (Quantifying CDS Pricing Power), Chapter 28 (Joint Probability of Load-Bearing Walls), Chapter 32 (CQ Closed Loop)

CQ-5: Arm China Deceleration (Weight 10%)

Question: What are the revenue and governance risks for Arm China (Anmou Technology)?
End State Judgment: Confidence level 65% (highest). Arm China accounts for 19% of revenue ($900M+), and Chinese policy promoting domestic substitution poses a long-term risk. Extreme scenario (revenue zeroed out) impacts enterprise value by -$15-25B; moderate scenario (revenue decline 30%) impacts by -$5-8B.
Key Uncertainties: Intensity of Chinese policy implementation; possibility of Arm China independent IPO; geopolitical risks in the Taiwan Strait (unknowability 9/10).
See Also: Chapter 19 (Arm China Spin-off Valuation), Chapter 25 (In-depth Analysis of Geopolitical Risks), Chapter 32 (CQ Closed Loop)

CQ-6: Phoenix Transformation Paradox (Weight 12%)

Question: Is ARM's transition from an IP licensor to a chip designer an opportunity or a risk?
Final State Assessment: Confidence Level 45% (lowest, highest uncertainty). The Five-Terminal-State Model shows a probability-weighted net effect of only +$1.75B (approximately neutral). Phoenix should not be a reason to buy or sell, but the backlash contagion chain (customers accelerating evaluation of RISC-V alternatives) is a tail risk that needs to be monitored.
Key Uncertainties: TSMC capacity allocation; Scale of SoftBank's internal demand; Whether customers will accelerate their RISC-V evaluation as a result.
See Also: Chapter 6 (Phoenix Plan), Chapter 32 (CQ Closed-Loop)

This is a self-limiting growth model: the faster royalty rates increase, the more attractive RISC-V alternatives become, and ultimately, the less room there is for further royalty rate increases. The market's optimistic expectations for ARM (FY2030 EPS $4.70) implicitly assume this self-limiting mechanism will not take effect within five years—a critical foundational assumption that requires serious scrutiny.

---

Chapter 2: Royalty Economics — The Four-Tiered Price Escalation from v8→v9→v10→CSS

2.1 Evolution of ARM's Pricing Power: Four Eras

ARM's pricing strategy has evolved through four phases, each significantly increasing per-unit value capture:

Phase 1: Armv8 Era (2011-2020) — "Wide Net, Low Royalty Rates"

• Royalty Rate: $0.03-0.07/chip (IoT), $0.30-0.60/chip (Mobile AP), $5-15/chip (Servers)
• Strategy: Maximize market share, become the default choice
• Result: ARM architecture penetration >95% in smartphones, >70% in embedded systems
• Annual Royalty Revenue: ~$1.5-1.6B (FY2022-FY2023)
• Philosophy: "Our royalty rates are so low that no one has an incentive to build alternatives"

Phase 2: Armv9 Era (2021-Present) — "Architectural Upgrade, Double Royalties"

• Royalty Rate: 2x+ that of v8 (Mobile AP from $0.35→$0.70+, Servers from $10→$20-30+)
• Strategy: Drive customer migration through mandatory architectural upgrades (new features only available on v9)
• New Features: Security Extensions (Confidential Compute), AI Acceleration Instructions (SVE2), Trace Mechanisms (BRBE)
• Adoption Progress: >50% of royalty revenue already from v9 (Q3 FY2026, rapidly rising from ~20% in FY2024)
• Revenue Impact: FY2025 royalties $2.168B (+20%), incremental growth mainly from v9 price increases

The logic behind v9's price increase: ARM added Security Extensions (Confidential Compute), AI Acceleration Instructions (SVE2), and Trace Mechanisms (BRBE). These features offer real value to data center customers but are a mandatory "bundling" for IoT/low-end mobile chip customers—who don't need these features, but v8 no longer receives support for new core designs, eventually forcing them to upgrade to v9 and accept higher rates.

Phase 3: CSS Era (2023-Present) — "Pre-integrated, Doubled Again"

• Royalty Rate: 2-3x that of TLA (using ARM core design)
• Strategy: ARM not only provides core IP but also pre-integrated chip subsystems (CPU+interconnect+memory controller)
• Customer Value: Reduces design time by 12 months, saves tens of millions in NRE
• Adoption Progress: 21 CSS licenses, 12 customers, 5 in mass production (latest as of Q3 FY2026)
• Management Goal: CSS to account for >50% of royalty revenue (within 2-3 years)
• CSS Royalty Rate: ~10% of chip ASP (vs historical v8 level 1-2%, v9 ~5%) — This represents a 3-4x increase in value capture

The logic for CSS pricing power is healthier than v9's—ARM genuinely performs more design work (from core IP to complete subsystems), making it reasonable for customers to pay a premium. However, CSS also carries an implicit risk: the deeper ARM delves into chip design, the more likely it is to directly compete with licensees (especially with the Phoenix project).

Phase 4: Armv10 (Timeline Uncertain) — "May not appear as v10 at all"

• Key Finding: Armv10 does not yet exist. ARM officially refuses to discuss the v10 roadmap (The Register, October 2024). The current strategy is to continuously enhance functionality through annual dot updates of Armv9 (v9.0→v9.7), rather than major version jumps.
• Historical Pace: v8 (2011)→v9 (2021) = 10-year interval. If the pace remains similar, v10 might not appear until ~2031.
• ARM claims a next-generation architecture will only be considered after v9 ships in "at least 300 billion devices" → implying v9's lifecycle will be much longer than the 10 years from v8→v9
• ARM's recent IP roadmap focuses on the Lumex brand (CPU C1+GPU G1 in 2026) and CSS expansion, rather than major architectural upgrades
• Important Correction for Analysis: Previous reports assumed v10 would be released in 2027-2028—this assumption now appears overly aggressive. If v10 is delayed to 2030+, it means v9+CSS will be ARM's sole royalty rate growth engine for the next 5 years, and royalty rate growth could be slower than consensus expectations.
• Impact on RISC-V: If ARM does not release v10, it might paradoxically reduce the urgency for RISC-V (as there would be no new mandatory upgrade cycle to trigger migration assessment). However, it also means ARM's royalty rate increases will entirely depend on CSS penetration (rather than architectural shifts).

2.2 Royalty Rate × Shipments Matrix: ARM's Revenue Formula (Detailed Version)

ARM's royalty revenue can be broken down as follows:

Based on management guidance, analyst estimates, and FMP data, we construct the following matrix:

FY2025 Royalty Matrix (Estimated):

End Market	Annual Shipments (M)	v8 Share	v9 Share	CSS Share	Weighted Royalty Rate	Annual Royalty ($M)	Share
Mobile AP	1,400	55%	40%	5%	$0.72	1,008	46%
IoT/Embedded	18,000	80%	15%	5%	$0.035	630	29%
Data Center	15	10%	50%	40%	$28.0	420	19%
Automotive	600	70%	25%	5%	$0.85	510	—
PC/Laptops	40	20%	60%	20%	$6.50	260	—
Networking/Other	2,000	60%	35%	5%	$0.12	240	—
Total	~22,000	—	—	—		~2,168	100%

FY2028E Royalty Matrix (Estimate — Base Case Scenario):

End Market	Annual Shipments (M)	v8 Share	v9 Share	CSS Share	v10 Share	Weighted Royalty Rate	Annual Royalty ($M)
Mobile AP	1,500	15%	50%	30%	5%	$1.35	2,025
IoT/Embedded	20,000	50%	30%	15%	5%	$0.055	1,100
Data Center	40	0%	30%	50%	20%	$42.0	1,680
Automotive	900	30%	40%	25%	5%	$1.50	1,350
PC/Laptop	80	5%	40%	40%	15%	$9.00	720
Networking/Other	2,500	30%	40%	25%	5%	$0.20	500
Total	~25,000	—	—	—	—	~$0.295 avg	~7,375

Key Insights: Decomposition of Royalty Growth from $2.17B to $7.38B (+240%) from FY2025 to FY2028E:

• Shipment Growth Contribution: 22B→25B = +14% → Contributes ~$300M
• Royalty Rate Increase Contribution: From $0.099 to $0.295 = +198% → Contributes ~$4,900M
• Conclusion: 95% of royalty growth comes from price increases (v9/CSS/v10 penetration), only 5% from shipments

This reinforces the assessment in Section 1.5, "The Second Growth Trap" — the market is pricing not shipment growth (5%), but pricing power (95%). Once price increases face obstacles (RISC-V competition, customer pushback, v10 delays), growth will decelerate significantly.

2.3 Royalty Rate Tier Quantitative Model: Detailed Estimation of Royalty Rates for Each Generation v8→v9→v10

To more precisely understand ARM's pricing trajectory, we have constructed a three-generation royalty rate tier for each end market:

Mobile AP Royalty Rate Evolution:

Version	Base Core IP Royalty	CSS Premium	Effective Royalty Rate	Corresponding ASP	% of ASP
v8 Cortex-A78	$0.30-0.50	N/A	$0.35 avg	$25	1.4%
v9 Cortex-A710/X3	$0.60-1.00	N/A	$0.70 avg	$30	2.3%
v9 CSS (N2/V2)	—	2-3x TLA	$1.50-2.10	$35	5.1%
v10 (Estimate)	$1.00-1.50	—	$1.20 avg	$35	3.4%
v10 CSS (Estimate)	—	2-3x TLA	$2.50-3.60	$40	7.5%

Data Center Royalty Rate Evolution:

Version	Core IP Royalty	CSS Premium	Effective Royalty Rate	Chip ASP	% of ASP
v8 Neoverse N1	$5-10	N/A	$8 avg	$500	1.6%
v9 Neoverse V2	$15-30	N/A	$20 avg	$800	2.5%
v9 CSS (V2/N2)	—	2-3x	$40-90	$1,000	6.5%
v10 Neoverse V3+	$25-50	—	$35 avg	$1,200	2.9%
v10 CSS (Estimate)	—	2-3x	$70-150	$1,500	7.3%

IoT/Embedded Royalty Rate Evolution:

Version	Core IP Royalty	Effective Royalty Rate	Chip ASP	% of ASP
v8 Cortex-M4/M7	$0.02-0.05	$0.03 avg	$1.50	2.0%
v9 Cortex-M55/M85	$0.04-0.10	$0.06 avg	$2.00	3.0%
v10 (Estimated)	$0.08-0.15	$0.10 avg	$2.50	4.0%

2.4 Pricing Ceiling: Triple Mathematical Constraints

ARM's royalty rate increases face three ceilings:

Ceiling 1: Chip ASP Constraint (Hard Cap)
The proportion of royalty rates to chip ASP has an implicit upper limit. Historically, IP licensing rarely exceeds 5-8% of chip ASP:

Ceiling Range	Mobile AP	Data Center	IoT	Automotive
Current % of ASP	2.3%	2.5%	2.0%	2.5%
v9 CSS % of ASP	5.1%	6.5%	N/A	4.5%
v10 CSS % of ASP	7.5%	7.3%	4.0%	6.0%
Historical Cap	~8%	~8%	~5%	~8%
Headroom	0.5pp	0.7pp	1.0pp	2.0pp

Conclusion: v10 CSS royalty rates will bring ARM close to or touch the historical upper limit of IP licensing as a percentage of ASP in multiple end markets. The room for further price increases is extremely limited—unless chip ASP itself significantly rises (which depends on Moore's Law continuing to reduce area/cost).

Ceiling 2: RISC-V Replacement Cost Constraint (Soft Cap)
RISC-V's core licensing fee = zero. ARM's royalty rate is essentially capped by the "RISC-V migration cost (amortized over expected shipments)":

End Market	Estimated RISC-V NRE	Annual Shipments	5-Year Amortization/Chip	ARM Current Royalty	Buffer
IoT	$5-20M	1B+	$0.004-0.004	$0.035	ARM 8x Costlier
Mobile AP	$200-500M	500M	$0.08-0.20	$0.72	ARM 3.6-9x Costlier
Data Center	$500M-2B	10M	$10-40	$28	ARM 0.7-2.8x
Automotive	$100-300M	100M	$0.20-0.60	$0.85	ARM 1.4-4.3x Costlier

Key Finding: In the IoT market, ARM's royalty is already 8 times the RISC-V migration cost—this is why RISC-V has already gained a 30% share in the IoT market. In the Mobile AP market, ARM still has a 3.6-9x "safety buffer"—primarily due to the massive migration costs of the Android/iOS software stack. However, if ARM continues to raise prices to v10 CSS levels, this safety buffer will shrink to 2-5x, beginning to touch the migration decision point for major customers.

Qualcomm spent $2.4B to acquire Ventana Micro for RISC-V IP. If Qualcomm ships 1 billion chips annually, amortized over 5 years, this amounts to $0.48/chip. If ARM's royalty rate to Qualcomm exceeds $0.50/chip, Qualcomm would have a financial incentive to use RISC-V.

Ceiling 3: Customer Concentration Constraint (Negotiation Ceiling)
ARM's top 5 customers (including Arm China) account for 56% of its revenue. This means ARM cannot aggressively price for major customers—the loss or downgrade of any major customer would result in a double-digit revenue impact:

Customer (Estimated)	% of ARM Revenue	Annual Payment to ARM ($M)	Replacement Incentive
Apple	~12-15%	$500-600	Low (ALA, In-house Core)
Qualcomm	~10-12%	$400-500	High (Ventana Acquisition)
Samsung	~8-10%	$300-400	Medium (Exynos Decline)
Arm China	~15-18%	$600-700	High (Policy-driven RISC-V Adoption)
MediaTek	~6-8%	$250-350	Medium (Follows Market)
Top 5 Total	~56%	~$2,250	—

Qualcomm's legal dispute with ARM (2020-2024) has demonstrated that major customers are willing to resist price increases through legal means. Arm China's 15-18% revenue exposure is directly affected by US-China relations.

2.5 ARM's "Pricing Window" Theory: How Much Time Remains?

ARM's current pricing strategy (v9 doubled, then CSS doubled again) is essentially to collect as much royalty as possible before "RISC-V becomes a viable alternative." How wide is this "pricing window"?

Estimated Window Width — By End Market:

End Market	Time for RISC-V to Reach Performance Parity	Time for Ecosystem Maturity	Total Window	ARM's Remaining Pricing Window
IoT/MCU	Achieved (2023+)	Achieved	0 years	Closed
Low-End Smartphones	2027-2028	2029-2030	3-5 years	1-3 years
DC/Servers	2026 (Tenstorrent)	2028-2029	2-3 years	0-1 years
Automotive Body	2027-2028	2028-2030	2-4 years	1-2 years
High-End Smartphones	2029-2030	2031-2033	5-7 years	3-5 years
Automotive ADAS	2029-2030	2031-2033	5-7 years	3-5 years
PC/Laptops	N/A (x86 competition)	N/A	—	—

Weighted Average Pricing Window: ~2.5 years (revenue-weighted)

This implies that ARM's "free pricing period" in most end markets may conclude in 2028-2029. After this, ARM will be forced to choose between "maintaining royalty rates (risk of losing customers)" and "reducing royalty rates (to remain competitive)" – a dilemma Qualcomm QTL faced from 2015-2020 (forced to cut mobile device royalties from 5% to 3.25%).

Implications for Valuation: If the market is pricing ARM based on "perpetual pricing power" (P/E 140x), but ARM's actual pricing window is only ~2.5 years, then EPS growth for FY2028-FY2030 will be significantly lower than market expectations. Specifically:

• Consensus FY2029 EPS: $3.70 (implying GAAP OPM ~30%, revenue $8.5B)
• If FY2029 royalty rate growth slows from +15%/year to +5%/year (pricing window closes): Revenue drops to $7.2B, EPS $2.50
• This means 140x × $2.50 = $350 → but the market will not assign a 140x multiple to a decelerating growth company → P/E could compress to 80-100x → stock price $200-250 (still upside from current $128, but far below the "perfect execution" path)

The More Critical Question: If ARM's growth rate slows from +25% to +10% after the pricing window closes, a P/E compression from 140x to 50-70x is reasonable. 50x × $2.50 = $125 — almost identical to the current share price. This implies that the current share price of $128 assumes the "pricing window will not close" – if this assumption proves false, investors buying at the current price will achieve near-zero returns.

2.6 v9 Penetration S-Curve: Accelerating but Peaking

Evolution of v9 royalty revenue share:

• FY2024: ~20% → FY2025: ~30% → Q3 FY2026: >50%

This acceleration is impressive – from 20% to 50%+ in 18 months. However, the latter half of an S-curve (50%→90%) is typically slower than the first half:

• Legacy v8 devices (IoT, automotive) have long replacement cycles (5-10 years)
• v8 still holds a pricing advantage in low-cost chips (royalty rate is half that of v9)
• Some customers opt for a v8+RISC-V hybrid strategy rather than upgrading to v9
• The ultimate penetration rate for v9 might be 75-80% rather than 100% (v8 will permanently exist in the low-end market)

Key Question: After v9 penetrates 70-80%, what will be the next growth engine? Management's answer is CSS (higher royalty rates) and v10 (not yet released). However, CSS currently only has 5 production designs (among 11 customers), and the v10 timeline is unclear – if v10 royalty rates continue to double (=4x of v8), the economic incentive for RISC-V migration will significantly strengthen.

2.7 The Ultimate Question of Royalty Economics: What is the "Terminal State" of Royalty Revenue?

If we assume ARM's royalties eventually reach a steady state (all pricing tiers exhausted, RISC-V erosion reaching equilibrium), where does ARM's terminal royalty stand?

Terminal State Modeling (FY2032E+):

End Market	Terminal Shipments (M)	ARM Share	RISC-V Share	Terminal Royalty Rate	Terminal Royalties ($M)
Mobile AP	1,500	85%	10%	$1.00	1,275
IoT	25,000	45%	45%	$0.04	450
DC	80	55%	20%	$35	1,540
Automotive	1,500	65%	25%	$1.20	1,170
PC	150	40%	5%	$7.00	420
Networking/Other	3,000	50%	30%	$0.15	225
Total	—	—	—	—	~5,080

Terminal Royalties ~$5.1B vs. Consensus FY2030E Royalties ~$7.5B — a gap of $2.4B (32%). The gap stems from:

1. RISC-V share assumptions (we are 10-15pp more conservative than consensus)
2. Royalty rate ceiling (we assume no significant price hikes after v10 CSS)
3. Shipment growth (relatively consistent)

This terminal state model implies: If RISC-V continues its current trajectory, ARM's royalty growth will reach an inflection point in FY2029-FY2030, followed by slow growth or even a plateau. The market-implied FY2030 EPS of $4.70 (corresponding to ~$11.2B revenue) likely overestimates the sustainability of royalties.

---

Chapter 3: Dissecting the Licensing Model — The Pricing Power Game of ALA/TLA/CSS

3.1 The Four-Tier Licensing Pyramid

ARM's licensing system is a meticulously designed value ladder:

ALA (Architecture License Agreement): Customers obtain a license for the ARM instruction set architecture, allowing them to design their own CPU cores. Approximately 15 companies hold an ALA. Royalty rates are relatively low because ARM only provides the architecture specification, and the design work is completed by the customer.

Known ALA Customers: Apple (A/M series), Qualcomm (Kryo→Oryon), Google (Tensor), Amazon (Graviton), Microsoft (Cobalt), Samsung (Exynos), NVIDIA (Grace), Fujitsu (A64FX), Ampere (Altra/AmpereOne), Huawei (Kunpeng), Alibaba (Yitian), Marvell, Broadcom, MediaTek (partial), Cavium (acquired by Marvell)

TLA (Technology License Agreement): Customers use off-the-shelf cores designed by ARM (Cortex series/Neoverse series). Royalty rates are higher than ALA because ARM undertakes all core design work. This is the choice of the vast majority of ARM customers—hundreds of chip companies use the Cortex-A/M/R series.

ATA (Total Access Agreement): Exclusive "all-inclusive package" access for large customers, obtaining the right to use ARM's entire product line with a one-time payment.

CSS (Compute Subsystems License): ARM's newest and most important license type. ARM not only provides CPU core IP but also offers pre-integrated subsystems comprising CPU + interconnect + memory controller + security unit. Royalty rates are 2-3 times that of TLA.

3.2 CSS: ARM's Value Capture Lever (In-depth Analysis)

CSS is ARM's most important strategic initiative in recent years. Understanding CSS means understanding ARM's growth narrative for the next 5 years:

Customer Pain Points Addressed by CSS:

• Traditional Process: Obtain core IP → Self-design subsystem (bus/interconnect/cache/PD) → Integration and verification → Tape-out (takes 24-36 months)
• CSS Process: Obtain pre-integrated subsystem → Direct integration into SoC → Tape-out (takes 12-24 months)
• Savings: Up to 12 months of design time + $20-50M NRE costs + Reduced risk of tape-out failure

CSS Pricing Logic and Customer Economics:

Dimension	TLA Model	CSS Model	Customer Net Benefit
ARM Royalty/chip	$X	$2-3X	-$X~2X
Customer NRE	$30-50M	$10-20M	+$10-30M
Time-to-market	24-36 months	12-24 months	+$100M+ opportunity cost
Tape-out risk	Medium	Low	Unquantifiable
Net Economic Effect	—	—	Significantly Positive

Even if CSS royalty rates are 2-3 times that of TLA, customer benefits from NRE savings and accelerated time-to-market far outweigh the incremental royalties. This is the underlying logic behind the accelerated adoption of CSS—ARM is increasing prices while creating real customer value, rather than purely implementing "monopoly surcharges." This fundamentally differs from the v9 price increase (feature bundling).

CSS Adoption Progress Tracking (as of Q3 FY2026):

Metric	Q1 FY25	Q2 FY25	Q3 FY25	Q4 FY25	Q1 FY26	Q2 FY26	Q3 FY26
Total CSS Licenses	12	14	15	16	17	19	21
CSS Customers	7	8	9	10	10	11	12
Designs in Mass Production	2	3	3	4	4	5	5

CSS Penetration Forecast (as % of Royalty Revenue):

Fiscal Year	CSS as % of Royalties	Driving Force
FY2025	~5-8%	Early mass production (2-3 customers)
FY2026E	~12-18%	Ramp-up of 5 mass production designs
FY2027E	~25-35%	Additional 5-8 mass production designs + DC acceleration
FY2028E	~35-45%	Management's path to 50% target
FY2030E	~50-60%	If CSS becomes the default choice

The timeline for CSS to reach 50% (management's target of "within 2-3 years") might extend to FY2029 instead of FY2028—because a large number of low-end TLA customers (IoT) do not require CSS.

3.3 SoftBank "Concentration" Issue in Licensing Revenue (In-depth Quantification)

ARM's licensing revenue structure exhibits a concerning concentration issue:

Annual Data:

Fiscal Year	Total Licensing Revenue ($M)	SoftBank-related Estimate ($M)	SoftBank Share	Growth excl. SB
FY2023	1,159	~200	~17%	—
FY2024	1,428	~350	~25%	+15%
FY2025	1,839	~550	~30%	+5%
FY2026E	~2,300	~800	~35%	~0%

Quarterly Details (FY2026):

Quarter	Total Licensing ($M)	SoftBank-Related Estimate ($M)	Excluding SB ($M)
Q1 FY26	350 (Est.)	~120	~230
Q2 FY26	390 (Est.)	~140	~250
Q3 FY26	505	~200	~305

BofA noted when it downgraded ARM in January 2026: Excluding SoftBank's contribution, FY2026 licensing revenue is projected to grow approximately 0% year-over-year or even decline. This implies that ARM's organic licensing growth has stalled.

Related-Party Transaction Risk Chain:

This self-reinforcing cycle is virtuous in an upswing (ARM stock price rises → SoftBank can borrow more → Invests more in AI → More ARM licenses → ARM revenue rises → Stock price rises). However, it will become a vicious cycle in a downswing (see SoftBank deep dive for details).

3.4 Analysis of Quarterly Volatility in Licensing Recognition

ARM's licensing revenue exhibits extremely high quarterly volatility, posing challenges for forecasting and valuation:

Licensing Revenue Quarterly Series (Derived):

Quarter	Total Revenue ($M)	Royalty (Est., $M)	Licensing (Est., $M)	Licensing YoY
Q1 FY25	939	508	431	+85%
Q2 FY25	844	486	358	-15%
Q3 FY25	983	550	433	+24%
Q4 FY25	1,241	624	617	+52%
Q1 FY26	1,053	569	484	+12%
Q2 FY26	1,135	590	545	+52%
Q3 FY26	1,242	737	505	+17%

Sources of Volatility: The timing of large contract (ATA/CSS) signings is unpredictable. A single ATA contract can exceed $100M, and a quarterly difference can lead to a 10% revenue fluctuation. Such volatility makes quarterly "beat/miss" almost random—the significant beat in Q4 FY25 might simply be due to a large customer (SoftBank?) signing early.

3.5 Game Theory Analysis of ARM's Pricing Strategy

ARM's pricing power is not a simple "supply-side monopoly pricing" issue—it is a multi-party game:

Game Participants:

1. ARM (Pricer): Goal = Maximize Royalty Rate × Royalty Base × Duration
2. Large Customers (ALA Holders): Apple/Qualcomm/Google/AWS — Possess in-house R&D capabilities, can threaten to switch to RISC-V
3. SME Customers (TLA/CSS Users): MediaTek/Broadcom, etc. — Dependent on ARM, lack in-house R&D capabilities
4. RISC-V Ecosystem: Zero-cost alternative, but immature ecosystem — a "threat" rather than a "realistic alternative"
5. SoftBank (Controlling Shareholder): Requires ARM revenue growth to maintain valuation/collateral value

Optimal Pricing Strategy (Nash Equilibrium Analysis):

ARM faces the classic "monopolist's pricing dilemma": Raising prices too quickly → accelerates RISC-V adoption → long-term revenue decline; Raising prices too slowly → insufficient short-term revenue → valuation decline → SoftBank pressure.

Strategy	Short-term Revenue Impact	Long-term Moat Impact	RISC-V Catalytic Effect	Overall Assessment
Aggressive Price Hikes (v9 Mandatory + CSS Bundling)	++	--	High	Current Strategy
Moderate Price Hikes (v9 Optional + CSS Voluntary)	+	Medium	Medium	Theoretically Optimal
Maintain Prices (v8→v9 Free Upgrade)	0	++	Low	Revenue Compression
Selective Pricing (Low Price for Large Customers/High Price for Small Customers)	+	+	Low-Medium	Difficult to Implement

Reasons ARM Chose the "Aggressive Price Hikes" Strategy:

1. SoftBank needs revenue growth to support valuation (collateral requirements)
2. RISC-V is not yet a sufficiently effective competitor in the short term (2025-2027)
3. The window for v9 mandatory upgrades + CSS bundling is limited (RISC-V narrows the gap annually)
4. Management KPIs (Non-GAAP EPS growth) are highly aligned with the price hike strategy

Game Theory Prediction: ARM's current strategy ("collect as much as possible before the window closes") is rational in the short term (2-3 years). However, this is a short-sighted equilibrium—similar to Intel's aggressive pricing utilizing its server monopoly from 2010-2015, which ultimately spawned two competitors, AMD Zen and ARM Neoverse. ARM is repeating Intel's path, but on a more compressed timeline (because RISC-V is open-source, its iteration speed may be faster than historical AMD).

3.6 Legal Structure and Audit Rights of Royalty Contracts

ARM's royalty contracts include several legal structural features important for analysis:

Contract Term: Licensing contracts typically last 5-15 years, with royalty obligations continuing throughout the contract period. Upon expiration, customers must renew or cease using ARM IP.

Audit Rights: ARM reserves the right to audit the shipment volumes of its licensees. This serves as the "enforcement mechanism" for royalty revenue—if a customer underreports shipment volumes, ARM can recover royalties through audits. Historically, audit recoveries have contributed $10-30M in unexpected revenue in individual quarters.

Minimum Royalty Guarantee (MRG): Some large contracts include an MRG, requiring a minimum annual royalty payment regardless of the customer's actual shipment volume. The presence of an MRG provides downside protection for royalty revenue but may also lead customers to decline renewal upon contract expiration (if the MRG significantly exceeds royalties corresponding to actual shipments).

Contract Assignability: The Qualcomm lawsuit (ruled in 2024) confirmed that ARM licenses acquired through M&A can be transferred to the acquirer. This weakens ARM's "customer-by-customer independent pricing" model—theoretically, new customers could acquire ARM licenses by purchasing small companies with existing licenses, rather than signing at ARM's current list price.

Geographic Restrictions: Arm China's contract structure is unique—Chinese customers obtain licenses through Arm China, and royalties are also remitted via Arm China. This creates geopolitical risks (see Ch7) and also increases collection delays (impacting DSO).

Contract Renewal Risk: Approximately 30% of ARM's license agreements are due for renewal between 2026 and 2028. Renewal negotiations represent a critical juncture for ARM to raise prices, but also a time for customers to evaluate RISC-V alternatives. If ARM demands significant price increases (+50-100%) at renewal, some customers may choose a "gradual migration to RISC-V post-expiration" rather than renewing. This renewal cycle creates a "pricing cliff" risk—a concentrated period of non-renewal upon contract expiration could occur between 2028 and 2030.

---

Chapter 4: End Market Landscape — TAM Modeling and Penetration Maps for Seven Battlegrounds

4.1 Mobile: Maturation of a Cash Cow

Current Status: ARM architecture's penetration in the mobile AP market exceeds 99%. Globally, approximately 1.2-1.4 billion smartphones are shipped annually, almost every one utilizing ARM-architecture chips.

TAM Modeling:

Variable	FY2025	FY2027E	FY2030E	Drivers
Global Smartphone Shipments (M)	1,250	1,300	1,350	Mature markets + slight increase in emerging markets
ARM Share	>99%	>98%	>95%	Marginal RISC-V erosion
ARM-served AP Shipments (M)	1,240	1,274	1,283	Slight increase
Average ASP($)	28	32	38	5G+AI features enhancement
ARM Royalty Rate as % of ASP	2.3%	3.5%	4.5%	v9→CSS migration
Mobile Royalties ($M)	~1,008	~1,427	~2,189	CAGR +17%

Growth Breakdown:

• Shipment Volume: +1.5% annually (contributing ~10% of growth)
• ASP Increase: +6% annually (contributing ~35% of growth) — 5G SoC average ASP is higher than 4G
• Royalty Rate Increase: +14% annually (contributing ~55% of growth) — v9→CSS→v10

Risk Factors:

• Smartphone replacement cycle extends from 2.5 years to 3.5 years → Annual shipments decrease by 5-10%
• Chinese OEMs (Xiaomi/Oppo/Vivo) pay royalties through Arm China, with growth of only +7.5%
• China may promote indigenous RISC-V mobile chips (though commercialization is difficult before 2029)
• Foldable/AI phones may increase ASP but not chip volume

4.2 Data Center: Growth Engine (See Chapter 5 for details)

TAM Quick Overview:

Variable	FY2025	FY2027E	FY2030E
DC ARM Chip Shipments (M)	15	40	100
Average Royalty ($)	28	35	42
DC Royalties ($M)	420	1,400	4,200
ARM DC Share (by shipment volume)	~22%	~35%	~50%

4.3 IoT/Embedded: High Volume, Low Price, Most Direct RISC-V Erosion

Current Status: The ARM Cortex-M series holds ~60-70% market share in the MCU market, but it is rapidly losing market share.

TAM Modeling:

Variable	FY2025	FY2027E	FY2030E	Drivers
Global MCU/IoT Chip Shipments (B)	25	28	35	IoT device growth
ARM Share	65%	55%	45%	RISC-V Erosion
RISC-V Share	30%	38%	45%	Accelerated Penetration
Others (MIPS, etc.)	5%	7%	10%	Fragmentation
ARM IoT Shipments (B)	16.3	15.4	15.8	Stable/slight decrease in volume
Average Royalty ($)	0.039	0.045	0.055	v9 Penetration
IoT Royalties ($M)	635	693	869	CAGR +6.5%

RISC-V Erosion Curve:

• Already Happened: Espressif's ESP32 series migrated from ARM Cortex-M to RISC-V, with shipments exceeding 1 billion units in 2024.
• Currently Happening: Chinese MCU manufacturers (GigaDevice/WCH) are launching numerous RISC-V MCUs.
• Will Happen: Next-generation MCUs from European manufacturers such as Infineon/STMicro will partially adopt RISC-V.
• By FY2030, RISC-V's share in IoT may increase from the current 30% to 45%.

Detailed Analysis of RISC-V's Main Drivers in IoT:

Driving Force	Quantified Estimate	Representative Companies/Products	Impact on ARM
China's "Self-Reliance and Controllability" Policy	50% of China's IoT chips will use RISC-V (by 2028)	GigaDevice (GD32V)/WCH/Nuclei	ARM China IoT Revenue -30-40%
Zero License Fee Cost Advantage	Savings of $0.02-0.05 per MCU	Espressif ESP32-C6	Complete Loss of Low-End Market
LLVM/GCC Maturity	Compiler Performance Gap <5% (2025)	GCC 14 RISC-V Backend	Developer Migration Cost Nearly Zero
Educational Penetration	RISC-V Teaching Adoption Rate >40% (Universities)	UC Berkeley, MIT, Tsinghua	New Generation Developers Default to RISC-V
Arduino/PlatformIO Support	Mainstream Development Frameworks Support RISC-V	Arduino RISC-V boards	Hobbyist/Maker Community Migration

Strategic Significance: Although IoT royalty rates are low ($0.03-0.05 per chip), it represents the "foundational soil" of the ARM ecosystem – a large number of embedded developers start with Cortex-M and then continue to use ARM in mobile/server projects. If IoT developers shift to RISC-V, it could long-term erode ARM's developer base in higher-value endpoints. This is an ecosystem funnel effect – today's IoT developers are tomorrow's data center architects.

Quantified Funnel Effect: If ARM's developer share in IoT drops from 70% to 45% (consistent with chip share), assuming 10% of IoT developers will advance to DC/mobile architecture design roles in 5 years, this means ARM's "developer pipeline" in DC/mobile will shrink by approximately:

• Current: 70% × 2M IoT developers × 10% advancement rate = 140K new DC/mobile ARM developers/year
• FY2030: 45% × 2.5M IoT developers × 10% advancement rate = 113K new DC/mobile ARM developers/year
• 19% reduction in new developer supply — a chronic but persistent ecosystem attrition.

By contrast: during the same period, RISC-V's "developer pipeline" will go from: 25% × 2M × 10% = 50K → 45% × 2.5M × 10% = 113K (+126%). This means that by FY2030, the number of new RISC-V and ARM developers will be close to 1:1 — the long-term foundations of the ecosystem barrier are being eroded.

4.4 Automotive: High-Value Incremental Market

Current Status: ARM's penetration in automotive chips continues to increase:

• ADAS/Autonomous Driving: NVIDIA (ARM architecture) + Qualcomm (ARM architecture) dominate
• Body/Cockpit: Migrating from traditional MCUs to ARM Cortex-A
• Battery Management: Still primarily dedicated MCUs (ARM/RISC-V hybrid)

TAM Modeling:

Variable	FY2025	FY2027E	FY2030E	Driving Factor
Global Vehicle Production (M)	85	88	92	Marginal increase
ARM Chips per Vehicle	7	12	22	Electrification + Intelligence
ARM Automotive Chip Shipments (M)	595	1,056	2,024	CAGR +26%
Average Royalty ($)	0.86	1.10	1.40	v9+ Premiumization
Automotive Royalties ($M)	512	1,162	2,834	CAGR +41%

Automotive is one of ARM's fastest-growing end markets, but also faces unique risks:

• Quintauris Alliance (Bosch/Infineon/NXP/Qualcomm/STMicro) driving automotive RISC-V standardization
• Automotive chip design cycles are long (5-7 years), so RISC-V penetration will be delayed but also stickier.
• ISO 26262 functional safety certification remains an obstacle for RISC-V (3-5 years behind ARM).

In-Depth Analysis of the Quintauris Threat to Automotive RISC-V:

Quintauris (originally named "Automotive RISC-V Alliance") is a joint initiative of five of the world's largest automotive semiconductor companies:

Member	Automotive Chip Market Share	Current ARM Usage	RISC-V Motivation
Bosch	~8%	Extensive (Cortex-M Body)	Reduce IP costs
Infineon	~13%	Extensive (Cortex-M/R)	Self-Reliance and Controllability
NXP	~12%	Extensive (Cortex-A/M/R)	Reduce IP dependency
Qualcomm	~7%	Extensive (Snapdragon Cockpit)	Existing Nuvia/Ventana capabilities
STMicro	~10%	Extensive (Cortex-M MCU)	Reduce IP costs
Total	~50%	—	—

Strategic Significance of Quintauris: 50% of the automotive chip market share jointly promoting RISC-V standardization — this is not a marginal experiment by a few small companies, but a systemic migration by industry mainstream players. Quintauris aims to release the RISC-V automotive instruction set standard (including functional safety extensions) by 2028 and mass-produce the first batch of compliant automotive RISC-V chips by 2030.

Timeline of Impact on ARM: Due to the automotive chip design cycle (5-7 years) + mass production ramp-up (2-3 years) + automotive platform lifespan (7-10 years), Quintauris's revenue impact on ARM will only fully materialize between 2032-2035. However, design-side impact will begin in 2028 — new vehicle designs will start to include RISC-V chip options, and ARM's automotive license agreements will begin to decline.

4.5 PC/Laptops: Windows on ARM: Dawn and Shadow

Current Status: ARM's market share in PCs surged from ~3% in 2024 to ~13% in 2025 (including Copilot+ PCs). This is ARM's third attempt in the PC market — the previous two (Windows RT 2012, Snapdragon 835 Always-on PC 2018) both ended in failure.

History of Three Attempts:

Attempt	Timeframe	Product	Outcome	Reason for Failure
Windows RT	2012-2015	Surface RT	Failed	Unable to run x86 applications, Microsoft abandoned it
Always-on PC	2018-2020	Snapdragon 835/8cx	Failed	50% weaker performance than x86, inefficient translation layer
Copilot+ PC	2024-Present	Snapdragon X Elite/Plus	To Be Verified	Performance close to x86, AI NPU differentiation

The Third Iteration's Distinction: Qualcomm Snapdragon X Elite's single-threaded performance has for the first time matched Intel Core Ultra (Meteor Lake), with multi-threaded performance exceeding it. Concurrently, Windows on ARM application compatibility has significantly improved through the Prism translation layer (vs. early WoA's mere 30-40% compatibility, now approximately 85-90%). Microsoft is tying AI Copilot features (which require an NPU) to ARM chips (x86 chips have weaker NPU performance) – this marks the first time ARM has achieved functional differentiation in the PC market (not just power efficiency advantages).

TAM Modeling:

Variable	FY2025	FY2027E	FY2030E	Drivers
Global PC Shipments (M)	270	285	300	Marginal Growth + Replacement Cycle
ARM PC Share	13%	20%	30%	Copilot+/Native Ecosystem
ARM PC Shipments (M)	35	57	90	CAGR +21%
Average Royalty ($)	6.5	8.0	10.0	CSS Penetration
PC Royalties ($M)	228	456	900	CAGR +32%

S-Curve Prediction for Windows on ARM:

Key Variable: The success or failure of Windows on ARM is highly dependent on the native application ecosystem:

1. Enterprise software (e.g., SAP/Oracle ERP) not yet natively ported
2. Gaming (significant performance loss of 30%+ for many DirectX 12 x86 games under Prism emulation)
3. Professional software (Adobe Creative Suite already supported, but AutoCAD/MATLAB, etc., are still x86-only)

Microsoft's stance is crucial but contradictory: Microsoft is simultaneously promoting ARM (Copilot+ PC) and x86 (Intel/AMD are key partners), making it unlikely to push ARM adoption with the "cut off a limb" approach like Apple. The ceiling for ARM in PCs is likely 30-40% rather than Qualcomm's claimed 50%.

Insights and Limitations from Apple Mac:
Apple completed a full migration from Intel x86 to ARM with the M1 in 2020. However, Apple's success is not replicable:

1. Vertical Integration: Apple controls hardware (Mac) + operating system (macOS) + app store (App Store) + chip design (Apple Silicon) — the Windows PC ecosystem lacks any single party with this level of control
2. Rosetta 2 Translation Layer: Apple's translation layer coverage is >95% with performance loss <20% — Windows Prism translation layer coverage is ~85-90% with performance loss of 25-40% (especially in gaming)
3. Application Ecosystem Control: Apple can compel developers to submit Universal Binaries to the App Store — Microsoft cannot mandate third-party ISV porting
4. Market Share: Mac's global share is ~15-18%, allowing it to "disregard" compatibility for certain enterprise software — Windows PCs need to support 40 years of x86 enterprise software legacy

Analysis of ARM's Real Ceiling in PCs:

Customer Segment	% of Total PCs	ARM Conversion Rate	ARM PC Share Contribution
Consumers (Students/Light Office Use)	35%	40-50%	14-18%
SMEs	25%	15-25%	4-6%
Large Enterprises	30%	5-10%	2-3%
Gaming/Creative Professionals	10%	3-5%	0.3-0.5%
Weighted Average	100%	—	20-28%

Conclusion: The achievable ceiling for ARM in PCs is approximately 20-28%, not Qualcomm's vision of 50%. Large enterprises (30% PC share) will largely not adopt ARM PCs (due to enterprise IT software compatibility requirements + IT department conservatism), and gaming/creative professionals even less so (due to x86 dependencies for DirectX 12/Vulkan/CUDA). ARM's growth in PCs will primarily come from consumers and SMEs in light office use scenarios – these are precisely the PC segments with the lowest ASP, further limiting the total volume of ARM PC royalties.

4.6 Network Infrastructure: Invisible but Stable

Current Status: ARM holds a stable position in network equipment (routers/switches/base stations):

• 5G Base Stations: Most use ARM architecture
• Network Processors: Marvell, Broadcom, etc., use ARM cores
• DPU/SmartNIC: NVIDIA BlueField (ARM core)

TAM Modeling:

Variable	FY2025	FY2027E	FY2030E
Networking Chip Shipments (B)	2.0	2.3	2.8
ARM Share	55%	58%	60%
Avg. Royalty ($)	0.12	0.15	0.20
Networking Royalties ($M)	132	200	336

4.7 Security/Wearables: Emerging but Niche

ARM's secure microcontrollers (TrustZone/SecurCore) and wearable device (Cortex-M33/M55) chips contribute a small but stable royalty stream. The strategic value of these markets to ARM lies in maintaining the breadth of the developer ecosystem, rather than direct revenue contribution.

4.8 End-Market Competitive Landscape Deep Dive Matrix

ARM faces significantly different competitive dynamics in each end market:

Competitive Depth Comparison Across End Markets:

End Market	Main Competitors	ARM Pricing Power	Customer Stickiness	Cycle Length	Replacement Time
Mobile AP	RISC-V (Long-term)	Very Strong	Very High (iOS/Android)	18 Months	5-7 Years
IoT	RISC-V (Already eroding)	Weak	Low (Toolchain Sharing)	6-12 Months	1-2 Years
DC	x86+RISC-V	Medium	Medium (Software Migration)	24-36 Months	3-5 Years
Automotive	RISC-V (Quintauris)	Strong	High (Safety Certification)	5-7 Years	5-10 Years
PC	x86 (Intel/AMD)	Weak	Low (New Entrants)	12-18 Months	N/A (Already Co-existing)
Networking	RISC-V/x86	Medium	Medium	24-36 Months	3-5 Years

Key Competitive Dynamics:

Mobile: ARM's "safe haven." iOS only supports ARM (Apple is unlikely to migrate), and Android has deep dependencies on the ARM software stack (millions of native applications). For RISC-V to challenge ARM in the mobile market, it first needs to address: (a) application binary compatibility (similar to a Rosetta translation layer), (b) GPU ecosystem (RISC-V lacks an equivalent to Mali/Adreno), and (c) baseband chip integration (all 5G basebands are currently ARM-based). It is challenging to overcome all these obstacles before 2030.

IoT: ARM's "bleeding zone." RISC-V's zero licensing fee plus sufficient performance (Cortex-M level) has led Chinese MCU manufacturers (GigaDevice, WCH) and Espressif (ESP32 series) to massively shift towards RISC-V. ARM's share in IoT has fallen from ~80% in 2020 to ~65% in 2025, a trend that is irreversible. ARM's counter-strategy is to enhance high-end MCU differentiation through Helium (MVE) and AI acceleration instructions, but the low-end MCU segment ($0.30-1.00 ASP range) has largely been ceded to RISC-V.

Data Center: ARM's "strategic high ground." ARM's growth in DC is not organic but rather driven by hyperscale customers actively choosing ARM to reduce their reliance on Intel/AMD (a procurement diversification strategy). The core reason AWS/Google/Microsoft choose ARM is not "ARM is better," but "we don't want to have only one supplier." This implies an implicit ceiling for ARM's growth in DC—once ARM's share exceeds 30-40%, customers' "diversification" incentive will weaken, and ARM will no longer be an "alternative option" but one of the "mainstream options," potentially leading to increased pressure on its pricing power.

Automotive: ARM's "time buffer." The design cycle (5-7 years) and safety certification cycle (ISO 26262, 2-3 years) for automotive chips provide ARM with the longest protection window. However, the Quintauris alliance (Bosch/Infineon/NXP/Qualcomm/STMicro) is standardizing automotive RISC-V—once standardization is complete (projected 2027-2028), new vehicle designs will begin adopting RISC-V. The full impact will only be seen between 2032-2035 (due to the lag from design to mass production).

4.9 Summary of Seven Battlegrounds: TAM Aggregation and Growth Decomposition

End-Market Royalty Forecast Summary:

Endpoint	FY2025($M)	FY2027E($M)	FY2030E($M)	CAGR
Mobile AP	1,008	1,427	2,189	17%
IoT	635	693	869	6.5%
DC	420	1,400	4,200	58%
Automotive	512	1,162	2,834	41%
PC	228	456	900	32%
Networking	132	200	336	20%
Total Royalties	2,935	5,338	11,328	31%

Shift in Royalty Growth Focus:

Share	FY2025	FY2027E	FY2030E
Mobile + IoT (Traditional Core)	56%	40%	27%
DC + Automotive + PC (New Growth)	40%	57%	70%
Networking / Other	4%	3%	3%

Key Insight: ARM's growth narrative hinges on a shift in focus from the left (Mobile + IoT) to the right (DC + Automotive + PC). The 140x P/E ratio awarded by the market essentially prices in the success of this transition. However, each step of this transition faces:

• RISC-V competition (already present in IoT, brewing in DC/Automotive)
• x86 defense (PC/Servers — Intel Lunar Lake/AMD Strix Point)
• Internal pricing strategy contradictions (price increases accelerate substitution)

If DC growth falls short of expectations (40% share instead of 70%): Royalty growth would decrease from a CAGR of 31% to ~18%, which would reduce total FY2030 revenue from $11.2B to ~$8B, requiring the P/E to compress from 140x to below 100x to maintain valuation discipline.

---

Chapter 5: The Data Center Journey — Credibility of the Path from 10% to 50%

5.1 Strategic Significance of Data Centers

Management has stated in multiple earnings calls: Data centers will surpass mobile to become ARM's largest royalty business "within a few years." This means data center royalties need to grow from the current ~$250M/quarter to exceed mobile's ~$300M/quarter.

Why data centers are so crucial for ARM:

1. Extremely High Per-Unit Royalties: Server chip ASP $500-2000+, ARM royalties $10-100/chip → 15-150 times that of mobile
2. Fastest CSS Penetration: Server customers most need CSS's rapid time-to-market advantage (AI race is time-sensitive)
3. Phoenix's Target Market: ARM's first self-designed chip is aimed at data centers
4. AI-Driven Structural Growth: Every AI data center requires a CPU (even if the primary computation is performed by GPUs)
5. Valuation Leverage: DC royalty growth has the greatest marginal impact on ARM's overall valuation (high unit price × high growth rate)

5.2 The ARM Server Market Share Conundrum

There are significant discrepancies in ARM server market share data:

Metric	Share	Source	Meaning	Confidence Level
Traditional CPU Shipments	20-23%	IDC/Omdia	Counts only standalone CPU servers	H
ARM's Official Claim	~50%	ARM Newsroom	Includes Grace in NVIDIA GB200	M
Revenue Share	<10%	Estimate	ARM royalties vs. Intel/AMD chip revenue	M
AI Training Nodes with ARM	>50%	NVIDIA	GB200 = Grace + Blackwell combination	H

The core of the discrepancy lies in NVIDIA GB200: Each GB200 supercomputing node contains one Grace ARM CPU + multiple Blackwell GPUs. ARM counts GB200 as an "ARM compute" node — which is technically correct (it indeed has an ARM CPU inside), but 99% of the value of these nodes comes from the GPUs, and the royalties ARM collects from the Grace CPU are negligible compared to the total value of the node.

5.3 Deep Dive into Four Hyperscale Customers

AWS Graviton Series (Flagship Customer — The Most Important Single Driver of ARM DC Royalties):

Generation	Core Count	Base Core	Process Node	Status	ARM License Type	Estimated Annual Shipments (M)
Graviton1	16	A72(v8)	16nm	Retired	ALA	0
Graviton2	64	N1(v8)	7nm	Mass Production (Decreasing)	ALA	~1.5
Graviton3	64	V1(v9)	5nm	Mass Production	ALA	~2.0
Graviton4	96	V2(v9)	4nm	Primary Mass Production	ALA	~3.5
Graviton5	192	Unknown(v9+)	3nm	2025-12 Preview	ALA	0 (2026 H2 Mass Production)
Total	—	—	—	—	—	~7.0

AWS Graviton's ARM Royalty Economics:

• AWS is an ALA customer (Annapurna Labs in-house core design), with relatively lower royalty rates.
• Estimated Royalty Rate: $5-15/chip (ALA tier, excluding CSS premium)
• Estimated Annual Royalty: 7M chips × $10 (mid-range estimate) = $70M/year
• In contrast, if AWS used CSS (assuming $30-50/chip): royalties would increase to $210-350M/year.
• This is the "pricing power dilution effect" of ALA — ARM's largest DC customer happens to use the license type with the lowest royalty rate.

Strategic Significance of AWS Graviton:

1. AWS states: Graviton instances account for the majority of new AWS instances, with ARM proportion >30% of EC2 usage.
2. Cost Advantage: Graviton4 offers approximately 20-30% better price-performance compared to equivalent generation Intel Xeon (AWS public benchmarks).
3. Iteration Speed: AWS launches a new generation of Graviton approximately every 18-24 months (faster than Intel's 2-3 year cycle).
4. Key for Graviton5: The 192-core design suggests AWS is pursuing higher compute density, specifically targeting AI inference workloads.

AWS Risk Signals:

• The Annapurna Labs team is also evaluating RISC-V for networking chips (AWS Nitro's next generation may use RISC-V).
• AWS joined RISC-V International in 2025.
• AWS's "dual-architecture" strategy: ARM for general compute/AI inference, RISC-V for networking/security, x86 retained for enterprise compatibility.

Google Axion (ARM's most RISC-V-exposed customer):

Dimension	Detail
Specifications	72 cores, Neoverse V2, TSMC 4nm, 2024 GA
Performance	Benchmark performance superior to AMD EPYC (at comparable power consumption).
Deployment	Google Cloud Compute Engine ARM instances
License Type	TLA/CSS (non-ALA) → Higher royalty rate than AWS
Estimated Shipments	~1-2M/year (Google Cloud market share < AWS)
Estimated Royalties	$20-40/chip × 1.5M = $30-60M/year

Google's Strategic Investments in RISC-V:

1. Android GKI kernel is already adapted for RISC-V (starting with Android 14).
2. Google is a Platinum member of RISC-V International.
3. Google has invested in SiFive (RISC-V IP vendor) and Ventana Micro (RISC-V server chips).
4. Google's internal video transcoding chip (Argos) already uses RISC-V cores.
5. ChromeOS is being adapted for RISC-V (long-term for education/low-end markets).

Google is the customer ARM needs to be most wary of: It is both an ARM DC customer (Axion) and the largest corporate driver of the RISC-V ecosystem (Android+ChromeOS). Google's dual-pronged strategy is essentially preparing a path for RISC-V to replace ARM — when RISC-V core performance matches ARM's (possibly 2028-2029), Google may be the first hyperscale customer to migrate DC chips from ARM to RISC-V.

Microsoft Cobalt (Most Aggressive CSS Adopter):

Dimension	Cobalt 100	Cobalt 200
Core Count	128	132
Base Core	Neoverse N2	Neoverse V3
Process Node	TSMC 5nm	TSMC 3nm
Status	Generally Available (GA)	2026 H1 Shipments
License	Likely CSS	CSS (Confirmed)
Estimated Royalties	$30-50/chip	$40-60/chip

• Used internally by Azure, gradually being opened to external customers.
• Copilot+ PCs are driving the establishment of an ARM ecosystem foundation on the client side as well.
• Microsoft is a benchmark customer for CSS in DC — if Cobalt succeeds, it will attract more Tier 2 DC customers to adopt CSS.
• Estimated Royalties: ~2M chips × $45 = $90M/year (Higher than AWS due to higher CSS royalty rates)
• Microsoft is also heavily investing in Qualcomm ARM PCs → Multi-dimensional ARM revenue contribution.

NVIDIA Grace (Highest Shipments but Lowest Royalty Density):

Dimension	Detail
Specifications	72 cores, Neoverse V2, TSMC 4nm
Configuration	Grace Hopper (Grace+H100), GB200 (Grace+Blackwell)
Shipments	Largest single source of ARM server shipments (estimated ~3-5M/year).
Standalone Deployment	Meta will begin large-scale deployment of Grace CPU standalone (without GPU) in February 2026.
License Type	ALA (NVIDIA in-house microarchitecture optimization)

Grace's Royalty Economics Dilemma:

• Grace CPU standalone price ~$3,500 (Grace Hopper configuration), ARM royalty estimated $15-30/chip (ALA level)
• GB200 supernode total price $30,000-50,000, Grace CPU accounts for only ~7-10% of value
• ARM collects only $15-30 royalty from each GB200 node (0.05-0.1% of node value)
• Estimated Royalty: 4M chips × $22 = $88M/year
• Impact of Meta Grace standalone: Meta's independent Grace deployment is a positive signal — proving that ARM CPUs are not just "accessories" to GPUs, but can independently handle AI agent/inference workloads. If 5-10% of Meta's planned $135B AI CapEx (2026) is allocated to ARM CPUs (i.e., $6.75-13.5B), it implies Grace shipments of 2-4M units → ARM royalties $30-60M
• Whether NVIDIA uses CSS or standard Neoverse licensing for Grace is not publicly confirmed

Summary of Royalty Contribution from Top Four Customers:

Customer	License Type	Estimated Shipments (M/year)	Royalty Rate ($/chip)	Estimated Royalty ($M/year)	Share of DC Royalty %
AWS	ALA	7.0	$10	$70	17%
Microsoft	CSS	2.0	$45	$90	21%
Google	TLA/CSS	1.5	$35	$53	13%
NVIDIA	ALA	4.0	$22	$88	21%
Others (Ampere, Oracle, etc.)	TLA/CSS	3.0	$35	$105	25%
DC Total	—	17.5	$23 (Weighted)	~$406	100%
Vs: ARM Reported DC Royalty	—	—	—	~$420	—

Key Insight: The top 4 customers contribute ~72% of DC royalties, but AWS and NVIDIA (38% share) use low-royalty-rate ALA licenses. The quality of DC royalty growth depends on: (a) whether CSS customers (Microsoft, etc.) can catch up in shipments with ALA customers, or (b) whether ALA customers can be persuaded to upgrade to CSS. Both face challenges—ALA customers have in-house development capabilities and don't need CSS, while CSS customer shipments are limited by cloud market share.

5.4 Data Center Royalty Trajectory: Three Scenarios Modeling

Key Variables: Data Center Royalty = DC ARM Chip Shipments × Royalty Rate Per Chip

Variable	Unit	FY25	Bull Case FY28	Base Case FY28	Bear Case FY28
Hyperscaler In-house Chip Shipments	M	5	20	15	8
NVIDIA Grace Shipments	M	5	15	10	5
Other ARM Servers	M	5	15	10	5
Total DC ARM Chip Shipments	M	15	50	35	18
Weighted Average Royalty	$/chip	28	45	38	30
DC Royalty	$M	420	2,250	1,330	540
DC Royalty CAGR	—	—	+75%	+47%	+9%

Bull Case Scenario ($2,250M by FY2028): AI infrastructure investment sustained at $200B+/year, all hyperscale customers in full production of ARM chips, CSS penetration >60%, no substantial progress for RISC-V in DC
Base Case Scenario ($1,330M by FY2028): AI investment growth slows to +15-20%, hyperscale customers gradually ramp up production, CSS penetration ~40%, Tenstorrent RISC-V chips begin small-batch shipments
Bear Case Scenario ($540M by FY2028): AI investment sees a dot-com bubble-like correction (-30%), hyperscale customers delay ARM procurement, RISC-V gains 5-10% share in DC

Key New Signals (from WebSearch 2026-02):

• Meta has deployed NVIDIA Grace CPU standalone systems (February 2026): This marks Meta's first large-scale adoption of ARM server CPUs for general-purpose + AI agent workloads, a pure CPU deployment without GPUs → Positive for ARM DC royalties
• Tenstorrent Ascalon-X: 8-wide decode, out-of-order superscalar RISC-V CPU, 22 SPECint2006/GHz, >2.5 GHz on Samsung SF4X. 2025 tape-out, 2026 product launch. Performance claimed to rival Neoverse V3 and AMD Zen 5 → RISC-V first time achieving performance parity at DC level
• Meta acquires Rivos: Meta is re-architecting its internal AI accelerator MTIA to RISC-V → First strategic signal of a hyperscale customer moving chip IP from ARM to RISC-V
• Meta AI CapEx Plan: Expected to reach $135B in 2026 (1.9x of $72B in 2025) → Huge potential increment for DC royalties, but Meta's pivot to RISC-V suggests this increment might bypass ARM

5.5 Neoverse Product Line Roadmap

Performance Leap per Generation: Approximately 15-25% single-thread performance improvement (at the same power consumption) or 30-40% energy efficiency improvement (at the same performance). ARM's annual iteration cadence (12-18 months/generation) is faster than Intel's (18-24 months/generation).

5.6 Five Drivers and Five Headwinds for ARM DC Royalty Growth

Understanding ARM's growth trajectory in data centers requires a balanced analysis of its drivers and headwinds:

Five Drivers (Bullish Factors):

#	Driver	Quantifiable Impact	Timeline	Confidence
P1	Explosion in AI Inference Demand	DC chip shipments ×3 (FY25→FY28)	Ongoing	H
P2	Hyperscaler In-house Chip Development Acceleration	G5/Cobalt200/Axion2	2026-2027	H
P3	CSS Penetration in DC	Royalty rate ×2-3 (TLA→CSS)	2026-2028	M
P4	ARM Performance/Power Advantage	Server chip TCO -20-30%	Continuous	H
P5	Edge AI/5G + DC Convergence	New DC-Edge ARM Chip Demand	2027+	M

Five Headwinds (Bearish Factors):

#	Headwind	Quantifiable Impact	Timeline	Confidence
R1	Low Royalty Rates for ALA Customers	2 out of top 4 customers use ALA	Structural	H
R2	RISC-V DC Entry	Tenstorrent mass production in 2026	2026-2028	M
R3	AI CapEx Cycle Risk	-30% adjustment = $90 billion reduction	Uncertain	H
R4	Intel/AMD Power Consumption Catch-up	Lunar Lake/Turin -40% power consumption	2025-2026	H
R5	Phoenix Loss of Trust	Customers accelerate evaluation of alternatives	2026+	M

Net Effect Assessment: The drivers will remain dominant from 2025-2027 (AI demand growth from P1+P2 is too strong). However, starting in 2028, headwinds R2+R4 will significantly strengthen. ARM's DC royalty CAGR may exhibit a "high first, then low" pattern: FY2025-FY2027 CAGR 50-80% → FY2027-FY2030 CAGR 15-30% → FY2030+ CAGR <10%. This contradicts the market's implied assumption of "long-term high growth."

5.7 Data Center ARM vs x86: Workload Suitability Matrix

ARM is not a "universal x86 replacement" in data centers – different workloads have significantly different architectural preferences:

Workload	ARM Advantage	x86 Advantage	Current Best Choice	ARM Future Share (FY2030E)
AI Training (Large Models)	None (GPU-dominated)	None (GPU-dominated)	GPU + Any CPU	30% (Host CPU)
AI Inference (LLM)	Power Efficiency + Core Count	Single-thread Performance	Each has its advantages	35-45%
Web Services	Power/Core Count	Mature Ecosystem	ARM Leads	40-50%
Databases (OLTP)	Core Count (Concurrency)	Single-thread (Latency)	Depends on DB type	25-35%
HPC/Scientific Computing	Partial (Fugaku)	Traditional Advantage	Mixed	15-25%
Enterprise IT (ERP/CRM)	Weak (Poor Compatibility)	Very Strong (40-year Ecosystem)	x86	5-10%
Virtual Desktop (VDI)	Weak	Strong	x86	5-10%
Containers/Microservices	Strong (K8s Support)	Strong	Even Split	35-45%
CDN/Edge	Very Strong (Low Power)	Weak	ARM Leads	50-60%
Storage Systems	Moderate	Moderate	Even Split	25-35%

Weighted Average ARM DC Share (FY2030E): ~30-35%

Key Insight: ARM's "golden workloads" in DC are AI inference, web services, containers/microservices, and CDN/edge – these are all the fastest-growing segments. However, in enterprise IT (ERP/CRM) – the largest single category of global IT spending ($500B+/year) – ARM has almost no presence. This means ARM's DC share could reach 40-50% in the "new economy" (cloud-native) but remain at 5-10% in the "old economy" (enterprise IT). The ceiling for overall DC share might be 30-35%, rather than the 50% expected by the market.

5.8 Acceleration Logic and Limitations of CSS in Data Centers

Reasons why CSS offers the greatest value to DC customers:

1. AI competition time pressure: 12-month time-to-market acceleration worth $100M+ in opportunity cost
2. 3nm/2nm design cost $500M+: CSS reduces NRE risk
3. Heterogeneous computing SoC complexity: CSS provides validated CPU subsystems

Limitations of CSS in DC:

• ALA customers (AWS/Google/Microsoft/NVIDIA) have strong in-house development capabilities and do not need CSS → The largest DC customers happen to have the lowest demand for CSS
• CSS's target DC customers are "second-tier players" (second-line products from cloud vendors/telecom equipment providers/small-to-medium DC chip design companies)
• If CSS is primarily adopted by second-tier players, its positive impact on DC royalties is limited by shipment volumes (second-tier shipments are significantly smaller than hyperscalers)

This implies: The primary driver for DC royalty growth is not an increase in CSS royalty rates, but rather the growth in chip shipments from ALA customers (with lower royalty rates). CSS's true contribution in DC may be overestimated.

5.9 The "Quality vs. Quantity" Paradox of DC Royalties

ARM faces a fundamental dilemma in the data center market:

**High-Quality Royalties (CSS, $40-90/chip)** come from Tier-2 DC customers (Oracle, Fujitsu, small and medium-sized design companies) – but these customers have limited shipments (<500K chips/year per customer).

**High-Volume Royalties (ALA, $10-22/chip)** come from Tier-1 hyperscale customers (AWS, Google, NVIDIA) – but these customers develop their own core designs, resulting in lower royalty rates.

Management Narrative Test: ARM management emphasizes "accelerated CSS adoption" to imply a continuous increase in DC royalty rates. However, if the primary adopters of CSS in DC are Tier-2 customers with limited shipments, while Tier-1 customers (contributing >60% of DC shipments) persist in using ALA/TLA, then the increase in the weighted average royalty rate for DC will be significantly slower than the CSS-specific royalty rate increase implied by management.

Quantitative Test: Assuming DC shipment distribution = ALA (65%) + TLA (20%) + CSS (15%):

• Weighted average royalty rate = 0.65×$15 + 0.20×$25 + 0.15×$55 = $23.0/chip
• vs. If entirely CSS: $55/chip → Actual weighted average royalty rate is only 42% of CSS
• Even if CSS penetration reaches 30% (optimistic): Weighted average royalty rate = 0.50×$15 + 0.20×$25 + 0.30×$55 = $29.0/chip — only a 26% increase

Conclusion: The increase in DC royalty rates is limited by the structural presence of ALA customers. Unless ARM can persuade AWS/Google/NVIDIA to upgrade to CSS (highly unlikely – they have in-house development capabilities), DC royalty growth will primarily depend on shipment volume rather than royalty rates – which is inconsistent with ARM's overall "price-driven growth" narrative.

---

Chapter 6: Project Phoenix — A Bold Bet: IP Licensor Transforms into Chip Designer

6.1 What is Project Phoenix

Project Phoenix represents ARM's most audacious strategic transformation in its history: from a pure IP licensor to a complete chip designer.

Project Phoenix Chip Specifications:

• 128 Neoverse V3 Cores
• TSMC 3nm, Multi-Chip Module (MCM) packaging
• 12-channel DDR5 Memory
• 96 PCIe Gen6 Lanes
• Target: Data Center AI Inference + General Purpose Computing
• Competitive Goal: Direct competition with NVIDIA Grace and AMD EPYC

Initial Customers: Meta (first customer) → OpenAI (via SoftBank Stargate) → Cloudflare

6.2 Why Project Phoenix is "SoftBank's Project" and not "ARM's Optimal Strategy"

The strategic rationale for Project Phoenix is not obvious under ARM's independent operating framework:

If ARM were an independent company (without SoftBank):

• Optimal Strategy: Maintain IP licensing neutrality, maximize number of licensees and royalties
• Phoenix Risk: Direct competition with customers (AWS/Google/Microsoft/NVIDIA) → potentially accelerate customer evaluation of RISC-V
• Capital Requirements: Chip design NRE $100M+, changes ARM's asset-light nature
• Valuation Impact: Chip design company P/S 5-15x vs. IP licensing P/S 25-30x

Under SoftBank's Control:

• SoftBank needs ARM chips to support its AI infrastructure vision (Stargate requires CPUs)
• SoftBank's $100B+ Izanagi plan explicitly includes ARM chip design
• Masayoshi Son claims ARM is the "foundational technology for ASI" → merely doing IP licensing does not fit this narrative
• SoftBank has acquired Ampere for $6.5B (an ARM server chip company) → integration logic

Core Contradiction: Project Phoenix might maximize SoftBank's strategic value (integrating the AI supply chain), but it sacrifices ARM's neutrality and licensee trust. This is not CEO Rene Haas's business judgment, but rather Masayoshi Son's AI vision projected onto ARM.

6.3 Project Phoenix Competitive Positioning Matrix

Dimension	Phoenix	NVIDIA Grace	AMD EPYC	Intel Xeon	AWS Graviton5
Cores	128	72	128(Bergamo)	144(Sierra Forest)	192
Architecture	ARM v9 V3	ARM v9 V2	x86 Zen5	x86	ARM v9
Process Node	TSMC 3nm	TSMC 4nm	TSMC 4nm	Intel 3	TSMC 3nm
Power Target	350W (est.)	500W (Superchip)	360W	350W	Unknown
Memory	12ch DDR5	LPDDR5X (integrated)	12ch DDR5	8ch DDR5	12ch DDR5 (est.)
PCIe	Gen6 x96	Gen5 x48	Gen5 x128	Gen5 x80	Unknown
Availability	2027 H1 (est.)	Currently shipping	Currently shipping	Currently shipping	2026 H1 (est.)
Differentiation	ARM's official optimal cores + CSS-level interconnect	GPU synergy (NVLink C2C)	Most mature ecosystem	Enterprise compatibility	AWS proprietary, not for external sale
Customers	Meta/OpenAI/Cloudflare	NVIDIA DC Customers	All data center customers	Enterprise IT	AWS only

Project Phoenix's Core Differentiation: ARM possesses the best core IP (V3 is ARM's strongest self-designed core) and utilizes CSS-level interconnect/cache designs. Theoretically, Phoenix should be the "purest ARM chip." However, this also presents a problem – if Project Phoenix performs exceptionally well, licensees will question whether ARM is keeping "the best" for itself.

6.4 Project Phoenix Financial Impact Modeling

Cost Model (Estimated):

Cost Item	Amount ($M)	Notes
Chip Design NRE (5-year amortization)	$100-200/year	3nm MCM design, incl. IP+PD+verification
TSMC 3nm Wafer	$15K-20K/wafer	~100-150 chips/wafer (MCM)
Packaging and Testing (OSAT)	$50-100/chip	Higher MCM packaging costs
Total Cost Per Chip	$200-350	Incl. amortized NRE
Expected Selling Price	$3,000-8,000	Referencing AMD EPYC/Intel Xeon
Gross Margin (Estimated)	75-90%	Higher than average for chip design companies (due to ARM's own IP)

Project Phoenix Revenue Scenarios:

Scenario	FY2028 Shipments (K)	Selling Price ($K)	Chip Revenue ($M)	Contribution to ARM Total Revenue
Bull Case	500	6.0	3,000	+43%
Base Case	200	5.0	1,000	+14%
Bear Case	50	4.0	200	+3%
Failure	<10	—	<50	Negligible

Valuation Paradox (Quantified):

Phoenix Outcome	Impact on Revenue	Impact on Blended Gross Margin	Impact on P/S Multiple	Net Effect on Market Cap
Major Success ($3B)	+43%	95%→90%	28x→20x	-10%
Moderate Success ($1B)	+14%	95%→93%	28x→25x	+5%
Failure (<$50M)	+0%	Unchanged	28x→26x (Trust Loss)	-7%

Optimal Path: Moderate Success (a few clients like Meta, $500M-1B revenue, without changing ARM's "IP licensing-centric" positioning) → This also appears to be the path management is pursuing.

6.5 Quantifying the Phoenix Cannibalization Effect

As ARM's own chip, if Phoenix adopts CSS-level licensing (from ARM's IP division to ARM's chip division), each Phoenix chip effectively "consumes" a CSS royalty revenue opportunity:

Assumption: CSS royalty $50-80/chip, Phoenix internal transfer price $0 (same company):

• If Phoenix ships 200K units: Cannibalization of $10-16M/year (insignificant)
• If Phoenix ships 500K units: Cannibalization of $25-40M/year (still small)
• If Phoenix ships 2M units: Cannibalization of $100-160M/year (starts to become meaningful)

Conclusion: The direct cannibalization of ARM's royalties by Phoenix is limited (due to relatively small shipment volumes). The real risk is not cannibalization, but rather loss of trust—licensees accelerating their evaluation of RISC-V due to ARM's entry into chip design. This indirect damage is difficult to quantify but could be an order of magnitude greater than direct cannibalization.

"Cannibalization vs. Trust Tax" Framework Comparison:

• Direct Cannibalization (Quantifiable): Phoenix shipments × CSS royalty rate = $10-160M/year — less than 5% of ARM's total royalties
• Trust Tax (Hard to quantify but potentially larger): Long-term customer attrition due to licensees accelerating RISC-V evaluation = $45-100M/year incrementally, potentially $200-500M in case of ecosystem collapse
• Net Conclusion: The market focuses on "how much Phoenix chips can sell for" (direct revenue), but should focus on "how much customer trust Phoenix makes ARM lose" (indirect cost)

Historical Analogy: When Intel launched its Xe GPU in 2019, AMD/NVIDIA did not panic due to Intel's entry into the GPU market. However, if ARM (as a CPU IP vendor) starts designing complete CPU chips, the reaction from licensees would be much stronger—because ARM, their "supplier," would become a "competitor," whereas Intel's entry into the GPU market did not affect AMD/NVIDIA's CPU supply chain. ARM's situation is more akin to TSMC announcing it will design its own chips—this would shake the trust foundation of the entire foundry business model.

6.6 Phoenix Timeline and Milestones

Milestone	Expected Time	Status	Impact
Phoenix Design Completion	2025 H2	Completed	TSMC 3nm Tape-out
Engineering Sample (A0)	2026 H1	In Progress	First Silicon Validation
Meta Deployment Testing	2026 H2	Pending	First Client Validation
B1 Revision (if needed)	2026 H2-2027 H1	Pending	Bug Fixes + Optimization
Mass Production	2027 H1	Pending	Revenue Recognition Begins
OpenAI/Stargate Adoption	2027+	Pending	Closing the SoftBank Strategic Loop

If the Phoenix timeline is delayed (chip design project delays are extremely common—refer to Intel Meteor Lake's 18-month delay), ARM will face a double loss: it will have no chip revenue, yet will have already alienated some licensees.

6.7 Phoenix's "Trust Tax": Licensee Reaction Matrix

The greatest risk of Phoenix lies not in technology or finance, but in loss of ecosystem trust. ARM's positioning as a "neutral IP vendor" is the cornerstone of its business model—when a supplier becomes a competitor, clients will re-evaluate alternatives:

Licensee	Competing with Phoenix?	Reaction Intensity	Possible Actions	Impact on ARM Royalties
AWS	Yes (DC CPU)	High	Accelerate Graviton in-house R&D + Evaluate RISC-V	-$10-20M/year
Google	Yes (DC CPU)	High	Accelerate RISC-V investment (already underway)	-$10-20M/year
Microsoft	Yes (DC CPU)	Medium	Maintain Cobalt but diversify suppliers	-$5-10M/year
NVIDIA	Partially (Grace competition)	Medium	NVIDIA has stronger GPU dependence, unlikely to leave ARM	Minimal
Qualcomm	Indirect (DC)	High	Already has Nuvia in-house cores, may accelerate Oryon-RISC-V evaluation	-$20-50M/year
MediaTek	No (Different market)	Low	Wait and see	None
Samsung	No (Different market)	Low	Wait and see	None
Chinese Customer Base	No (Policy isolation)	Low	Continue through Arm China	None
Total Trust Tax Estimate	—	—	—	-$45-100M/year

Non-linearity of the Trust Tax: The above estimates are for a "gradual scenario" (each customer makes independent decisions). However, if 2-3 hyperscale customers simultaneously announce "accelerated RISC-V strategies" (e.g., Google announcing Axion's next generation will use RISC-V + AWS announcing edge chips will switch to RISC-V), it could trigger a **synergistic effect**—other customers would perceive the "RISC-V ecosystem as having reached critical mass" and accelerate their evaluations. The impact of such a "synergistic collapse of trust" could be 3-5 times that of the gradual scenario.

6.8 Phoenix and the Linkage with SoftBank's AI Empire

Phoenix is not an isolated chip project—it is a piece in SoftBank's larger AI strategic puzzle:

The role Phoenix plays in this strategy: SoftBank needs its own controlled AI chips (not relying on NVIDIA) to deploy in Stargate data centers. Without Phoenix, all of SoftBank's AI infrastructure would have to pay a "GPU tax" to NVIDIA—similar to the "IP tax" ARM collects from chip manufacturers. Phoenix transforms SoftBank's approach from "ARM IP → other companies' chips → SoftBank data centers" to "ARM IP → ARM chips → SoftBank data centers," capturing more vertical integration value.

Implications for Public Shareholders: Phoenix's business logic is sound within SoftBank's overall strategy—but for ARM's public shareholders (10% free float), this means ARM's resources (R&D, management time, TSMC capacity allocation) are being utilized to serve SoftBank's AI vision, rather than to maximize the interests of all shareholders. This is a classic issue with Controlled Company governance structures—the interests of the controlling shareholder are not perfectly aligned with those of the public shareholders.

---

Chapter 7: Industry Chain Niche — ARM's Unique Position in the Semiconductor Value Chain

7.1 ARM's Position in the Industry Chain

ARM is in the **second segment** of the semiconductor value chain — IP licensing. Its P/E multiple is the highest in the entire industry chain (140x vs. industry average 20-40x).

7.2 ARM TAM Ceiling: A Mathematical Constraint of 1-2%

If the global semiconductor market size is $600 billion (2025), what "tax rate" can ARM achieve?

IP Licensor	Annual Revenue	Relevant Market Size	Penetration Rate
ARM	$4.7B	$600 billion	0.78%
Qualcomm QTL	~$5B	~$150 billion	~3.3%
Dolby	~$1.3B	~$300 billion	~0.4%
MPEG LA	~$2B	~$200 billion	~1.0%
ARM Consensus FY2030	$11.2B	~$750 billion	1.49%

Ceiling Calculation: IP licensing's share of semiconductor revenue has never exceeded 3-4% (total for the entire industry). For ARM to grow from 0.78% to 1.5%+, it would need to simultaneously increase royalty rates across multiple end markets—which is becoming increasingly difficult amidst RISC-V competition.

7.3 Comparison with EDA: Valuation Anchoring

Dimension	ARM	Synopsys	Cadence
Revenue Model	Licensing + Royalties	Subscription + Maintenance	Subscription + Maintenance
Revenue	$4.7B	$6.1B	$4.3B
Gross Margin	95.4%	80.2%	89.4%
Operating Margin (GAAP)	18.7%	21.4%	30.9%
P/E (TTM)	140x	55x	75x
Growth Rate	26%	38%	6%
Substitution Risk	RISC-V (High)	Open-Source EDA (Low)	Open-Source EDA (Low)
R&D/Revenue	56%	32%	25%
SBC/Revenue	~18%	~12%	~10%

Key Differences: EDA tools do not have zero-cost open-source alternatives (open-source EDA tools like OpenROAD are not available for commercial-grade designs). ARM has RISC-V. ARM's P/E is 2-2.5 times that of its EDA peers, but it faces higher substitution risk—this premium entirely stems from the market's pricing of ARM for "growth" rather than "security."

7.4 Global Semiconductor IP Market Landscape

Company	IP Revenue (FY2025E)	Market Share	Core IP	Relationship with ARM
ARM	$4.0B+	~40-45%	CPU Cores	—
Synopsys IP	~$1.5B	~15-18%	Interface/Security/Processor	CSS Competition
Cadence IP	~$0.8B	~8-10%	Interface/Memory	Complementary
Imagination	~$0.3B	~3%	GPU IP	Previously competed with ARM
CEVA	~$0.2B	~2%	DSP/AI/Connectivity	Complementary
Alphawave	~$0.3B	~3%	High-speed Interface	Complementary
Global IP Market	~$8-10B	100%	—	—

7.5 Supply Chain Risk Transmission: ARM's Four Vulnerable Points

Path 1: TSMC Capacity → Phoenix
Phoenix makes ARM directly dependent on TSMC's 3nm capacity for the first time. If cross-strait tensions escalate or TSMC prioritizes capacity allocation to other customers (Apple/NVIDIA), Phoenix shipments will be constrained. Ironically, Phoenix has transformed ARM from "zero manufacturing risk" to "having manufacturing risk."

TSMC's 3nm capacity allocation priority sequence (based on historical patterns and revenue contribution):

1. Apple (single largest customer, ~25% TSMC revenue) — iPhone/Mac SoC
2. NVIDIA (fast-growing, ~10-15% TSMC revenue) — GPU/Grace
3. AMD (stable major customer, ~10%) — EPYC/Ryzen
4. Qualcomm (~7-8%) — Snapdragon
5. MediaTek (~5-6%) — Dimensity
6. ARM/Phoenix (new customer, <1%) — Lowest priority

As a new wafer customer for TSMC (distinct from ARM's long-standing relationship as an IP vendor), ARM has virtually no bargaining power during periods of tight capacity. This represents a structural shift "from never waiting in line to now having to wait in line."

Path 2: Qualcomm Lawsuit → ARM Pricing Power
The dispute concerns the transfer of CPU design licenses obtained by Qualcomm through its acquisition of Nuvia. A victory for Qualcomm (with a ruling expected in 2024) would affirm the transferability of ARM's licensing agreements, meaning major customers could acquire ARM licensees to circumvent price increases.

The profound implications of this ruling are underestimated by the market:

• Short-term: ARM's royalty claims on Qualcomm Oryon chips are limited (potential loss of $50-100M/year in royalties)
• Mid-term: Other major customers (Google/Amazon) may also acquire smaller ARM licensees to obtain lower rates
• Long-term: The foundation of ARM's "per-customer independent license" pricing model is undermined—the precedent of contract transferability will reduce ARM's pricing power with each new customer
• Quantitative Impact: If the top 10 customers achieve a 15-20% royalty discount through M&A, ARM's annual royalty revenue could be impacted by approximately -$200-400M

Qualcomm Oryon's special status: Qualcomm uses Nuvia's self-designed core (not ARM's Cortex core) for its PC/data center chips. If these chips only require ALA-level royalties (instead of TLA-level royalties), ARM's PC royalty ceiling will be significantly reduced.

Path 3: AI Investment Cycle → ARM Royalties
ARM's Data Center (DC) growth is highly dependent on AI infrastructure investment (>$150B/year). Should a sharp contraction similar to the post-2000 dot-com bubble occur, DC royalties would decline accordingly.

AI CapEx cycle historical mapping:

Cycle	Peak Investment	Investment Decline Magnitude	Recovery Time	Analogous ARM Impact
2000 Dot-com Bubble	~$110B	-60%	5 years	DC Royalties -50-60%
2007 Financial Crisis	~$50B(IT)	-30%	3 years	DC Royalties -25-35%
2016 Cloud Computing	~$120B	-15%	1 year	DC Royalties -10-15%
2025 AI	~$300B+	?	?	DC Royalties -??%

The current scale of AI CapEx (>$300B/year) far exceeds any historical cycle. Should a 30% pullback occur (historical median), an investment contraction of over $90B would directly impact ARM's DC chip shipments. ARM management's guided target of "DC royalties surpassing mobile" would be delayed by 2-3 years.

Key Vulnerable Point: ARM's DC royalty growth (>100% YoY) is highly leveraged to the growth rate of AI CapEx. If the growth rate slows from +40% to +10%, ARM's DC royalty growth could decline from >100% to 20-30%—because royalties lag chip shipments by 1-2 quarters, and chip shipments lag CapEx decisions by 6-12 months.

Path 4: Geopolitics → Arm China
If U.S.-China tech decoupling intensifies, Arm China may be unable to obtain the latest ARM IP. Arm China's $749M revenue (accounting for 19% of ARM) faces policy risks. Chinese customers will accelerate their migration to RISC-V.

Arm China's complex legal structure:

• Arm China (Anmou China) is an independent Chinese entity, with ARM holding approximately 48% economic interest (but limited control)
• The control dispute between ARM and former CEO Allen Wu from 2020-2022 has been resolved, but it exposed the fragility of Arm China's governance
• When Chinese customers obtain ARM IP through Arm China, the versions typically lag the latest global versions by 1-2 generations
• The Chinese government has designated RISC-V as a core component of its "indigenous and controllable" technology roadmap

Three geopolitical scenarios and their impact on ARM revenue:

Scenario	Probability	Arm China Revenue Impact	ARM Total Revenue Impact	Timeline
De-escalation (Status Quo Persists)	40%	+5-10%/year	+1-2%/year	—
Gradual Decoupling	40%	-5-15%/year	-1-3%/year	3-5 years
Sharp Decoupling	20%	-30-50%/year	-6-10%/year	1-2 years
Weighted Average	—	-2-5%/year	-0.5-1%/year	—

Under a sharp decoupling scenario, Arm China's $749M revenue could shrink to $375-525M within 2 years, while the RISC-V ecosystem in China would rapidly mature (due to a forced effect). The impact on ARM's global valuation would not be limited to the loss of Chinese revenue but also include a second-order effect of "RISC-V maturing in China and then being exported back to the global market."

7.6 Assessing ARM's Irreplaceability in the Global Semiconductor Supply Chain

ARM is often described as "irreplaceable"—but this assertion needs to be examined market by market:

End Market	ARM Current Share	Alternatives	Replacement Difficulty (1-10)	Replacement Timeline	Irreplaceability
High-end Mobile AP	>99%	RISC-V	9	5-7 years	Extremely High
Mid-to-low-end Mobile AP	>98%	RISC-V	7	3-5 years	High
DC/Server CPU	~22%	x86(Intel/AMD)	3	Already Exists	Low
IoT/MCU	~65%	RISC-V	4	1-3 years	Low
Automotive ADAS	~80%	RISC-V(Quintauris)	7	5-7 years	High
Automotive Body	~50%	RISC-V	5	3-5 years	Medium
PC/Laptop	~13%	x86(Intel/AMD)	2	Already Exists	Extremely Low
Networking/Base Station	~55%	RISC-V/x86	6	3-5 years	Medium

Overall Irreplaceability Score: 5.4/10 — ARM is almost irreplaceable in high-end mobile and automotive ADAS (within 5-7 years), but faces existing or rapidly developing alternatives in data centers, IoT, and PCs.

Key Insight: ARM's irreplaceability exhibits a U-shaped distribution — strongest at the highest end (flagship mobile SoCs) and lowest end (ultra-low-power IoT), but weakest in the middle tier (data centers, PCs). However, ARM's growth narrative (DC royalties > mobile) is precisely concentrated in the areas of lowest irreplaceability — this is a hidden contradiction in its valuation logic.

---

Chapter 8: Historical Analogy — Intel's Moat Collapse (2005-2020) vs. Visa's Deepening Network Effect

8.1 Intel's Moat Collapse Timeline: ARM's Mirror Risk

From an absolute hegemon in 2005 to a 67% market cap decline by 2024, Intel experienced a series of seemingly minor but cumulatively fatal moat erosions. The lessons ARM can draw from this go far deeper than "don't make Intel's mistakes":

Intel Moat Collapse Timeline:

Year	Event	Moat Impact	Analogy to ARM
2005	Apple switches from PowerPC to Intel	Intel at its peak — Mac using Intel proved it was the "best choice"	ARM's current stage: 99% penetration = peak
2006	Intel sells XScale ARM business	Abandoned its own ARM chip foothold — strategically exited the mobile market	If ARM abandons a certain end market
2007	Apple asks Intel to manufacture iPhone chips, CEO Otellini refuses	Gravest strategic error in history: handed over the smartphone era	—
2007	iPhone uses ARM	Mobile market chooses ARM over x86	— (ARM is the beneficiary)
2010	ARM Cortex-A15 approaches PC-level performance	Began to prove ARM was "more than just a phone chip"	RISC-V is currently at a similar stage
2013	Intel abandons Atom mobile chips	Acknowledged inability to compete with ARM in power efficiency	RISC-V has already started replacing ARM in IoT
2015	10nm delay (originally planned for 2016)	Manufacturing advantage shaken for the first time	If ARM's "architectural advantage" is matched by RISC-V
2017	AMD Zen released	Performance gap narrowed from 50% to <10%	Tenstorrent Ascalon performance close to Neoverse V3
2018	ARM Neoverse enters servers	Intel's server monopoly challenged head-on for the first time	RISC-V has already gained 30% share in embedded systems
2020	Apple M1 released	Proved ARM chips comprehensively surpassed Intel	If RISC-V has an "M1 moment" in 2028
2021-24	Intel 7nm/4nm continued delays	Manufacturing advantage completely lost	—
2024	Intel market cap drops 67% from 2020 peak	From $250B → $85B, P/E from 15x → loss	—

Key Pattern: Intel's collapse did not happen overnight — it took 15 years from its peak in 2005 to Apple M1 in 2020 (confirming the moat collapse). However, the real valuation erosion was concentrated in the last 3-5 years (2020-2024, -67%).

Timeline mapped to ARM:

• 2020-2025: ARM is at its peak (similar to Intel 2005-2010), RISC-V is equivalent to early ARM
• 2025-2028: RISC-V performance parity + initial ecosystem maturity (similar to Intel 2015-2018, AMD Zen achieving parity)
• 2028-2030: A potential "M1 moment" for RISC-V (similar to Intel 2020)
• 2030+: If RISC-V succeeds, ARM's valuation will face an Intel-style compression

Intel Lesson 1: The Non-Linearity of Moat Collapse
Intel's market share barely changed between 2005 and 2017 (servers >95%). But "unchanged share ≠ unchanged moat" — the development of AMD Zen, the preparation of ARM Neoverse, and the design of Apple M1 were all quietly underway during this period. When the moat truly began to collapse (2017-2020), the speed exceeded everyone's expectations. ARM might also experience a similar pattern: market share data might look good in 2025-2028 (>95% mobile, >20% servers), but RISC-V's preparatory work is already underway beneath the surface.

Intel Lesson 2: Price Hikes as a Catalyst for Moat Collapse
Intel continuously raised server chip prices (from $1000 to $5000+ for high-end) between 2005 and 2015, leveraging its monopolistic position. This created an economic incentive for AMD (and later ARM) to enter — if Intel had maintained a low-price strategy, AMD's Zen project might never have received AMD board approval (insufficient ROI). ARM is repeating Intel's mistake: v9 doubling → CSS doubling again → v10 potentially doubling again → the economic incentive for RISC-V to enter continues to rise.

Intel Lesson 3: "Insurmountable Ecosystem Barrier" is an Illusion
Intel's most confident moat was the x86 software ecosystem — "billions of lines of x86 code cannot be rewritten." Apple M1 proved this assumption wrong: the Rosetta 2 translation layer achieved 95%+ x86 → ARM compatibility. RISC-V faces a similar narrative ("ARM software ecosystem cannot migrate"), but LLVM/GCC toolchains have already eliminated compilation layer barriers, and Android GKI has adapted the kernel for RISC-V. Ecosystem barriers can be circumvented by technological innovation.

8.2 Visa's Deepening Network Effect Comparison: Why ARM is Not a Payment Network

Section 1.4 briefly discussed ARM vs. Visa. Here, a more systematic structural analysis is provided:

Network Effect Type Comparison:

Dimension	Visa	ARM	Difference
Network Type	Bilateral Platform (Merchants ↔ Cardholders)	Unilateral Ecosystem (Developers → Software → Devices)	Visa's network effect is stronger
Direct Network Effect	More Cardholders → More Valuable to Merchants	More ARM Devices → More Valuable to Developers	Similar
Indirect Network Effect	More Merchants → More Valuable to Cardholders	More Software → More Valuable to Device Users	Similar
Network Exclusivity	High (Merchants typically do not build their own payment networks)	Low (Developers can code for both ARM + RISC-V simultaneously)	Key Difference
Migration Cost	Extremely High (Tens of millions of merchant terminals globally)	Medium (Compiler switching + library porting)	ARM is more vulnerable
Alternatives	No zero-cost alternative	RISC-V = zero-cost alternative	Fundamental Difference
Regulation	Fees subject to regulatory restrictions	No fee regulation	ARM's current advantage
Historical Stability	50 years (from BankAmericard to present)	35 years (but RISC-V only exists for 15 years)	Visa is more mature

Key Insight: Visa's network is exclusive — you cannot build a Visa-compatible payment network yourself (requiring multilateral agreements from banks, merchants, and regulators). ARM's ecosystem is forkable — RISC-V uses the same compilers (LLVM/GCC), the same operating systems (Linux), and the same development languages. Developers don't need to "leave ARM" to "simultaneously support RISC-V". This means ARM's network effect moat is an order of magnitude thinner than Visa's.

Valuation Implications: Visa's P/E (33x) already includes a stability premium for "irreplaceable infrastructure" and a discount for fee regulation. ARM's P/E (140x) includes the assumption of "irreplaceable infrastructure" + a high growth premium, but does not factor in RISC-V substitution risk. If the market starts pricing ARM like "infrastructure with alternatives" (P/E 40-60x) rather than "irreplaceable infrastructure" (P/E 25-35x), the valuation difference represents the downside.

8.3 Comprehensive Assessment of Six Historical Analogies: Who Does ARM Resemble Most?

ARM's business model has several potential analogies in tech history. Let's evaluate their applicability one by one:

Analogy	Core Similarities	Core Differences	Implications for ARM's Valuation	Applicability
Visa/Mastercard	Transactional revenue, asset-light, high gross margin	Visa has no open-source alternative, bilateral network	P/E 30-35x ("Irreplaceable")	3/10
Microsoft Windows	OS-level lock-in, licensing + royalties	Windows has no zero-cost alternative (Linux desktop failed)	Historical P/E 25-35x	5/10
Qualcomm QTL	IP licensing + royalties, chip shipment driven	QTL faces continuous legal challenges, declining royalty rates	QTL P/E 15-20x (includes discount)	7/10
Intel (2005-2010)	Architectural monopoly, high P/E, "irreplaceable"	Intel had no open-source alternative at the time (AMD was a commercial competitor)	Peak P/E 25-30x → 2024 Loss	6/10
Dolby	IP licensing, embedded in standards, high gross margin	Dolby market is narrower, slower growth	P/E 30-40x (stable growth)	5/10
ASML	Critical semiconductor link, no alternative	ASML is physical equipment (truly irreplaceable)	P/E 35-50x	4/10

Best Analogy: Qualcomm QTL — identical IP licensing + royalty model, facing similar pricing power challenges (customer legal challenges + open standard threats), and a history of forced royalty rate reductions. QTL's P/E is in the 15-20x range. If ARM's long-term steady-state P/E converges towards QTL (considering RISC-V poses a greater threat to ARM than Wi-Fi standards did to QTL), the current 140x P/E implies a 7-9x overvaluation.

Quantitative Bridging for QTL Analogy:

Dimension	QTL	ARM	Reason for ARM Premium	Reasonable Premium?
P/E	17x	140x	ARM grows faster (+24% vs +5%)	Partially reasonable (2-3x)
Revenue Growth	+5%	+24%	—	—
Gross Margin	~100%	95%	Similar	Does not support premium
Substitution Threat	Legal challenges/standardization	RISC-V (more severe)	ARM should be discounted	Supports discount
Customer Concentration	Similar (Top 5 >50%)	Similar (Top 5 56%)	Similar	Does not support premium
Related Party Transactions	None	SoftBank 30%	ARM should be discounted	Supports discount
Ownership Structure	Dispersed Ownership	SoftBank 90%	ARM should be discounted	Supports discount

Reasonable P/E Derivation: QTL 17x × Growth Premium 2.5x × Moat Discount 0.85x × Governance Discount 0.90x = ~32x — This implies that if the market re-prices ARM according to the "best analogy", the target share price would be $32x × EPS($0.91) = $29, representing a 77% drop from the current $128.

This figure is clearly an extreme scenario — the market is unlikely to re-price ARM as "another QTL" in the short term. However, it provides a useful "valuation floor" reference: even in the most extreme re-pricing scenario, ARM's core IP licensing business is still worth $30-40B (vs current $136B).

8.4 "Boiling Frog" Scenario: ARM's Path of Slow Moat Erosion

Combining lessons from Intel and comparisons with Visa, ARM's most probable moat erosion path is not "sudden collapse" but rather a "boiling frog" scenario:

24-Month Path (March 2026 → March 2028):

Month	Event	Impact	ARM Share Change	Market Reaction
0-6 Months	Tenstorrent Ascalon mass production, Meta RISC-V MTIA deployment	RISC-V achieves performance parity in DC for the first time	IoT -2pp, DC -1pp	Ignored ("Niche")
6-12 Months	ARM v9 penetration to 65%+, CSS to 20%+, royalties continue to grow	ARM revenue growth masks share loss	IoT -3pp	Optimistic ("Growth exceeds expectations")
12-18 Months	Quintauris releases automotive RISC-V standard, Google announces Android 16 native RISC-V support	RISC-V upgrades from "available" to "supported"	IoT -4pp, automotive designs begin migration	Concerned ("Long-term risk")
18-24 Months	2-3 Chinese mobile OEMs launch RISC-V mobile chips (low-end), ARM renewal negotiations hit obstacles	First cracks appear in the mobile "iron curtain"	IoT -5pp, Mobile -1pp	Valuation begins to adjust

Characteristics of this path:

1. Each month appears to be "nothing major" — RISC-V's progress is gradual
2. ARM's revenue may still be growing throughout this process (price hike effect > share loss)
3. However, looking back at the end of 24 months, ARM's moat has transformed from "irreplaceable" to "has alternatives but still leads"
4. Once this perception shift is accepted by the market, the P/E compression from 140x to 60-80x could be completed within 3-6 months

The "boiling frog" scenario is the most difficult for ARM to defend against — because each individual RISC-V advancement is not enough to trigger an alarm, but the cumulative effect is fatal. Intel experienced the exact same pattern from 2005-2020.

---

Chapter 9: Management and Organizational Capabilities — The Challenge of Transforming from IP Licensing to a Versatile Player

9.1 In-depth Profile of CEO Rene Haas

Professional Experience:

Period	Position	Key Experiences
1998-2013	NVIDIA, various IP/Partner roles	Established close cooperation with ARM
2013-2017	NVIDIA, Head of IP Licensing and Partnerships	Managed NVIDIA's IP licensing business, including ARM licenses
2017-2022	ARM, Head of IP Products Division	Responsible for the full range of Cortex/Neoverse IP products
Feb 2022-Present	ARM CEO	Led IPO + v9/CSS/Phoenix strategies

Haas's Strategic Positioning: He is a "product-oriented" CEO (vs. SoftBank's Masayoshi Son's "vision-oriented"). Haas's core contributions at ARM include driving CSS (incremental pricing power) and accelerating the commercialization of Neoverse in DC. Phoenix is more of a SoftBank-led strategic direction.

Haas's Trilemma:

1. Maintain IP Licensing Ecosystem — Retain the trust of licensees (including former employer NVIDIA)
2. Advance Phoenix — Enter chip design without alienating licensees
3. Serve SoftBank Strategy — Execute Masayoshi Son's AI infrastructure vision

These three objectives are inherently contradictory: advancing Phoenix (#2) will inevitably damage IP licensing trust (#1), and SoftBank's strategy (#3) is the fundamental driver of Phoenix. Haas's effectiveness as CEO largely depends on his ability to find a balance among the three — but the room for balance is being squeezed by SoftBank's aggressive AI strategy.

9.2 CEO Compensation and Insider Trading

Confirmed CEO Compensation Data:

• Annual Total Compensation: $24.46M (latest proxy data)
  • Base Salary: ~$1.35M (~5.5%)
  • Equity/Options/Bonus: ~$23.1M (~94.5%)
• IPO Special Bonus (FY2024): ~$69M (one-time IPO-related award)
• Direct Stock Ownership: 0.033% of ARM, valued at ~$43.7M based on recent stock price
• CEO Rule 144 Systematic Sales: ~6,152 ADRs per month, ~$0.7-1.0M/month
  • This is a standard RSU vesting automatic sale and does not constitute a bearish signal
  • However, the CEO has not increased holdings in the open market
• New Rule: Effective March 18, 2026, the SEC requires directors and officers of FPIs (including ARM) to file Form 4 insider trading reports within 2 business days → Previously, ARM as an FPI enjoyed an exemption, and this will significantly increase insider trading transparency

Information Limitations on ARM Insider Trading:
As a Foreign Private Issuer + Controlled Company, ARM enjoys dual disclosure exemptions:

1. Not required to file Section 16(a) insider trading reports (US standard)
2. Disclosure of management compensation details is below SEC requirements (the UK 20-F standard is more lenient)
3. Transactions by SoftBank as the 90% controlling shareholder do not require the same disclosure as that for <10% shareholders

9.3 Organizational Structure and Capability Matrix

ARM has approximately 7,600 employees across 50+ countries. Assessing organizational capabilities is critical to understanding the feasibility of Phoenix:

Capability Dimension	IP Licensing Company Needs	Chip Design Company Needs	ARM's Level	Gap Assessment
CPU Core Design	Extremely Strong	Strong	Extremely Strong (World-class)	No Gap
Physical Design (PD)	Medium	Extremely Strong	Medium (CSS improving)	Medium Gap
Chip-level Verification	Low (IP-level)	Extremely Strong (DFT/BIST)	Low	Large Gap
Supply Chain Management	No Demand	Critical (TSMC relationship)	Low (Indirect via SoftBank)	Very Large Gap
Mass Production Management	No Demand	Critical (Yield/Quality)	None	Very Large Gap
Software Ecosystem	High (SDK/Toolchain)	High (Drivers/Firmware)	High	Low Gap
Sales/BD	IP Sales (B2B)	Chip Sales (B2B)	High	Medium Gap

Key Observation: ARM is world-leading in CPU core design and software ecosystem but severely lacking in chip-level verification, supply chain management, and mass production management. The acquisition of Ampere ($6.5B) partially mitigates these gaps (Ampere has experience mass producing ARM server chips), but Ampere's scale (~1,000 employees) is still small relative to Phoenix's ambition.

9.4 Ampere Acquisition: Bridging Capability Gaps or a Strategic Burden?

In 2024, SoftBank acquired Ampere Computing (an ARM server chip company) for approximately $6.5B and subsequently integrated it into ARM. This is a crucial step in understanding ARM's transformation:

Ampere's Asset Value:

Asset	Valuation	Contribution to ARM
Server Chip Design Experience (Altra/AmpereOne)	High — 4 years of mass production experience	Addresses shortcomings in chip-level verification and mass production
TSMC Foundry Relationship	Medium — 5nm/4nm tape-out experience	Phoenix requires 3nm, only partially applicable
DC Customer Relationships (Oracle/Microsoft)	High — Direct DC sales channel	ARM previously only had an IP sales channel
Team (~1,000 people)	Medium — Led by former Intel executives	Chip design and verification talent
Intellectual Property	Medium — AmpereOne self-developed core (non-Neoverse)	Not directly used for Phoenix

Three Problems with the Acquisition:

1. Overvaluation? $6.5B for a company with annual revenue of ~$200-300M and not yet profitable, implying >20x P/S. This exceeds ARM's own P/S (28x) growth premium.
2. Integration Risk: Ampere's AmpereOne core (self-developed v8 microarchitecture) and ARM's Neoverse V3 (Phoenix foundation) represent different design philosophies. Integrating the two design teams could lead to internal competition or wasted resources.
3. SoftBank-Led: The acquisition was initiated by SoftBank (Ampere was originally a SoftBank Vision Fund investment) — essentially SoftBank selling its portfolio company to ARM (a related-party transaction).

Ampere's Contribution to Phoenix: Ampere's greatest value is not its technical IP, but its mass production experience. Ampere has gone through the complete cycle from RTL design to TSMC tape-out to customer deployment — something ARM (as an IP licensor) has never done internally. This tacit knowledge (yield optimization, power design, thermal management, customer support) is crucial for Phoenix's success.

9.5 Key Management Profile (Excluding CEO)

Executive	Position	Background	Impact on ARM's Strategy
Jason Child	CFO	Joined in 2023, former Splunk/Opendoor CFO	Extensive capital market experience, but lacks semiconductor industry background
Chris Madhavapeddy	CPO	20+ year ARM veteran	Custodian of core IP product lines, CSS architect
Dipti Vachani	SVP Automotive	Former Intel/NXP	Key driver for automotive endpoint growth
Mark Hambleton	SVP Tech Strategy	Promoted internally at ARM	Technology roadmap decision-maker
Mohamed Awad	SVP Infrastructure	Recalled from Arm China	Head of DC/5G/Edge Strategy

Overall Management Assessment:

• Strengths: World-class technical team (undisputed CPU core design capability)
• Weaknesses: CFO lacks semiconductor experience (Splunk/Opendoor are software/real estate tech) — as ARM transitions from IP licensing to chip design + sales, the CFO needs to understand hardware company issues such as COGS, inventory management, and channel management.
• Potential Conflicts: Potential resource competition between ARM's CPO (Head of IP Products) and the Phoenix team — every $1 of engineering resources invested in Phoenix is $1 not invested in core IP improvements.
• SoftBank Interference: Key strategic decisions (Phoenix direction, Ampere acquisition, Stargate participation) are led by SoftBank, limiting management's execution autonomy.

9.6 Employees and Talent Competition

ARM has approximately 7,600 employees globally (an increase of about 18% from ~6,400 at IPO). The talent competition landscape in the semiconductor IP industry:

Company	Employees	Revenue per Employee ($K)	R&D per Employee ($K)	SBC per Employee ($K)
ARM	7,600	615	346	115
Synopsys	19,500	313	180	68
Cadence	11,500	374	165	52
Qualcomm QTL (IP division estimate)	~2,000	~2,500	~500	~150

ARM's SBC per employee ($115K/year) is the highest among its EDA peers (69% higher than Synopsys, 121% higher than Cadence). This reflects ARM's reality of needing to compete with NVIDIA/Apple/Google for top CPU design talent — these companies offer higher total compensation packages (NVIDIA ~$200K+ SBC per employee).

Talent Retention Risk: ARM's CPU design team (possibly 300-500 people) is the company's most critical asset. If NVIDIA/Apple/Google continue to attract ARM's core designers, ARM's technological lead will be diminished. The rapid growth in ARM's SBC in recent years (from FY2023 $350M to FY2026E $1.1B) partly reflects the urgency of talent retention.

9.7 In-depth R&D Investment Analysis

Fiscal Year	R&D ($M)	R&D/Revenue	Estimated SBC Content	Adjusted R&D/Revenue	YoY Growth
FY2021	768	37.9%	~$100M	~33%	—
FY2022	945	35.0%	~$150M	~29%	+23%
FY2023	1,080	40.3%	~$200M	~33%	+14%
FY2024	1,932	59.8%	~$600M	~41%	+79%
FY2025	2,009	50.1%	~$500M	~38%	+4%
TTM	2,628	56.3%	~$550M	~44%	+30%

R&D Growth Breakdown:

• The +79% jump in FY2024 primarily comes from concentrated SBC recognition post-IPO (~$600M SBC recorded in R&D)
• After adjusting for SBC, "real" R&D grew from ~$700M (FY2022) to ~$2,078M (TTM) — still +197%
• This growth rate significantly outpaces revenue growth (+73%), indicating ARM is investing heavily in multiple growth vectors (v9/CSS/Phoenix/Neoverse)

R&D Efficiency Issues: ARM's R&D-to-revenue ratio (56%) is significantly higher than its EDA peers (Synopsys 32%, Cadence 25%). This can be interpreted optimistically (investing in the future) or cautiously (projects like Phoenix permanently altered ARM's cost structure, reducing profit margins). If R&D does not lead to a corresponding increase in revenue (i.e., declining R&D ROI), the narrative of ARM's profit margin expansion will be disproven.

R&D ROI Analysis (Key Addition):

Metric	Calculation Method	ARM	Synopsys	Cadence	Assessment
R&D ROI (3-year lag)	ΔRevenue(t+3)/R&D(t)	$1.1B/$768M=1.43x	$2.5B/$1.2B=2.08x	$1.1B/$0.6B=1.83x	ARM Lowest
R&D ROI (Current Period)	Incremental Revenue/R&D	$1.0B/$2.6B=0.38x	$1.7B/$2.0B=0.85x	$0.2B/$1.1B=0.18x	ARM Medium
R&D per Employee	R&D/Employee ($K)	$346K	$180K	$165K	ARM 2x Peers
Patent Output	Number of Patents/R&D ($M)	Low (IP Licensing Model)	High	High	ARM Patents Not a Core Moat

ARM's R&D ROI (3-year lag) is 1.43x—for every $1 of R&D invested, $1.43 in future incremental revenue is generated—lower than Synopsys (2.08x) and Cadence (1.83x). This indicates that ARM's R&D investments are facing diminishing marginal returns:

1. The design complexity of core IP (Cortex/Neoverse) grows exponentially as process technology shrinks.
2. New initiatives such as Phoenix/CSS require substantial R&D but have long revenue conversion cycles (3-5 years).
3. The classification of SBC as R&D inflates nominal R&D (approximately $550M of SBC is included in R&D).

If R&D ROI is adjusted for SBC: Actual cash R&D is approximately $2.08B (TTM, excluding $550M SBC), ROI = $1.0B/$2.08B = 0.48x — which is still relatively low. Management needs to demonstrate that R&D investments in Phoenix and CSS can generate returns between FY2028-FY2030; otherwise, a sustained decline in R&D efficiency will compress long-term profit margins.

9.8 ARM's IP Product Roadmap and Lumex Brand Strategy

ARM launched the Lumex brand in late 2025—marking an attempt to move from Cortex/Neoverse towards a unified consumer brand:

Lumex Product Line:

• Lumex C1: Next-generation CPU core (replacing Cortex-X5), Q4 2026 samples
• Lumex G1: GPU core (potentially challenging Qualcomm Adreno/ARM's own Mali)
• Lumex S1: System IP (interconnect/security/accelerator)

Strategic Intent of Lumex:

1. To build consumer brand recognition (similar to "Intel Inside")
2. To provide a brand vehicle for CSS packaging
3. To increase influence with end consumers (bypassing chip manufacturers to establish a direct brand)

Risks of Lumex:

1. "Intel Inside" succeeded because Intel is a chip manufacturer (with a physical product)—ARM, as an IP licensor, lacks a suitable vehicle to establish a consumer brand.
2. Licensees (Qualcomm/MediaTek/Apple) are unlikely to cooperate in promoting the Lumex brand (they prefer to promote their own brands).
3. Lumex brand investment will increase SG&A expenses with uncertain returns.

Implications for Analysis: The Lumex brand is likely an attempt by ARM management to create an "Intel Inside"-style brand premium for the company. If successful (Lumex becomes a factor in consumer purchasing decisions), ARM's pricing power will be significantly enhanced; if it fails (the most probable outcome), it will only increase marketing expenses without generating returns.

9.9 Incentive Alignment Analysis

SoftBank vs. Public Shareholder Interest Divergence:

Dimension	SoftBank Interests	Public Shareholder Interests	Conflict?
Phoenix	Promote (Support AI Ecosystem)	Cautious (Harm Neutrality)	High
Related-Party Licensing	More agreements (Increase ARM Revenue)	Independent verification (Ensure Fairness)	High
Stock Price	Maintain high level (Pledged Value)	Reflect fundamentals reasonably	Medium
Dividends/Buybacks	Low priority (Needs capital for AI investment)	High priority (Shareholder returns)	Medium
Long-Term Strategy	Full AI Ecosystem Integration	Maintain asset-light, high ROIC model	High

As a Controlled Company (SoftBank holding ~90%), ARM is exempt from most NYSE governance rules (including independent director majority requirements, compensation committee independence requirements, etc.). Public shareholders have extremely limited governance influence at ARM.

9.10 SoftBank's Control: Seven Governance Exemptions

As a "Controlled Company" and "Foreign Private Issuer," ARM enjoys the following NYSE and SEC governance rule exemptions:

#	Exemption Content	US Standard Public Company Requirements	ARM's Actual Practice	Impact on Public Shareholders
1	Independent Director Majority	Board >50% independent	Not Required	SoftBank can control the Board
2	Compensation Committee Independence	All independent directors	Not Required	CEO compensation determined by SoftBank
3	Nominating Committee Independence	All independent directors	Not Required	New directors nominated by SoftBank
4	Insider Trading Disclosure	Section 16(a), Form 4 within 2 days	Applicable from March 18, 2026	Upcoming Improvement
5	Annual Governance Certification	CEO/CFO Sarbanes-Oxley certification	Simplified FPI Version	Less disclosure
6	Related-Party Transaction Audit	Independent review by audit committee	Limited	SoftBank transactions lack independent oversight
7	Shareholder Proposal Rights	Any 1%+ shareholder can submit a proposal	UK Law Standard (More Lenient)	Public shareholders find it difficult to submit proposals

Practical Implications: SoftBank can, with virtually no external checks and balances:

1. Determine ARM's strategic direction (Phoenix, Ampere acquisition, Stargate participation)
2. Set management compensation (CEO $24.46M + $70M IPO bonus)
3. Sign licensing agreements with ARM through related-party transactions (approximately 30% of licensing revenue)
4. Pledge ARM shares to obtain loans ($20 billion) without shareholder approval
5. Engage in insider trading without public notification (until March 18, 2026)

"Transparency Improvement" on March 18, 2026: The SEC will require directors and 10%+ shareholders of FPIs (including ARM) to disclose insider transactions (Form 4) within 2 business days. This will significantly enhance the transparency of ARM's insider trading—investors will for the first time be able to see real-time changes in SoftBank's and management's holdings. This rule change could be a short-term catalyst (if data shows insiders are buying) or a negative catalyst (if data shows large-scale divestments).

---

Chapter 10: Deep Dive into Earnings Quality — The Triple Distortion of SBC/DSO/OCF

10.1 Quarterly Breakdown of SBC and Dilution Effect

SBC Overview:

Metric	FY2023	FY2024	FY2025	FY2026E(TTM)
SBC ($M)	~$350	$1,037	~$820	~$870
SBC/Revenue	~13%	32%	20.5%	~18.6%
GAAP OPM	25.3%	3.1%	20.6%	18.7%
Non-GAAP OPM (Est)	~38%	~35%	~41%	~37%
GAAP-Non-GAAP Difference	~13pp	~32pp	~21pp	~18pp
Diluted Shares (M)	1,026	1,044	1,063	1,069
Net Share Growth	—	+1.8%	+1.8%	+0.6%
Buyback Offset Rate	0%	0%	~41%	~33%

Economic Substance of SBC:

• FY2026 TTM SBC ~$870M → ~100% of Operating Profit ($871M)
• This means: If SBC is treated as a true cost (which it is), almost 100% of ARM's GAAP operating profit is consumed by SBC
• Buybacks ($355M TTM) only offset 41% of SBC → Net dilution ~0.4% per year
• For a 140x P/E stock, every 1% dilution = $1.36B market cap loss

SBC Allocation by Segment (Estimate):

Segment	SBC % (Est)	SBC Amount (TTM, $M)	Rationale
R&D (CPU Design + Software)	~65%	~$565	Core technical talent is most expensive
SG&A (Sales + G&A)	~25%	~$218	Sales + Executive Incentives
COGS (minimal)	~10%	~$87	Production Support

10.2 DSO Deep Dive: Worst Accounts Receivable Quality in the Industry

DSO (Days Sales Outstanding) Quarterly Tracking:

Quarter	DSO (Days)	Industry Benchmark (SNPS/CDNS)	Difference	Possible Reasons
Q4 FY24	109	70/80	+35 Days	Year-end contract settlement
Q1 FY25	121	—	—	Normal fluctuation
Q2 FY25	174	—	+100 Days	Delayed collection for large Q1 licensing agreements
Q3 FY25	146	—	—	Partial recovery
Q4 FY25	138	—	—	Continued improvement
Q1 FY26	184	—	+110 Days	Year-end contracts + SoftBank extended payment terms?
Q2 FY26	157	—	—	Seasonal decline
Q3 FY26	146	—	—	Stable
Average	147	~75	+72 Days	—

ARM's DSO is approximately twice that of its EDA peers. This is not a normal industry phenomenon—Synopsys and Cadence also do software licensing, but their DSOs are only 70-80 days. ARM's prolonged DSO suggests:

1. Lax licensing contract terms: Large ALA/TLA contracts may have payment collection staggered over 3-5 years, but revenue is recognized at the time of signing.
2. Special payment terms for SoftBank related-party transactions: SoftBank may enjoy extremely long payment terms (>180 days).
3. Cross-border collection friction: Collection through intermediary entities like Arm China increases delays.

The most concerning metric: Accounts Receivable growth / Revenue growth ratio

Fiscal Year	Revenue Increment($M)	AR Increment($M)	AR/Rev Increment Ratio	Normal Range
FY2024	+554	+135	24%	Normal
FY2025	+774	+632	82%	Extremely Abnormal
FY2026 3Q YTD	+664	+83	12%	Improved

The 82% ratio in FY2025 means that for every $1 of new revenue, $0.82 turned into accounts receivable – this is a warning sign in any industry. The improvement in FY2026 (12%) is positive and may reflect collections starting for contracts signed intensively at the end of FY2025. However, continuous monitoring is required.

10.3 Operating Cash Flow Quality Breakdown

OCF vs. Net Income Comparison (Quarterly):

Quarter	Net Income($M)	OCF($M)	OCF/NI	Abnormal?	Primary Detractor
Q4 FY24	224	667	2.98	High (Good)	Significant decrease in deferred tax assets
Q1 FY25	223	-290	-1.30	Very Poor	AR +$723M consumed all cash
Q2 FY25	107	6	0.06	Poor	AR +$111M + Working Capital -$269M
Q3 FY25	252	423	1.68	Good	AR collections + SBC contribution
Q4 FY25	210	258	1.23	Normal	AR -$377M drag
Q1 FY26	130	332	2.55	High (Good)	SBC $241M add-back
Q2 FY26	238	567	2.38	High (Good)	SBC $265M + AR collections
Q3 FY26	223	365	1.64	Good	AR -$96M minor drag

FY2025 Full Year: OCF/NI = $397M/$792M = 0.50 — This is an extremely low ratio for an IP/software company (normally should be >1.0). The core reason for the poor cash conversion in FY2025 was a sharp increase in accounts receivable of $632M.

FY2026 TTM (first 3 quarters): OCF/NI = $1,264M/$591M = 2.14 — A significant improvement. However, the improvement mainly comes from the "non-cash add-back" of SBC ($506M confirmed + $250M estimated = ~ $756M). If we exclude the OCF-boosting effect of SBC:

• Adjusted OCF = $1,264M - $756M = $508M
• Adjusted OCF/NI = $508M/$591M = 0.86 — Still relatively low

10.4 Free Cash Flow (FCF) Quality

FY/TTM	OCF($M)	CapEx($M)	FCF($M)	FCF/Revenue	FCF/NI
FY2023	883	-68	815	30.4%	156%
FY2024	775	-131	644	24.0%	210%
FY2025	397	-234	163	4.1%	21%
TTM(FY26 3Q)	1,522	-553	969	20.8%	121%

Key Change: CapEx soared from $68M in FY2023 to $553M in TTM (+713%). This surge in CapEx reflects:

1. Phoenix chip design NRE (TSMC 3nm tape-out costs + design tools)
2. New R&D facilities (UK/US/France)
3. IT infrastructure (AI/ML development platform)

FCF Yield: FCF ($969M) / Market Cap ($136B) = 0.71% — significantly lower than the treasury yield (~4.5%). Investors holding ARM receive a cash return that is only 1/6 of the treasury yield – this is entirely paying for growth.

10.5 Capital Return Analysis: The Cost of Transitioning from Asset-Light to Asset-Heavy

ARM's capital return metrics are undergoing a structural change:

Metric	FY2023	FY2024	FY2025	TTM	Trend
ROIC	31.2%	18.5%	17.2%	23.4%	Volatile
ROA	8.1%	6.2%	7.8%	8.6%	Stable
ROE	9.4%	7.0%	9.5%	10.2%	Slow Improvement
Asset Turnover	0.28	0.23	0.28	0.32	Stable
CapEx/Revenue	2.5%	4.9%	5.8%	11.8%	Sharply Rising
CapEx/Depreciation	1.3x	2.1x	3.2x	5.1x	Extremely Rising

The sharp change in CapEx intensity is the most noteworthy metric:

• CapEx/Revenue rose from 2.5% in FY2023 to 11.8% in TTM — a nearly 5x increase
• A CapEx/Depreciation ratio >5x means ARM is rapidly building new assets (rather than just maintaining existing ones)
• Comparison: Synopsys CapEx/Revenue ~4%, Cadence ~3% → ARM has already surpassed its EDA peers by 2-3 times

What this means: ARM is transforming from a "pure IP company" with $50-70M/year CapEx to an "IP + chip design company" with $500M+/year CapEx. This transformation will:

1. Lower long-term ROIC (due to a larger capital base)
2. Reduce FCF conversion rate (due to more CapEx consumption)
3. Lower valuation multiples (market values "asset-light" companies higher than "asset-heavy" ones)

DuPont Analysis (TTM):

ARM's ROE (10.2%) is significantly lower than Synopsys's (22.4%)—the gap stems from Net Profit Margin (ARM 12.6% vs SNPS 19.8%, dragged down by SBC) and Asset Turnover (ARM 0.32 vs SNPS 0.41). ARM's P/E is 2.5 times that of Synopsys, but its ROE is only 45% of Synopsys's—this gap requires an extremely high growth rate to bridge.

10.6 Peer Profitability Quality Benchmarking

Where does ARM's profitability quality stand among its IP/EDA peers?

Metric	ARM	Synopsys	Cadence	QCOM-QTL	Worst Performer
Gross Margin	95.4%	80.2%	89.4%	~100%	SNPS
GAAP OPM	18.7%	21.4%	30.9%	~80%	ARM
Non-GAAP OPM	40.7%	33.8%	39.5%	~85%	SNPS
GAAP vs Non-GAAP Difference	22pp	12pp	9pp	~5pp	ARM
SBC/Revenue	~22%	12%	10%	~3%	ARM
DSO (Days)	147	70	80	45	ARM
OCF/NI	2.14	1.35	1.28	1.15	ARM (apparently high but inflated by SBC)
FCF/NI	1.21	1.05	1.10	1.05	ARM (same as above)
ROIC	23.4%	22.1%	29.7%	>50%	SNPS

ARM ranks worst in four key quality metrics: GAAP OPM, GAAP-NonGAAP Difference, SBC/Revenue, and DSO. If these quality discounts were reflected in the valuation (applying a P/E discount of 10-20%), ARM's "fair" P/E should decrease from 140x to 112-126x — the quality discount alone implies a market capitalization compression of $12-30B ($11-28/share).

Most unfavorable finding: ARM's OCF/NI appears high (2.14x), but this is primarily a mathematical effect of adding back SBC. If "Adjusted OCF (excluding SBC add-back) / Adjusted NI (GAAP)" is used to measure true cash conversion:

• ARM Adjusted OCF/NI = ($1,264M - $791M) / $591M = 0.80 — close to the industry worst
• Synopsys: ~1.05 | Cadence: ~1.08 — normal level

Conclusion: ARM's profitability quality is the worst among its peer group—GAAP profit is heavily diluted by SBC, accounts receivable turnover is extremely slow, and true cash conversion rate is low. These quality issues are masked by a 95% gross margin and high growth (+24%), but they will be exposed when growth decelerates.

10.7 Three-Statement Combined Anomaly Scan

Anomaly #1: "Other Investments" in Investing Cash Flow

Quarter	Other Investing Activities ($M)	Description
Q1 FY26	-$6M	Normal
Q2 FY26	-$2M	Normal
Q3 FY26	+$43M	Unknown

Q3 FY26 saw a $43M inflow from "Other Investing Activities" – nature unknown. This could be proceeds from investment recovery or asset sales, but it was not detailed in the 6-K. Such opaque cash flows are harder to track under ARM's FPI (Foreign Private Issuer) disclosure framework – US public companies (SEC-mandated) explain each investing activity in detail in their 10-Q/10-K, but ARM's 6-K (UK standard) allows for greater aggregation flexibility.

Anomaly #1B: Hidden Growth in Capitalized Expenses

ARM's "Capitalized Development Costs" continue to grow on the balance sheet:

Period	Capitalized Development Costs ($M)	% of R&D	YoY
FY2023	~$120M	~11%	—
FY2024	~$180M	~9%	+50%
FY2025	~$280M	~14%	+56%
Q3 FY2026 (Est.)	~$350M	~13%	+25%

ARM capitalizes a portion of its R&D expenditures (in compliance with IFRS standards, IAS 38 – development costs should be capitalized when criteria are met). However, the continuous rise in the capitalization rate warrants attention:

• Optimistic Interpretation: More R&D projects entering a "clearly commercializable" phase (e.g., Phoenix, CSS) → Capitalization is reasonable.
• Cautious Interpretation: Management is reducing current period expenses/improving profit margins by increasing the capitalization ratio → Earnings quality issue.
• Comparison: Synopsys's capitalized development costs account for <5% of R&D, Cadence <3% → ARM's 13-14% is unusually high.

If the entire $280M capitalized in FY2025 were expensed:

• Adjusted R&D: $2,009M + $280M = $2,289M (R&D/Revenue increases from 50%→57%)
• Adjusted GAAP OPM: From 20.6%→13.6% (further compressed)

This implies that ARM's reported profit margins could be 7 percentage points higher than in a "fully expensed" scenario – representing the second layer of earnings quality distortion, beyond SBC.

Anomaly #2: The Dual Nature of Deferred Revenue

Quarter	Current Deferred Revenue ($M)	Non-Current Deferred Revenue ($M)	Total	YoY Change
Q3 FY25	176	750	926	—
Q3 FY26	331	720	1,051	+$125M

Deferred revenue growth of $125M (+13.5%) is lower than total revenue growth (+26%). This means ARM is "consuming" previously accumulated deferred revenue – the pace of upfront payments for new licensing agreements is not keeping up with the pace of revenue recognition. This could be a positive signal (accelerated contract execution) or a neutral signal (contract structure shifting from upfront payments to installments).

Anomaly #3: Rapid Expansion of PP&E

Period	PP&E ($M)	YoY Change	% of Total Assets
FY2024 Q4	420	—	5.3%
FY2025 Q4	714	+70%	8.0%
FY2026 Q3	1,189	+67% (9 months)	11.7%

PP&E increased from $420M (FY2024) to $1,189M (Q3 FY2026) – a 2.8x increase. This is the most direct evidence that ARM's "asset-light" positioning is changing. The proportion of PP&E to total assets rose from 5.3% to 11.7%, moving closer to "chip design companies" (AMD ~8%, QCOM ~15%).

PP&E Growth Breakdown (Estimated):

PP&E Component	Estimated Amount ($M)	Growth Driver
R&D Facilities (UK/US/France)	~$350	New construction + Expansion (Cambridge, San Jose, Sophia Antipolis)
IT Infrastructure (AI/ML Development)	~$250	GPU Clusters + EDA Tool Servers + Cloud Infrastructure
Phoenix Tape-out + Design Tools	~$350	TSMC 3nm NRE + Synopsys/Cadence EDA Licenses
Office Equipment / Renovation	~$120	Hiring Expansion (6,400→7,600 employees)
Lease Assets (IFRS 16)	~$120	Right-of-Use Assets (Non-cash)
Total PP&E	~$1,189	—

Key Insight: Phoenix-related PP&E (tape-out + design tools) is estimated to account for ~30% of total PP&E. If the Phoenix project is canceled or significantly scaled back, ~$350M of PP&E would face impairment risk. This is another sign that ARM is transitioning from a "zero impairment risk" (IP company) to a "with impairment risk" (chip design company).

Anomaly #4: The Ampere Goodwill Issue

After ARM acquired Ampere Computing for ~$6.5B in 2024, a significant amount of Goodwill appeared on its balance sheet:

Period	Goodwill ($M)	% of Total Assets
FY2024 Q4 (Pre-acquisition)	~$2,700	34%
FY2025 Q4 (Post-acquisition)	~$5,200 (Est.)	58% (Est.)
Q3 FY2026	~$5,200	51%

The Ampere acquisition contributed approximately $2,500M in new Goodwill. If ARM's market capitalization significantly declines from $136B (e.g., -50% to $68B), management might need to perform an impairment test on Ampere's Goodwill – if Ampere's "recoverable amount" falls below its carrying value, a Goodwill impairment would be triggered, further depressing GAAP earnings. This represents a self-reinforcing downside risk: Stock price decline → Goodwill impairment → Earnings decrease → Further stock price decline. Although Goodwill impairment is a non-cash item and does not affect operating cash flow, its impact on GAAP earnings and investor sentiment cannot be overlooked.

---

Chapter 11: Summary of Core Findings

11.1 Seven Key Findings

Finding 1: ARM's Business Model Is More Fragile Than It Appears [CQ-1, H1]

• 95% gross margin masks SBC (~$870M/year = 100% of GAAP operating profit), deteriorating accounts receivable quality (DSO averaged 147 days), and concentration of related-party transactions (SoftBank 30%+ licensing)
• GAAP operating margin is only 18.7%, far below the 41% Non-GAAP frequently cited by the market
• FY2025 OCF is only 50% of net income—poor earnings quality

Finding 2: 95% of Royalty Growth Comes from Price Hikes, Only 5% from Volume [CQ-1, H2]

• v9+CSS royalty rate tiers are the core engine of current growth
• But this is "self-limiting growth"—the faster prices rise, the stronger the incentive for RISC-V migration
• v10 CSS royalty rates will approach historical ASP share ceilings (7-8%) across multiple end markets

Finding 3: Data Center Growth Is Real But Exaggerated [CQ-2]

• ARM's official announcement of "50% server market" vs. independent analysts' 20-23%
• DC royalties are only "slightly over 10%" of the royalty mix, from a very low base
• CSS's largest customer in DC (ALA) coincidentally does not need CSS → CSS price hike effectiveness may be overestimated
• Base case FY2028 DC royalties $1.33B, bull case $2.25B, significant difference

Finding 4: Phoenix Is SoftBank's Strategy, Not ARM's Optimal Choice [CQ-3, CQ-6]

• Phoenix's optimal path is "moderate success" ($500M-1B) — great success would instead compress ARM's valuation premium
• Cannibalization effect is limited ($25-40M/year @ 200K shipments), but loss of trust could be greater
• PP&E soared from $420M to $1,189M — ARM's "asset-light" positioning is changing

Finding 5: Intel's Moat Collapse Provides a 15-Year Mapping Reference [H2]

• Non-linearity of moat collapse: constant market share ≠ constant moat
• Price hikes are the catalyst: Intel price hikes → AMD Zen gains economic incentive → ARM price hikes → RISC-V gains incentive
• Ecosystem barriers can be circumvented: Apple Rosetta 2 proved translation layers are effective → LLVM/GCC already support RISC-V compilation

Finding 6: ARM Is Not Visa (Network Effects Are an Order of Magnitude Weaker) [H3]

• Visa's network is exclusive (irreplicable), ARM's ecosystem is forkable (RISC-V shares toolchains)
• Visa's P/E (33x) is priced as "irreplaceable infrastructure", ARM's P/E (140x) should not be higher
• If the market re-prices ARM as "infrastructure with alternatives" (P/E 40-60x), downside of 57-71%

Finding 7: Governance Structure Creates Systemic Risk [CQ-3]

• SoftBank 90% ownership + Controlled Company + Foreign Private Issuer = Triple governance discount
• SoftBank pledges ARM shares → margin loan → AI investment → ARM licensing = Self-reinforcing cycle
• Public shareholders hold only 10%, governance influence is near zero

11.2 Quantified Findings Dashboard

Revenue Breakdown Summary Table:

Revenue Source	FY2025 ($M)	FY2027E ($M)	FY2030E ($M)	CAGR	Confidence	Key Risks
Royalties - Mobile	1,008	1,427	2,189	17%	H	Price hike self-limiting
Royalties - IoT	635	693	869	6.5%	M	RISC-V erosion
Royalties - DC	420	1,400	4,200	58%	L	AI Cycle + RISC-V
Royalties - Automotive	512	1,162	2,834	41%	M	Quintauris
Royalties - PC	228	456	900	32%	M	x86 counterattack
Royalties - Network	132	200	336	20%	M	Mature market
Licensing (incl. SB)	1,839	2,500	3,500	14%	M	SoftBank concentration
Licensing (excl. SB)	1,289	1,750	2,800	17%	M	Organic growth stalled?
Phoenix Chips	0	0	1,000	—	L	Tech + Execution + Trust
Total Revenue	4,774	7,838	15,828	27%	—	—
Consensus Total Revenue	4,007	5,910	11,170	23%	—	—

Valuation Implied Assumptions List:

Implied assumptions for current market cap of $136B:

Assumptions	Market Implied Value	Current Assessment	Fragility
FY2030 Revenue	$11.2B+	Likely Overstated (RISC-V not factored in)	High
Terminal GAAP OPM	35-40%	Challenging (SBC % difficult to reduce)	Medium
Terminal Royalty Rate CAGR	+15%/year (5 years)	Self-limiting (Price increases → Substitution)	High
DC Becomes Largest Endpoint	FY2028-2029	Possible (but CSS limited in DC)	Medium
RISC-V DC Share FY2030	<5%	Likely Underestimated (Tenstorrent+Meta)	High
SoftBank Maintains Control	Yes	Uncertain (Debt Pressure)	High
P/E Stays 100x+	5 years	Highly Unrealistic (Historically Rare)	Extremely High

Contradiction Identification (Requires Further Resolution):

1. Market pricing ARM as "irreplaceable infrastructure" (P/E 140x) vs. ARM having higher substitutability in DC/IoT/PC (Irreplaceability 5.4/10)
2. Royalty growth driven by price increases (v9/CSS) vs. a self-limiting cycle where price increases accelerate substitution (RISC-V)
3. ARM as a "neutral IP vendor" vs. direct competition from Phoenix chip design
4. SoftBank related-party transactions supporting growth vs. liquidity risk created by SoftBank's leverage

Chapter 12: A-Score v2.0 Moat Rating — 10 Dimensions × Detailed Justification

12.1 A-Score Rating Framework Overview

A-Score v2.0 assesses a company's barrier strength (0-10 points per item) across 10 moat dimensions, calculating a weighted total score. As a pure IP licensor, ARM's moat characteristics are distinctly different from manufacturing semiconductor companies—ARM has no fabs, no products, no end customers. Its moat is entirely built upon the instruction set ecosystem and the "irreplaceability" of its design IP.

Special Considerations for A-Score:

• ARM's moat is not "economies of scale" or "brand loyalty" in the traditional sense, but rather ecosystem lock-in—a positive feedback loop formed by billions of lines of code, millions of developers, and thousands of vendors.
• The advantage of this moat lies in its "depth" (extremely high migration costs), while its disadvantage is that when facing structural open-source alternatives (RISC-V), the moat may be "circumvented" rather than "breached."
• A-Score must differentiate between "current moat width" and "moat evolution direction"—ARM's current score may be high, but the trend points towards narrowing.

12.2 A-Score Detailed Item-by-Item Rating

A1: Brand/Standard Lock-in — 9/10

Reason for Rating: The ARM Instruction Set Architecture (ISA) is the global de facto standard for the most widely used processor architecture.

Quantitative Evidence:

• Smartphone SoCs: >99% use ARM architecture (Apple A-series, Qualcomm Snapdragon, MediaTek Dimensity)
• Embedded/IoT: >70% market share (ARM Cortex-M series dominant)
• Cumulative Shipments: >280 billion ARM chips (as of FY2025)
• New Chips/Year: >28 billion (FY2025 shipment volume)
• Licensees: 500+ companies hold ARM licenses

Why 9 instead of 10: The ARM instruction set is no longer the sole option. RISC-V, as an open-source ISA, has already achieved ~25% penetration in the IoT sector (measured by core shipments). While "ARM Compatible" remains a product selling point, in the IoT/embedded domain, "RISC-V Based" is also starting to gain market acceptance. If RISC-V achieves substantial adoption in the data center and mobile sectors during 2028-2030, this rating could drop to 7-8.

Barrier Type: N (Network Effect) — ARM's standard status is not maintained by brand advertising, but by the installed base of billions of devices and the ecosystem built around it.

A2: Intellectual Property — 8/10

Reason for Rating: ARM possesses 30+ years of CPU microarchitecture design accumulation and a vast patent portfolio.

Quantitative Evidence:

• Patents: ~6000+ active patents (covering CPU microarchitecture, interconnect, security, GPU)
• Architecture Generations: v1 (1985)→v9 (2021)→CSS (2023)→v10 (estimated 2027-2028), continuous innovation
• R&D Investment: FY2025 $2.01B (50.1% of revenue), absolute value is top-tier among IP companies
• Talent: ~5000 engineers out of ~7600 employees; the CPU design team is one of the strongest globally
• Design IP Breadth: Full coverage from Cortex-M (microcontrollers) to Cortex-X/Neoverse (high-performance computing)

Why 8 instead of 9: ARM's IP barrier faces challenges from two directions:

1. Open-source Alternatives: The RISC-V instruction set itself is free and open-source. ARM's ISA advantage is shifting from "exclusive" to "shared." While ARM's microarchitecture implementations remain proprietary, the barrier at the ISA level is decreasing.
2. Talent Attrition Risk: Jim Keller (formerly of Apple/AMD/Intel/Tesla) leading Tenstorrent demonstrates that top-tier CPU architects can design ARM-level performance using RISC-V (Ascalon-X SPECint ~22/GHz ≈ Neoverse V3). ARM's IP barrier increasingly relies on "team depth" rather than "architectural exclusivity."

Barrier Type: T (Technology) — but transitioning from T (Exclusive Technology) to T (Team Depth).

A3: Switching Costs — 8/10

Reason for Rating: The full-stack switching cost from ARM to RISC-V is substantial but not insurmountable.

Layered Quantification of Switching Costs:

Layer	Switching Cost	Time	Description
ISA Layer (Instruction Set)	Low	1-2 years	RISC-V toolchain is mature (full GCC/LLVM support)
Compiler/OS Layer	Medium	2-3 years	Linux/Android already support RISC-V; Windows does not yet
Application Software Layer	High	3-5 years	Billions of lines of ARM native code require recompilation + retesting
Driver/Firmware Layer	Extremely High	4-7 years	Hardware drivers for millions of device types require rewriting + certification
Certification Layer	Extremely High	5-10 years	Safety certifications like ISO 26262 (automotive), IEC 61508 (industrial) require re-acquisition
Enterprise Inertia	Extremely High	Ongoing	CTOs are unwilling to risk "replacing a proven solution"

Why 8 instead of 7 or 9: Switching costs represent ARM's most crucial short-term moat. Even if RISC-V achieves technical parity, a full-stack migration would require 3-7 years and billions of dollars in investment (for the entire industry). However, switching costs are a depreciating asset—with each successful RISC-V case (e.g., Alibaba Cloud C930 server deployment), the cost for the next migrator decreases (as software/driver/certification resources accumulate).

Analogy: This is similar to the enterprise IT transition from Windows to Linux—almost unimaginable in 2010, yet by 2025, most backend services are running on Linux. ARM's switching cost barrier might follow a similar path: appearing unshakeable for the first 5 years, then rapidly disintegrating after a critical tipping point.

A4: Network Effect — 7/10

Reason for Rating: The ARM ecosystem exhibits a one-sided network effect (developers → software → devices → developers), but it is not a two-sided platform.

Deconstruction of Network Effect Mechanism:

Strengths and Weaknesses of the Positive Feedback Loop:

• Strengths: The loop has been operating for 35 years, with immense inertia. The addition of 28 billion ARM chips annually further deepens ecosystem lock-in.
• Weaknesses: This is a one-sided network effect (more developers → more software → more devices), rather than a two-sided platform effect (e.g., Visa's merchants ↔ cardholders). One-sided network effects are more easily eroded when facing "good enough" alternatives.
• RISC-V's Network Effect: While RISC-V is much smaller in scale, its open-source nature provides a network effect that ARM lacks—the contributor network effect (more enterprise contributions → better public IP → more enterprise adoption). This effect has been proven in the Linux ecosystem.

Why a 7 instead of an 8: If ARM's network effect were two-sided (like an app store), it could be rated a 9. However, ARM's network effect is indirect (through software compatibility) and is gradually being eroded by the parallel ecosystem established by RISC-V. RISC-V International's 4000+ members (vs. ARM's 500+ licensees) indicate that the "network effect around ARM" is no longer the sole ecosystem choice.

A5: Economies of Scale — 6/10

Rating Rationale: ARM's R&D can be amortized over extremely high shipment volumes, but RISC-V's open-source model fundamentally weakens this advantage.

Economies of Scale Calculation:

• ARM FY2025 R&D: $2.01B → Amortized over 28 billion chips/year → R&D Amortization per chip: $0.072
• ARM FY2025 Total OpEx: $2.97B → Amortized over 28 billion chips → Operating Amortization per chip: $0.106
• RISC-V Foundation: No centralized R&D costs, borne by the community + individual enterprises

ARM's Economies of Scale Paradox: ARM's R&D amortized to just $0.07 per chip across 28 billion chips—an extremely strong scale advantage. However, RISC-V charges no ISA licensing fees at all, and its "R&D" is distributed (each company bears its own microarchitecture design costs). For large customers (e.g., Apple/Google/QCOM), the R&D cost of designing their own RISC-V cores might be only slightly higher or on par with ARM licensing fees—ARM's economies of scale advantage is greatest for small customers (MCU vendors) and smallest for large customers (hyperscale/SoC giants).

Why a 6 instead of a 7: In traditional comparisons (ARM vs. x86 from Intel/AMD), ARM's economies of scale were overwhelming (no need to build factories). However, in the ARM vs. RISC-V comparison, ARM's scale advantage has significantly diminished—because RISC-V's "R&D" is community-driven, with marginal costs of zero.

A6: Barriers to Entry — 5/10

Rating Rationale: 30 years of architectural accumulation constitutes a huge barrier, but RISC-V has proven that "starting from scratch" is feasible.

Two Sides of Barriers to Entry:

• For new entrants: Creating a commercial ISA to compete with ARM (as MIPS once attempted) is nearly impossible—requiring decades of ecosystem building and billions of dollars in investment.
• For RISC-V: RISC-V chose a different path—not a commercial ISA, but an open-source standard. This bypasses the "barrier to entry" issue. You don't need to "enter" ARM's market; you just need to build a parallel ecosystem.

Why a 5 instead of a 6: RISC-V's existence proves that barriers to entry can be circumvented. While building a complete ISA ecosystem still requires 5-10 years, this process has already begun (starting with the RISC-V project at Berkeley in 2010) and has made substantial progress (IoT ~25%, Automotive Quintauris, Data Center Tenstorrent).

A7: Pricing Power — 7/10

Rating Rationale: ARM has demonstrated strong pricing power in recent years (v9=2×v8, CSS=2-3×TLA), but RISC-V constitutes a long-term ceiling.

Three Tiers of Pricing Power:

1. Proven Pricing Power (FY2023-2025): The successful doubling of royalties from v8 to v9 was executed, with no large-scale customer attrition despite complaints. Royalties per chip increased from ~$0.04-0.05 to ~$0.07-0.09. Proof: ARM possesses short-term pricing power under current ecosystem lock-in.

2. Pricing Power Under Test (FY2026-2028): CSS further pushes royalties to 2-3 times that of TLA. However, CSS also means ARM has entered the system design domain, creating a competitive relationship with some licensees (e.g., Qualcomm). This "supplier and competitor" identity conflict will gradually erode pricing power.

3. Potentially Constrained Pricing Power (FY2029+): Can v10 (expected) double again? This depends on the maturity of the RISC-V ecosystem in 2027-2028. If RISC-V achieves scaled deployment in data centers and Android mobile phones, ARM's v10 pricing will face a hard constraint—it cannot be priced above the cost of migrating to RISC-V.

"Elasticity Function" of Pricing Power: For details, refer to the CDS Pricing Power Quantification Model in Chapter 14A. In brief:

• ARM's current pricing power elasticity ε ≈ -0.3 (10% price increase, ~3% drop in shipments, ~7% increase in net revenue)
• After the RISC-V ecosystem catches up, elasticity may become ε ≈ -1.5 (10% price increase, ~15% drop in shipments, ~5% decrease in net revenue)
• Tipping Point: When ε = -1.0, ARM loses pricing power (price increases do not increase revenue)

A8: Customer Lock-in — 7/10

Rating Rationale: Large customers are deeply bound through ALA/TLA, but the largest customers are "buying insurance" with RISC-V.

Customer Lock-in Spectrum Analysis:

Customer	Lock-in Depth	License Type	RISC-V Hedging	Switching Risk
Apple	Extremely High (10/10)	ALA Perpetual	Extremely Low (Self-developed but ARM-based)	Extremely Low
Samsung	High (8/10)	TLA+CSS	Medium (Exploring RISC-V IoT)	Low
MediaTek	High (7/10)	TLA	Medium (Under Evaluation)	Medium-Low
Qualcomm	Medium (6/10)	TLA	High (Ventana)	Medium
Google	Medium (5/10)	TLA	High (Android GKI)	Medium-High
Alibaba/T-Head	Low (3/10)	TLA	Extremely High (C930 Deployment)	High
Tenstorrent	Zero	None	N/A (Pure RISC-V)	N/A

Key Insight: Among ARM's top 5 largest customers, two (Qualcomm and Google) are actively investing in RISC-V. This is not because they plan to leave ARM (not in the short term), but because they are "buying bargaining chips"—with a credible RISC-V alternative, these customers' negotiating power in ARM renewal/upgrade talks will be significantly enhanced.

Apple Exception: Apple is ARM's most deeply locked-in customer in the ARM ecosystem (ALA perpetual license), but Apple is also the customer ARM relies on the least (Apple's ALA means that royalties collected by ARM are extremely low relative to Apple's chip ASP). Apple is unlikely to migrate to RISC-V (having invested billions of dollars in custom ARM cores), but it is also unlikely to be a growth engine for ARM's price increases.

A9: Ecosystem — 8/10

Reason for Rating: The ARM Total Design ecosystem is currently the most complete chip design ecosystem.

Ecosystem Quantitative Comparison:

Dimension	ARM Ecosystem	RISC-V Ecosystem	ARM Advantage
ISA Maturity	35 years (v1-v9)	16 years (2010-)	+19 years
Licensees/Members	500+	4000+	-3500
Full EDA Support	Full (S/C/M)	Full (S/C/M)	Tie
Operating Systems	Linux/Android/iOS/Windows	Linux/Android (in progress)	ARM Leads
Number of Safety Certifications	>1000	<50	ARM Significantly Leads
Number of Commercial IP Cores	>200 Cortex cores	~50+ commercial cores	ARM Leads
Reference Designs (SoC)	ARM Total Design complete reference	Fragmented	ARM Significantly Leads

Defensive Value of ARM Total Design: ARM's Total Design ecosystem alliance, launched in 2023, integrates EDA (Synopsys/Cadence), foundries (TSMC/Samsung), and IP providers (Imagination/Rambus) into a "one-stop SoC design platform." This lowers the barrier to designing chips using ARM IP while increasing ARM's "ecosystem lock-in" – if your entire design flow is within the ARM Total Design ecosystem, switching to RISC-V means abandoning a complete set of validated reference designs.

RISC-V Ecosystem Catch-up: Although the number of members (4000+) far exceeds ARM (500+), the "depth" of the RISC-V ecosystem is far less than ARM's. Key gaps are in safety certifications (ISO 26262/IEC 61508) and enterprise-grade operating system (Windows) support. It is estimated that it will take 3-5 years (until 2029-2030) to catch up with ARM in most dimensions.

A10: Sustainability — 5/10

Reason for Rating: The direction of the moat is uncertain, and sustainability over 5-10 years is questionable.

Three Sustainability Scenarios:

Scenario	Probability	Moat Trend	A10 Score
Optimistic: RISC-V only has substantial penetration in IoT/Embedded	25%	Stable (7/10)	7
Baseline: RISC-V penetrates IoT/DC/Automotive	50%	Slowly Narrowing (5/10)	5
Pessimistic: RISC-V fully penetrates Mobile/DC/Automotive	25%	Rapidly Narrowing (3/10)	3

Probability-Weighted A10: 25%×7 + 50%×5 + 25%×3 = 5.0

Why Sustainability Received One of the Lowest Scores: ARM's moat itself is wide (A-Score 7.17), but its direction is uncertain. RISC-V is not a threat that will "disappear" – it is an open-source movement supported by 4000+ global enterprises/institutions, driven by a triple force of geopolitics (China's independent control) + economic rationality (free ISA) + technological advancement (Tenstorrent performance parity). Even if RISC-V's development is 2-3 years slower than expected, its direction will not change.

12.3 A-Score Weighted Calculation

A-Score = 7.17/10 — Moat "Wide but in Transition"

12.4 A-Score Industry Benchmarking and Implications

Company	A-Score	Main Moat Type	Trend	P/E(TTM)	Valuation/Moat
ASML	~9.0	T (Technological Monopoly)	Stable/Expanding	~38x	4.2x
TSMC	~8.5	T+S (Technology + Scale)	Stable	~25x	2.9x
NVIDIA	~8.0	N+T (CUDA+GPU)	Expanding	~45x	5.6x
ARM	7.17	N (Network)	Declining	142x	19.8x
Synopsys	~7.5	T+N (EDA)	Stable	~55x	7.3x
Cadence	~7.5	T+N (EDA)	Stable	~76x	10.1x
Qualcomm	~6.5	T+I (Technology + Institutional)	Declining	~33x	5.1x
AMD	~5.5	T (Technology)	Expanding	~25x	4.5x
Intel	~4.7	T+S (Technology + Scale)	Significantly Declining	NM (Loss-making)	N/A

Key Finding: ARM's Valuation/Moat Ratio (P/E ÷ A-Score = 19.8x) is the highest in the entire semiconductor industry. This means the market is paying a significantly higher price for each unit of ARM's moat compared to ASML (4.2x), TSMC (2.9x), and even NVIDIA (5.6x).

Interpretation: This indicates that the market is not paying for the "width" of ARM's moat (in which case ASML would be the choice), but rather for ARM's growth rate + expected moat expansion. The market implicitly assumes that ARM's moat will expand from 7.17 (through CSS/DC penetration) rather than narrow (due to RISC-V erosion). If this assumption is incorrect—i.e., if the moat stabilizes or narrows—the 19.8x Valuation/Moat Ratio would rapidly compress.

12.5 A-Score Sensitivity: RISC-V Penetration Scenarios

How will ARM's A-Score change if RISC-V reaches different penetration levels across different timelines?

Time Point	RISC-V Penetration Rate (IoT/DC/Automotive/Mobile)	A-Score Impact	Projected A-Score
2026 (Current)	25%/3%/5%/0%	Baseline	7.17
2028 (Baseline)	35%/10%/15%/2%	A4-0.5, A5-0.5, A7-0.5	~6.4
2030 (Pessimistic)	45%/20%/25%/10%	A1-1, A3-1, A4-1, A5-1, A6-1, A7-1, A10-1	~5.2
2030 (Optimistic)	30%/5%/10%/1%	A4-0.3	~6.9

Implication: Even in the "baseline" scenario (where RISC-V develops at its current pace), ARM's A-Score is projected to decline from 7.17 to approximately 6.4 by 2028. This is roughly Qualcomm's current moat level. If ARM's valuation still maintains a 142x P/E at that time, it would imply the market is paying 22.2x P/E/A-Score for a 6.4-point moat—an unprecedented level historically.

12.6 Synergy and Conflict Among A-Score Moat Dimensions

The 10 dimensions of the A-Score are not independent—there are complex synergistic (positive feedback) and conflicting (negative feedback) relationships among them:

Moat Synergy Matrix (Positive Feedback):

Dimension A	Dimension B	Synergy Mechanism	Strength
A1 (Standard)	A3 (Switching Costs)	Standard lock-in → Increased switching costs	Strong
A3 (Switching Costs)	A4 (Network Effects)	High switching costs → User retention → Sustained network effects	Strong
A4 (Network Effects)	A9 (Ecosystem)	More developers → Stronger ecosystem → Stronger network effects	Strong
A2 (IP)	A7 (Pricing Power)	IP leadership → Basis for price increases	Medium
A9 (Ecosystem)	A8 (Customer Lock-in)	Comprehensive ecosystem → Increased customer dependence	Medium

Moat Conflicts (Negative Feedback/Self-Undermining):

Dimension A	Action	Negative Impact on Dimension B	Mechanism
A7 (Pricing Power)	v9/CSS price increase	A8 (Customer Lock-in) ↓	Excessive price increases → Customers seek alternatives → Lock-in weakens
A7 (Pricing Power)	v9/CSS price increase	A10 (Sustainability) ↓	Price increases → Accelerate RISC-V investment → Long-term moat erosion
A9 (Ecosystem/Phoenix)	In-house chip development	A8 (Customer Lock-in) ↓	Competing with customers → Reduced customer trust → Lock-in weakens
A6 (Entry Barriers)	Bypassed by RISC-V	A1 (Standard) ↓	Establishment of parallel ecosystem → ARM's standard status relativized

Key Finding: ARM's moat has a "self-erosion loop":

Key characteristics of this self-eroding cycle:

• Short-term: Price Hikes → Revenue Growth (Positive) → Stock Price Increase
• Long-term: Price Hikes → RISC-V Acceleration → Pricing Power Decline → Forced Price Hikes (to compensate for volume loss) → More RISC-V Investments...
• Cycle Speed: Each "Price Hike - RISC-V Investment" cycle is approximately 2-3 years (corresponding to ISA generations)
• Interruptibility: If ARM stops price hikes (does not release v10), the cycle slows down, but growth also stagnates → valuation will still compress

Investment Implications: ARM faces an impossible trinity: (1) high growth + (2) stable pricing power + (3) un-eroded moat → at most, only two can be achieved simultaneously. Current market prices imply that all three are simultaneously valid.

12.7 Cross-Analysis of A-Score and SGI (Specialist Premium Index)

SGI = 9.35 — ARM exhibits extremely strong specialist characteristics

The SGI (Specialist Genius Index) measures a company's "focus" and "irreplaceability in its area of expertise." ARM's SGI = 9.35 signifies an extremely high specialist premium in the CPU IP licensing domain—no other company globally can offer a CPU IP portfolio of equivalent depth and breadth.

A-Score × SGI Matrix:

Quadrant	Characteristics	Representative Companies	Valuation Implications
High A-Score + High SGI	Broad and Deep Moat	ASML, TSMC	Highest Valuation Justification
High A-Score + Low SGI	Broad but Shallow Moat	Google, Microsoft	High Valuation but Competitive Replacement Risk
Low A-Score + High SGI	Narrow but Deep Moat	Niche Industrial Companies	Stable but Low Growth
Low A-Score + Low SGI	No Apparent Moat	Commoditized Businesses	Low Valuation

ARM's Position: A-Score 7.17 (High) × SGI 9.35 (Extremely High) → Located in the first quadrant (Broad and Deep). This should ideally support a high valuation.

However, the trend is key: ARM's A-Score is declining (7.17 → forecast 6.4), and its SGI may also be declining (as RISC-V's open-source cores gradually approach ARM's design quality, and ARM's entry into Phoenix dilutes its "pure specialist" characteristic in IP licensing).

If ARM transforms into A-Score 6.4 × SGI 8.0 in 3 years: Still in the first quadrant but moving towards the fourth quadrant → valuation support significantly weakens.

---

Chapter 13: Moat Migration Analysis — T→N Transformation Roadmap and Vulnerability Window

13.1 Moat Migration Detection (EVO-INTC-001 Methodology)

Moat migration analysis (derived from INTC report's EVO-INTC-001) detects historical changes and future trends in a company's barrier types. ARM's migration trajectory is particularly worth studying—it is undergoing a barrier type transformation that may determine its long-term fate.

ARM Barrier Type Evolution Timeline:

13.2 Barrier Migration Matrix: 10 Dimensions × 3 Periods

Dimension	2010 Barrier Type	2020 Barrier Type	2026 Barrier Type	Migration Direction
A1 (Standard)	T (Technical Standard)	N (De Facto Standard)	N (Inertial Standard)	T→N
A2 (IP)	T (Exclusive)	T (Leading)	T (Team)	T→T'
A3 (Switching)	T (Toolchain)	N (Ecosystem Lock-in)	N (Ecosystem + Certification)	T→N
A4 (Network)	—	N (Weak)	N (Medium)	—→N
A5 (Scale)	S (R&D Amortization)	S (R&D Amortization)	S (Diminishing)	S→S'
A6 (Entry)	T (Difficult to Replicate)	T (Still Difficult)	T (Already Bypassed)	T→↓
A7 (Pricing)	T (Technological Monopoly)	N+T (Ecosystem + Technology)	N (Ecosystem Dependence)	T→N
A8 (Lock-in)	T (Tools)	N (Design Flow)	N (but Customers Hedging)	T→N↓
A9 (Ecosystem)	T (Design Resources)	N (Total Design)	N (Mature but Being Challenged)	T→N
A10 (Durability)	T (10-year Lead)	T (5-year Lead)	? (3-year Lead?)	T→?

Migration Detection Result: Across 10 dimensions, the primary barrier type in at least 6 dimensions (A1/A3/A4/A7/A8/A9) has migrated from Technology (T) to Network (N). This signifies that ARM's moat has completed a fundamental transformation **from "unreplicable technology" to "irreplaceable ecosystem"**.

13.3 Quantitative Assessment of Migration Speed and Completeness

Indicator	Value	Meaning	Benchmark (Intel)
Migration Rate	6/10 Dimension Type Change	60% — High-speed Migration	Intel also 6/10
Migration Completeness	~70%	Old barrier (T) weakened but not vanished, new barrier (N) established but not fully complete	Intel ~30% (Stuck mid-migration)
Migration Direction	T→N	From "irreproducible technology" to "irreplaceable ecosystem"	Intel is T→? (No clear direction)
Migration Health	Moderately Favorable	N barrier is more durable than T barrier (Ecosystem is harder to replicate than technology)	Intel is "Poor" (Migration failed)

ARM vs Intel Migration Comparison: Both companies experienced moat migration, but with vastly different paths and outcomes:

• Intel (2005-2020): Attempted to migrate from T (process leadership) to T' (new process), but the migration failed (10nm delayed by 7 years). The old barrier (T) collapsed, and a new barrier was not established → A-Score dropped from ~9 to ~4.7. Took 15 years.
• ARM (2015-2026): Is currently migrating from T (ISA technological exclusivity) to N (ecosystem network lock-in). The migration direction is correct (N is more durable than T), and the new barrier (N) has been partially established. However, the open-source nature of RISC-V means that the N barrier also faces erosion risks.

Key Difference: Intel's migration failure was due to its goal (new process) being thwarted by insufficient execution. ARM's migration risk is different—ARM's execution is not an issue (v9/CSS/Total Design are all executed very well), the risk is that the goal itself (ecosystem lock-in) might be circumvented by RISC-V's open-source model.

Broader Historical Analogy Spectrum: Moat migration is not limited to semiconductors—the following cases provide references for different migration outcomes:

Company	Old Barrier→New Barrier	Migration Outcome	Time Span	Implications for ARM
Microsoft (Office→365)	T (Technology)→N (Subscription Lock-in)	Success	2011-2018	Success Story: Proactive migration to a stronger barrier
Nokia (Hardware→Platform)	T (Manufacturing)→N (Ecosystem)	Failure	2007-2013	Warning: Migration window is short, irreversible once missed
Adobe (Desktop→Cloud)	T (Functionality)→N (Subscription+Data)	Success	2013-2017	Success Story: But requires enduring short-term pain
Qualcomm (Patents→Platform)	I (Institutional/Patents)→T (Integration)	Partial Success	2015-Present	Reference: But faces antitrust + substitution pressure
Kodak (Chemical→Digital)	T (Chemistry)→T' (Digital)	Failure	1990-2012	Ultimate Warning: Possessed technology but dared not disrupt itself
ARM (ISA→Ecosystem)	T (ISA Exclusivity)→N (Ecosystem Lock-in)	In Progress	2015-Present	Correct direction but faces open-source circumvention risk

ARM's migration most closely resembles Microsoft's successful path: Proactively migrating from an old barrier (ISA technological leadership) to a new barrier (ecosystem subscription lock-in/CSS). CSS is essentially ARM's version of "Office 365"—shifting customers from one-time licenses (TLA = outright purchase of Office) to continuous subscriptions (CSS = Office 365 annual fee). If ARM successfully executes this transition like Microsoft, its moat could become stronger rather than weaker through the T→N migration.

However, ARM faces threats that Microsoft did not: When Microsoft migrated from Office→365, there was no "open-source Office" that could achieve parity in functionality and compatibility (LibreOffice is still far inferior to Microsoft Office). In contrast, as ARM migrates from ISA→Ecosystem, it faces a rapidly catching-up open-source alternative (RISC-V) with national-level support. This is the core difference in ARM's migration risk.

13.4 Vulnerable Window: 2026-2028 Analysis

ARM is currently in an "intermediate state" of moat migration—the old barrier (technological leadership) has been eroded by RISC-V's performance parity, and the new barrier (ecosystem lock-in) has not yet fully solidified:

Current State of Old Barrier (Technology, T):

Technological Dimension	ARM Current	RISC-V Best	Gap	Trend
SPECint (Single-core Performance)	Neoverse V3 ~22/GHz	Ascalon-X ~22/GHz	0%	Parity Achieved
Power Efficiency (perf/W)	Cortex-X4/X5 Leading	Catching Up (~80-90%)	10-20%	Rapidly Narrowing
Vector Computing (SVE2)	Mature	RVV 1.0 Mature	Parity	Parity Achieved
AI Inference (SME)	Leading	Custom Extensions Flexible	Minimal	RISC-V More Flexible
Overall Technological Barrier	—	—	~5/10	Declined from 8/10 to 5/10

Current State of New Barrier (Ecosystem, N):

Ecosystem Dimension	ARM Current	RISC-V Current	ARM Lead	Estimated Parity Time
Operating Systems	Linux+Android+iOS+Windows	Linux+Android (In progress)	2 OSs	2028-2029
Development Tools	DS-5/Keil/Mature	Rapidly Improving	Medium	2027-2028
Security Certifications	>1000 items	<50 items	Very Large	2030+
Application Software	Tens of billions of lines	Rapid Growth	Very Large	2030+
Reference Designs	Total Design Complete	Fragmented	Large	2028-2029
Overall Ecosystem Barrier	—	—	~7.5/10	—

Fragile Window Analysis: 2026-2028 represents the most fragile period for ARM's moat. The technological barrier has decreased from 8/10 to 5/10 (RISC-V performance parity has been achieved), while the ecosystem barrier is approximately 7.5/10 but is declining. ARM's strategy within this window (v9 price increase + CSS bundling + Total Design + Phoenix) will determine:

1. If ARM succeeds: Significantly increases customers' full-stack switching costs through CSS + Total Design → Ecosystem barrier strengthens from 7.5 to 8.0+ → Even if RISC-V technology catches up, customer migration costs remain prohibitively high → A-Score stabilizes at 6.5-7.0
2. If ARM fails: v9/CSS price increases trigger customer backlash + Phoenix alienates licensees + RISC-V ecosystem catches up beyond expectations → Ecosystem barrier decreases from 7.5 to 5.0 → A-Score drops to ~5.0-5.5 by 2030

13.5 Moat Migration "Health Metrics" Dashboard

To track the progress of ARM's moat migration, we have established a continuously monitored "health metrics" dashboard:

Metric (KS)	Current Value	Healthy Threshold	Warning Threshold	Data Source	Frequency
KS-M01: CSS Penetration Rate	~30-35% (new signings)	>50%	<20%	ARM Quarterly Report	Quarterly
KS-M02: RISC-V IoT Share	~28%	<30%	>40%	RISC-V Intl	Annually
KS-M03: RISC-V DC Share	~3%	<5%	>10%	Industry Reports	Semi-annually
KS-M04: ARM DC Share	~23%	>25%	<18%	Mercury Research	Quarterly
KS-M05: Royalty/Chip ASP	~$0.065	Rising	2 consecutive quarters of decline	Estimated from ARM Quarterly Reports	Quarterly
KS-M06: Number of New ALA/TLA Signings	—	Stable/Growing	2 consecutive quarters of decline	ARM Quarterly Report	Quarterly
KS-M07: Deferred Revenue	$911M	Stable/Growing	3 consecutive quarters of decline	ARM 20-F	Quarterly
KS-M08: Quintauris Progress	Pre-commercial	No mass-produced RISC-V automotive chips	First mass production	Public Information	Semi-annually
KS-M09: QCOM RISC-V Shipments	Zero	Zero	First shipment announcement	QCOM Quarterly Report	Quarterly
KS-M10: Google RISC-V Phones	Android GKI Ready	Not commercialized	First Chinese RISC-V phone	Public Information	Semi-annually

Warning Trigger Rule: If any 2 of KS-M02/M03/M09/M10 simultaneously hit the warning threshold, ARM's moat migration status will be upgraded from "Fragile Window" to "Migration Crisis" → Re-evaluate A-Score and valuation.

Current Overall Status: 🟡 Fragile Window (0 out of 10 metrics have hit the warning threshold, but several are moving towards the warning direction)

13.6 Quantitative Impact of Moat Migration on Valuation

Moat Status	A-Score	Reasonable P/E Range	Current P/E	Implied Assumption
Wide and Stable (TSMC/ASML)	8-9	25-40x	—	Sustained Growth + Low Risk
Wide but Migrating (ARM Current)	7.17	30-60x	142x	Requires Extremely High Growth Compensation
Moderate and Declining (QCOM)	6-7	15-30x	33x	Moderate Growth
Narrow and Unstable (INTC)	4-5	8-15x	NM	Under Restructuring

Even under the most optimistic assumption of a "wide" moat (A-Score maintained at 7.17), ARM's 142x P/E requires extremely strong growth expectations to justify. If the moat declines from 7.17 to 5-6 (RISC-V parity scenario), the reasonable P/E could fall to 15-30x — even with sustained revenue growth, valuation could compress significantly (-57% to -79%).

Investment Implications of Moat Migration:

• When A-Score is stable or expanding, a high P/E is sustainable (NVIDIA's CUDA moat expands → 45x P/E is sustainable)
• When A-Score is declining, a high P/E is fragile — any slowdown in growth will trigger a "double whammy" (growth downgrade + valuation compression)
• ARM faces the latter: A-Score declining trend + 142x P/E = Highly fragile valuation combination

13.7 10-Year Moat Decay Probability Model

Combining the A-Score migration path with probability analysis, we establish ARM's moat 10-year trajectory under three scenarios:

Scenario A: Successful Migration (Probability 30%)
ARM successfully establishes new barriers (CSS penetration >60% + Total Design becomes industry standard + RISC-V stalls at <10% in DC).

Year	A-Score	Key Events	P/E Support Range
2026	7.2	v9 price increase completed, CSS 35% penetration	50-70x
2027	7.0	DC ARM reaches 25%, RISC-V stagnates	45-65x
2028	7.0	v10 released, royalties successfully doubled	40-60x
2029	7.3	CSS >50%, ecosystem lock-in solidified	40-55x
2030	7.5	DC ARM reaches 30%, NVIDIA's moat solidifies	35-50x
2033	7.5-8.0	ARM becomes the "Visa of semiconductors"	30-45x

Scenario B: Gradual Erosion (50% Probability)
RISC-V continues to penetrate IoT/China, DC market competition intensifies, ARM's pricing power gradually limited but not collapsing.

Year	A-Score	Key Events	P/E Support Range
2026	7.2	v9 price increase triggers some customer dissatisfaction	45-60x
2027	6.8	QCOM's first RISC-V product mass-produced	35-50x
2028	6.5	v10 price increase below expectations (only 1.2x)	30-45x
2029	6.2	RISC-V DC market share reaches 10%	25-40x
2030	5.8	Google RISC-V phone launched	22-35x
2033	5.0-5.5	ARM remains the largest architecture but "just one option"	18-28x

Scenario C: Accelerated Collapse (20% Probability)
Multiple catalysts erupt simultaneously (China policy + major client defection + SoftBank sell-off), ARM's moat rapidly narrows.

Year	A-Score	Key Events	P/E Support Range
2026	6.8	China mandates RISC-V procurement policy	35-50x
2027	6.0	QCOM+Samsung simultaneously promote RISC-V	25-35x
2028	5.0	Android RISC-V fully ready	18-25x
2029	4.5	Mobile RISC-V penetration >10%	15-22x
2030	4.0	ARM becomes similar to Qualcomm (has a moat, but it's not wide)	12-18x
2033	3.5-4.0	ARM shrinks to embedded + legacy mobile	10-15x

Probability-Weighted A-Score Path: 2026(7.1) → 2028(6.4) → 2030(5.9) → 2033(5.4)

Investment Implications: After probability weighting, ARM's A-Score is projected to drop to ~5.9 by 2030 (a decline of -1.3 from the current 7.17). This corresponds to a reasonable P/E range of approximately 22-38x – significantly lower than the current 142x. Even in the most optimistic "Successful Transition" scenario (30% probability), an A-Score of 7.5 in 2030 can only support a P/E of 35-50x.

Moat Decay Speed vs. Earnings Growth Speed: Can ARM "outrun" its moat's decay? If Net Income (NI) grows from $0.8B to $4.5B (+460%) while the P/E compresses from 142x to 30x (-79%), the market capitalization would still fall from $136B to $135B—effectively treading water. Investors only achieve positive returns if NI growth surpasses the P/E compression. Based on the probability-weighted path, this would require FY2030 NI >$4.0B + P/E sustained >35x → Market Cap $140B+. This implicitly assumes that all of B1-B3 materialize.

---

Chapter 14: RISC-V Special Topic — Challenger or Credit Default Swap?

14.1 Unconventional Framework: RISC-V as ARM's "CDS"

Traditional analysis positions RISC-V as a "long-term substitute threat" to ARM — a zero-sum game where RISC-V wins and ARM loses. However, this binary framework overlooks a more nuanced and crucial dynamic:

Hypothesis H2: RISC-V is ARM's Credit Default Swap (CDS)

In financial markets, the existence of a CDS (Credit Default Swap) does not require a default to occur – merely its presence and pricing alter a bond's risk premium and the issuer's behavior. Similarly:

1. RISC-V doesn't need to replace ARM; it only needs to exist — Its existence as a credible alternative permanently limits ARM's ability to raise prices. Even if RISC-V only ever achieves 10% market share, its constraining effect on ARM's pricing power could be worth billions of dollars.

2. CDS Premium = Total RISC-V Investment: Global annual investment in RISC-V (corporate R&D + government subsidies + VC) is estimated at $5-8B/year. This is the "premium" – the fee customers pay for "insurance" against ARM's monopolistic pricing.

3. CDS Stock → Constrains ARM Behavior: By 2026, cumulative RISC-V investment will exceed $30B. Each additional $1B in investment reduces ARM's pricing freedom by a notch. This is the "CDS Stock Effect."

4. CDS Doesn't Need to Be Exercised: Qualcomm spent $2.4B acquiring Ventana but may never ship RISC-V chips at scale. The value of this $2.4B is not in producing RISC-V chips, but in Qualcomm being able to say 'we have an alternative' the next time it negotiates royalties with ARM.

14.2 RISC-V Penetration Curve: A Market-by-Market Analysis of Seven End Markets

In-Depth Penetration Analysis by Market:

IoT/Embedded (~25-30%, High Penetration)

Current Status: RISC-V is no longer a "threat" but a "reality" in the IoT sector. Chinese companies such as GigaDevice and Espressif have already shipped RISC-V MCUs at scale, squeezing ARM Cortex-M out of the low-end market.

Impact on ARM's Royalties: IoT/Embedded is ARM's "low-price, high-volume" market (royalty per chip $0.01-0.03). Even if RISC-V completely replaces ARM's IoT share, the impact on ARM's revenue would be approximately $100-200M/year (less than 5% of total revenue). However, the signaling significance far outweighs the financial impact: IoT is a "leading indicator" – if RISC-V succeeds in IoT, penetration into the next market (automotive MCUs) will be faster.

Automotive MCUs (~10-15%, Medium Penetration)

Current Status: The Quintauris alliance (Bosch/Infineon/NXP/Qualcomm/STMicro) is driving automotive RISC-V standardization. The first batch of RISC-V automotive chips is expected to enter mass production between 2025-2027.

Impact on ARM's Royalties: Automotive is a high-value growth market for ARM (royalty per chip $0.10-0.30, significantly higher than IoT). 15% RISC-V penetration in automotive would mean ARM loses approximately $150-300M/year in potential revenue (FY2030 estimate).

Key Constraints: Safety certification for automotive chips (ISO 26262, ASIL-D) has extremely high requirements, and RISC-V needs 3-5 years to establish a complete certification infrastructure. This is ARM's strongest barrier in the automotive market.

Data Centers (~2-5%, Low Penetration but Fastest Growing)

Current Status:

• Tenstorrent Ascalon-X: High-performance RISC-V server core, SPECint ~22/GHz ≈ Neoverse V3. Led by Jim Keller, with $370M+ in funding. However, commercial shipments remain very small.
• Alibaba/T-Head C930: China's first large-scale RISC-V server deployment. 512-bit vector unit + 8 TOPS AI matrix engine. Commercial delivery starting March 2025; Alibaba Cloud has already deployed it.
• SOPHGO SG2044: Chinese startup, 64-core RISC-V server chip, targeting the Chinese data center market.

Impact on ARM's Royalties: Data centers are ARM's core engine for royalty growth (royalty per chip $0.50-2.00+). If RISC-V gains a 10-20% share in data centers by 2030, ARM's data center royalty revenue could be $500M-1B/year lower than in a "no RISC-V" scenario. This represents the largest potential impact of the H2 hypothesis (CDS effect).

Mobile Phones (<1%, Pre-Commercial)

Current Status: Google has integrated RISC-V into the Android Generic Kernel Image (GKI). Chinese OEMs are expected to launch the first RISC-V Android phones in 2027-2028. However, high-end mobile SoCs (Snapdragon/Dimensity/Tensor) are expected to remain ARM-dominated before 2030.

Impact on ARM: If RISC-V in mobile phones reaches 5-10% penetration by 2030, ARM will face a direct threat to its most critical market (mobile royalties account for ~60% of total royalties). However, due to the extremely high software compatibility requirements of the mobile ecosystem (millions of apps), RISC-V mobile phone penetration will take longer.

AI Accelerator Co-processors (~5-10%, Emerging Market)

Current Status: RISC-V is rapidly penetrating AI accelerators as a "control plane processor" (non-compute CPU). Meta MTIA v3 and several AI chip startups (e.g., Cerebras/Graphcore) use RISC-V co-processors in their AI chips instead of ARM Cortex-R/M series.

Impact on ARM: RISC-V penetration in the AI accelerator market (estimated $50B+ by 2026) primarily impacts ARM's low-value royalties (Cortex-R/M $0.01-0.03 per chip). The financial impact is limited ($50-100M/year), but the strategic significance is immense: it cultivates a cohort of "ARM-free" chip design teams and toolchain experience, lowering the barrier for these teams to adopt RISC-V in larger future chips (main CPUs).

Infrastructure/Networking (~2-5%, Early Stage)

Current Status: Embedded processors in network switches/routers have traditionally been dominated by MIPS (now in decline) and ARM. RISC-V is beginning to penetrate this understated but stable market, primarily through Chinese network equipment providers (Huawei/ZTE) and some Western startups.

Impact on ARM: Infrastructure represents ARM's "hidden revenue" (royalty per chip $0.05-0.15, stable volume), not a growth engine. 5-10% RISC-V penetration would have a financial impact of approximately $30-50M/year.

14.3 In-Depth Profiles of Key RISC-V Players

Tier 1: Strategic Investors ("CDS Buyers")

Qualcomm + Ventana Micro ($2.4B, Dec 2025):

Qualcomm is one of ARM's largest mobile chip customers (estimated to pay ARM approximately $500-600M in royalties in FY2025). The strategic rationale behind acquiring Ventana is not to "replace ARM"—Qualcomm will not use RISC-V in Snapdragon in the short term. The true value lies in:

1. Negotiating Leverage: The next time v10 royalties are negotiated, Qualcomm can say, "we have our own high-performance RISC-V core."
2. Server Market: Qualcomm can launch RISC-V chips in the data center (non-mobile) market, avoiding ARM's continuous royalty price increases in DC.
3. Long-Term Insurance: If ARM continues to raise prices in the future or if Phoenix competes with Qualcomm, Qualcomm has a "Plan B."
4. Ventana Veyron V2: High-performance RISC-V server core, with performance comparable to Neoverse V2/V3 levels.

Timeline: Qualcomm RISC-V products are expected to first ship in 2027-2028 (server/embedded, non-mobile)
Impact on ARM: Pricing Power Ceiling — ARM's royalty negotiations with Qualcomm will henceforth face a hard constraint.

Google (Android GKI + Tensor + Cloud Chips):

Google's RISC-V strategy is broader than Qualcomm's:

• Android GKI: Full integration of RISC-V into the Android Generic Kernel by 2025. This means future RISC-V phones can directly run Android applications (via compatibility layers or native compilation).
• Tensor Chips: Some components (security co-processors) of Google's self-developed mobile/tablet SoCs already use RISC-V.
• Cloud Chips: Google is evaluating the use of RISC-V in its self-developed server chips (as an alternative to Axion).
• Strategic Intent: Ensuring the Android ecosystem is not locked into ARM instruction sets, maintaining negotiating power with ARM.

Timeline: RISC-V Android phones are expected in 2027-2028 (Chinese OEMs to launch first)
Impact on ARM: Ecosystem Barrier Erosion — Android's shift from "ARM-only" to "ARM+RISC-V" will weaken ARM's A4 (network effect) rating.

Tier 2: Performance Verifiers ("CDS Pricing Validation")

Tenstorrent (Jim Keller + Ascalon-X):

Jim Keller's resume (Apple A-series → AMD Zen → Intel 5-core → Tesla FSD → Tenstorrent) makes him one of the most respected chip architects globally. Ascalon-X's performance (SPECint ~22/GHz) demonstrates RISC-V's viability in high-performance computing.

Funding: $370M+ (2025 Series C at $693M valuation, led by Samsung)
Clients: Undisclosed, but already has HPC client samples (presumed defense/government)
Significance: Tenstorrent does not require massive commercial success – its very existence proves that "ARM-level performance can be achieved on RISC-V," which is a permanent blow to ARM's pricing power.

Alibaba/T-Head (C930 + Alibaba Cloud Deployment):

• Server-grade RISC-V Processor: 128 cores, 512-bit RVV vector unit + 8 TOPS AI matrix engine
• Commercial delivery from March 2025; Alibaba Cloud has begun small-scale RISC-V server cluster deployments
• Motivation: China's indigenous control (to counter US chip sanctions) + Reduce reliance on ARM
• Significance: China's first large-scale RISC-V server deployment, paving the way for RISC-V in Chinese data centers

Tier 3: Ecosystem Builders ("CDS Infrastructure")

Quintauris Automotive Alliance: Bosch/Infineon/NXP/Qualcomm/STMicro jointly promoting automotive RISC-V standards. First chips expected 2025-2027. Directly threatens ARM in the high-value incremental market for automotive MCUs.

RISC-V International: Headquartered in Switzerland (to circumvent unilateral sanctions), 4000+ members, 8 Chinese government agencies promoting adoption. Standardization progress is faster than ARM's early history.

14.4 Quantitative Impact Model of RISC-V on ARM's Pricing Power

Three-Scenario Quantitative Analysis:

Scenario	RISC-V Positioning	ARM Pricing Power	v10 Royalty Rate	FY2030 Royalty Revenue	vs. Baseline
A: No RISC-V	Non-existent	Unconstrained	v9×1.5-2x	$5.5B	Baseline
B: CDS Effect (Baseline)	Exists but no large-scale substitution	Constrained	v9×1.0-1.3x	$3.5-4.5B	-18~36%
C: Substantial Substitution	Multi-market penetration >20%	Severely constrained	v9×0.8-1.0x	$2.5-3.5B	-36~55%

CDS Effect Quantification (Scenario B):

• Baseline Royalty (No RISC-V): $5.5B
• Royalty erosion due to CDS effect: $1.0-2.0B/year
• Sources of erosion: ①v10 unable to double (~$0.5B) + ②CSS pricing discount (~$0.3B) + ③Client negotiation leverage (~$0.2-0.5B) + ④Chinese clients switching to RISC-V (~$0.2-0.5B)
• Current Market Cap Impact: $1.0-2.0B/year erosion × 25-30x market multiple = $25-60B market cap erosion (18-44% of current $136B)

This is the essence of H2 ("RISC-V=CDS"): It's not the risk of ARM disappearing (which would require complete RISC-V substitution, probability <10%), but the risk that ARM cannot raise prices as the market expects (probability >50%). The market's current 142x P/E implies the assumption of continuous ARM price increases (see Ch16 Reverse DCF) – if pricing power is constrained by RISC-V, valuation will be significantly compressed.

14.5 ARM's RISC-V Defense Strategy Assessment Matrix

Strategy	Effectiveness	Durability	Side Effects	ROI Assessment
Forced v9 Upgrade	High (Short-term 2 years)	Low (Accelerates RISC-V)	Client resentment, QCOM litigation	Short-term+/Long-term-
CSS Bundling (Core Strategy)	High (Medium-term 3-5 years)	Medium (Increases switching costs)	Competition with licensees	Optimal Balance
Total Design Ecosystem	Medium (Ongoing)	High (Ecosystem lock-in)	Requires continuous investment	Positive
DreamBig Acquisition (chiplet)	Medium (Differentiation)	Medium	Increases CapEx	Neutral
Phoenix (In-house chip development)	Low (for RISC-V defense)	Low	Alienates licensees → Accelerates RISC-V	Counterproductive
Patent Litigation (Nuclear Option)	High (Deterrent)	Low (Escalates confrontation)	May drive anti-ARM alliances	High risk, high reward

Comprehensive Assessment: ARM's defense strategy mix is effective in the short term (2-3 years) but decreasingly effective in the medium term (5 years). CSS is the optimal balancing strategy (increases client integration + increases switching costs), but Phoenix is a strategic own goal – it transforms ARM from a "neutral IP provider" into "a chip designer competing with clients," driving clients to more actively evaluate RISC-V alternatives.

14.6 RISC-V Penetration Rate Modeling: S-Curve vs. Linear

Historical Analogy: Technology substitution penetration rates typically follow an S-curve (Sigmoid) rather than linear growth. We use the closest historical technology substitution cases to calibrate the possible penetration rate for RISC-V:

Technology Replacement	Time to Reach 10%	Time from 10% to 30%	Time from 30% to 50%	Total Time (0→50%)	Driving Force
x86→ARM (Mobile)	~8 years (2005-2013)	~3 years (2013-2016)	~2 years (2016-2018)	~13 years	iPhone/Android
ARM→x86 (Server)	~4 years (2020-2024)	In Progress	—	Estimated 10-12 years	Cloud Computing CapEx
MIPS→ARM (Embedded)	~10 years (1995-2005)	~5 years (2005-2010)	~3 years (2010-2013)	~18 years	Cost + Ecosystem
x86→ARM (PC)	~5 years (2020-2025)	—	—	Estimated 15+ years	Apple M-series
RISC-V→ARM (IoT)	~5 years (2018-2023)	In Progress (2023-?)	—	Estimated 12-15 years	Open Source + China Policy
RISC-V→ARM (DC)	~6 years (2022-2028E)	~5 years (2028-2033E)	—	Estimated 15+ years	Cost + Geopolitics

Special Factors for RISC-V Penetration Rate:

1. Accelerating Factors: China national policy push (similar to 4G/5G deployment speed) + Enterprise "CDS insurance" procurement (Qualcomm/Google) + Open-source community speed (faster iteration than commercial IP)
2. Decelerating Factors: Software ecosystem depth inferior to ARM + Security certification gap (automotive/industrial) + Large customer migration inertia + Lack of Windows support

Calibrated RISC-V Penetration Forecast (weighted by end markets):

Key Findings:

• Base Case: RISC-V reaches the pricing power threshold (weighted penetration rate ~22%) around 2033.
• Bear Case (for ARM): Threshold reached as early as 2031 → ARM loses pricing power after 2031.
• Bull Case (for ARM): Threshold not reached until after 2035 → ARM has a longer high-growth window.

Investment Implications: In any scenario, ARM has at least a 5-year (2026-2031) "high pricing power window." The question is whether the market will preemptively price in the long-term threat of RISC-V—if so, ARM's P/E might start to compress before the actual loss of pricing power. This is a "time discount" issue in valuation.

14.7 RISC-V Penetration Threshold Analysis

Question: What penetration level does RISC-V need to reach to transform from a "fringe threat" to a "systemic risk"?

S-Curve Model: Technology penetration typically follows an S-curve—the first 10% is slow (requires infrastructure buildout), 10-50% accelerates (infrastructure ready + network effects), and 50%+ decelerates (residual lock-in).

Tipping Point	RISC-V Penetration	Event Trigger	Impact on ARM	Estimated Timing
IoT Tipping Point	IoT >30%	Achieved	Royalty impact <5%, significant signal	2025 (Passed)
Automotive Tipping Point	Automotive >15%	Quintauris Mass Production	High-value market penetration begins	2027-2028
Data Center Tipping Point	DC >10%	Hyperscale Customer RISC-V Adoption	ARM growth engine threatened	2028-2030
Mobile Tipping Point	Mobile >5%	Chinese OEM RISC-V Handsets	ARM core market shaken	2029-2031
Systemic Tipping Point	Weighted >20%	Multi-market simultaneous penetration	Fundamental shift in ARM's pricing power	2030+

Definition of Systemic Tipping Point: When RISC-V simultaneously reaches >15% penetration in more than three end markets, the RISC-V software ecosystem will transition from "fragmented" to "self-sustaining"—no longer dependent on the promotion of individual enterprises, but driven by the ecosystem's own positive feedback loop. This represents the greatest long-term risk for ARM.

14.8 RISC-V Investment Landscape: Breakdown of $30B+ Cumulative Investment

As of early 2026, the global cumulative investment in RISC-V is estimated to exceed $30B:

Category	Estimated Investment	Key Players	Intent
Corporate Internal R&D	~$10-15B	Google/QCOM/Samsung/Huawei/Alibaba	Strategic Hedging + Self-reliant Control
Startup Funding (VC/PE)	~$5-8B	Tenstorrent/SiFive/Esperanto/StarFive	Commercializing RISC-V Cores
Chinese Government Subsidies	~$3-5B	National Integrated Circuit Fund/Local Governments	Self-reliant Control + De-ARM Reliance
Academic/Open-Source Community	~$1-2B	Berkeley/ETH/Various Universities	Basic Research + Standardization
Corporate Acquisitions	~$5-7B	QCOM ($2.4B Ventana)/Intel (SiFive Investment)	Strategic Layout
Total	~$25-37B	—	—

Investment Growth Rate: RISC-V's annual new investment grew from ~$2B in 2020 to ~$7-9B in 2025, a CAGR of approximately 35%. Even if this growth rate halves (to 15-20%), cumulative investment will exceed $80B by 2030.

CDS Premium Analogy Extension: If RISC-V investment is viewed as a "CDS premium," then the cumulative premium of $30B+ signifies that the global semiconductor industry has purchased substantial insurance against an "ARM pricing power default." This is not a decision by a single company, but a collective industry action—the fact that the industry collectively spent $30B to ensure ARM cannot monopolize pricing is the strongest evidence that ARM's pricing power faces long-term constraints.

14.9 The China Dimension of RISC-V: A Geopolitical Catalyst

China is the fastest-penetrating market for RISC-V, and also an "amplifier" of RISC-V's threat to ARM:

China RISC-V Policy Timeline:

• 2018: Chinese Academy of Sciences releases RISC-V roadmap
• 2019: China RISC-V Alliance established (currently 200+ members)
• 2020: 8 Chinese government agencies (including MIIT/MOST) jointly issue documents supporting RISC-V
• 2022: RISC-V integrated into the National 14th Five-Year Plan for Science and Technology
• 2023: T-Head C930 server core released, Alibaba Cloud begins deployment
• 2025: China's annual RISC-V chip shipments are projected to exceed 10 billion units (primarily IoT MCUs)

Unique Drivers for RISC-V in China:

1. Policy-Driven: Unlike free markets where RISC-V is primarily driven by economic rationale, RISC-V adoption in China is also propelled by policy forces (to reduce reliance on ARM/US technology). This may lead to a faster penetration rate for RISC-V in China compared to other global markets.
2. Huawei Effect: Huawei HiSilicon (a former major ARM client) shifted to in-house R&D due to U.S. sanctions, accelerating the "vigilance" of the entire Chinese semiconductor industry towards ARM. Even if other Chinese companies were not directly sanctioned, the fear that "what happened to Huawei could happen to us" has driven the preemptive adoption of RISC-V.
3. Talent Inflow: The "returnee wave" in China's semiconductor industry (engineers returning to China from Western companies like ARM/Intel/AMD) has provided a talent base for Chinese RISC-V design.

Quantitative Impact on ARM: China currently accounts for ~20-22% ($800-900M) of ARM's total revenue. If RISC-V penetration in China is 2x faster than globally:

• China's weighted RISC-V penetration rate could reach ~20% by 2030 (vs. ~12% globally)
• Corresponding Arm China revenue loss: $150-250M/year
• Impact on ARM's total market capitalization: -$4-8B (calculated at 30x P/E)

However, the China dimension also has positives: China's "demand" for ARM is structural—Chinese chip design companies still require ARM's high-end cores (Cortex-X/Neoverse) to design competitive products, and RISC-V cannot yet fully substitute in the high-end market. Arm China's revenue is more likely to experience "slower growth" rather than a "sharp decline."

---

Chapter 15: RISC-V CDS Pricing Power Quantification — Modeling ARM Royalty Elasticity Function

15.1 Pricing Power Elasticity Function Framework

Candidate Insight (CI): ARM's pricing power is not a fixed binary state (present/absent), but rather a continuous function of the RISC-V ecosystem's maturity. This chapter establishes a "pricing elasticity function" model that can be applied to any IP company facing open-source alternatives.

Core Formula:

15.2 Parameter Estimation and Calibration

ε₀ (Baseline Elasticity) Estimation:

ARM's pricing behavior before the rise of RISC-V (2015-2020) provides a natural experiment for ε₀:

• ARM increased prices by approximately 50% from v7→v8 (but primarily through architectural upgrades rather than same-version price increases)
• Customers had virtually no alternative options (MIPS declining, x86 uncompetitive in low-power applications)
• Estimated ε₀ ≈ -0.15 ~ -0.25 (approximates inelasticity, price increases do not significantly affect shipment volume)

Weighted Calculation of R(t) (RISC-V Penetration Rate):

Each market is weighted by its ARM royalty contribution:

End Market	ARM Royalty Weight	2026 RISC-V	2028 Forecast	2030 Forecast
Mobile	55%	0.5%	2%	5%
Data Center	15%	3%	8%	15%
IoT/Embedded	10%	28%	35%	40%
Automotive	8%	8%	15%	20%
Consumer Electronics	7%	5%	10%	15%
Infrastructure	5%	2%	5%	10%

Weighted R(t):

• R(2026) = 55%×0.5% + 15%×3% + 10%×28% + 8%×8% + 7%×5% + 5%×2% = 4.6%
• R(2028) = 55%×2% + 15%×8% + 10%×35% + 8%×15% + 7%×10% + 5%×5% = 8.1%
• R(2030) = 55%×5% + 15%×15% + 10%×40% + 8%×20% + 7%×15% + 5%×10% = 12.4%

15.3 Elasticity Function Calculation Results

Elasticity Interpretation:

Year	Elasticity ε	Meaning	Net Effect of 10% ARM Price Increase
2026	-0.158	Approximately Inelastic	Revenue +8.4% (Net Increase)
2028	-0.126	Increased Elasticity	Revenue +8.7% (Net increase but growth slowing)
2030	-0.087	Further Increased Elasticity	Revenue +9.1% (Loss in volume starts to be felt)
Critical Point	-1.000	Perfectly Elastic	Revenue ±0% (Price increase yields no revenue growth)

Important Finding: Even by 2030, ARM's pricing elasticity remains negative (price increases lead to revenue growth)—this implies that RISC-V is unlikely to fully eliminate ARM's pricing power before 2030. However, elasticity is continuously increasing, meaning the "net incremental gain" from each ARM price increase is diminishing. By the time the weighted penetration rate R(t) reaches 22% (likely between 2033-2035), ARM will lose its pricing power (price increases will not generate additional revenue).

15.4 Impact of Pricing Elasticity Function on Valuation

Translating the Elasticity Function into Valuation Impact:

Assuming ARM increases prices by 15% every two years in the base case scenario (through architectural upgrades):

Price Increase Event	Revenue without RISC-V	Revenue after CDS Effect	Difference	Cumulative Loss
FY2027 v9+	$5.9B	$5.7B	$0.2B	$0.2B
FY2029 v10	$8.7B	$7.6B	$1.1B	$1.3B
FY2031 v10+	$11.2B	$8.9B	$2.3B	$3.6B

Terminal Valuation Impact:

• FY2031 Difference of $2.3B/year × 25x P/E = $57.5B Market Cap Erosion (42% of current $136B)
• This is consistent with the estimated CDS effect in Section 14.4 ($25-60B), passing cross-validation.

15.5 CI Registry: Quantifying RISC-V CDS Pricing Power

Non-Consensus Insight (CI-01): ARM's pricing power is not binary (on/off) but a continuous function of RISC-V penetration. Current elasticity ε=-0.16 means ARM still has strong pricing power, but by 2030, elasticity will increase to -0.087 (net incremental gain from each price increase decreasing by 40%). The "gradual erosion" rather than "sudden loss" of pricing power is a more probable path—meaning ARM's revenue growth will gradually slow down (rather than a sudden cliff-edge drop), corresponding to a "slow decline" rather than a "collapse" in valuation.

Transferability: This elasticity function framework can be applied to any IP company facing open-source alternatives:

• Oracle vs PostgreSQL
• Red Hat vs Community Linux
• Unity vs Godot
• ARM vs RISC-V

Key Parameters: ε₀ (Base Elasticity) × R_crit (Critical Penetration Rate) × α (Decay Coefficient)

15.6 Sensitivity Analysis of the Elasticity Function

The elasticity function model relies on three key parameters, and the uncertainty of each parameter can affect the conclusions:

Parameter Sensitivity Matrix:

Parameter	Low Valuation	Base Case	High Valuation	Impact on Conclusion
ε₀ (Base Elasticity)	-0.10	-0.20	-0.30	Lower ε₀ → Stronger ARM Pricing Power → Smaller CDS Effect
R_crit (Critical Penetration Rate)	15%	22%	30%	Higher R_crit → Longer Window for ARM → Delayed CDS Effect
α (Decay Coefficient)	0.5	1.0	1.5	Higher α → Faster Elasticity Decay → More Rapid Decline in ARM Pricing Power

Extreme Scenario Combinations:

Scenario	ε₀	R_crit	α	2030 Elasticity	ARM Pricing Power Status
Most Optimistic (for ARM)	-0.10	30%	0.5	-0.066	Very Strong (Hardly Affected)
Baseline	-0.20	22%	1.0	-0.087	Moderate (Net Increment Decreases by 40%)
Most Pessimistic (for ARM)	-0.30	15%	1.5	-0.009	Extremely Weak (Virtually No Pricing Power)

Key Finding: Even under the most optimistic parameter combination, ARM's pricing elasticity by 2030 changes from -0.10 to -0.066 (a 34% weakening). The core conclusion of the elasticity function (ARM's pricing power will gradually weaken) is robust within a reasonable range of parameter variations—the only differences lie in the speed and magnitude of the weakening.

15.7 Specific Forecasts of v10 Royalty Rates Based on the Pricing Elasticity Function

ARM's next-generation ISA (v10) is expected to be launched in 2027-2028. The v10 royalty rate will be a critical "real-world test" for the CDS elasticity function:

Scenario	v10 Royalty Rate (vs v9)	Annualized Royalty Increment	Probability	Implied Elasticity
v9 Doubled (e.g., v8→v9)	v9 × 2.0	+$800M-1.2B	15%	ε < -0.10
v9 Price Increase of 50%	v9 × 1.5	+$400-600M	35%	ε ≈ -0.12
v9 Price Increase of 20%	v9 × 1.2	+$160-240M	30%	ε ≈ -0.15
v9 Flat	v9 × 1.0	$0	15%	ε ≈ -0.20
v9 Price Decrease (Forced)	v9 × 0.8	-$160-240M	5%	ε > -0.20

Probability-Weighted v10 Royalty Rate: 15%×2.0 + 35%×1.5 + 30%×1.2 + 15%×1.0 + 5%×0.8 = 1.36x v9

Implication: The model predicts that the v10 royalty rate will be approximately 1.36x v9—significantly lower than the 2.0x seen from v8→v9. This implies that ARM's "price increase cycle" is decelerating. v10 will be an "experimental validation" of the CDS effect: If v10 can only increase prices by 36% (vs. 100% from v8→v9), this will confirm that the constraining effect of RISC-V is substantially limiting ARM's pricing power.

Tracking Signal: The announcement of the v10 royalty rate (expected 2027-2028) will be a critical watershed for the ARM investment thesis:

• v10 ≥ v9×1.5: CDS effect weaker than expected → ARM pricing power solid → Bull case strengthened
• v10 = v9×1.0-1.3: CDS effect in line with expectations → ARM growth slows → Valuation compression begins
• v10 < v9×1.0: CDS effect stronger than expected → ARM faces severe challenges → Significant valuation compression

---

Chapter 16: x86 Competition and the Dual Effects of Custom Silicon

16.1 Data Center CPU Tripartite Game: ARM/x86/RISC-V

The data center CPU market is not a binary competition, but a three-way (and evolving into a four-way) game:

Share Evolution Data:

Year	Intel x86	AMD x86	ARM	RISC-V	Source of ARM Share
2020	~90%	~8%	~2%	~0%	AWS Graviton2 first large-scale deployment
2022	~75%	~15%	~10%	~0%	Graviton3+Ampere Altra
2024	~65%	~18%	~16%	~1%	Google Axion+NVIDIA Grace
2026E	~55%	~20%	~23%	~2%	Microsoft Cobalt+CSS Diffusion
2028E	~48%	~22%	~27%	~3%	China + Southeast Asia DC Adoption
2030E	~42%	~23%	~30%	~5%	RISC-V begins to divert ARM share

Quality Analysis of ARM's Data Center Growth: ARM's share in the data center market grew from ~2% in 2020 to ~23% in 2026, which is ARM's most compelling growth story over the past five years. However, the quality (sustainability) of this growth needs to be examined in layers:

1. Royalty Growth vs. Profit Growth: ARM's royalties from DC indeed grew at a high rate (>100% YoY), but Phoenix investments and CSS R&D are also significantly increasing. DC royalty growth ≠ DC profit growth.
2. Source of Market Share: Almost all of ARM's DC market share comes from Intel (not AMD). This growth might slow down as Intel stabilizes.
3. Customer Concentration: ARM's DC royalties are highly concentrated among 3 hyperscale customers (AWS/Google/Microsoft). If 1-2 of these three companies switch to RISC-V (Google is already evaluating it), ARM's DC growth will significantly slow down.

16.2 Custom Silicon: ARM's Double-Edged Sword

The custom silicon trend among hyperscale customers is ARM's biggest growth engine over the past 5 years, but also its largest long-term risk.

Short-Term Positive (2024-2028): +$1-2B Royalty Increments

Customer	Chip Project	ARM Contribution	Estimated Royalties (Annual)	Status
AWS	Graviton 4/5	Neoverse V2/V3 Cores	~$300-500M	In Production
Google	Axion	Neoverse V3 Cores	~$200-400M	In Production
Microsoft	Cobalt 100/200	Neoverse N3/CSS	~$150-300M	Production/R&D
NVIDIA	Grace	Neoverse V2 Cores	~$100-200M	In Production
Oracle	Post-Ampere Acquisition	Ampere Cores (ARM)	~$50-100M	Expanding
Total			~$800M-1.5B

Long-Term Risk (2028+): Custom silicon → RISC-V Path

Custom silicon customers are ARM's most valuable clients in the data center, but also the most capable of migrating to RISC-V:

1. Accumulation of Design Capability: AWS/Google/Microsoft each employ thousands of chip engineers. Designing RISC-V cores is not a question of 'whether they can' for them, but 'whether there is an economic incentive'.
2. Testing ARM Royalty Ceiling: Each time ARM raises prices (v9→CSS→v10), the option of 'in-house RISC-V development' becomes more attractive in these customers' ROI analyses.
3. Precedent: Meta has already used RISC-V coprocessors in its AI accelerator MTIA v3. Although not the main CPU, this proves that hyperscale customers are willing to adopt RISC-V in critical chips.

Key Turning Point: If AWS, Google, or Microsoft announce a RISC-V server CPU (even an experimental one), it would be a 'black swan' event for ARM's valuation. This is not because it immediately impacts revenue (migration takes 3-5 years), but because it fundamentally alters market confidence in ARM's DC growth narrative.

16.3 ARM vs AMD/Intel: Workload Suitability Analysis

ARM's competitive advantage in data centers is not universal, but rather workload-specific:

Workload	ARM Advantage	x86 Advantage	ARM Share (2026E)	Trend
Cloud Native/Web	High energy efficiency, 15-20% lower TCO	—	~30-35%	📈 Rapid Growth
AI Inference	Grace+GPU combo (GB200)	AMD MI300 CPU+GPU	~20-25%	📈 Growth
AI Training	Auxiliary (CPU with GPU)	Auxiliary (AMD CPU with GPU)	~15%	Even
Database (OLTP)	Adapting	Strong historical advantage	~10-15%	📈 Slow Growth
HPC (Scientific Computing)	Fugaku validation	Still dominant (AVX-512)	~5-10%	Slow
Enterprise On-Prem	Very Low	x86 absolute dominance	<5%	Stagnant
Edge Computing	Energy efficiency advantage	Operational consistency	~15-20%	📈 Growth

Key Insight: ARM's success in data centers is highly concentrated in the "Cloud Native/Web Services" workload category. In enterprise on-prem (still 40%+ of the server market), ARM has almost no penetration—this is because enterprise IT departments' software stacks, operational processes, and procurement relationships are all built around x86. ARM's DC growth forecasts need to distinguish between two distinct markets: 'cloud native' (ARM-friendly) and 'enterprise' (x86-locked).

16.4 Where is the 'Ceiling' for ARM DC Growth?

ARM's data center growth story is the core support for its 142x P/E. But does this growth have a ceiling?

Server CPU TAM Estimate (2026):

• Global Server CPU Market: ~$40-45B/year (incl. Intel+AMD+ARM)
• ARM Share ~23%: ~$9-10B (but this is 'chip-level' TAM, ARM only collects royalties)
• ARM DC Royalty Revenue: ~$600-800M (royalty rate approx. 6-8% of chip ASP)
• ARM DC Licensing Revenue: ~$200-300M
• ARM DC Total Revenue: ~$0.8-1.1B (approx. 15-18% of ARM's total revenue)

ARM DC Ceiling Analysis:

Time	ARM DC Share	DC CPU TAM	ARM DC Royalties	Royalty Rate Assumption	Limiting Factors
2026E	23%	$42B	~$700M	~7.5%	Normal Growth
2028E	27%	$50B	~$1.2B	~9%(CSS Price Increase)	Nascent RISC-V Competition
2030E	30%	$58B	~$1.7B	~10%(v10)	RISC-V Pricing Constraint
2032E	32%(Ceiling?)	$65B	~$2.0B	~10%(Unable to continue raising prices)	RISC-V + Enterprise x86 Inertia

Ceiling Estimate: ARM's market share ceiling in DC might be around 30-35% (assuming ARM has almost no penetration in the enterprise on-prem market, which accounts for 35-40% of the DC market). The DC royalty ceiling is approximately $2.0-2.5B/year (assuming royalty rates cap at 10%).

Comparison with Market Expectations: Consensus implies ARM's DC revenue reaching $4.5B by FY2030 (accounting for 40% of total revenue). This requires DC royalty rates significantly exceeding 10% or DC market share significantly exceeding 30%—both of which are constrained by RISC-V. The market's DC growth assumption may be overly aggressive.

16.5 Impact of Project Phoenix on the Competitive Landscape

Phoenix (ARM's self-developed chip, targeting the DC custom silicon market) not only impacts ARM's business model (see Ch6) but also alters the game theory structure of x86/ARM competition:

Before Phoenix: ARM is a neutral IP vendor → A friend to all chip designers → Maximizes IP sales
After Phoenix: ARM is an IP vendor + chip designer → Competes with Qualcomm/NVIDIA/Ampere → Strained IP client relationships

Game Theory Perspective: Phoenix's Nash Equilibrium

	Clients Retain ARM	Clients Switch to RISC-V
ARM Does Not Pursue Phoenix	(Stable, ARM earns royalties)	(ARM loses clients)
ARM Pursues Phoenix	(Tense, ARM earns on two fronts)	(ARM's chip business loses money, royalties also decrease)

ARM's Dilemma: If ARM does not pursue Phoenix, it fears being excluded from the high-value DC market. However, if ARM does pursue Phoenix, clients are pushed towards RISC-V. This is a classic Prisoner's Dilemma: ARM's "optimal" individual strategy (pursuing Phoenix) leads to a worse industry equilibrium (accelerated RISC-V adoption).

Most Likely Outcome: ARM may invest heavily in R&D for Phoenix but ultimately achieve only limited commercial success (DC market share <5%), while permanently damaging some client relationships. Phoenix's "trust tax" (an implicit cost of $45-100M/year analyzed in Ch6.7) may exceed its direct revenue contribution.

---

Chapter 17: Reverse DCF Belief Inversion — What Does a 141.6x P/E Imply?

17.1 Belief Inversion Methodology

Reverse DCF is not a forward valuation answering "How much is ARM worth?", but rather answers: "If ARM is worth $136.1B (current market cap), what must the market believe?"

This method is particularly important for ARM because:

1. ARM has never had a "normal" P/E (142x-1073x since IPO), and forward DCF lacks a valuation anchor.
2. Traditional DCF yields a massive range of $6.66-$80, indicating extreme model sensitivity to assumptions.
3. Reverse DCF translates "price" into a "set of beliefs," each of which can be independently verified.

17.2 Reverse DCF: Three-Scenario Framework

Input Parameters:

• Current Market Cap: $136.1B
• Net Cash: $1.73B → EV ≈ $134.4B
• Discount Rate: 10% (standard WACC for high-growth tech companies)
• Terminal Growth Rate: 3% (inflation + GDP growth)
• Projection Period: 10 years (FY2026→FY2035)

Scenario A: 25x Terminal P/E (Mature Software Company)

Simplified Calculation (without discounting terminal value, only looking at steady-state value after 10 years):

Scenario B: 30x Terminal P/E (High-Quality Platform Company)

Scenario C: 35x Terminal P/E (Visa/MC-level Network Platform)

17.3 Implied Belief Set: 6 Independently Verifiable Sub-Beliefs

#	Sub-Belief	Implied Value (Scenario B)	Current Actual Value	Gap	Verification Method	Fragility
B1	Revenue CAGR (10 years)	~14.1%	FY25: +23.9% YoY	First year exceeds, but needs to be sustained	Historically, very few IP licensors sustain >15%	High
B2	GAAP Net Margin Expansion	17.2%→30%	FY25: 19.8%	Needs +10.2pp expansion	Trend of SBC/R&D as % of revenue	High
B3	Pricing Power Not Constrained by RISC-V	v10+ can continue to raise prices	v9 price increases successful	H2 hypothesis challenged	RISC-V penetration tracking	Extremely High
B4	DC Contribution 10%→40%+	DC becomes largest business	DC ~15% rev	Requires +25pp	CSS penetration + Phoenix	High
B5	Arm China Sustained Growth	China does not decelerate	+7.5% vs Group +24%	Already decelerating	Geopolitical risk tracking	Medium
B6	SB Does Not Reduce Holdings / No Liquidity Crisis	90% stake stable	Current 90.6%	$250B+ debt	SB quarterly report tracking	Medium

17.4 In-Depth Analysis of Belief Fragility

B1: Revenue CAGR ~14.1% (10 years) — Fragility: High

Historical Benchmarking: In the IP licensing industry (pure IP, non-product), very few companies can sustain >15% revenue CAGR for more than 5 years:

• Qualcomm QTL (closest comparable): FY2014-2024 CAGR approximately +3% (affected by antitrust and smartphone growth deceleration)
• Synopsys: FY2015-2025 CAGR approximately +12% (benefiting from AI EDA demand)
• Cadence: FY2015-2025 CAGR approximately +14% (similar)

ARM requires a 14.1% CAGR sustained for 10 years, implying FY2035 revenue of ~$14.9B (3.7x of current $4.0B). This requires: ① sustained royalty rate increases (v9→CSS→v10→v10+) + ② data center revenue contribution growing from ~15% to 40%+ + ③ continuous expansion in IoT/Automotive.

Calibration: Analyst consensus FY2030 revenue of $11.2B (4-year CAGR ~23%) is more aggressive than Reverse DCF Scenario B (consensus implies ARM will achieve most of its growth in the first 4 years). However, the consensus projected FY2030 net margin of 45% is significantly higher than our 30% assumption.

B2: GAAP Net Margin Expansion from 19.8% to 30% — Vulnerability: High

ARM's margin expansion faces two structural headwinds:

1. Persistent SBC (Largest Headwind): ARM FY2024 SBC = $1.037B (32.1% of revenue). FY2025 FMP reports SBC = $0 (clear data error), but Non-GAAP adjustments show SBC still at $700-900M/year. If SBC remains at 15-20% of revenue, GAAP net margin will struggle to exceed 25%.

2. R&D Investment Requirements: ARM FY2025 R&D = $2.01B (50.1% of revenue), FY2024 R&D = $1.93B (59.8%). If ARM is to maintain its CPU architecture leadership (v10+Phoenix), R&D as a percentage of revenue is unlikely to drop below 40%.

Margin Expansion Path:

Driver	Impact (pp)	Probability	Timing
SBC as % of Revenue Declines (IPO effect fades)	+5-8pp	70%	FY2027-2028
Revenue Growth Faster than OpEx	+3-5pp	60%	Ongoing
Royalty Contribution Increases (High Margin)	+2-3pp	50%	FY2027+
R&D as % of Revenue Declines	+3-5pp	30%	Uncertain (vs Phoenix investment)
Potential Total Expansion	+13-21pp	—	—
Target Expansion	+10.2pp(19.8%→30%)	~50%	FY2035

Conclusion: Margin expansion from 19.8% to 30% has a plausible path but faces structural headwinds (SBC + R&D). The consensus projected 45% net margin assumes almost all positive factors materialize simultaneously with no negative impacts – overly aggressive.

B3: ARM's Pricing Power Unconstrained by RISC-V — Vulnerability: Extremely High

This is the most vulnerable of the six beliefs – see Chapter 14 and Chapter 14A for detailed analysis. Core logic:

• ARM achieved a doubling of royalties from v8 to v9, leading the market to extrapolate sustained price increases from v9→v10→v10+.
• However, RISC-V's CDS effect (elasticity function in Chapter 14A) predicts ARM's net incremental price increases will decrease by 40% by 2030.
• If v10 cannot achieve another doubling of price increases from v9→v10, ARM's 10-year CAGR will fall from 14.1% to 10-12%.
• The difference between a 10-year CAGR of 10% vs 14.1%: Terminal Revenue $10.4B vs $14.9B → Terminal Net Profit $2.6B vs $4.5B → Market Cap Difference **$52B** (vs 38% of current $136B).

B3 is the "Kingmaker Belief": If B3 holds true (ARM can continue to raise prices), then the other five beliefs can partially fail, and ARM can still maintain a market cap of $100B+. If B3 proves false (RISC-V constrains pricing power), even if all other five beliefs hold true, ARM would struggle to maintain a market cap of $136B.

B4: Data Center Contribution Increases from ~15% to 40%+ — Vulnerability: High

Data center growth is a core pillar of ARM's "growth story":

Current DC Revenue Breakdown (FY2026E estimate):

• DC Royalties: ~$600-800M (rapid growth, >80% YoY)
• DC Licensing: ~$200-300M (CSS+design licensing)
• DC as % of Total Revenue: ~15-18%

FY2030E Required DC Revenue: If ARM's FY2030 revenue is $11.2B (consensus) and DC accounts for 40% → DC revenue needs to reach $4.5B → requires 4-5x growth from current ~$0.8-1.1B.

Three Drivers vs. Three Headwinds for DC Growth:

Driver	Valuation Quantification	Probability
CSS Penetration (Higher Royalty Rates)	+$1.0-1.5B	70%
ARM DC Share Expansion (from 23%→30%)	+$0.5-1.0B	60%
Phoenix In-house Chip Royalties	+$0.3-0.5B	40%

Headwind	Valuation Quantification	Probability
RISC-V DC Penetration (Diverting ARM Growth)	-$0.3-0.7B	50%
AI Training → Inference Shift Slows DC Growth	-$0.2-0.5B	30%
Hyperscale Customers Develop In-house RISC-V	-$0.5-1.0B (if it occurs)	15-20%

17.5 Belief Reversal Analysis: How Many Belief Failures Would Change the Rating?

Reversal Scenario	Failed Beliefs	Terminal Net Profit	Terminal Market Cap (30x)	Downside
Zero Reversals (Currently Implied)	All Hold True	$4.5B	$136B	Baseline
Single Belief Reversal	B3 (Pricing Power)	$2.5-3.0B	$75-90B	-34~45%
Double Belief Reversal	B3+B2 (Pricing + Margins)	$1.5-2.0B	$45-60B	-56~67%
Triple Belief Reversal	B3+B2+B4 (+DC Growth)	$1.0-1.5B	$30-45B	-67~78%
Triple Belief + Event	B3+B2+B6 (+SB Liquidation)	$1.0-1.5B×0.7x (discount)	$21-32B	-77~85%

Conclusion: ARM's valuation is extremely sensitive to B3 (Pricing Power). Just one belief reversal could lead to a 34-45% downside. This is consistent with A-Score's moat migration analysis – ARM's moat is shifting from technology to ecosystem; if RISC-V bypasses the ecosystem barrier (the root cause of B3 failure), the entire valuation framework will collapse.

17.6 Detailed Reversal Paths for the Six Key Beliefs

The failure of each belief is not "on-off" – it has specific triggers, time paths, and observable signals. This provides more valuable investment guidance than simple probability assignments:

B1 (Revenue Growth >14% CAGR) Reversal Path:

• Trigger Conditions: Revenue growth <15% YoY for 2 consecutive quarters + Full-year guidance cut
• Most Likely Trigger Factors: Saturation in mobile royalty growth (v8 installed base shipments beginning to decline) + DC growth falling short of expectations (end of Graviton/Grace procurement cycles)
• Time Window: FY2027-2028 (after the v9 mobile pricing increase effect fades)
• Leading Indicators: KS-M02 (DC RISC-V share) + Quarterly disclosure of ARM licensing (ATA) pipeline + Mobile chip shipments (IDC/Counterpoint)
• Reversal Consequences: B1 failure → 10% revenue CAGR → FY2030 revenue $7.8B (vs implied $14.9B) → Even at 30% NM, only $2.34B Net Income → $70B market cap (-49%)

B2 (GAAP Margin Expansion to 30%) Reversal Path:

• Trigger Conditions: FY2027 GAAP Operating Margin still <25% + SBC/Revenue sustained >12%
• Most Likely Trigger Factors: Slower-than-expected SBC normalization (still 15%+ five years post-IPO) + Continued growth in Phoenix R&D expenditure + v10 development team expansion
• Time Window: FY2027 (3 years post-IPO, when SBC "should" begin to normalize)
• Leading Indicators: Quarterly Non-GAAP vs. GAAP delta + Employee headcount growth rate + R&D/Revenue ratio
• Reversal Consequences: B2 failure → Margin stagnation at 20% → Net Income growth solely dependent on revenue growth → P/E multiple de-rating → 30-50% valuation compression

B3 (Pricing Power Maintained — RISC-V Not Materially Constraining) Reversal Path:

• Trigger Conditions: v10 royalty rate increase <0.8x of v9's + Weighted RISC-V penetration exceeds 15%
• Most Likely Trigger Factors: Qualcomm RISC-V product mass production (2027E) + Google Android RISC-V kernel maturity (2028E) + China policy mandates (2026-2028)
• Time Window: 2027-2029 (v10 pricing negotiation period)
• Leading Indicators: RISC-V International member growth rate + China mobile RISC-V SoC announcements + Changes in ARM licensing contract renewal terms
• Reversal Consequences: B3 failure → CI-01 elasticity function begins to dominate → Annual royalty growth rate drops from 15% to 5-8% → FY2030 royalty revenue $4.5B (vs implied $7B+) → Greatest valuation impact (-34~45%)
• This is the single most sensitive belief: The downside from B3 reversal far exceeds other beliefs because royalties are ARM's core profit engine

B4 (DC Growth Path: From 10% → 25%+) Reversal Path:

• Trigger Conditions: ARM DC market share stagnates in the 15-18% range for 4 consecutive quarters
• Most Likely Trigger Factors: Difficulties in enterprise on-prem migration (poor Windows Server ARM compatibility) + Intel's counter-attack (Sierra Forest cost-performance ratio) + NVIDIA Grace remaining for internal consumption (not for external sale)
• Time Window: 2027-2028 (critical validation period for whether current growth momentum is sustainable)
• Leading Indicators: Cloud vendor ARM instance growth rate + Number of newly released ARM server chips + Enterprise ARM procurement intention surveys

B5 (Competitive Landscape Not Dramatically Changing) Reversal Path:

• Trigger Conditions: 3+ of ARM's top 10 customers simultaneously invest in in-house RISC-V R&D teams + Any single customer releases a mass-produced RISC-V chip
• Most Likely Trigger Factors: Samsung/MediaTek follow Qualcomm's RISC-V strategy + Apple evaluating RISC-V for Mac/iPad
• Time Window: 2028-2030 (results of current RISC-V investments begin mass production)

B6 (SoftBank Not Forced to Liquidate) Reversal Path:

• Trigger Conditions: SoftBank Net Debt/EBITDA > 10x + ARM stock price falling below $100 triggers a margin call + Masayoshi Son's health/retirement event
• Most Likely Trigger Factors: Sustained high global interest rates (SoftBank refinancing difficulties) + Deterioration in SoftBank's other investment portfolios (requiring ARM monetization)
• Time Window: 2026-2028 (SoftBank debt maturity peak)
• Leading Indicators: SoftBank quarterly results + ARM block trade announcements + Changes in SoftBank's ARM ownership stake
• Reversal Consequences: B6 reversal → Does not change ARM's intrinsic value but leads to supply-demand imbalance → Short-term stock price decline of 30-50% → If occurring simultaneously with B3 → May trigger a liquidation spiral (-50~70%)

17.7 Implied Beliefs vs. Analyst Consensus

Dimension	Reverse DCF (Scenario B)	Analyst Consensus (FMP)	Difference	Which is more reliable
FY2030 Revenue	$14.9B	$11.2B	DCF+33%	Consensus is more conservative
FY2030 Net Profit	$4.5B	$5.0B	DCF-10%	Similar
FY2030 Net Margin	30%	45%	Consensus+15pp	Consensus is overly aggressive
FY2026-30 CAGR	~14.1% (10 years)	~23% (4 years)	Consensus front-loaded	Not comparable (different periods)
FY2030 P/E	30x (terminal state)	~26x (implied)	—	—

Key Finding: The analyst consensus's FY2030 net margin assumption (45%) requires ARM's GAAP net margin to expand by +25.2pp from 19.8%—this demands SBC falling below 5% of revenue (currently 15-20%) + R&D as a percentage of revenue dropping to 30% (currently 50%) + full realization of operating expense leverage. Against the backdrop of ARM's sustained high investment (Phoenix/v10/CSS), this is an extremely aggressive assumption.

If net margin stagnates at 25% (instead of 45%):

• FY2030 Net Income: $11.2B × 25% = $2.8B (vs. Consensus $5.0B, -44%)
• 30x P/E: Market Cap $84B → Current $136B → -38% downside

17.8 Joint Probability of Pillar Assumptions

The 6 beliefs are not independent—they have conditional dependencies:

Joint Probability Analysis:

• P(B3 holds) = 40% (Probability that RISC-V competitive displacement effect limits pricing power >50%)
• P(B1 holds|B3 holds) = 60% (If pricing power is maintained, revenue CAGR might be achieved)
• P(B1 holds|B3 fails) = 20% (Without pricing power, revenue CAGR will be difficult to achieve)
• P(B2 holds) = 35% (Probability of margin expansion to 30%)
• P(All beliefs hold) ≈ 40% × 60% × 35% × 60% × 70% × 80% ≈ 2.8%

Implication: The market's current $136.1B valuation implies that "all 6 beliefs simultaneously hold," but the joint probability is only about 2.8%. Even if we are overly pessimistic about some probabilities (assigning 50% instead of 40% for B3), the joint probability would still only be ~5-8%. This means ARM's "bet" at the current price is highly asymmetric—requiring nearly perfect execution to justify the valuation.

Joint Probability Sensitivity Matrix: Impact of changing P(B3) and P(B2) on joint probability – these two beliefs are the largest sources of uncertainty for the valuation:

P(B3)↓ \ P(B2)→	25%	35%	45%	55%
30%	1.5%	2.1%	2.7%	3.3%
40%	2.0%	2.8%	3.6%	4.4%
50%	2.5%	3.5%	4.5%	5.5%
60%	3.0%	4.2%	5.4%	6.6%

Interpretation: Even in the most optimistic corner (P(B3)=60%, P(B2)=55%), the joint probability is only 6.6%. To achieve a joint probability of >15% (providing reasonable support for a $136B valuation), the probabilities for B3 and B2 would need to be raised to >70% respectively – but neither the analysis in CI-01 (RISC-V CDS) nor the SBC analysis in Section 17.4 supports such optimistic assumptions.

vs. Bull Case Argument: Bullish investors might argue that B1-B6 do not need to "all hold true" – ARM's growth potential could still provide positive returns even if some beliefs fail. However, the reversal analysis in Section 16.5 shows that a standalone failure of B3 leads to a 34% to 45% downside, implying that even if B1/B2/B4/B5/B6 all hold true, a reversal of B3 would still be catastrophic. Pricing power is the load-bearing pillar of ARM's valuation, not an optional embellishment.

17.9 Reverse DCF vs Analyst Price Target Distribution

Analyst Price Target Distribution (FMP Estimates, 2026-02):

Analyst Consensus	Value	Implied Market Cap	vs. Current
Target Price (Median)	~$145	~$154B	+13%
Target Price (Low End)	~$100	~$106B	-22%
Target Price (High End)	~$200	~$213B	+56%

Analyst Estimates vs. Reverse DCF Comparison:

Metric	Analyst Consensus (FY2030)	Reverse DCF (Scenario B)	Variance Analysis
Revenue	$11.2B	$14.9B	Consensus more conservative (-25%)
Net Income	$5.0B	$4.5B	Consensus higher (+11%)
Net Margin	45%	30%	Key Disagreement
EPS	$4.70	$4.22	Consensus slightly higher
Implied P/E	~26x	30x	Similar

Core Disagreement Lies in Net Margin: Analyst consensus believes ARM can achieve a 45% net margin by FY2030 (an expansion of 25.2pp from the current 19.8%). Our Reverse DCF Scenario B assumes 30%. A 45% net margin implies:

• SBC decreasing to 3-5% of revenue (currently 15-20%) → requires IPO effects to fully dissipate + restricted stock programs to scale back
• R&D/Revenue decreasing to 30-35% (currently 50%) → requires revenue growth to significantly outpace R&D (but Phoenix/v10 require continuous investment)
• Tax rate maintained at ~12-15% (current effective tax rate is -10%, but this benefits from non-recurring tax advantages)

If Both Ends of the Net Margin Disagreement Are Wrong:

Net Margin Assumption	FY2030 NI	30x P/E	vs. Current Market Cap
20% (Current Level)	$2.24B	$67B	-51%
25%	$2.80B	$84B	-38%
30% (Our Scenario B)	$3.36B	$101B	-26%
35%	$3.92B	$118B	-13%
40%	$4.48B	$134B	-1%
45% (Consensus)	$5.04B	$151B	+11%

Key Finding: Under a 30x terminal P/E assumption, ARM needs at least a 40% net margin to justify its current $136B market capitalization. This is a full 20.2pp higher than the current actual 19.8%. Net margin is the 'hidden variable' in ARM's valuation equation – the market focuses on growth rate, but profit margin is the critical parameter determining valuation.

17.10 "If the Market Is Right" — Steelman Version of the Bull Case Argument

To avoid confirmation bias, the bull argument must be taken seriously. If ARM's 142x P/E is justified, what might the market believe that we have not fully considered?

Bull Case Steelman Argument:

1. ARM is not a traditional IP company; it's a "Computational Tax Authority": If you believe ARM's business model is closer to Visa (global payment tax) rather than Synopsys (EDA licensing), then a terminal P/E of 35-40x is more reasonable, with lower growth requirements. Visa's historical average P/E is ~30-35x, even during "steady-state" growth periods.

2. The Threat of RISC-V is Overestimated: It took Linux 25 years to gain ~4% market share on desktops. RISC-V might face a similar "always a step away" dilemma – every year it's "about to threaten ARM," but every year it fails to scale due to software ecosystem gaps. If RISC-V remains perpetually in niche markets like IoT + China DC, ARM's pricing power would remain effectively unaffected.

3. AI is a Structural Accelerator for ARM's Growth: AI inference (rather than training) is increasingly running on ARM CPUs (e.g., Grace CPU in GB200). If AI workloads grow at a 50%+ CAGR and ARM captures most of the incremental growth, ARM's FY2030 revenue could significantly exceed $11.2B.

4. Margin Expansion is Inevitable: The decline of SBC from $1.037B in FY2024 (IPO year) to $400-500M (steady-state) is a highly certain event. SBC normalization alone could free up +10pp in GAAP margin.

Rebuttal Assessment:

Bull Case Argument	Our Assessment	Potentially Underestimated?
"Computational Tax Authority"	Partially valid, but Visa has no open-source alternative	Partially
RISC-V threat overestimated	Possible, but $30B+ investment cannot be ignored	Possible
AI structural acceleration	Valid, but AI royalty rates may not be higher than traditional DC	Moderate
Margin expansion inevitable	SBC normalization is certain, but R&D is uncertain	Yes

Conclusion: The strongest point in the bull case argument is SBC normalization. We may have underestimated the speed and magnitude of SBC's decline from its IPO peak to a steady state. If SBC declines to $400M by FY2027 (7% of revenue), GAAP operating margin could increase from 20.6% to 33-35%, which would significantly improve ARM's earnings quality score.

However, even if we accept the strongest bull case arguments (SBC normalization + AI acceleration), ARM's 142x P/E still requires the B3 (pricing power) belief to hold true – and this is precisely what the RISC-V CDS effect directly challenges.

17.11 Reverse DCF "Payment Network" Valuation Bridge

Previous sections (Ch2.3/Ch7A) introduced the analogy of ARM as a "computational payment network" (similar to Visa/Mastercard). If this analogy holds, ARM's Reverse DCF should use the payment network valuation framework:

Visa/MC vs ARM Valuation Comparison:

Metric	Visa	Mastercard	ARM	ARM vs V/MC
P/E (TTM)	~33x	~38x	142x	ARM 3.7-4.3x multiple
P/E (FY+2)	~27x	~30x	~46x (FY28)	ARM 1.5-1.7x multiple
Revenue CAGR (5Y)	~11%	~13%	~19%	ARM faster
Gross Margin	~80%	~80%	95%	ARM higher
Net Margin	~55%	~46%	20%	ARM significantly lower
FCF Yield	~3.5%	~3.0%	0.13%	ARM significantly lower
Network Effect Strength	Two-sided (strong)	Two-sided (strong)	One-sided (moderate)	ARM weaker
Faces open-source alternative?	No	No	Yes (RISC-V)	ARM disadvantaged

ARM's "Fair" Valuation under the Payment Network Framework:

If ARM's business model indeed approximates Visa's (charging transaction taxes on computation), then its terminal valuation multiple should be close to Visa's:

• Current Visa P/E ~33x
• ARM growth premium (CAGR 19% vs Visa 11%): +40-50% → ~46-50x P/E
• ARM open-source alternative discount (Visa has no RISC-V equivalent): -20-30% → ~32-40x P/E
• ARM margin gap discount (NM 20% vs Visa 55%): -15-25% → ~24-34x P/E

Payment Network Framework Conclusion: Even when applying the "payment network" valuation framework to ARM (which is the most favorable analogy for ARM), a fair P/E is only around 24-34x → current 142x is still 4-6x higher.

The 142x P/E can only be justified under the payment network framework if:

1. ARM's net margin reaches 45%+ within 5 years (matching Visa) + 2. Revenue CAGR remains >20% (exceeding Visa) + 3. RISC-V poses no substantial threat (no "open-source alternative" for Visa)

• Probability estimate: (1)35% × (2)30% × (3)40% = 4.2%

17.12 Valuation Synthesis: Five-Engine Preliminary Exploration

A complete five-engine valuation will further validate these findings; here, the Reverse DCF provides a preliminary valuation range:

Preliminary Valuation Judgment:

• Midpoint Estimate: $75-100B (per share $71-94) vs Current $128.14 → Downside 22-45%
• Confidence Level: Moderate (40%, subject to full five-engine valuation confirmation)
• Initial Rating Assessment: Leaning towards "Cautious Watch" (expected return <-10%)
• Key Variables: B3 (pricing power) and B2 (profit margin) determine whether the midpoint falls at $75B or $100B.

However, this valuation requires a complete DCF and option value adjustment: ARM's option value (AI inference market explosion + Phoenix success + Arm China IPO) could add $15-25B → adjusted midpoint $90-125B → downside narrows to -8~34%.

---

Chapter 18: In-Depth Financial Statement Analysis and Earnings Quality Audit

18.1 5-Year Income Statement Vertical Analysis

Metric ($M)	FY2021	FY2022	FY2023	FY2024	FY2025	5-Year CAGR
Revenue	2,027	2,703	2,679	3,233	4,007	+18.6%
COGS	236	220	195	234	206	-3.3%
Gross Profit	1,791	2,483	2,484	2,999	3,801	+20.7%
Gross Margin	88.4%	91.9%	92.7%	92.8%	94.9%	+6.5pp
R&D	768	945	1,080	1,932	2,009	+27.2%
SG&A	781	858	726	968	965	+5.4%
Operating Expenses	1,549	1,803	1,806	2,900	2,974	+17.7%
Operating Income	242	680	678	99	827	+36.0%
Operating Margin	11.9%	25.2%	25.3%	3.1%	20.6%	—
Net Income	388	549	524	306	792	+19.5%
Net Margin	19.1%	20.3%	19.6%	9.5%	19.8%	—
EPS (Diluted)	$0.38	$0.54	$0.51	$0.29	$0.75	+18.6%

5-Year Trend Analysis:

1. Gross Margin Expansion (88.4%→94.9%): ARM's gross margin is among the highest in the semiconductor industry (second only to the pure software segments of Synopsys (~80%) and Cadence (~86%)). A gross margin of nearly 95% reflects ARM's extremely low marginal cost (selling IP licenses, not manufacturing anything). However, note that if the Phoenix project leads ARM into chip design/testing, the gross margin could fall from 95% to 85-90%.

2. FY2024 Operating Income Plunges to $99M (3.1%): This is a one-time impact from $1.037B in SBC during the first year of IPO ($1.037B ÷ $3.233B revenue = 32.1% of revenue consumed by SBC). Non-GAAP operating income of $1.072B (33.2%) better reflects recurring profitability.

3. R&D Growth Rate (27.2%) Significantly Faster than Revenue Growth (18.6%): R&D as a percentage of revenue surged from 37.9% (FY2021) to 50.1% (FY2025). This reflects ARM's continued high investment in v9/CSS/v10/Phoenix. If the R&D-to-revenue ratio remains at 50%, margin expansion will be very limited.

4. The "Illusion" of Net Margin: The FY2025 net margin of 19.8% appears healthy but masks two issues:

• FY2025 effective tax rate was -10% (tax benefits), the normalized tax rate should be 15-20%
• FMP reported SBC=$0 (data error), actual SBC is estimated at $700-900M

18.2 Balance Sheet 5-Year Longitudinal + Quality Analysis

Metric ($M)	FY2021	FY2022	FY2023	FY2024	FY2025	Trend/Signal
Cash + Short-term Investments	—	1,635	2,215	2,923	2,825	Stable with slight decrease
Accounts Receivable	—	1,290	1,153	1,117	1,749	⚠️FY25 surged
Total Current Assets	—	3,092	3,537	4,197	4,830	Growth
PP&E	—	417	391	420	714	FY25 +70%⬆️
Goodwill	—	1,636	1,620	1,625	1,620	Stable
Intangible Assets	—	205	138	152	151	Amortizing
Total Assets	—	6,510	6,866	7,927	8,932	Continuous growth
Accounts Payable	—	57	65	26	63	Very low
Deferred Revenue (Current)	—	334	293	198	209	⚠️FY24-25 decrease
Deferred Revenue (Non-current)	—	792	807	717	702	⚠️Continuous decrease
Total Deferred Revenue	—	1,126	1,100	915	911	-19.1%
Total Liabilities	—	2,962	2,815	2,632	2,093	Decreasing (Good)
Total Shareholders' Equity	—	3,548	4,051	5,295	6,839	Growth
Net Cash	—	743	1,335	1,697	1,729	Consistently increasing

Key Warning Signals:

⚠️ Signal 1: Accounts Receivable Surge: FY2025 AR=$1,749M (vs FY2024 $1,117M, +57%). Revenue growth during the same period was only 24%. The significant increase in accounts receivable compared to revenue growth suggests that ARM's revenue recognition might be becoming more aggressive, or that payment terms for major customers have been extended.

DSO (Days) Longitudinal:

FY	DSO (Days)	Peer Reference	Gap
FY2022	174	SNPS 59, CDNS 44, QCOM 32	ARM is 4-5x peers
FY2023	159	SNPS 59, CDNS 57, QCOM 32	ARM is 3-5x peers
FY2024	127	SNPS 56, CDNS 54, QCOM 37	ARM is 2-3x peers
FY2025	173	SNPS 78, CDNS 65, QCOM 36	ARM is 2-5x peers

ARM's DSO is extremely anomalous within the industry. 173 days means ARM, on average, takes nearly 6 months to collect payment for a sale. In contrast, Synopsys is 78 days, Cadence 65 days, and Qualcomm 36 days.

Possible Reasons for Abnormal DSO:

1. Installment Payments for Licensing Agreements: ARM's TLA/ALA contracts are typically multi-year and paid based on milestones. Revenue recognition (IFRS 15) may occur upon contract signing, but actual cash collection could take several quarters.
2. Related Party (SoftBank) Delays: Payments from SoftBank and its portfolio companies might have preferential terms.
3. Accounts Receivable from China: Collection of accounts receivable from Arm China might be affected by foreign exchange controls and operational independence.
4. Concentrated Recognition of Large Contracts at Year-End: FY2025 Q4 revenue of $1,241M (vs Q3 $983M, +26% QoQ) suggests concentrated recognition of large contracts at year-end, with the corresponding cash not yet collected.

⚠️ Signal 2: Continuous Decline in Deferred Revenue: Total deferred revenue decreased from $1,126M in FY2022 to $911M in FY2025 (-19.1%). A decrease in deferred revenue could mean: ① ARM is accelerating revenue recognition (declining profit quality) or ② the rate of new contract signings is slower than the rate of existing contract recognition (weakening growth prospects). These two possibilities need to be distinguished.

⚠️ Signal 3: Significant Increase in PP&E: FY2025 PP&E=$714M (vs FY2024 $420M, +70%). For a "pure IP company," a 70% increase in PP&E is unusual. This could reflect: ①Construction of Phoenix chip design labs ②New office facilities (data shows capital leases of $316M) ③Investment in server/computing infrastructure. This suggests ARM is transitioning from a "light-asset IP model" to a "heavy-asset design model."

18.3 5-Year Longitudinal Deep Dive into Cash Flow

Metric($M)	FY2021	FY2022	FY2023	FY2024	FY2025	Trend
Net Income	388	549	524	306	792	Volatile
D&A	201	185	170	162	183	Stable
SBC	54	26	79	1,037	0*	*Data Error
Deferred Tax	-33	-76	-34	-273	-218	Highly Volatile
Change in Working Capital	1,010	-188	-73	-195	-1,465	⚠️ Deterioration
OCF	1,233	458	739	1,090	397	Extremely Volatile
CapEx	-165	-75	-93	-143	-219	Increasing
FCF	1,068	383	646	947	178	FY25 Plummeted
OCF/NI	3.18	0.83	1.41	3.56	0.50	⚠️ Unstable
FCF/NI	2.75	0.70	1.23	3.09	0.22	⚠️ Highly Unstable

Breakdown of FY2025 OCF Plunge:

ARM's FY2025 OCF was only $397M (vs FY2024 $1,090M, -64%). Breakdown of reasons:

Item	FY2024	FY2025	Change	Explanation
Net Income	306	792	+486	Positive
D&A	162	183	+21	Normal
SBC	1,037	0(Error)	-1,037	FMP Data Missing
Deferred Tax	-273	-218	+55	Slightly Better
Other Non-Cash Items	53	1,105	+1,052	May include SBC
Working Capital	-195	-1,465	-1,270	Main Culprit
OCF	1,090	397	-693	Plunge

SBC Data Correction: FMP reported FY2025 SBC=$0, but "Other Non-Cash Items" of $1,105M is unusually high (FY2024 was only $53M). It is highly probable that SBC was misclassified under "Other Non-Cash Items." If SBC≈$800-900M, then the remaining $200-300M in "Other Non-Cash Items" would come from other sources.

Breakdown of Working Capital -$1,465M:

• Increase in Accounts Receivable: -$743M (AR from $1,117M→$1,749M)
• Other Working Capital: -$722M (possibly including increase in contract assets, prepayments, etc.)

Diagnosis: The core reason for the sharp decline in FY2025 OCF is the deterioration of working capital (mainly a surge in accounts receivable). This could be:

1. One-time event: Large year-end licensing contracts (revenue recognized but not yet collected) → Should rebound strongly in FY2026
2. Trend of deterioration: Aggressive revenue recognition (revenue recognized early but cash collection delayed) → Persistent decline in earnings quality

TTM OCF=$1,523M Suggests it Might be a One-Time Event: OCF performance for Q1-Q3 FY2026 (estimated at approximately $1,126M) offset the weakness of FY2025. However, at least two quarters are needed to confirm the trend.

18.4 Accounts Receivable Forensic Analysis: ARM's Biggest Financial "Mystery"

ARM's Accounts Receivable (AR) surged to $1,749M (+57% YoY) in FY2025, and DSO reached 173 days—2-5 times that of peers. This anomaly warrants a deeper dive:

AR Growth vs. Revenue Growth Comparison:

FY	Revenue Growth	AR Growth	AR Growth/Revenue Growth	DSO	Signal
FY2022	+35.0%	+14.7%	0.42x	91	🟢 Healthy
FY2023	+0.8%	+11.3%	14.1x	102	🟡 Slight Disconnect
FY2024	+20.7%	+24.3%	1.17x	138	🟡 Marginal Deterioration
FY2025	+23.9%	+56.6%	2.37x	173	🔴 Significant Disconnect

AR growth rate is 2.37 times revenue growth—this is highly unusual in the IP licensing industry. Normally, AR and revenue should grow roughly in sync (ratio close to 1.0x).

Possible Reasonable Explanations:

1. Large Contract Timing Mismatch (60% probability): ARM signed a large multi-year licensing contract in FY2025 Q4 (December-March). IFRS 15 requires recognition of the full licensing revenue upon signing (if "distinctly identifiable"), but cash is collected over 3-5 years. This leads to immediate revenue recognition but a significant increase in AR. If true, AR should significantly decrease in FY2026 Q1-Q2 (upon receipt of initial contract payments).

2. Customer Payment Delays (25% probability): Some major customers (possibly Chinese clients or those negotiating renewals) have delayed payments. ARM's top 5 customers account for 50%+ of revenue, and a single customer delay can significantly push up DSO.

3. Aggressive Revenue Recognition (15% probability): ARM may have adopted a more aggressive stance on revenue recognition judgments—for example, recognizing revenue that should have been recognized in installments upfront upon contract signing. In this scenario, AR growth could be a "false" signal of revenue quality.

Cross-Verification: Inverse Movement of Deferred Revenue:

Metric	FY2023	FY2024	FY2025	Trend
Deferred Revenue (Total)	$1,183M	$1,122M	$911M	-23%
AR (Total)	$899M	$1,117M	$1,749M	+95%
AR/Deferred Revenue Ratio	0.76x	1.00x	1.92x	⚠️ Reversal

The AR/Deferred Revenue ratio soaring from 0.76x to 1.92x is a strong signal. In a healthy IP licensing model, AR and deferred revenue should be roughly balanced (contract signing → AR increases; installment revenue recognition → deferred revenue increases). When AR surges while deferred revenue declines, it implies ARM is more aggressively "front-loading" revenue recognition.

Investment Implications of Abnormal DSO:

1. If it is a timing factor (contract cadence): FY2026 DSO should fall back to 120-140 days, OCF strongly recovers → Buy signal
2. If it is a structural factor (increased customer bargaining power): DSO may persist at 150-170 days → ARM's royalty collection capability might be eroded.
3. If it is aggressive recognition (accounting quality issue): Potential future revenue restatement risk → Major red flag

Monitor Key Statistics: Changes in DSO and deferred revenue in the FY2026 Q3/Q4 10-K are the most critical data points. If FY2026 DSO does not fall below 140 days, the confidence level for H1 (triple distortion of earnings quality) should be raised from 65% to 75%+.

18.5 SBC: ARM's Most Underestimated Shareholder Cost

ARM's SBC is key to understanding its true profitability—and is also the most easily misunderstood:

SBC Triple Cross-Verification:

Method	FY2024 SBC	FY2025E SBC	FY2026E SBC
Method 1: Direct Data	$1,037M (FMP)	$0M (FMP, Error)	—
Method 2: Non-GAAP Adjustment	~$880M (GAAP/Non-GAAP Diff × Estimate)	~$700-900M	~$800-950M
Method 3: Share Dilution	~$750-1000M	~$600-800M	—
Combined Estimate	$880-1,037M	$700-900M	$800-950M

FY2024 SBC=$1,037M Breakdown:

• First full fiscal year post-IPO (Sept 2023), includes one-time accelerated amortization of RSU grants
• Represents 32.1% of revenue — one of the highest levels among all semiconductor/IP companies
• Comparison: Synopsys FY2024 SBC/Revenue ≈ 7%, Cadence ≈ 9%, Qualcomm ≈ 6%

SBC Post-IPO Decay Curve: Based on peer post-IPO SBC trajectories, ARM's SBC normalization path may be as follows:

Year	SBC Estimate	SBC/Revenue	Drivers
FY2024	$1,037M	32.1%	First year post-IPO, one-time RSU acceleration
FY2025	$700-900M (Est)	17-22%	One-time effects fade, but ongoing grants
FY2026E	$800-950M	16-19%	New grants begin to amortize
FY2027E	$600-750M	10-13%	IPO batch fully amortized
FY2028E	$500-650M	7-9%	Approaching steady state (aligned with peers)

SBC Normalization is Highly Certain: IPO batch RSUs have a clear vesting period (typically 4 years), and will be largely fully amortized by FY2028 (5 years post-IPO). This means that for FY2028+, SBC/Revenue will decrease to 7-9% (aligned with Synopsys/Cadence).

Impact of SBC Normalization on P/E:

• FY2028E Revenue: ~$7.1B (Consensus)
• FY2028E SBC: $550M (Steady State, 8% Revenue)
• vs FY2024 SBC: $1,037M → GAAP Profit Increase of $487M
• FY2028E GAAP NI: ~$2.97B (Consensus) → P/E: ~46x (vs Current 142x)

Conclusion: SBC normalization will transform ARM from "extremely expensive on a GAAP basis" to "justifiable on a Non-GAAP basis". Among the bull arguments, this is one of the most certain ones — no assumptions are needed to confirm SBC will normalize to a steady state within 3 years. This is the "strongest bull argument" we acknowledged in Ch16.9.

Data Correction for FY2025 SBC=$0:

• FMP CashFlow reports Stock-Based Compensation=0, but "otherNonCashItems" surged from $53M to $1,105M
• Highly likely FMP reclassified SBC into "otherNonCashItems"
• SBC should be separately disclosed in 20-F/Annual Report, but FMP API failed to extract it correctly
• Revised Estimate: FY2025 SBC ≈ $700-900M (IPO one-time effects subside)

Impact of SBC on ARM Valuation:

Valuation Method	Without SBC Adjustment	After SBC Adjustment	Impact
GAAP P/E	142x	N/A	Benchmark
Non-GAAP P/E	~65x	N/A	Looks Reasonable
True FCF Yield	—	~0.5-0.8%	Extremely Low
True P/E After SBC Adjustment	—	~95-110x	Still Extremely Expensive

True FCF Calculation: OCF $397M - CapEx $219M = FCF $178M. However, this includes SBC (non-cash but a true cost). If we treat SBC as a "true cash cost" (because it is part of employee compensation, replacing cash wages):

• True FCF = FCF + SBC (since OCF already adds back SBC) - SBC (as a true cost) = $178M
• However, FMP actually reports SBC=0, so SBC might not have been added back in OCF: True FCF ≈ $178M + $800M (SBC added back) - $800M (deducting true cost) = $178M
• FCF Yield = $178M / $136.1B = 0.13% — Extremely Low

Even using TTM FCF of $970M (more normalized): Yield = 0.71% — Still extremely low.

18.6 Peer Comparison Analysis (4 Companies)

Logic for Selecting Peer Companies:

• Synopsys (SNPS): Semiconductor IP/EDA company, high-margin, partially similar IP licensing model
• Cadence (CDNS): Semiconductor IP/EDA company, similar to SNPS
• Qualcomm QTL: Qualcomm's patent licensing division, most similar to ARM's "IP licensing" model
• (Note: Qualcomm's overall data, QTL accounts for approx. 20-25% of revenue)

Metric (Latest FY)	ARM (FY2025)	SNPS (FY2025)	CDNS (CY2025)	QCOM (FY2025)
Revenue ($B)	4.01	—	—	—
Gross Margin	94.9%	77.0%	85.6%	55.4%
Operating Margin (GAAP)	20.6%	13.0%	31.1%	27.9%
Net Margin	19.8%	18.9%	20.9%	12.5%
R&D/Revenue	50.1%	—	—	—
SBC/Revenue	~18-25%	~7%	~9%	~6%
DSO (Days)	173	78	65	36
OCF/NI	0.50	—	—	—
P/E (TTM)	142x	55x	76x	33x
P/S	28.0x	10.3x	16.0x	4.1x
ROE	11.6%	—	—	—

ARM's 4 "Worst" Metrics in Peer Comparison:

1. Highest SBC/Revenue (18-25% vs. peers 6-9%): ARM's SBC ratio is 2-4x that of its peers. This reflects IPO legacy effects (significant RSUs) and intense competition for high-wage talent in the UK/Cambridge. It is expected to gradually decline to 12-15% after FY2027, but will still remain higher than peers.

2. Longest DSO (173 days vs. peers 36-78 days): ARM's accounts receivable collection speed is 2-5x slower than its peers. This is a structural characteristic of ARM's model (large multi-year licensing agreements + China receivables), which is difficult to improve in the short term.

3. Highest P/E (142x vs. peers 33-76x): Even compared to Cadence (76x), the most expensive in the industry, ARM is 1.87 times more expensive. The premium the market pays for ARM is not an "IP model premium" (which would favor SNPS/CDNS), but rather a "Growth+AI+DC" premium.

4. Lowest and Most Volatile OCF/NI (0.50x, historical range 0.22x-3.56x): ARM's cash flow conversion rate is highly unstable and significantly below 1.0 in FY2025. Peers' OCF/NI is typically stable in the 1.0-1.5 range.

18.7 Overall Earnings Quality Score

Dimension	Score (1-10)	Weight	Weighted	Reason
Revenue Quality	5	25%	1.25	Related-party transactions 30%+, surge in AR, extremely high DSO, decline in deferred revenue
Profit Quality	4	25%	1.00	Largest GAAP/Non-GAAP gap (2.6x)+SBC data missing+non-recurring item volatility
Cash Flow Quality	5	25%	1.25	FY25 OCF/NI=0.50+Working Capital -$1.5B+CapEx increase
Balance Sheet	8	25%	2.00	Net cash $1.7B+zero leverage+strong liquidity (5.2x CR)
Overall Earnings Quality	—	—	5.5/10	Relatively Low — Inconsistent with the image of 95% gross margin and 142x P/E

H1 Hypothesis Validation Progress: "ARM's earnings quality is 50% worse than it appears"

Data Support:

• GAAP operating profit is only 38% of Non-GAAP (FY2025: $827M vs. ~$1,500M Non-GAAP estimate)
• FY2025 OCF/NI is only 0.50 (only $0.50 cash generated per $1 profit)
• AR increase / Revenue increase = 743/774 = 96% → 96% of revenue growth "locked" in receivables
• SBC accounts for 15-25% of revenue, 2-4x that of peers
• DSO of 173 days is 2-5x that of peers

H1 Confidence Update: Initial 40% → 65% (Upgraded from 60%, further supported by new data: the 96% AR/revenue growth ratio and the decline in deferred revenue)

18.8 In-depth Audit of Revenue Recognition Policies

ARM's revenue recognition policy (IFRS 15) is key to understanding its earnings quality:

Differences in Recognition for Two Revenue Types:

Revenue Type	Recognition Point	Cash Collection Point	Time Lag	Quality Assessment
Royalty Revenue (~60%)	1-2 quarters after customer reports shipments	Customer report + payment terms	1-3Q	Higher: Linked to actual shipments
Licensing Revenue (~40%)	Contract signing / milestone achievement	Installment payments (2-5 years)	2-5 years	Lower: Can be recognized upfront

Key Audit Findings:

1. Risk of "Upfront Recognition" for Licensing Revenue: Under IFRS 15, ARM can recognize "distinct licensing revenue" at the time of contract signing (even if the customer pays over 5 years). This means ARM could sign a large TLA contract in Q4, immediately recognize all licensing revenue, but it would take 5 years to collect all the cash. The unusual surge in FY2025 Q4 revenue to $1,241M (vs. Q3 $983M, +26% QoQ) may partly stem from this.

2. SoftBank Related-Party Transactions: Are ARM's licensing transactions with SoftBank and its portfolio companies (e.g., WeWork previously) priced at fair value? SoftBank holds a 90.6% stake in ARM and could also be an ARM licensing customer (or indirectly procure ARM licenses through its portfolio companies). The estimated ~30% of revenue from related-party transactions needs detailed verification in the 20-F.

3. Impact of Contract Modifications: If ARM modifies contract terms during the performance period (e.g., expanding licensing scope or adjusting royalty rates), IFRS 15 may require the modification to be treated as a new contract or an adjustment to an existing contract—both approaches have significant differences in revenue timing. Disclosures on "contract modifications" in ARM's 20-F are limited, making it difficult to assess their impact on revenue timing.

4. Abnormal Quarterly Revenue Distribution: ARM's Q4 (January-March) revenue consistently exceeds other quarters, with FY2025 Q4 accounting for 31.0% of annual revenue. While this may reflect a peak in licensing contract signings at fiscal year-end, it also raises the possibility of "year-end stuffing" (revenue management). Healthy IP companies should have more evenly distributed revenue (approximately 25% per quarter).

5. Two Interpretations of Deferred Revenue Decline:

• Positive Interpretation: ARM is accelerating the recognition of existing contracts (releasing deferred revenue as current revenue), but this is a one-off event—growth will slow once existing contracts are fully recognized.
• Negative Interpretation: The rate of new contract signings is insufficient to replenish the consumption from existing contract recognition → the licensing revenue growth pipeline is narrowing.
• Actual Situation: Likely a mix of both. FY2025 licensing revenue grew +21% while deferred revenue decreased -19%, suggesting ARM is "consuming inventory."

18.9 DuPont Analysis: Why is ARM's ROE So Low?

ARM's ROE (11.6%) is relatively low in the semiconductor/IP industry—intuitively, an IP company with a 95% gross margin and no factories should have extremely high ROE. DuPont analysis reveals the reasons:

Three Reasons for ARM's Low ROE:

1. Extremely Low Asset Turnover (0.45): Of ARM's $8.9B in assets, $1.6B is goodwill (SoftBank acquisition legacy), $1.7B is investments, and $2.8B is cash. These non-operating assets drag down the asset turnover ratio. If non-operating assets are excluded, ARM's "operating asset turnover" is approximately 0.9x—more reasonable but still relatively low (due to significant uncollected receivables).

2. Extremely Low Equity Multiplier (1.31): ARM has almost zero leverage (only $356M total debt vs. $6.8B equity). Low leverage is positive for ARM's credit rating but lowers ROE. If ARM used moderate leverage (2x equity multiplier, similar to QCOM), ROE could increase from 11.6% to ~18%.

3. Net Profit Margin Depressed by SBC (19.8%): True operating profit margin (Non-GAAP ~35-40%) is significantly higher than GAAP. SBC is the biggest drag on ARM's ROE.

ROE Improvement Path: If ARM achieves the following by FY2028:

• Net profit margin expands to 30% (SBC normalization + economies of scale) → +10.2pp
• Asset turnover remains unchanged (0.45)
• Equity multiplier remains unchanged (1.31)
• ROE ≈ 30% × 0.45 × 1.31 = 17.7% (Still lower than QCOM/TSMC/ASML)

Conclusion: Even with margin expansion, ARM's ROE may never match the levels of TSMC/ASML (because ARM's asset-light model implies low asset turnover + low leverage). This has implications for ARM's long-term valuation—low ROE means ARM has limited ability to create value through reinvestment, and growth primarily relies on "price increases" (external) rather than "efficiency" (internal).

18.10 Quarterly Trend Analysis: FY2026 Q1-Q3 Progress

While FMP primarily provides annual data, the difference between TTM data and annual data in checkpoint.yaml allows us to estimate the performance for FY2026 Q1-Q3:

FY2026 Progress Estimation:

Metric	FY2025 Full Year	FY2026E Full Year	Q1-Q3 FY2026 Estimated	Q3 FY2026 (Single Quarter)
Revenue	$4,007M	~$4,906M (Consensus)	~$3,665M	~$1,241M
Royalties	~$2,400M	~$3,200M (Est.)	~$2,400M	~$800M+
Licensing	~$1,607M	~$1,700M (Est.)	~$1,265M	~$440M
Gross Margin	94.9%	~95.2% (Projected)	~95.4%	~95.4%
Operating Profit (Non-GAAP)	~$1,500M (Est.)	~$2,000M (Est.)	~$1,500M	~$505M
Royalties/Chip (ASP)	~$0.065	~$0.075 (Est.)	—	—

Key Tracking Metrics:

• Royalties Growth Rate: FY2025 Royalties +38% (Est.), FY2026E Royalties +33% (Est.) → Decelerating but still strong
• CSS Penetration Rate: Q3 FY2026 CSS as a percentage of new licensing agreements (Undisclosed, but estimated 30-40%)
• DC Revenue Contribution: Rising from ~15% in FY2025 to ~18-20% in FY2026 (Estimated)
• Non-GAAP Operating Margin: Q3 FY2026 ~40.7% (checkpoint data) — Steady expansion

FY2026E Consensus Verification (FMP Estimates):

• Revenue: $4,906M (25 analysts) → vs FY2025 $4,007M = +22.4%
• EPS: $1.75 (21 analysts) → vs FY2025 $0.75 = +133%
• Reasons for significant EPS growth: SBC normalization (IPO first-year effect fading) + Revenue growth + Operating leverage

Consensus FY2026 EPS Implied Net Income: $1.75 × 1,063M shares = $1,860M (vs FY2025 $792M, +135%)

If achieved, the FY2026 P/E will decrease from 142x to ~73x (based on a $128.14 share price). This marks ARM's transition from "extremely expensive" to "expensive but growth-supported" — and is the core logic behind many analysts rating ARM as "Hold/Buy" (looking at forward P/E instead of TTM P/E).

18.11 Altman Z-Score and Financial Safety

Despite ARM's lower earnings quality score (5.5/10), its financial safety is among the highest in the semiconductor industry:

Altman Z-Score Breakdown (FY2025):

Z-Score = 33.9 — Extremely Safe (>3.0 is considered the "safe zone")

Meaning of Z-Score: ARM will not go bankrupt. This sounds obvious (ARM has $2.8B in cash, zero long-term debt), but it has significant investment implications — ARM is not a "potential zero" risk, but rather a "potential fall from $128 to $50" valuation risk. This means:

1. Downside Protection: Even in the most pessimistic scenario (full RISC-V substitution), ARM still has a core business generating $4B+ in revenue (mobile royalties) → Calculated at 15x P/E, the floor is approximately $30-40 (vs. current $128)
2. Downside Path is "Gradual Decline" Not "Collapse": ARM will not suddenly go to zero like Wirecard, but rather like Intel — with continuous valuation compression over 5-10 years, occasional rebounds, but a downward trend.
3. High Cost to Short: Extremely high Z-Score + Strong short-term performance → The "carry" cost of shorting ARM (stock borrowing cost + counter-trend risk) is very high

18.12 Key Financial Figures Python Validation Checklist

The following figures need to be cross-validated using Python:

#	Validation Item	Data Source 1	Data Source 2	Priority
V1	FY2025 SBC Actual Value	ARM 20-F (Original Filing)	Non-GAAP Adjustment Estimation	P0
V2	FY2025 OCF Components	FMP CashFlow	ARM 20-F	P0
V3	5-Year DSO Calculation	FMP Balance+Income	Manual Verification	P1
V4	Reverse DCF CAGR Requirement	Calculated in this Chapter	Python DCF Model	P0
V5	A-Score Weighted Calculation	Calculated in this Chapter	Python Validation	P2

---

Chapter 19: Arm China Separate Valuation — Three Geopolitical Scenarios

19.1 Arm China (Anmou China) Background

Arm China is ARM's exclusive licensing partner in China and one of ARM's most complex assets. In 2022, ARM regained control of Arm China (dismissing co-founder Allen Wu), but Arm China continues to operate as an independent legal entity:

Basic Financials (FY2025 Est.):

• Revenue Contribution: Accounts for ~20-22% of ARM's total revenue ($800-900M)
• Growth Rate: +7.5% YoY (vs. ARM overall +24%)
• Royalties Contribution: Chinese mobile phones (Huawei/OPPO/vivo/Xiaomi) + Chinese IoT + Chinese automotive
• Licensing Contribution: Chinese semiconductor design companies (e.g., Unisoc/Allwinner/ZTE Microelectronics)

19.2 Structural Reasons for Arm China's Growth Deceleration

Arm China's +7.5% growth rate (vs. ARM Group's +24%) is not a temporary slowdown, but rather reflects structural headwinds:

1. China's Policy Leaning Towards RISC-V: Eight Chinese government agencies are promoting RISC-V adoption as part of the "indigenous and controllable" strategy. Geopolitical motivations (reducing reliance on ARM/US technology) + Economic rationale (RISC-V is free) = Dual drivers.

2. Huawei Restriction Effect: Huawei HiSilicon (once one of ARM's largest clients in China) has shifted towards in-house development + alternative architectures due to US sanctions. While Huawei still uses ARM (through existing licenses), new designs are increasingly favoring RISC-V or hybrid architectures.

3. Chinese Semiconductor "Going Global" Restricted: Exports by Chinese chip design companies are subject to US technology control restrictions, indirectly limiting the growth of these companies (and the royalties ARM receives from them).

4. Arm China Governance Legacy Issues: Following the Allen Wu incident in 2022, some Chinese customers remain cautious about Arm China's stability.

19.3 Arm China Three-Path Geopolitical Scenario Analysis

Detailed Analysis of Three Scenarios:

Scenario 1: Détente (20% Probability)

Assumption: US-China relations significantly improve in 2027-2028, technology restrictions ease, and Huawei regains access to ARM's advanced core licenses.

Impact:

• Arm China growth recovers to ~15% CAGR (aligning with ARM Group)
• FY2030 Arm China revenue of $1.5-1.8B (vs. base case $1.1B)
• Incremental $400-700M/year × 30x → +$12-21B market capitalization impact
• However, this scenario currently has a low probability: US-China tech competition shows a trend of structural intensification.

Scenario 2: Stalemate (55% Probability)

Assumption: The current state persists—controls neither intensify nor relax, Arm China continues to operate independently, and RISC-V gradually penetrates the Chinese market.

Impact:

• Arm China growth ~5-8% CAGR (continuation of current trend)
• FY2030 Arm China revenue of $1.1-1.3B
• Largely priced in by the market
• RISC-V Penetration in China: China may become the fastest-growing market for RISC-V penetration globally (driven by both policy and economics).
• Arm China's market share in China will slowly yield to RISC-V (from ~80% → 65-70%).

Scenario 3: Decoupling (25% Probability)

Assumption: US-China tech decoupling intensifies—new export controls restrict ARM from providing advanced cores (v9+/CSS) to China, or China proactively shifts massively to RISC-V.

Impact:

• Arm China revenue stagnates or declines (-5%~0% CAGR)
• FY2030 Arm China revenue of $0.6-0.85B (vs. current $0.85B)
• Revenue shortfall of $250-500M/year × 30x → -$7.5-15B market capitalization impact
• Extreme Scenario: If China fully shifts to RISC-V (probability <10%), Arm China's revenue could shrink to $300-500M after 2030 → market capitalization impact -$15-25B.

Trigger Factors to Monitor:

1. Whether the U.S. Department of Commerce lists ARM architecture under export controls (currently not restricted)
2. Whether China introduces mandatory RISC-V procurement policies (currently encouragement/recommendation)
3. Whether Huawei HiSilicon releases fully self-developed RISC-V SoCs (currently unconfirmed)

19.4 Arm China Standalone Valuation

Valuing Arm China as a standalone business:

Method	Valuation Range	Assumption
Revenue Multiple (P/S)	$8.5-17B	$850M × 10-20x (IP company P/S range)
EV/EBIT	$5-10B	Estimated EBIT $200-400M × 25x
DCF (Base Case Scenario)	$7-12B	8% CAGR + 25% Margin + 25x Terminal Multiple
Composite	$7-13B	—

Arm China's Share of ARM Total Market Cap: $7-13B / $136.1B = 5-10%

Implication: Arm China's share of ARM's total market capitalization (5-10%) is significantly lower than its revenue share (20-22%). This reflects a "geopolitical discount" already applied by the market to the China business. However, if Scenario 3 (decoupling) occurs, this discount may prove insufficient—Arm China's valuation could drop from $7-13B to $3-7B, impacting ARM's total market capitalization by approximately $4-10B (-3~7%).

19.5 Arm China Governance Risks and Aftermath of the Allen Wu Incident

Recap of the 2022 Allen Wu Incident:

• Allen Wu (Wu Xiongang), co-founder and CEO of Arm China, was removed from his position by the ARM board in 2020.
• However, Wu refused to resign and leveraged the Chinese legal system and Arm China's independent legal entity status to legally obstruct ARM's exercise of control.
• It was not until 2022 that ARM regained control of Arm China through a series of legal and corporate governance actions.
• This incident exposed fundamental flaws in Arm China's governance structure.

Aftermath:

1. Customer Trust Erosion: Some Chinese customers remain cautious about Arm China's stability, concerned that ARM headquarters might restrict licenses for Chinese customers due to geopolitical reasons.
2. Employee Turnover: Arm China experienced talent drain during the incident, with some engineers moving to Chinese RISC-V startups (e.g., T-Head/Nuclei System Technology).
3. Contract Scrutiny: Following the incident, ARM increased its financial and operational scrutiny of Arm China, which could lead to delays in contract approvals.
4. SoftBank's Role: SoftBank, as a common shareholder of both ARM and Arm China, had a complex role in the incident—its investment interests in China might conflict with ARM's business interests.

Governance Risk Assessment: Although the Allen Wu incident has been resolved, it exposed a structural issue—ARM's control over its China operations is not as direct and certain as it is in other regions. Amid intensifying geopolitical tensions, this governance uncertainty could resurface.

---

Chapter 20: SoftBank Related-Party Transaction Segregation and Governance Risks

20.1 SoftBank-ARM Relationship Overview

SoftBank Group (9984.T) holds approximately 90.6% of ARM's shares through Kronos II LLC. This control structure not only impacts corporate governance but also directly affects ARM's revenue quality and valuation framework. SoftBank acquired ARM in 2016 for $32B (at a 43% premium at the time), and its current equity value is approximately $123B (+285%).

Control Structure Overview:

SoftBank Shareholding Timeline:

Date	Event	Ownership Change	ARM Market Cap
2016-07	SoftBank acquires ARM (£24.3B)	100%	$32B
2016-2023	SoftBank wholly owns (ARM delists)	100%	—
2023-09	ARM IPO (NASDAQ)	~96.3%	$65B
2024-03	IPO Lock-up Expiration (Partial)	~92%	$130B
2025-01	Minor Sell-down + Employee Stock Dilution	~90.6%	$136B
2026-02	Current	90.6%	$136B

Key Figures: SoftBank holds 90.6% of ARM, corresponding to a stake value of approximately $123B. This makes ARM approximately 54.6% of SoftBank's NAV (Net Asset Value) – ARM is SoftBank's largest single asset, significantly exceeding any other investment.

20.2 Double Exemptions: Structural Flaws in the Governance Framework

ARM enjoys two key exemptions that allow it to bypass core safeguards of US public company governance:

Exemption 1: Controlled Company (Nasdaq Rule 5615)

As SoftBank holds >50% ownership, ARM qualifies for the "Controlled Company" exemption:

• Exempt from independent director majority requirement → Only 4 of ARM's 9 board members are independent
• Exempt from independent compensation committee requirement → CEO compensation is substantially influenced by SoftBank
• Exempt from independent nomination committee requirement → Director nominations are dominated by SoftBank

Exemption 2: Foreign Private Issuer (FPI)

As a UK-registered company, ARM enjoys the FPI exemption:

• Exempt from Section 16 insider trading disclosure → SoftBank's transactions in ARM shares do not require prior public announcement
• Exempt from proxy solicitation rules → Public shareholders' voting rights are effectively rendered meaningless
• Exempt from detailed quarterly 10-Q disclosures → Only semi-annual reports (6-K) are required, leading to lower information transparency

The Compounded Effect of Dual Exemptions:

In the US equity market, it is extremely rare for large-cap companies to simultaneously enjoy both Controlled Company and FPI exemptions. This means ARM's governance framework effectively resembles an "unconstrained controlling shareholder model" – SoftBank can influence ARM's strategic decisions, revenue recognition, and related-party transactions with virtually no external checks and balances.

Governance Benchmark:

Company	Controlling Structure	Exemptions	P/E	Governance Discount	Information Disclosure
ARM	SoftBank 90.6%	Dual (CC+FPI)	142x	0%	Semi-annual + FPI
Alphabet (GOOGL)	Page/Brin ~51% Voting Power	CC Only	24x	5-10%	Standard 10-Q
Meta (META)	Zuckerberg ~60% Voting Power	CC Only	28x	5-15% (Historical)	Standard 10-Q
Alibaba (BABA)	SoftBank→Exited 2024	FPI Only	12x	30-50%	20-F Only
Sea (SE)	Tencent formerly held stake	FPI Only	25x	10-20%	20-F Only

Key Insight: ARM's 142x P/E completely disregards any governance risk discount. In other companies with similar controlling structures, the market typically applies a 10-30% governance discount. ARM's zero governance discount could mean: (a) the market ignores this risk, or (b) the market perceives SoftBank's presence as positive. However, given SoftBank's own financial pressures (see Section 18.6), assumption (b) is becoming increasingly untenable.

20.3 Related-Party Transaction Segregation: "True" Licensing Revenue

This is the core analysis for CQ-3 – what is ARM's "organic" licensing growth if licensing revenue from the SoftBank ecosystem is excluded?

Identification of SoftBank Related Licensing:

ARM's annual report (20-F) discloses transactions with SoftBank and its affiliates, but the granularity of disclosure is insufficient for precise segregation. Based on BofA research and ARM's IPO prospectus information:

Source	Related-Party Licensing Revenue Share	FY2025 Estimated Amount	FY2026E Amount	Trend
SoftBank Direct (e.g., SB Arm Ltd)	~10-15%	$180-275M	$200-300M	Growth
Vision Fund Portfolio Companies	~10-15%	$180-275M	$150-250M	Stable/Decline
SoftBank Strategic Investments	~5-10%	$90-180M	$80-150M	Decline
Total	~25-35%	$450-730M	$430-700M	Slight Decline

BofA's Key Findings (2025 Downgrade Report):

• Excluding SoftBank related licensing, FY2026 licensing revenue is projected to decline by ~5%
• ARM's licensing revenue growth is primarily driven by the SoftBank ecosystem
• Non-related party licensing growth may only be in the single digits

Segregated "True" Licensing Revenue:

Metric	Reported Value (FY2025)	Excluding Related Parties	Difference	FY2026E (Reported)	FY2026E (Excluding Related Parties)
Licensing Revenue	$1,839M	$1,109-1,389M	-24~40%	~$2,100M	~$1,350-1,600M
Licensing YoY Growth	+28.7%	+5~15%	-14~24pp	+14%	+2~8%
Licensing/Total Revenue	45.9%	28-35%	-11~18pp	~43%	~28-33%

Three-Layer Impact Quantification:

1. Revenue Quality Discount: Related-party transaction licensing receives a 10-30% quality discount → Effective licensing revenue decreases by $100-200M/year
2. Growth Rate Downgrade: Organic licensing growth reduced from +29% to +5-15% → Overall revenue growth rate reduced from +24% to +15-20%
3. Sustainability Risk: If SoftBank reduces related-party procurement (due to funding constraints), licensing revenue might experience negative growth

The "Circular" Problem of Related-Party Transactions:

There is a subtle but significant cycle here: SoftBank holds 90.6% of ARM shares → SoftBank has an incentive for ARM to perform well (stock price) → SoftBank's subsidiaries might procure ARM licenses at above-market prices → ARM's licensing revenue growth "looks" good → ARM's stock price increases → SoftBank's stake value increases. This cycle is virtuous when SoftBank is not short on cash (SoftBank is willing to "subsidize" ARM), but when SoftBank faces liquidity pressure (see Section 18.6), the cycle might reverse—SoftBank reduces related-party procurement → ARM licensing growth slows → ARM stock price falls → SoftBank faces greater pressure.

Combined Impact: $14-27B EV Impairment (10-20% of current EV)

20.4 SoftBank Margin Loans: The Sword of Damocles

Margin Loan Structure:

Parameter	Value	Source	Update Time
Loan Facility Size	$20B	SoftBank FY2024 Annual Report	2025-03
Amount Drawn	$8.5B (as of Sep 2025)	SoftBank H1 Report	2025-11
Collateral	ARM Shares (partial)	SoftBank Disclosure	2025-11
Participating Banks	33	Media Report	2024-12
LTV Ratio (Est.)	~40-50% (Industry Standard)	Analyst Estimate	—
Interest Rate	SOFR + 150-200bps Est.	Industry Standard	—
Annualized Interest Cost	~$400-500M (at current drawn level)	Derived	—

Margin Call Trigger Price Calculation:

20.5 SoftBank Governance Risk: Three-Layer Valuation Framework

Risk Layer	Description	Probability	EV Impact	Expected Value
L1: Related-Party Transaction Discount	25-35% of licensing revenue from related parties	80%	-15%	-12%
L2: Sell-Down Overhang	SoftBank orderly sell-down over the next 3-5 years	40%	-20%	-8%
L3: Margin Call Crisis	Full amount drawn + ARM crash → Forced liquidation	5%	-50%	-2.5%
Total	—	—	—	-22.5%

20.6 SoftBank Leverage Deep Dive: $250B+ Debt Ecosystem

SoftBank's financial health directly dictates its actions regarding ARM. Below is a comprehensive overview of SoftBank Group's debt:

SoftBank Group Debt Structure (Sep 2025, FY2025 H1):

Debt Type	Amount ($B)	Interest Rate Range	Maturity Distribution	Collateral
Corporate Bonds (JPY/USD)	~$45B	2.5-5.5%	2025-2034 Rolling	Unsecured
ARM Margin Loan	$8.5B (drawn)	SOFR+~175bps	Revolving	ARM Shares
Other Margin Loans (SVF etc.)	~$15B est.	Various	Various	SVF Portfolio
Project Izanagi Financing	Planned $100B+	To be determined	2025-2030	To be determined
Stargate Commitment	Planned $100-500B (share unknown)	To be determined	2025-2029	To be determined
Bank Loans + Others	~$20B	Various	Various	Various
Total Explicit Debt	~$90B	—	—	—
Planned Funding Needs	$200-400B	—	Within 5 years	—

SoftBank's Funding Sources vs. Funding Needs:

Key Leverage Ratios:

Indicator	Value	Assessment
Total Debt/NAV	~45-50%	Moderately High
ARM % of NAV	54.6%	Extremely Concentrated
Debt Interest/Operating CF	~100%	⚠️ Barely Covered
5-Year Funding Gap	$100-200B	Far Exceeds Available Assets
If ARM Drops 50%: NAV Decrease	$62B(-38%)	Severely Weakened

SoftBank's "ARM Dependency Syndrome":

SoftBank's financial health is highly tied to ARM's stock price:

1. ARM accounts for 54.6% of NAV → A decline in ARM directly weakens SoftBank's credit.
2. ARM is the largest source of liquidity → Large investments (Izanagi/Stargate) require ARM monetization.
3. ARM margin loans → A decline in ARM could trigger margin calls.
4. SoftBank corporate bond credit → Rating agencies focus on the value of ARM holdings.

Vicious Cycle Risk: ARM decline → SoftBank NAV deterioration → Credit rating downgrade risk → Increased debt financing costs → SoftBank forced to reduce ARM holdings → ARM further decline. This spiral could become self-reinforcing when ARM falls 30%+.

20.7 Margin Call Trigger Scenario Matrix

Extending the calculation from 18.4 to a systematic trigger matrix — margin call trigger prices at different withdrawal levels and collateral ratios:

Trigger Matrix: Critical ARM Share Prices for Margin Call Triggers

Withdrawal Amount↓ \ Collateral Ratio→	20% Holding Pledged	30% Holding Pledged	50% Holding Pledged	100% Holding Pledged
$8.5B (Current)	$78	$52	$31	$16
$12B	$110	$73	$44	$22
$15B	⚠️$137	$92	$55	$27
$20B (Full Amount)	⚠️$183	⚠️$122	$73	$37

Key Findings:

1. Current $8.5B Withdrawal + 30% Collateral: Trigger price $52 → ARM needs to drop 59% → Distant and Safe
2. $15B Withdrawal + 20% Collateral: Trigger price $137 → ARM's current price of $128 is already below the trigger! → ⚠️If SoftBank simultaneously increases withdrawal and reduces collateral, the current price is already unsafe.
3. $20B Full Withdrawal + 30% Collateral: Trigger price $122 → ARM only needs to drop 5% → Extremely Dangerous

Most Likely Path: To fund Izanagi/Stargate, SoftBank will gradually increase margin loan withdrawals (from $8.5B → $12-15B) + potentially pledge additional ARM shares as collateral (from 20% → 30-50%). During this process, the margin call trigger point will gradually approach from being "distant".

Signals Investors Should Monitor:

• Changes in margin loan balance in SoftBank's quarterly reports
• ARM block trade announcements
• Changes in SoftBank's credit ratings (S&P: BB+, Moody's: Ba3)
• Other asset sales by SoftBank (e.g., T-Mobile stake reduction)

20.8 SoftBank-ARM Divestment Timeline & Market Impact Model

Evolution of Divestment Probability Over Time:

Time Window	Divestment Probability	Divestment Scale	Divestment Method	ARM Share Price Impact
0-6 Months	5%	<2%	Increased margin loans	-2~5%
6-12 Months	10%	2-5%	Placement/Margin loans	-5~10%
1-2 Years	25%	5-10%	Orderly placement	-10~15%
2-3 Years	35%	10-20%	Large placement/Secondary offering	-15~25%
3-5 Years	50%	20-40%	Multiple divestment rounds	-20~30%

Market Impact Model of Divestment:

ARM's current free float is approximately 103M shares (9.7% of total shares 1,063M). SoftBank holds 961M shares.

Divestment Scale	Shares Released	% Increase in Free Float	Supply Shock	Equilibrium Discount
2%	~19M	+18%	Moderate	-3~5%
5%	~48M	+47%	Significant	-8~12%
10%	~96M	+93%	Large	-15~20%
20%	~192M	+186%	Massive	-25~35%

Key Insight: ARM's current free float is extremely small (9.7%). A 5% divestment by SoftBank would increase the free float by 47% — such a supply shock has historically led to a short-term discount of 8-12%. If the divestment scale reaches 20% (SoftBank would still hold 70%+), the shares in circulation would nearly triple, and the medium-term discount could reach 25-35%.

20.9 Historical Analogy: Mirroring SoftBank-Alibaba Divestment Path

SoftBank's divestment history with Alibaba (2016-2024) is the most important historical reference for understanding ARM's future — same controlling shareholder, same motivation (debt service), same structure (super-major shareholder divesting shares with low free float).

Alibaba Divestment Timeline:

Year	SoftBank's BABA Holding	Action	Amount	Impact on BABA Share Price
2016	~32%	Began small-scale divestment	$10B Placement	-5~8%
2017-2019	~26%	Phased Sales + Prepaid Forwards	~$14B Cumulative	Moderate (-3~5% each time)
2020 Q4	~24%	Accelerated after Ant IPO cancellation	$9.5B	-10~15%
2021 Q1-Q2	~20%	Global Antitrust + China Regulation	$10B	-15%
2022 Q2	~14%	Sold via ADSs + Forwards	$22B	-20%+
2022 Q3	~3.6%	Massive Liquidation	$14B	-25~30%
2023 Q2	0%	Full liquidation completed	Remaining $7.2B	Minor impact (already priced in)
Total	32%→0%	Exited completely over 8 years	~$87B Cumulative	BABA -65% during period (from peak)

Mapping from Alibaba to ARM:

Dimension	BABA Divestment	ARM (Expected Path)	Differentiating Impact
Initial Holding	32%	90.6%	ARM's free float is scarcer, impact is greater
Reason for Divestment	Debt + WeWork Losses	Debt + Izanagi + Stargate	ARM faces greater pressure (higher capital needs)
Regulatory Catalyst	China Antitrust	No direct catalyst (but AI bubble?)	ARM might be more at ease
Share Price Trajectory	Declined throughout divestment (-65%)	Unknown	Depends on fundamentals
Change in Free Float	From 22%→100% (Hedge fund accumulation)	From 9.7%→?	ARM's starting point is more vulnerable
Market Liquidity	BABA average daily $3-4B	ARM average daily $0.8-1.2B	ARM has lower liquidity, same divestment causes greater impact

Three Key Lessons from BABA's Experience:

Lesson 1: "Orderly Divestment" Eventually Turned into "Accelerated Liquidation"
SoftBank's divestment of BABA, starting in 2016, was initially a "strategic fine-tuning," which evolved into a "panic liquidation" by 2022. The turning point was the cancellation of the Ant IPO in 2020—SoftBank realized BABA's "certainty premium" had vanished, shifting its decision from "selling at the optimal time" to "monetizing as quickly as possible." For ARM, a similar turning point could be: an AI bubble correction (-30%+) causing ARM's market capitalization to fall below SoftBank's loan collateral requirements, triggering the same "orderly to panic" switch.

Lesson 2: The Self-Fulfilling Spiral of Divestment
BABA fell from $320 (October 2020 peak) to $58 (October 2022 trough)—SoftBank's divestment was one of three major factors (the other two being regulation + fundamentals). Crucially, SoftBank's divestment accelerated BABA's valuation discount (ECM placement → depressed prices → SoftBank received less cash → needed to sell more shares). ARM's free float is lower (9.7% vs BABA's 22%), so this spiral effect could be more severe:

Lesson 3: Governance Valuation Reappraisal due to Decreased Controlling Stake
When SoftBank's stake in BABA decreased from 32% to 20%, the market began to assign BABA a "SoftBank governance discount removal premium"—losing control was seen as a positive. ARM's situation is the opposite: SoftBank reducing its stake from 90%→70% would not change control, only increase free float + liquidity (potentially positive); however, a reduction from 90%→below 50% would trigger significant changes in the governance structure (when would dual-class voting rights be triggered?), which could be a double-edged sword for ARM.

Key Differences between BABA and ARM—Why ARM Could Be Better or Worse:

Factor	Favorable for ARM (vs BABA)	Unfavorable for ARM (vs BABA)
Fundamental Quality	ARM growth +24% (vs BABA decline)	ARM 142x P/E (vs BABA 15x)
Divestment Motivation	SoftBank currently has no urgent need	Izanagi/Stargate demand is greater
Free Float	—	ARM Free Float 9.7% (vs BABA 22%)
Regulatory Catalyst	No external catalyst (positive)	No external constraints (SB can sell anytime)
Alternatives	ARM has no direct listed competitors	RISC-V concept companies may divert capital
Market Liquidity	—	ARM average daily $0.8B (vs BABA $3-4B)

Probabilistic Path Model (5 Years):

Path	Description	Probability	Impact on ARM Valuation
A: "BABA Path"	Accelerated Liquidation (90%→30%/5 years)	15%	-25~40%
B: "Orderly Divestment"	Phased Sales (90%→60%/5 years)	35%	-10~15%
C: "Maintain Control"	Minor Adjustments Only (90%→80%)	30%	-3~5%
D: "Izanagi Spin-off"	ARM integrated into Izanagi/Special Purpose Entity	10%	-15~25% (uncertain)
E: "Status Quo"	No Divestment (Locked)	10%	0%
Probability-Weighted	—	—	-9~15%

Cross-Analysis with SoftBank Debt Instruments: SoftBank's current margin loans collateralized by ARM are estimated at $15-25B (based on Section 18.6 analysis). If ARM's share price drops from $128 to $90 (-30%), the LTV of these loans would rise from ~25% to ~36%—nearing the margin call threshold (typically 40%). This is precisely the trigger point for a shift "from orderly to panic."

20.10 Global Benchmarking of SoftBank Group Holding Company Discount

As a mega holding company, how much of a "holding company discount" should ARM bear?

Holding Group	Held Company	Stake	Holding Discount	Catalysts
SoftBank→BABA	Alibaba	32%→0%	30-50%	China Regulation + Stake Sale
SoftBank→Sprint/T-Mobile	T-Mobile	67%→40%→0%	15-25%	Merger + Stake Sale
Samsung Group→Samsung Elec	Samsung Elec	20% (Direct)	20-30%	Korea Governance Discount
Alphabet→Waymo	Waymo	100%	N/A (Not Publicly Traded)	Continued Cash Burn
Berkshire→GEICO/BNSF	Subsidiaries	100%	Premium	Buffett Trust Premium
SoftBank→ARM	ARM	90.6%	?	See Below

ARM Holding Discount Estimate:

• SoftBank credit rating (BB+/Ba3) below investment grade → +5~10% discount
• 90.6% stake = extremely low free float → +5~10% discount (liquidity)
• SoftBank history (Alibaba divestment + WeWork disaster) → +5~8% discount (trust cost)
• Majority of ARM independent directors → -3~5% discount reduction
• Net Holding Discount: 12-23% → i.e., $16-31B should be deducted from market capitalization

Implication: If $16-31B of the $136B market capitalization represents an "implied but not fully reflected holding discount," then ARM's "fundamental fair value" would be $105-120B – highly consistent with our E5 engine's $96-120B range. In other words, the market may have partially reflected the holding discount, but not sufficiently.

---

Chapter 21: Royalty Revenue Reverse Engineering — Terminal TAM Five-Scenario Waterfall Model

21.1 Reverse Engineering Methodology

Royalty revenue is the gravitational center of ARM's business model. Reverse engineering reconstructs royalty revenue from the bottom up (chip shipments × per-chip royalty rate) and projects the upper limit of the terminal TAM.

Basic Formula:

Methodology Upgrade:

• Upgraded from three scenarios (Optimistic/Base/Pessimistic) to a five-scenario waterfall model, adding "H2 CDS Effect Scenario" and "Breakthrough Scenario"
• Each scenario has a detailed end market × version × penetration matrix

21.2 Global Chip Shipments and ARM Penetration — Seven Major End Markets

2025 Base vs 2030 Forecast:

End Market	2025 Shipments (100M units)	ARM Penetration (2025)	2030E Shipments (100M units)	ARM Penetration (2030E)	RISC-V Erosion
Mobile AP/Modem	13-14	~99%	14-15	97-99%	Low
IoT/Embedded	180-200	~60%	220-250	45-55%	High
Data Center CPU	0.5-0.8	~23%	1.0-1.5	25-40%	Medium
Automotive	15-18	~50%	22-28	45-65%	Medium
PC/Laptops	2.5-2.8	~13%	2.8-3.2	18-30%	Low
Networking/Infrastructure	8-10	~40%	10-12	35-50%	Medium
Consumer Electronics (Other)	20-25	~30%	22-28	20-30%	Medium
Weighted Average	~240-280	~50%	~290-340	~43-52%	—

Important Discovery: ARM's total weighted penetration may **slightly decline** from ~50% to 43-52% – IoT penetration is eroded by RISC-V (falling from 60% to 45-55%), partially offsetting growth in DC/PC/Automotive. This is a negative factor overlooked by the market.

21.3 Royalty Growth Driver Decomposition

Royalty growth comes from three multiplicative factors:

Factor 1: Shipments Growth (~5-8% CAGR)

End Market	Shipments CAGR (5-Year)	Drivers	Constraints
Mobile	+2-3%	5G→6G Device Upgrades	Mature Market Saturation
IoT	+8-12%	Smart Devices / Industrial IoT	RISC-V Competition
Data Center	+15-25%	AI/Cloud Computing Boom	Supply Chain / Power Consumption
Automotive	+10-15%	Electrification / ADAS	Semiconductor Cycle
PC	+3-5%	AI PC Upgrade Wave	Market Saturation
Weighted Average	~6-8%	—	—

Factor 2: Penetration Growth (~-1% to +3% CAGR)

End Market	Current ARM Penetration	FY2030E Penetration	Direction	Net Effect
Mobile	99%	97-99%	Slight decrease (RISC-V edge)	-0.5%
IoT	60%	45-55%	Decrease	-3~5%
Data Center	23%	25-40%	Increase	+2~5%
Automotive	50%	45-65%	Uncertain (RISC-V competition)	-1~+3%
PC	13%	18-30%	Increase	+1~3%
Weighted Net Effect	—	—	—	~-1% to +2%

Factor 3: ASP Growth (~8-15% CAGR, Core Driver but H2 Constrained)

Upgrade Path	Royalty Rate Increase	Time Window	Sustainability	H2 Constraint
v8→v9	2x	2023-2027	High (Executed)	Already priced in
TLA→CSS	2-3x	2024-2028	Medium (11 customers signed)	Low constraint
v9→v10	1.0-1.5x	2028+	Low (H2 core)	High constraint
Chip ASP Increase	~5-10%/year	Ongoing	Medium	Indirect

21.4 Royalty End State: Five-Scenario Waterfall Model

Upgraded to Five Scenarios (vs. previous three scenarios) to more accurately reflect the possibility space:

Five Scenarios Detailed:

Scenario 1: Breakthrough (Probability 10%)

Assumptions: v10 significant price increase (2.0x v9) + DC penetration 40%+ + Automotive/PC surge + RISC-V stagnation

End Market	FY2030 Royalty	Driving Assumption
Mobile	$2.5B	v10 full penetration, ASP +30%
DC	$2.5B	40% penetration, CSS $100+/chip
IoT	$1.5B	Share maintained at 55%, v9 penetration
Automotive	$1.0B	ADAS surge, ARM content per vehicle doubles
PC/Other	$1.5B	Snapdragon X dominates AI PC
Total	$9.0B	—

Scenario 2: Optimistic (Probability 20%)

Assumptions: v10 moderate price increase (1.5x v9) + DC penetration 30% + Gradual growth

End Market	FY2030 Royalty	Driving Assumption
Mobile	$2.0B	v9→v10 transition, ASP+20%
DC	$1.8B	30% penetration, increased competition
IoT	$1.0B	Share 50%, v9 penetration
Automotive	$0.8B	Steady growth
PC/Other	$0.9B	Moderate growth
Total	$6.5B	—

Scenario 3: Baseline (Probability 35%) — Moderate CDS Effect

Assumptions: v10 price increase constrained by RISC-V (1.2x v9) + DC penetration 25% + IoT share loss

End Market	FY2030 Royalty	Driving Assumption
Mobile	$1.7B	v9 penetration complete, v10 moderate
DC	$1.2B	25% penetration, balanced competition
IoT	$0.7B	Share drops to 50%, RISC-V erosion
Automotive	$0.5B	Steady but below expectations
PC/Other	$0.6B	Moderate
Total	$4.7B	—

Scenario 4: CDS Dominant (Probability 25%)

Assumptions: v10 unable to significantly increase price (1.0x v9) + DC growth slows + Significant IoT loss

End Market	FY2030 Royalties	Driving Assumptions
Mobile	$1.4B	v9 existing, v10 minimal price increase
DC	$0.8B	15-20% penetration, RISC-V competition
IoT	$0.5B	Share drops to 40%, significant loss
Automotive	$0.4B	Increased RISC-V competition
PC/Other	$0.4B	Low growth
Total	$3.5B	—

Scenario 5: Pessimistic (10% Probability)

Assumptions: Substantial RISC-V replacement + China's full shift to RISC-V + Multiple customer defections

End Market	FY2030 Royalties	Driving Assumptions
Mobile	$1.1B	Share erosion begins (97%→92%)
DC	$0.4B	RISC-V catches up, share stagnates at 15%
IoT	$0.3B	Share drops to 30%
Automotive	$0.2B	RISC-V + Domestic substitution
PC/Other	$0.3B	Low growth + competition
Total	$2.3B	—

Summary of Five Scenarios and Probability Weighting:

Scenario	Probability	FY2030 Royalties	5-Year CAGR	Contribution
Breakthrough	10%	$9.0B	+33%	$0.90B
Optimistic	20%	$6.5B	+25%	$1.30B
Base Case	35%	$4.7B	+17%	$1.65B
CDS-Dominated	25%	$3.5B	+10%	$0.88B
Pessimistic	10%	$2.3B	+1%	$0.23B
Probability-Weighted	100%	—	—	$4.95B

Probability-Weighted FY2030 Royalties: $4.95B — Close to analyst consensus implied royalty expectations ($5-6B), but below the optimistic end of consensus.

21.4.1 In-depth Validation of End Markets Across Five Scenarios

Mobile (FY2025: ~$1.5B → FY2030: $1.1-2.5B)

ARM's dominant position in mobile (97%+ market share) means growth primarily comes from ASP increases rather than market share expansion:

Driving Factor	Current Level	FY2030E	Certainty
Global Smartphone Shipments	~1.2 billion units	~1.35 billion units (+12%)	High
ARM Penetration Rate	~97%	95-97% (RISC-V minor erosion)	High
Average Royalty/Handset	~$1.20	$1.40-1.90 (v9/v10)	Medium
Flagship to Mid-range Ratio	30/70	35/65	Medium

Mobile Royalty Ceiling Analysis: Processors account for ~$15-50 of phone BOM (flagship $40-50, mid-range $15-25). ARM royalties account for 1-2% of SoC price → $0.15-1.0/chip. If v10 increases the royalty rate from 1.5% to 2.5%, mid-range SoC royalties would rise from $0.30 to $0.50—this magnitude has already met OEM resistance. The hard ceiling for mobile royalties is approximately $2.5-3.0B/year.

Data Center (FY2025: ~$300M → FY2030: $0.4-2.5B)

ARM's growth in DC is the end market with the largest difference across the five scenarios (6.3x gap):

Scenario	FY2030 DC Royalties	ARM DC Penetration Rate	CSS Market (for ARM)	Driver
Breakthrough	$2.5B	40%	Full adoption	Grace/Graviton dominance + China DC all ARM
Optimistic	$1.8B	30%	Rapid growth	Multi-customer CSS + v10 success
Base Case	$1.2B	25%	Stable growth	Graviton dominance + RISC-V constraint
CDS-Dominated	$0.8B	20%	Slower growth	x86 counterattack + RISC-V competition
Pessimistic	$0.4B	15%	Stagnation	RISC-V catches up + Intel counterattack

Core Assumptions Test for DC Growth: ARM's penetration in DC from current ~10% to a base case of 25% requires:

1. Continued expansion of AWS Graviton (current AWS internal ~25%→40%)
2. Azure Cobalt/Google Axion catching up with Graviton
3. NVIDIA Grace gaining share in HPC/AI training
4. At least 2 Chinese cloud vendors (Alibaba/Tencent/ByteDance) adopting ARM (currently still x86-dominated)

Step 4 is the most uncertain—the Chinese DC market accounts for ~20% of the global market. If China favors RISC-V over ARM, the global penetration rate ceiling would fall from 30%+ to below 25%.

IoT (FY2025: ~$700M → FY2030: $0.3-1.5B)

IoT is the fastest-penetrating end market for RISC-V, and also ARM's most challenging battleground for defense:

IoT Segment	ARM Current Share	RISC-V Penetration Speed	FY2030 ARM Share (Base Case)	Key Competitors
Microcontrollers (MCU)	~60%	Fast	45-50%	SiFive, Espressif, GigaDevice
Smart Sensors	~50%	Medium	40-45%	RISC-V + Custom ASICs
Edge AI Inference	~70%	Medium	55-60%	RISC-V + NPU Combo
Wearables	~80%	Slow	70-75%	ARM Still Dominant (Ecosystem Barrier)
Industrial Control	~45%	Medium	35-40%	RISC-V + Traditional MCU

IoT Royalty Dilemma: IoT chip ASP is typically $0.50-5.00, while ARM royalties are only $0.02-0.10/chip. Even with shipment volume growth (+40%/5 years), low ASP means limited royalty increment—IoT is not an ARM growth engine, but rather a "defensive battlefield."

Automotive (FY2025: ~$200M → FY2030: $0.2-1.0B)

Automotive is ARM's largest but most uncertain end market for growth:

Automotive Segment	ARM Chips Per Vehicle (Current)	FY2030E	Royalty/Vehicle	Driver
ADAS/Autonomous Driving	2-5	5-10	$0.50-2.00	Mobileye/NVDA Platforms
Infotainment	1-2	2-4	$0.20-0.60	Qualcomm 8295+
Body Control	3-8	5-10	$0.05-0.15	Domain Controller Integration
Power Management	0-1	1-3	$0.03-0.10	BEV Penetration
Total	$0.80-1.50/Vehicle	$1.50-3.50/Vehicle	—	—

Global annual automotive production is ~90 million vehicles. $2.00/vehicle × 90 million = $180M royalties (consistent with current). Growing to $3.50/vehicle × 95 million = $333M → This is the base case scenario. To reach $1.0B would require ARM royalties of >$10 per vehicle—almost impossible unless ARM secures high royalty rates on ADAS chips.

End Market Growth Contribution Ranking (Base Case Scenario):

Rank	End Market	FY2025→FY2030 Increment	% of Total Increment	Certainty
1	Data Center	+$900M	35%	Medium
2	Mobile (ASP Uplift)	+$200M	8%	High
3	v9/v10 Penetration (Cross-End Market)	+$1,000M	39%	Medium-High
4	Automotive	+$300M	12%	Low
5	PC	+$150M	6%	Low-Medium
	IoT	+$0 (Net Loss)	0%	—
	Total Increment	$2,550M	100%	—

Key Findings: 74% of the increment comes from DC (35%) + v9/v10 penetration (39%). This means our valuation heavily relies on two assumptions: (1) whether DC penetration can truly reach 25%+; (2) whether v10 can still achieve a 30%+ price increase under RISC-V constraints. If either of these assumptions fails, the increment will shrink by 30-40%.

21.5 Royalty "Waterfall Bridge" — Increment Breakdown from FY2025 to FY2030

Decomposing FY2025→FY2030 royalty growth ($2,168M→probability-weighted $4,950M) into specific increment sources:

Increment Details:

Incremental Source	Amount ($M)	Certainty	Dependencies
v9 Penetration (Remaining Upgrades)	+$600	High	Existing v9 customers continue to upgrade (already in progress)
Shipment Volume Growth (5 years)	+$400	High	Normal growth in global chip shipments
Data Center (CSS)	+$800	Medium	DC ARM penetration 23%→30%+, CSS $100+/chip
Automotive + PC	+$350	Medium	Driven by ADAS + AI PC
v10 Price Increase	+$500	Low	H2 Core: Is RISC-V a constraint?
IoT Share Loss	-$200	High	RISC-V's continuous erosion in IoT (irreversible)
China Slowdown	-$150	Medium	Geopolitics + RISC-V Substitution
Probability-Weighted Discount	-$520	—	Discount for scenarios with probability <100%
Net Increment	+$2,780	—	—
FY2030E	$4,950	—	—

Key Findings:

1. Certainty Stratification: High-certainty increments (v9 + shipment volume) only contribute $1,000M (36% of net increment). Medium-low certainty (DC + v10) contributes $1,300M (47%). This means nearly half of the growth relies on uncertain assumptions.

2. v10 price increase is the largest source of uncertainty: $500M increment (after probability weighting), but it can reach $1,000-1,500M in a "no CDS" scenario, and close to $0 in a "CDS-dominated" scenario. A single variable determines the royalty growth range of $0-1,500M.

3. IoT share loss is the most certain negative factor: -$200M (RISC-V's penetration in IoT is almost irreversible). However, the impact is relatively small—ARM's strategic focus has shifted from IoT to high-ASP endpoints (DC/Automotive).

4. DC is the "sweet spot" for growth: +$800M (29% of net increment). However, DC growth depends on CSS pricing (if customers choose TLA instead of CSS, ASP significantly decreases) and ARM vs x86 competition (Intel Sierra Forest counterattack).

Three Sub-scenarios for v10 Price Increase:

Sub-scenario	v10/v9 Ratio	Royalty Increment	Probability	Weighted Increment
Successful Price Increase	2.0x	+$1,500M	15%	$225M
Moderate Price Increase	1.3x	+$650M	45%	$293M
Price Maintained	1.0x	+$0M	30%	$0M
Price Decrease	0.8x	-$400M	10%	-$40M
Probability-Weighted	—	—	—	$478M

The v10 probability-weighted increment is ~$478M, close to the $500M used above. This figure is crucial for the internal consistency of the report—it anchors the bridge between CI-01 (RISC-V CDS elasticity function) and royalty TAM analysis.

21.6 Python Verification Requirements for Royalty Reverse Engineering

The following figures need to be cross-verified using Python:

#	Verification Item	Method	Priority
V-R1	Five-scenario probability weighting	Confirm correct weight multiplication	P0
V-R2	Bottom-Up vs Top-Down Cross-check	Difference <20%?	P1
V-R3	v10 three-sub-scenario expected value	Confirm $478M	P0
V-R4	IoT share loss trajectory	S-Curve fitting	P2
V-R5	Global Semiconductor TAM forecast source	SIA vs WSTS Cross-check	P1

21.7 Licensing Revenue End-State Projection

The growth logic for licensing revenue differs from royalties:

Factor	Growth Contribution	Sustainability	Uncertainty
New CSS Customers	+15-20%/year	High (only 11 customers currently)	Low
v9/v10 Upgrades	+10-15%/year	Medium (migration of existing base)	Medium
New End-Market Entry (Automotive/PC)	+5-10%/year	Medium	Medium
SoftBank Association	+5-10%/year	Low (non-organic)	High
RISC-V Substitution Erosion	-3-5%/year	Medium	High

FY2030 Licensing Revenue Five Scenarios:

Scenario	FY2030 Licensing	CAGR	Key Assumptions
Breakthrough	$4.5B	+20%	CSS Boom + Phoenix Licensing + SB Sustained
Optimistic	$3.5B	+14%	CSS Robust + SB Stable
Base Case	$2.8B	+9%	CSS Moderate + RISC-V Moderate Erosion
CDS Dominant	$2.2B	+4%	Clients Shift to RISC-V Evaluation → Delayed Signings
Pessimistic	$1.5B	-4%	RISC-V Acceleration + Reduced SB Linkage

21.8 Royalties + Licensing: Terminal Total Revenue and TAM Ceiling

Scenario	FY2030 Royalties	FY2030 Licensing	FY2030 Total Revenue	CAGR	vs. Consensus
Breakthrough (10%)	$9.0B	$4.5B	$13.5B	+27%	+21%
Optimistic (20%)	$6.5B	$3.5B	$10.0B	+20%	-11%
Base Case (35%)	$4.7B	$2.8B	$7.5B	+13%	-33%
CDS Dominant (25%)	$3.5B	$2.2B	$5.7B	+7%	-49%
Pessimistic (10%)	$2.3B	$1.5B	$3.8B	-1%	-66%
Probability-Weighted	$5.0B	$2.7B	$7.7B	+14%	-31%
Analyst Consensus	—	—	$11.2B	+23%	Benchmark

Probability-Weighted Total Revenue $7.7B vs. Consensus $11.2B → 31% Gap

Key Findings:

1. Analyst consensus ($11.2B) falls between our "Optimistic" (20% probability) and "Breakthrough" (10% probability) scenarios—market pricing implies a scenario with ~30% probability.
2. Probability-weighted ($7.7B) is 31% lower than consensus—the discrepancy arises from: (a) royalty price increases constrained by H2, (b) IoT share loss, (c) SoftBank related transactions exclusion.
3. Only in the "Breakthrough" scenario (10% probability) does ARM's revenue exceed consensus.

TAM Ceiling Analysis:

ARM's realistic royalty ceiling is approximately $8-9B—even under the most optimistic assumptions, royalty growth will decelerate at some point.

21.9 Quarterly Royalty Rhythm Analysis

Royalty revenue exhibits a clear quarterly fluctuation pattern:

FY2025-2026 Quarterly Royalty Trends:

Quarter	Total Revenue ($M)	Royalties (Est.)	Licensing (Est.)	Royalty % of Total	QoQ
FY25 Q1	939	~540	~399	57%	—
FY25 Q2	844	~470	~374	56%	-13%
FY25 Q3	983	~560	~423	57%	+19%
FY25 Q4	1,241	~598	~643	48%	+7%
FY26 Q1	1,053	~580	~473	55%	-3%
FY26 Q2	1,135	~620	~515	55%	+7%
FY26 Q3	1,242	~680	~562	55%	+10%

Quarterly Pattern Interpretation:

1. Q4 Anomaly: FY2025 Q4 total revenue was $1,241M (31% of full year), but licensing revenue's share jumped to ~52%. This suggests large licensing contracts are concentrated towards the fiscal year-end—a possible "year-end push".
2. Stable Royalty Growth: Royalty revenue shows a stable upward trend (~$540M→$680M, +26% over 6Q), which is a natural result of increased shipments + v9 penetration.
3. FY2026 Q1-Q3 total $3,430M → implying Q4 needs to reach $1,476M to meet the consensus of $4,906M. Q4 requires +19% QoQ—higher than normal historical fluctuations, suggesting consensus might be high or significant licensing agreements will be signed in Q4.

21.10 Triple Constraints on Royalty "Tax Rate" Long-Term Ceiling

ARM's royalties are essentially a "tax" on global semiconductor sales. However, this "tax rate" is subject to a triple constraint:

Constraint 1: Customer Affordability — ARM's customers (chip design companies) typically have net margins of 10-25%. If ARM's royalty rate increases from 0.5% to 2%, some low-margin customers might switch to RISC-V. The maximum "tax rate" customers can bear is approximately 1.0-1.5%.

Constraint 2: Threat of Substitution — RISC-V, as a free alternative, sets a hard ceiling on ARM's "tax rate". When the "ARM tax" exceeds the annualized cost of migrating to RISC-V (including toolchain investment, verification costs, etc., estimated at $5-15M per year for a medium-sized chip company), rational customers will begin to migrate.

Constraint 3: Regulatory Risk — If ARM's "tax rate" is too high (>1.5%), it could trigger antitrust scrutiny (similar to Qualcomm's 2017 FTC lawsuit). Due to its dual identity of SoftBank control and UK registration, ARM currently receives less antitrust attention in the US, but this could change.

Triple Constraint Intersection: ARM's "tax rate" ceiling is approximately 0.8-1.2% (probability-weighted), corresponding to a royalty ceiling of $5.6-8.4B (based on $1T taxable semiconductors). This is consistent with the five-scenario analysis in 19.4 (probability-weighted $4.95B).

21.11 v10 Pricing Power: The Core Variable for ARM Royalty Growth

The v10 architecture (expected release 2027-2028) is the most important single variable for ARM's royalty growth over the next five years. The pricing power of v10 will determine which of the five scenarios ARM falls into:

Historical Royalty Rate Increases Comparison:

Architecture Generation	Release Year	Royalty Rate vs. Predecessor	Driving Factors	Customer Acceptance
v7→v8(A-class)	~2013	+50-80%	64-bit + Performance Leap	High (No alternatives)
v8→v8.2(Update)	~2016	+15-25%	Incremental Improvements	High
v8→v9	~2021	+80-120%	DSU + CSS Enablement + AI Extensions	Medium-High
v9→v10(Expected)	2027-28	+30~100%?	See analysis below	Medium (RISC-V constraint)

Three Scenarios for v10 Pricing:

Scenario	v10/v9 Royalty Ratio	Total Royalty Impact (FY2030)	Dependencies	Probability
Significant Price Hike	2.0x	+$1.5-2.0B	RISC-V stagnation + huge v10 performance leap	15%
Moderate Price Hike	1.3-1.5x	+$0.5-1.0B	Moderate RISC-V competition + clear v10 advantage	45%
Limited Price Hike	1.0-1.2x	+$0-0.3B	Strong RISC-V competition → ARM forced to be conservative	40%

v10 Pricing Constraint Equation:

Game Theory Analysis of v10 Pricing:

ARM faces a classic "pricing prisoner's dilemma":

• Raise prices too much: Short-term revenue surge → Long-term customers accelerate RISC-V migration → Pricing power erodes
• Raise prices too little: Short-term growth falls short of expectations → Market disappointment → Valuation compression
• Optimal strategy: "Boiling the frog slowly" price increase – +30-50% per generation, sufficient for growth but not enough to trigger large-scale migration

However, ARM's ability to execute the optimal strategy depends on RISC-V's progress:

• If RISC-V is still immature in 2027: ARM can safely raise prices by 50-80%
• If RISC-V has its first mass-produced server chip in 2027: ARM is forced to limit price increases to 20-30%
• If 3+ companies mass-produce RISC-V products in 2028: ARM may be forced to maintain prices or lower them

Our Baseline Assumption: v10 moderate price increase of 1.3x (+30%) – this is historically the most conservative price hike, but under RISC-V competition, it may be the most realistic. If v10 can indeed only increase prices by 30% (vs +100% for v8→v9), this would be the first quantitative evidence confirming Hypothesis H2 (RISC-V CDS).

---

Chapter 22: Five-Engine Multi-Method Valuation

22.1 Valuation Methodology Selection

Given ARM's specific characteristics (pure IP licensing, unnormalized P/E, PW=6 hybrid model), five independent methods are employed:

Engine #	Method	Reason for Application	Independence	Weight
E1	Reverse DCF Extension	"What is the market betting on?"	High (Price → Assumption)	15%
E2	Peer Relative Valuation	Anchoring to Comparable Companies	High (Horizontal Comparison)	25%
E3	Royalty TAM Terminal Value Valuation	Based on Ch19 Reverse Engineering	High (Bottom-up)	15%
E4	Scenario-Weighted DCF	Integrated Probability Judgment	Medium (Dependent on E3)	25%
E5	Payment Network Framework	H3 Hypothesis Test	High (Cross-Industry)	20%

22.2 Engine 1: Reverse DCF Extension

The Reverse DCF extension from Ch16 presents a Dual-Margin Sensitivity Matrix:

Matrix A: 25% Terminal Net Margin (Benchmark)

Terminal P/E↓ \ 10-Year Rev CAGR→	10%	12.5%	15%	17.5%	20%
20x	-67%	-57%	-44%	-27%	-5%
25x	-59%	-47%	-31%	-9%	+20%
30x	-51%	-36%	-15%	+12%	+48%
35x	-43%	-24%	+2%	+36%	+80%
40x	-35%	-12%	+19%	+60%	+114%

Matrix B: 40% Terminal Net Margin (Bull Case Scenario)

Terminal P/E↓ \ 10-Year Rev CAGR→	10%	12.5%	15%	17.5%	20%
20x	-47%	-31%	-10%	+17%	+52%
25x	-34%	-14%	+13%	+47%	+90%
30x	-20%	+3%	+36%	+78%	+128%
35x	-7%	+20%	+59%	+108%	+167%
40x	+7%	+37%	+82%	+139%	+205%

Engine 1 Key Conclusions:

• 25% Net Margin: Requires 15% CAGR + 35x P/E to be "reasonable" (0%)
• 40% Net Margin: Only requires 10% CAGR + 40x P/E to be "reasonable"
• Margin assumption is at the core of valuation divergence: 25%→40% significantly lowers the "justification threshold"
• However, Ch17 analysis indicates that a 40% net margin requires SBC to fall below 5% + R&D to 30% – which is extremely aggressive given ARM's sustained high investment.

22.2.1 Reverse DCF Belief Set: What is the market "betting on"?

Translating matrix conclusions into market implied belief set:

Belief	Market Implied Assumption	Our Assessment	Gap
B1: Revenue Growth	10-Year CAGR 17-20%	12-15% (Probability-Weighted)	5-8pp too high
B2: Terminal Net Margin	35-40% (Non-GAAP)	25-27% (GAAP)	8-15pp too high
B3: Terminal P/E	30-35x	22-28x	7-12x too high
B4: RISC-V Threat	Negligible (0-5% impact)	Medium (10-20% impact)	Underestimated
B5: SBC Path	Normalize to 5% within 3 years	Still 10%+ in 5 years	Underestimated

Belief Fragility Ranking (Which belief is most easily falsified):

1. B5 (SBC Normalization) — Fragility: High — Will be directly shown in the actual SBC/Revenue ratio in FY2027-FY2028. If FY2027 SBC/Rev is still >12%, B5 will be falsified, and the Non-GAAP narrative will be damaged.
2. B1 (Growth Rate) — Fragility: Medium — Requires sustained DC penetration + v10 success for dual validation. DC growth is verifiable quarterly (Mercury Research).
3. B4 (RISC-V) — Fragility: Medium — RISC-V DC market share growth is not linear and could accelerate suddenly (Tenstorrent mass production + Android GKI). A single event could change the narrative.
4. B2 (Margin) — Fragility: Low-Medium — Margin expansion path is longer (visible in 3-5 years), not easily falsified in the short term. However, if B5 fails, B2 will also be reassessed.
5. B3 (Terminal P/E) — Fragility: Low — Terminal multiple is a long-term judgment, not verifiable in the short term.

Belief Reversal Analysis: If the most fragile belief B5 is falsified (SBC FY2027 still >12%):

• B2 (Terminal Margin) revised down from 35-40% to 28-32% → Valuation reduced by 15-20%
• A single belief reversal leads to $136B→$109-116B
• This is "minimal cost falsification" – no need for the RISC-V threat to materialize, no need for macroeconomic shocks; merely SBC not normalizing is enough to make ARM's current valuation unsustainable.

Mermaid Mapping of Belief Reversal:

22.3 Engine 2: Peer Relative Valuation

22.3.1 Semiconductor IP Peers (Most Directly Comparable)

Metric	ARM	SNPS	CDNS	Median	ARM Premium
P/E(TTM)	142x	55x	76x	65x	+118%
EV/Sales	28.0x	10.3x	16.0x	13.2x	+112%
EV/EBITDA	122x	33.6x	45.2x	39.4x	+210%
Rev Growth	+24%	~+15%	+6%	+11%	+13pp
Gross Margin	94.9%	77%	86%	82%	+13pp
Net Margin	19.8%	19%	21%	20%	-0.2pp
R&D/Rev	50.1%	~35%	~35%	35%	+15pp
DSO	173 days	78 days	65 days	72 days	+101 days

ARM Fair Valuation Based on EDA Peers:

Method	Calculation	Fair Market Cap	vs Current
P/E Benchmarking (incl. Growth Premium)	$1.60 Non-GAAP EPS × 65x × 1.15	$127B	-7%
EV/Sales Benchmarking (incl. Growth + GM Premium)	$4.67B × 13.2x × 1.3	$83B	-39%
EV/EBITDA Benchmarking (incl. Growth Premium)	$1.05B × 39.4x × 1.3	$57B	-58%
Median	—	$83B	-39%

22.3.2 Payment Network Peers (H3 Framework)

Metric	ARM	Visa	Mastercard	V/MA Median	ARM Premium
P/E(TTM)	142x	33x	34x	33.5x	+324%
EV/Sales	28.0x	16.7x	15.9x	16.3x	+72%
Net Margin	19.8%	~55%	~46%	50.5%	-31pp
ROE	11.6%	53%	193%	123%	-111pp
Rev Growth	+24%	+10%	+14%	+12%	+12pp

H3 Test: If ARM Were Visa

Method	Calculation	Fair Market Cap	vs Current
Terminal Margin (FY2030 45% NM × 33.5x P/E × 1.5 Growth Premium → Discounted)	$3.5B NI × 50x / 1.46	$120B	-12%
EV/Sales (incl. Growth Premium)	$4.67B × 16.3x × 1.4	$108B	-21%
TAM Ceiling Discount (ARM $15B Terminal vs Visa $100B+)	V/MA P/S 16.3x × 0.6 Discount × $4.67B	$49B	-64%
Median	—	$96B	-29%

H3 Final Judgment: ARM's business model (transaction taxation) is indeed closer to payment networks, but the current valuation has prematurely reflected this transition. Even utilizing the most favorable payment network framework, ARM remains overvalued by 12-29%.

22.3.3 Semiconductor IP Peer Valuation Deep Dive: Justifiable Premium Limits

Reasons and upper limits for ARM's 118% (P/E) premium over EDA peers (SNPS/CDNS):

Premium Factor Decomposition:

Premium Factor	Estimated Contribution	Justification	Sustainability
Higher Growth Rate (+24% vs +11%)	+35~50% P/E Premium	High	Medium (Growth will decelerate)
Higher Gross Margin (95% vs 82%)	+10~15%	High	High (Inherent to IP model)
AI/Data Center Concept Premium	+30~50%	Low	Low (Concept > Substance)
"Visa of Chips" Narrative Premium	+15~25%	Medium	Medium (Contingent on Proof)
Low Float Technical Premium	+10~20%	Low	Low (Eliminated by SoftBank's Sell-down)
Total Theoretical Premium	+100~160%	—	—
Current Actual Premium	+118%	—	—

Key Judgment: The current +118% premium falls within the theoretical range (+100~160%), but approximately 40% (i.e., ~50pp premium) of it originates from low-sustainability factors (AI concept + low float). Should these factors revert:

• Reasonable Premium: +60~80% (Excluding low-sustainability factors)
• Corresponding P/E: 65x × 1.7 = 110x (vs current 142x)
• Implied Downside: 142x → 110x = -23%

Horizontal Time-Series Comparison: ARM vs SNPS/CDNS Premium Evolution Post-IPO

SNPS and CDNS also enjoyed high multiples during their high-growth periods (2018-2021 with +30% growth):

• SNPS Peak P/E (2021): ~75x → Currently reverted to 55x (-27%)
• CDNS Peak P/E (2021): ~90x → Currently reverted to 76x (-16%)
• Mean Reversion Pattern: P/E compression of approximately 20-30% when growth decelerates

ARM's FY2028E growth is projected to decelerate to +21% (FY2026: +24%). According to peer reversion patterns:

• 2028 Reasonable P/E: 142x × 0.75 ≈ 107x (Conservative)
• 2028 EPS (Consensus): $2.80 → Implied Market Cap: $300B?
• Pitfall: This high multiple can only be sustained if ARM demonstrates a sustainable Non-GAAP EPS growth of 20%+. GAAP EPS growth may only be 10-15% (due to SBC drag).

22.3.4 Peer Deep Dive: QUALCOMM QTL Division Benchmarking

Qualcomm's QTL (Technology Licensing) division is the most direct comparable for ARM's royalty model:

Metric	ARM (Royalty Segment)	QCOM QTL	Implication
FY2025 Royalty Revenue	~$2.2B	~$5.3B	QCOM Royalty Scale 2.4x
Royalty Gross Margin	~95%	~95%	Nearly Identical
Royalty EBIT Margin	~60-65% (Est.)	~70%	QCOM Higher (Scale Effect)
Royalty Growth (3Y CAGR)	+15-20%	+3-5%	ARM Growth Significantly Higher
TAM Ceiling	~$15B (Mobile+DC+IoT)	~$10B (Primarily Mobile)	ARM Larger
Competitive Threat	RISC-V (Medium)	RISC-V+ARM's Own (Low)	QCOM More Secure
Implied EV (e.g., if Spun-off Independently)	—	$80-100B	See Below

QCOM QTL Spin-off Valuation:
QTL contributes approximately 30% of QCOM's revenue but >70% of its profit. If QTL were to be listed independently:

• QTL EBIT ~$3.7B × 25x Multiple = ~$93B
• Corresponding P/S (Royalty $5.3B) = 17.5x

ARM Royalty Segment Valuation Benchmarked against QTL:

• ARM Royalty $2.2B × 17.5x (QTL) × 1.3 (Growth Premium) = $50B
• Plus Licensing Revenue ($2.0B × 10x) = $70B
• Total Valuation Benchmarked against QTL: $70B → Lower than E2 Median $83B, further confirming the "overvalued" judgment

Explanation of Differences: ARM's growth premium (×1.3) is already very aggressive. QTL has higher royalty rates (3-5% vs ARM <1%), higher profit margins (70% vs 60%), and a larger scale — but slower growth. ARM trades margin for growth; whether this is worthwhile in the long run depends on the sustainability of that growth.

22.4 Engine 3: Royalty TAM Terminal Valuation

Based on Chapter 19's five-scenario waterfall:

Scenario	FY2030 Revenue	Net Margin	NI	Terminal P/E	Market Cap (2030)	Discounted (10%)	vs. Current
Breakthrough	$13.5B	38%	$5.1B	35x	$179B	$122B	-10%
Optimistic	$10.0B	35%	$3.5B	32x	$112B	$77B	-43%
Baseline	$7.5B	25%	$1.9B	28x	$53B	$36B	-74%
CDS-led	$5.7B	20%	$1.1B	22x	$25B	$17B	-88%
Pessimistic	$3.8B	15%	$0.57B	18x	$10B	$7B	-95%
Probability-Weighted	—	—	—	—	—	$55B	-60%

Note: E3 uses GAAP Net Margin (including SBC), resulting in a conservative valuation. If using Non-GAAP:

Scenario	Non-GAAP NI (Est.)	P/E	Market Cap (2030)	Discounted	vs. Current
Baseline	$2.7B(35%NM)	28x	$76B	$52B	-62%
Optimistic	$4.5B(45%NM)	32x	$144B	$99B	-27%

Even with Non-GAAP + optimistic assumptions, the discounted value is still below the current market cap.

22.5 Engine 4: Scenario-Weighted DCF – Full Three-Statement Build-out

Common Assumptions:

• WACC: 10% (ARM beta~2.0, unlevered beta~1.8, zero debt → unlevered WACC approximation)
• Terminal Growth Rate: 3%
• CapEx/Rev: 4-5% (IP model asset-light)
• Tax Rate: 15-18% (ARM's effective tax rate historically low)
• Depreciation/Rev: 3-4%

Scenario A: Bull Case (Probability 20%)

Income Statement Build-out:

Year	Revenue ($B)	YoY	GM%	R&D%	SGA%	SBC%	GAAP OM	NI	EPS
FY2026E	5.2	+30%	96%	48%	18%	14%	16%	0.7	0.64
FY2027	6.5	+25%	96%	44%	16%	12%	24%	1.2	1.13
FY2028	8.1	+25%	96%	40%	14%	10%	32%	2.1	1.97
FY2029	9.7	+20%	96%	37%	13%	8%	38%	3.0	2.80
FY2030	11.4	+18%	96%	35%	12%	7%	42%	3.9	3.64
FY2031	13.0	+14%	95%	34%	12%	6%	43%	4.6	4.30
FY2032	14.4	+11%	95%	33%	11%	5%	46%	5.5	5.14
FY2033	15.6	+8%	95%	33%	11%	5%	46%	5.9	5.51
FY2034	16.6	+6%	95%	33%	11%	5%	46%	6.3	5.89
FY2035	17.4	+5%	95%	33%	11%	5%	46%	6.6	6.17

Cash Flow Build-out:

Year	NI	D&A	SBC	WC Change	OCF	CapEx	FCFF
FY2026E	0.7	0.2	0.7	-0.2	1.4	-0.2	1.2
FY2027	1.2	0.2	0.8	-0.1	2.1	-0.3	1.8
FY2028	2.1	0.3	0.8	-0.1	3.1	-0.3	2.8
FY2029	3.0	0.3	0.8	0.0	4.1	-0.4	3.7
FY2030	3.9	0.4	0.8	0.0	5.1	-0.5	4.6
FY2031-35	Increasing	—	Decreasing	—	—	—	5.0-6.5

Scenario B: Base Case (50% Probability)

Income Statement Build-out:

Year	Revenue ($B)	YoY	GM%	R&D%	SGA%	SBC%	GAAP OM	NI	EPS
FY2026E	4.9	+22%	96%	50%	19%	15%	12%	0.5	0.47
FY2027	5.7	+16%	95%	48%	18%	13%	16%	0.7	0.66
FY2028	6.5	+14%	95%	46%	17%	11%	21%	1.1	1.03
FY2029	7.2	+11%	95%	44%	16%	10%	25%	1.5	1.40
FY2030	7.5	+4%	95%	43%	16%	9%	27%	1.6	1.50
FY2031	7.9	+5%	95%	42%	15%	8%	30%	1.9	1.78
FY2032	8.3	+5%	95%	42%	15%	7%	31%	2.1	1.96
FY2033	8.7	+5%	95%	42%	15%	7%	31%	2.2	2.06
FY2034	9.0	+3%	95%	42%	15%	7%	31%	2.3	2.15
FY2035	9.3	+3%	95%	42%	15%	7%	31%	2.3	2.15

Scenario C: Bear Case (30% Probability)

Income Statement Build-out:

Year	Revenue ($B)	YoY	GM%	GAAP OM	NI	EPS
FY2026E	4.8	+20%	95%	10%	0.4	0.37
FY2027	5.3	+10%	95%	12%	0.5	0.47
FY2028	5.5	+4%	94%	13%	0.5	0.47
FY2029	5.4	-2%	94%	12%	0.5	0.47
FY2030	5.4	0%	94%	12%	0.5	0.47
FY2031-35	+2-3%	—	94%	13-15%	0.5-0.7	—

Probability Weighting of Three Scenarios:

Scenario	Probability	Market Cap	Share Price	Contribution
Bull Case	20%	$175B	$164	$35.0B
Base Case	50%	$80B	$75	$40.0B
Bear Case	30%	$35B	$33	$10.5B
Probability-Weighted	—	—	—	$85.5B

Engine 4 Conclusion: Probability-Weighted DCF = $85.5B → The current $136.1B is 59% overvalued

22.6 DCF Sensitivity Analysis: 3×3×3 Matrix

A traditional 2-dimensional sensitivity matrix (WACC × Terminal Growth Rate) is insufficient to capture the valuation uncertainty for ARM. ARM's three core variables are: (1) Revenue Growth Path, (2) Margin Path, (3) Terminal Multiple. Below is the 3×3×3 matrix:

Dimension Definitions:

• Revenue CAGR: Low(8%), Mid(14%), High(20%)
• Terminal OP Margin: Low(20%), Mid(30%), High(42%)
• Terminal EV/FCFF: Low(18x), Mid(22x), High(28x)

Result Matrix (Market Cap $B):

Panel A: Low Terminal Multiple (18x)

OM↓ \ Rev CAGR→	8%	14%	20%
20%	$22	$35	$55
30%	$33	$53	$83
42%	$46	$74	$116

Panel B: Medium Terminal Multiple (22x)

OM↓ \ Rev CAGR→	8%	14%	20%
20%	$27	$43	$67
30%	$40	$65	$101
42%	$56	$90	$142

Panel C: High Terminal Multiple (28x)

OM↓ \ Rev CAGR→	8%	14%	20%
20%	$34	$55	$86
30%	$51	$83	$130
42%	$72	$116	$181

Current Market Cap of $136B Position in the Matrix: Only "reasonable" (> $130B) in the following combinations:

1. 20% CAGR + 42% OM + 22x (=$142B) — Requires all optimistic assumptions
2. 20% CAGR + 30% OM + 28x (=$130B) — Requires high growth + high multiple
3. 14% CAGR + 42% OM + 28x (=$116B) — Still below $136B

Only 2 out of 27 combinations support the current valuation — approximately 7% probability (2/27). This is consistent with the order of magnitude of joint probability analysis (~3%).

WACC Sensitivity Supplement: The above assumes 10% WACC. Impact of WACC changes:

• WACC 8% (Loose): Valuation +25~35%
• WACC 10% (Baseline): Baseline
• WACC 12% (Tight): Valuation -20~25%
• WACC ±1% impacts approximately ±15-18% → In the current interest rate environment, WACC upside risk > downside risk

22.7 DCF Key Assumption Audit

The reliability of the DCF model depends on the reasonableness of its key assumptions. Below is a cross-validation of the most important assumptions:

Assumption 1: FY2030 Revenue (Bull Case $11.4B / Baseline $7.5B / Bear Case $5.4B)

• Consensus: $11.17B → Close to Bull Case
• Ch19 Five-Scenario Probability-Weighted: $7.7B → Close to Baseline
• Validation: The baseline assumption was validated through Ch19 reverse engineering, with moderate credibility

Assumption 2: Terminal OP Margin (Bull Case 42% / Baseline 27% / Bear Case 15%)

• Current GAAP OM: 20.6% (FY2025)
• Peers: SNPS 13%, CDNS 31%, QCOM 28%
• Bull Case 42% OM requires: R&D/Rev decreases to 33% (current 50%) + SBC/Rev decreases to 5% (current 15-20%)
• Validation: Baseline 27% is credible (comparable to QCOM), Bull Case 42% requires multiple assumptions to hold simultaneously (probability <30%)

Assumption 3: WACC 10%

• ARM beta: ~2.0 (since IPO, high volatility)
• Unlevered beta: ~1.8 (ARM zero debt)
• Risk-free rate: 4.3% (10-year Treasury)
• Market risk premium: 5.5%
• CAPM WACC: 4.3% + 1.8 × 5.5% = 14.2% → Our use of 10% is actually conservative (favorable to ARM)
• If 14% WACC is used: All DCF valuations would decrease by another 25-30%
• We intentionally used a low WACC to benefit ARM → Even so, it is still overvalued

Assumption 4: Terminal EV/FCFF Multiple (Bull Case 25x / Baseline 22x / Bear Case 18x)

• Current ARM: ~760x (due to extremely low FCFF)
• Mature IP companies: 15-25x
• Mature payment networks: 25-30x
• Validation: Baseline 22x is between IP companies and payment networks, which is reasonable

22.8 Engine 5: Payment Network Framework Deep Dive

22.8.1 Business Model DNA Comparison

Dimension	ARM	Visa/MC	Similarity
Revenue Model	Per-chip transaction royalties + upfront licensing	Proportional fee per transaction	90%
Customer Relationship	B2B (chip companies) → B2B2C	B2B (banks) → B2B2C	85%
Asset Intensity	Extremely light (pure IP, no fabs)	Extremely light (pure network, no banks)	95%
Network Effect	Developers → Software → Devices → Developers	Cardholders → Merchants → Cardholders	70%
Switching Costs	High (code + toolchain)	High (brand + terminal protocols)	80%
Gross Margin	95%	~80%	ARM is higher
Net Margin	20%	~50%	ARM is much lower
TAM Ceiling	$8-16B (royalties)	$100B+ (global payments)	V/MA 6-12x
Competitive Threat	RISC-V (open-source alternative)	Digital wallets (partial alternative)	Similar but V/MA's moat is stronger

DNA Similarity: 78/100

22.8.2 Two Fatal Differences in the Payment Network Framework

Difference 1: TAM Ceiling
Visa's TAM (0.1% levy on global payments of $100T+) has virtually no ceiling — global economic growth equals Visa's growth. ARM's TAM ($680B semiconductors × 1-2% royalty rate) has a clear ceiling. ARM cannot sustain steady growth after maturity like Visa.

Difference 2: Durability of Competitive Moat
Visa's moat (two-sided network + regulatory licenses + global agreements) has not been eroded in 50 years. ARM faces a substantial threat from RISC-V. Visa has never faced an "open-source Visa".

Quantifying the Differences: These two differences suggest ARM should trade at a 30-50% discount relative to V/MA:

• TAM Discount: -20~30% (TAM is only 1/10th of V/MA's)
• Competition Discount: -10~20% (RISC-V has no equivalent threat to V/MA)

Difference 3: Fundamental Difference in Margin Trajectory

Visa's net margin steadily increased from its IPO (2008, ~30%) to ~55% (2025) — a stable increase of 25 percentage points within 17 years. ARM's net margin is expected to increase from 10% after its IPO (2023, ~10%) to 25-35% — but the path is entirely different:

Dimension	Visa's Margin Improvement Path	ARM's Margin Improvement Path
Main Drivers	Organic transaction volume growth + Economies of scale	SBC normalization (not operational improvement)
Sustainability	Very high (operating leverage)	Uncertain (SBC deceleration?)
Risks	Very low (no competitor alternative)	Medium (RISC-V might limit pricing power)
Time to reach 50% NM	Already achieved	May never be achieved (GAAP)

This means: Even if ARM's business model DNA is similar to Visa's (78/100), the "quality" of its margin path is far lower than Visa's. Visa's margin improvement = true operational efficiency; ARM's margin improvement (from 10% → 25%) primarily = SBC reduction from IPO peak. Visa earns "true profits", while ARM might earn "apparent profits" — this difference justifies a 10-15 percentage point valuation discount.

Payment Network Valuation Summary:

Method	Fair Market Cap	vs. Current
Revenue Multiple + Growth Premium	$116B	-15%
Terminal Margin (33x)	$79B	-42%
Terminal Margin (40x)	$96B	-29%
TAM Ceiling Discount	$49B	-64%
Median	$96B	-29%

22.9 Five-Engine Cross-Validation and Integrated Valuation

Engine	Method	Core Valuation	vs. $136B	Weight
E1	Reverse DCF	"Conditionally Reasonable"	~0% (Conditional)	15%
E2	Semiconductor IP Peers	$83B	-39%	25%
E3	Royalty TAM Terminal State	$55B (Probability-Weighted)	-60%	15%
E4	Scenario-Weighted DCF	$85.5B	-37%	25%
E5	Payment Network Framework	$96B	-29%	20%

Probability-Weighted Integrated Valuation:

Independence Audit: E1/E3/E4 share growth + margin assumptions (not entirely independent). Merged into "Endogenous Anchor":

• Endogenous Valuation Anchor (E1/E3/E4 Average): $75B × 55% Weight
• Semiconductor IP Peers (E2): $83B × 25% Weight
• Payment Network (E5): $96B × 20% Weight
• Adjusted: $75B×0.55 + $83B×0.25 + $96B×0.20 = $41.3B + $20.8B + $19.2B = $81.2B

22.10 Valuation Dispersion Analysis

Metric	Value
Highest Valuation	$175B (E4 Bull Case)
Lowest Valuation	$7B (E3 Bear Case)
Median	$82B
Standard Deviation	~$24B
Coefficient of Variation	~29%
Inter-Engine Dispersion (max/min)	25x
GAAP vs. Non-GAAP Dispersion	1.5-2.0x

Dispersion Source Breakdown:

Dispersion Type	Source	Contribution	Reducible?
Methodology Dispersion	Peers vs. DCF vs. TAM	30%	Low (different perspectives)
Anchor Dispersion	GAAP vs. Non-GAAP Margins	40%	Medium (pending SBC normalization)
Scenario Dispersion	Bull Case $175B vs. Bear Case $35B	30%	Low (inherent uncertainty)

Key Insight: 40% of ARM's valuation dispersion stems from "GAAP vs. Non-GAAP margins" – this is not a divergence in methodology or scenarios, but rather a difference in accounting treatment. Once SBC normalizes in FY2028-2029 (decreasing from 15-20% to 5-8%), this portion of dispersion will significantly narrow. Until then, investors using Non-GAAP perceive ARM to be "cheaper" by 1.5-2.0x than those using GAAP.

22.11 Option Value: ARM's "Possibility Premium"

Does ARM's 142x P/E include "option value" that traditional DCF cannot capture? PW=6 (hybrid model) requires us to consider this possibility:

ARM's Three "Options":

Option	Description	Exercise Conditions	Intrinsic Value (if exercised)	Probability	Expected Value
O1: Visa-Level Network	ARM becomes a $300B+ platform company	Net Margin >45% + TAM Expansion	+$100-200B	8%	$8-16B
O2: Chip Design Giant	Phoenix success → ARM becomes NVDA	Phoenix achieves $5B+ revenue	+$50-100B	5%	$2.5-5B
O3: AI Era Tax Collector	AI chips must use ARM → Royalty explosion	$100+ royalty per AI chip	+$80-150B	10%	$8-15B

Estimated Option Value: $18-36B

ARM Market Cap Breakdown:

Implication: Even considering the option value (upper bound of $36B), ARM is still overvalued by approximately $18B (~13%). The "Sentiment Premium" may stem from: (1) FOMO (fear of missing out on the next NVIDIA), (2) Scarcity (only 9.7% of shares in free float), (3) AI narrative halo.

However, option value has two important limitations:

1. Non-Independent: O1, O2, O3 cannot occur simultaneously (ARM cannot be both a Visa-type platform and an NVDA-type chip company)
2. Inverse Option: If ARM can become a $300B company, it can also become a $20B company (due to full RISC-V replacement). Options are bidirectional – negative option value is approximately -$10-20B. Net option value may only be $15-25B.

22.12 Valuation Summary

Metric	Value
Base Fair Value (5 Engines)	$80-85B
+Option Value	+$15-25B
Adjusted Fair Value	$95-110B
Implied Share Price	$89-103
Current Market Price	$128.14
Implied Downside	-20~30%
Expected Return (incl. Options)	-24%

Rating: Expected Return < -10% → Cautious Watch ($128 → Fair Value $89-103)

However, conditional rating (PW=6 blended model):

• If SBC normalizes to below 5% by FY2028 + DC penetration reaches 30%: → Rating could be upgraded to "Neutral Watch"
• If RISC-V breaks 10% in DC + v10 price increase fails: → Rating remains at the "Cautious Watch" downside

22.13 Valuation Sensitivity: Under What Conditions Is ARM Undervalued?

This is the fairest test of the bull case—if our framework has systemic biases, ARM's fair valuation could be higher:

Our Potentially Underestimated Assumptions	If Adjusted Higher	Impact	Current Probability
GAAP Margin reaches 35%+ by FY2028+ (SBC rapid normalization)	E3/E4 Valuation +30-40%	+$25-35B	20%
DC Penetration reaches 40%+ by FY2030+ (vs our 20-25%)	Royalty Revenue +$1.5-2B	+$30-40B	15%
v10 Royalty Price Increase 80%+ (vs our 30-50%)	Royalty Base +40%	+$20-30B	10%
ARM becomes the AI chip standard ($100+ royalty per chip)	TAM Ceiling Shifts Upward	+$40-60B	10%
Phoenix Success + No Cannibalization Effect	Additional $5B+ Revenue	+$30-50B	8%

Fairly Addressing the Bull's "Bull Market Bridge":

Key Insight: ARM's current valuation implies not "one optimistic assumption coming true," but "multiple optimistic assumptions coming true simultaneously." This is why we issue a "Cautious Watch"—not because ARM is a bad company, but because $136B has already priced in most of the good news.

22.14 Limitations Statement of the Valuation Framework

Reasons why our valuation might be systematically low:

1. GAAP Bias: We use GAAP margins (including SBC) as a basis. If the market consistently prices based on Non-GAAP (as in the Software industry), our valuation would systematically undervalue ARM by 30-40%. This is not an ARM issue, but a choice of our framework.

2. Linear Extrapolation Bias: We assume RISC-V penetrates along a linear path. However, technological substitution typically follows an S-curve—slower than expected initially (we might overestimate early threats), and faster than expected later (we might underestimate terminal threats).

3. Static Competition Bias: Our model assumes ARM reacts passively. In reality, ARM is actively evolving (v10 architecture + Phoenix + CSS). If ARM successfully upgrades its business model (from IP licensing → platform + chip + services), our valuation framework might need complete restructuring.

4. Excessive Holding Company Discount: Our estimate for the SoftBank holding company discount (12-23%) might be too high—SoftBank currently does not interfere with ARM's operations, and ARM has a majority of independent directors.

Honest Statement on Framework Choices: We chose GAAP + conservative assumptions because this has historically been more accurate for "high-growth + high-SBC" technology companies. However, if ARM indeed proves the sustainability of Non-GAAP margins in FY2028-2030, we should be prepared to revise our valuation upward by 30-40%. This is not an admission of error, but an honest calibration of framework limitations.

---

Chapter 23: Risk Topology Mapping

23.1 Risk Node Checklist

No.	Risk Node	Type	Probability (5-year)	EV Impact (Median)	CQ Link
R1	RISC-V Pricing Ceiling	Structural	60%	-20%	CQ-4, H2
R2	SoftBank Share Sale/Margin Call	Event-driven	40%	-22%	CQ-3
R3	Earnings Quality Deterioration	Cyclical	50%	-15%	CQ-1, H1
R4	DC Growth Below Expectations	Structural	35%	-28%	CQ-2
R5	Arm China Geopolitical Risk	Institutional	25%	-12%	CQ-5
R6	Phoenix Failure/Licensee Conflict	Execution	40%	-10%	CQ-6, H4
R7	Macro/Interest Rate Shock	Systemic	30%	-40%	—
R8	v9→v10 Price Hike Failure	Structural	45%	-15%	CQ-1, CQ-4
R9	SBC Continuous Dilution	Structural	80%	-7%/year	H1
R10	Customer Concentration Risk	Structural	20%	-20%	CQ-1

23.2 Risk Synergy Matrix

23.3 Three Synergistic Clusters + MVP

MVP: Minimum Viable Worst-Case Combination

"What is the minimum number of simultaneous risks that would render the $136B valuation unreasonable?"

Answer: Only a single risk, R1 (RISC-V pricing ceiling), with a 60% probability, is sufficient to cause an overvaluation of 30-45%.

MVP Combination	Number of Risks	Joint Probability	EV Impact	Assessment
R1 Alone	1	60%	-20~35%	Current Valuation is Unreasonable
R1+R8	2	54%	-30~45%	Pricing Power Erosion
R1+R8+R4	3	29%	-40~55%	Growth Narrative Reversal
R1+R8+R4+R2	4	12%	-55~70%	Full-Blown Crisis

MVP Conclusion: ARM's valuation exhibits extremely high fragility – a single risk (R1) with a 60% probability is enough to render the current price unreasonable. In traditional "stable companies," it typically takes 3-4 simultaneous risks to make the valuation unreasonable. ARM's fragility stems from its exceptionally high P/E (142x) – at such elevated valuation levels, any negative factor tends to have an amplifying effect.

Cluster 1: "Pricing Power Erosion Spiral" (R1+R8+R4)

• Trigger Chain: RISC-V performance catches up (R1) → v10 unable to raise prices (R8) → DC growth slows (R4)
• Joint Probability: 60% × 80%(R8|R1) × 60%(R4|R8) = 29%
• Joint EV Impact: -40~55%
• 24-Month Progression Path: See 21.5 "Boiling the Frog"

Cluster 2: "SoftBank Liquidity Crisis" (R2+R7+R3)

• Trigger Chain: Macro shock (R7) → ARM stock plummets → Margin call triggered (R2) → Forced share sale → Earnings quality exposed (R3)
• Joint Probability: 30% × 50%(R2|R7) × 70%(R3 exposure|R2) = 10.5%
• Joint EV Impact: -50~70%
• Characteristics: Low-probability but high-impact "black swan" event, and irreversible once triggered (forced share sale = real selling pressure)

Cluster 2 In-Depth: Five-Stage Deterioration Path of Liquidity Crisis

Five-Stage Timeline:

Stage	Time	Event	ARM Stock Price	SoftBank Action
1	T+0	AI CapEx Cycle Correction (-25%)	$128→$96	Observe
2	T+2 Weeks	Semiconductor Sector Contagion (-15%)	$96→$82	Additional Cash Collateral
3	T+1 Month	LTV Exceeds 40%, Margin Call Triggered	$82	Forced Placement of 5%
4	T+2 Months	Placement Impact + Analyst Downgrades	$82→$65	Considering Second Round of Placement
5	T+3-6 Months	Earnings Quality Exposure + P/E Compression	$65→$45-55	Continuous Sell-off

Key Amplifiers: The danger of this cluster lies in a positive feedback loop—stock price decline → margin call → forced selling → further stock price decline → further margin call. ARM's unique vulnerability (9.7% free float + 90% single shareholder) amplifies the impact of traditional "major shareholder sell-offs" by 3-5 times.

Cluster 3: "Accelerated China Substitution" (R5+R1)

• Trigger Chain: Escalation of US-China Tech Cold War (R5) → China Accelerates RISC-V Adoption (R1 Enhancement) → Arm China Revenue Decline
• Joint Probability: 25% × 80% (R1 Enhancement | R5) = 20%
• Joint EV Impact: -15~25%

Cluster 3 Deep Dive: The Irreversibility of China's RISC-V Autonomy

The particular danger of the Accelerated China Substitution cluster lies in its irreversibility—once China's RISC-V ecosystem achieves "critical mass," even if US-China relations improve, ARM will be unable to reclaim the market:

Stage	China RISC-V Status	Reversibility	ARM Countermeasures
Current: Policy-driven	IoT initial substitution	Reversible	Maintain technology gap
Tipping Point 1: Ecosystem takes shape	Android RISC-V available	Partially Reversible	Price competition
Tipping Point 2: Talent autonomy	Chinese engineers RISC-V > ARM	Irreversible	Abandon China low-end
End State: Independent ecosystem	50%+ of Chinese chips use RISC-V	Completely Irreversible	Retain only high-end licensing

Time Estimate: Approximately 3-5 years (2028-2030) from "Current" to "Tipping Point 2". Key observation indicators:

• Number of RISC-V engineer hires in China (currently ~20,000/year, needs to reach 50,000/year)
• Huawei HiSilicon RISC-V product launch rhythm
• Mandatory proportion of RISC-V in Chinese government procurement (currently 0%, tipping point = 30%+)

Impact on ARM FY2030 Revenue (vs. No Substitution Baseline):

Scenario	Arm China Royalty Impact	Impact on ARM Total Revenue	Probability
Mild Substitution (IoT only)	-$200M	-3%	40%
Moderate Substitution (IoT + Low-end Mobile)	-$500M	-7%	30%
Aggressive Substitution (incl. Mobile + DC)	-$1.2B	-15%	15%
Extreme (Full Substitution)	-$2.0B	-25%	5%
Probability-Weighted	-$480M	-6%	—

Cluster 4: "Phoenix Backlash" (R6+R1+R10)

• Trigger Chain: Phoenix Competes with Licensees (R6) → Licensees Accelerate RISC-V Adoption (R1) → Key Customers Reduce Licensing (R10)
• Joint Probability: 40% × 50% × 30% = 6%
• Joint EV Impact: -30~50%

Cluster 4 Deep Dive: The "Identity Crisis" from IP to Chip

ARM's Phoenix initiative (in-house chip development) creates an unprecedented business model conflict: ARM is simultaneously the "tool seller" (IP licensing) and the "tool user for product creation" (chips). The closest historical analogies are:

Analogy	Outcome	Lesson
Intel McAfee Acquisition (2010)	Security chip integration failed, ultimately sold for $4.2B	OEMs distrust suppliers = competitors
Google Pixel vs. Android OEM	OEM investment ↓, Samsung threatens to fork	Platform provider creating hardware = trust crisis
AWS vs. SaaS Companies	SaaS companies accelerate multi-cloud adoption	Platform provider vertical integration = customer defection
ARM Phoenix vs. Licensees	To be verified	Core of H4 Hypothesis

Quantified Pathways of Backlash:

1. Phoenix announces server SoC → Qualcomm/MediaTek increase RISC-V investment by 15% (M1-6)
2. Phoenix achieves first design win (e.g., cloud vendor custom chip) → 3-5 licensees initiate RISC-V adoption plans (M6-12)
3. Phoenix revenue reaches $500M+ → Licensees cut ARM licensing budget by 10-15% + increase RISC-V budget (M12-24)
4. Net Effect: Phoenix revenue of $500M × 40% profit = +$200M, but licensing revenue loss of $200-300M × 70% profit = -$140-210M → Net Loss of $0-100M

H4 Validation: Is Phoenix worthwhile? This strategy only creates net value if Phoenix's ultimate annual revenue > $2B AND licensee RISC-V migration ≤ 10%. Current probability estimate: 30%.

23.4 "Horrifying Combinations" of Logical Contradictions

While some risks are logically unlikely to occur simultaneously (anti-synergistic), if they do happen concurrently, their impact would far exceed a simple sum:

Horror Combination	Why Logically Contradictory	If It Actually Happens	Joint Probability	Joint Impact
R4 (DC Growth Deceleration) + R1 (Strong RISC-V)	DC growth deceleration might be due to x86 counterattack, not RISC-V	Failure on both fronts simultaneously	15%	-45~60%
R7 (Macroeconomic Shock) + R1 (RISC-V Acceleration)	Recessions typically reduce RISC-V investment	ARM faces a double blow of "recession + substitution"	10%	-55~70%
R6 (Phoenix Success) + R1 (RISC-V Acceleration)	Phoenix success should reduce pressure from RISC-V	ARM's own success paradoxically accelerates RISC-V	8%	-15~25% (Small net effect)

23.4.1 Deep Dive into Horror Combinations: "Perfect Storm" Scenarios

Horror Combination Alpha: "AI Winter + RISC-V Rise" (R4+R1+R7)

This is the most logically contradictory yet deadliest combination – three risk factors activated simultaneously:

• R4: Data center ARM penetration stagnates due to AI CapEx reduction
• R1: RISC-V accelerates in cost-sensitive scenarios (IoT/automotive) precisely due to cost advantages
• R7: Macroeconomic recession suppresses all semiconductor valuation multiples

Why "Logically Contradictory": AI winter (R4+R7) should simultaneously hit RISC-V investment (startups face financing difficulties). But the contradiction lies in: During a recession, companies focus more on cost → RISC-V's "zero royalty" becomes more attractive → existing replacement accelerates.

Impact Pathway if it Occurs:

1. AI CapEx cycle peaks (2027H2) → ARM DC royalty growth drops from +40% to +5%
2. Chip company budgets tighten → Accelerated evaluation of RISC-V alternatives (cost-driven)
3. Macroeconomic recession → ARM P/E drops from 142x to 60-80x (semiconductor mean reversion)
4. Triple blow simultaneously: Growth downgrade × P/E compression × narrative collapse

Probability: Joint probability of all three = 15% × 25% (R7|R4) × 40% (R1 acceleration|R7) = 1.5%
Impact: -65~80% (Almost complete repricing)

Horror Combination Beta: "SoftBank Spiral + Earnings Exposure" (R2+R3+R9)

• R2: SoftBank forced divestment (float shock)
• R3: Earnings quality exposure (GAAP vs Non-GAAP gap magnified)
• R9: Accelerated SBC dilution (new round of RSU vesting)

Elaboration: When SoftBank is forced to divest, the stock price decline triggers sellers to re-examine financials. If it's then discovered that GAAP Net Income is significantly lower than Non-GAAP (a gap of 2-3x), the narrative will shift from "high-growth IP giant" to "false profits subsidized by SBC". Simultaneously, SBC dilution means that even with Non-GAAP growth, per-share value is being eroded. The superposition of the three:

Factor	Independent EV Impact	Superimposed Impact	Magnification Factor
R2 (Divestment Shock)	-15~25%	-25~35%	1.5x
R3 (Earnings Exposure)	-20~30%	-30~40%	1.4x
R9 (SBC Dilution)	-5~10%	-10~15%	1.5x
Superposition (Not simple addition)	—	-50~65%	—

Probability: 15% × 60% (R3|R2) × 70% (R9 acceleration|R2+R3) = 6.3%

23.5 "Boiling Frog" Scenario: Detailed 24-Month Pathway

This is ARM's most probable yet most easily overlooked downside pathway – not a single "black swan" event, but a gradual erosion of pricing power from Cluster 1:

Detailed 24-Month Pathway:

Month	Event	Market Narrative	Cumulative Impact	Majority Investor Reaction
M1-3	Tenstorrent RISC-V server chip mass production	"Volume too small, ARM still holds 95%+"	-2%	Ignored
M4-6	QCOM launches RISC-V+ARM dual-architecture	"Diversifying risk, not replacing"	-4%	Slight concern
M7-9	Android GKI full support for RISC-V	"Support ≠ Adoption"	-8%	Moderate concern
M10-12	2 hyperscalers evaluate RISC-V	"Evaluation ≠ Procurement"	-15%	Beginning to get nervous
M13-15	v10 royalty rate only +30% (vs v9 +100%)	"ARM's pricing power is constrained!"	-25%	Narrative reversal
M16-18	3 customers announce RISC-V roadmap	"This is a trend, not an isolated case"	-35%	Panic selling
M19-21	IoT ARM share drops to 40%	"RISC-V proven viable"	-42%	Institutional offloading
M22-24	RISC-V automotive SoC mass production	"ARM no longer irreplaceable"	-50%	Repricing

Core Characteristics of the "Boiling Frog" Scenario:

1. First 12 months cumulative only -15% — Each individual event was "not severe enough"
2. M13 (v10 Pricing) is the turning point — This is the first directly quantifiable evidence of pricing power erosion
3. Next 12 months accelerate to -50% — After the narrative flipped, previously ignored "benefit of doubt" signals were reinterpreted
4. Probability 29% — This is not an extreme tail event, but a relatively probable path

23.6 Anti-Synergy (Natural Hedge)

Risk A	Risk B	Relationship	Explanation
R4 (Slow DC Growth)	R6 (Phoenix Failure)	Anti-Synergy	Phoenix failure → No competition with clients → Clients more willing to use ARM → DC growth better instead
R7 (Macroeconomic Shock)	R1 (RISC-V)	Weak Anti-Synergy	Recession → RISC-V investment slowdown → ARM's short-term pricing power maintained
R1 (RISC-V)	R9 (SBC Dilution)	Independent	No correlation

Natural Hedge Value: The anti-synergy between R4×R6 means ARM has partial protection in both "success" and "failure" scenarios for Phoenix — this is a built-in hedge in ARM's strategy (though possibly unintentional).

23.7 Risk Transmission Pathway Map: The Complete Link from Trigger to EV Impact

Each risk does not directly "jump" to EV impact — they are transmitted through a series of intermediate steps. Understanding the transmission pathway helps identify early warning signals:

R1 (RISC-V Pricing Ceiling) Transmission Chain:

• T1→M1 Lag: 6-12 months (Performance parity to client initiating evaluation)
• M1→M2 Lag: 12-18 months (Evaluation to contract negotiation)
• M2→M3 Lag: 6-12 months (Negotiation to final pricing)
• M3→O1: Immediate (Pricing determined → Revenue expectations updated)
• Total Transmission Time: 2-4 years (From RISC-V performance parity to visible ARM revenue impact)
• Early Warning Nodes: M1 (Client evaluation) → KS-4/KS-5 triggered

R2 (SoftBank Divestment) Transmission Chain:

• Total Transmission Time: 3-12 months (From funding needs to market impact)
• Early Warning Nodes: Changes in SoftBank's margin balance in quarterly reports

R7 (Macroeconomic Shock) Transmission Chain:

• R7's Specificity: Macroeconomic shocks impact ARM simultaneously through three paths (Revenue + Valuation Multiples + SoftBank) — this is why R7's EV impact (-40%) is much larger than implied by its probability (30%)
• Early Warning: PMI/ISM indices, US 10-year Treasury yield, Monthly semiconductor shipment data (SIA)

23.8 Risk Dashboard: "Distance" of Each Risk from Triggering

Risk	Current Status	Trigger Threshold	Distance to Trigger	Warning Signal	Monitoring Frequency
R1 RISC-V	<2% DC Share	>10% DC	Far (5+ years)	QCOM/Tenstorrent Mass Production	Quarterly
R2 SoftBank	0% Divestment	>5%/6 months	Medium	Margin Balance	Monthly
R3 Earnings Quality	DSO 173 days	DSO>200 days	Close	Q4 10-K Data	Quarterly
R4 DC Growth	~23% Share	Growth Stagnation 2Q	Medium	Mercury Research	Quarterly
R5 China	Encouraging RISC-V	Mandatory Procurement	Medium	Policy Documents	Quarterly
R6 Phoenix	Not Mass Produced	Customer Conflict	Far	ARM Earnings Call Wording	Semi-annually
R7 Macro	Expansion Phase	PMI<45	Medium	ISM/PMI	Monthly
R8 v10 Price Increase	v10 Not Released	Price Increase <1.2x	Medium (2028)	ARM Management Guidance	Annually

Nearest "Landmines": R3 (Earnings Quality, DSO only 27 days away) + R2 (SoftBank, dependent on Izanagi financing progress)

23.9 Probability-Adjusted Expected Loss

Risk	Probability	EV Impact (Median)	Expected Loss
R1 Pricing Ceiling	60%	-20%	-12.0%
R2 SoftBank	40%	-22%	-8.8%
R3 Earnings Quality	50%	-15%	-7.5%
R4 DC Growth	35%	-28%	-9.8%
R5 China	25%	-12%	-3.0%
R6 Phoenix	40%	-10%	-4.0%
R7 Macro	30%	-40%	-12.0%
R8 v10 Price Hike	45%	-15%	-6.8%
R9 SBC	80%	-7%	-5.6%
R10 Customer Concentration	20%	-20%	-4.0%

Simple Sum: -73.5% — However, due to synergy/anti-synergy + overlap, Net Risk Adjustment: -35~45% → Highly consistent with the five-engine valuation's "overestimation by 24-43%".

23.10 Risk-Adjusted Valuation: From Expected Loss to Risk Discount

Applying the probability-adjusted expected loss from 21.7 to the valuation framework:

Method: Risk-Adjusted Valuation = Base Fair Value × (1 - Risk Discount%)

Level	Risks Included	Risk Discount	Adjusted Valuation
L0: Zero-Risk Base	None	0%	$136B (Current)
L1: High Probability Risks	R1+R3+R9 (>50%)	-20%	$109B
L2: +Medium Probability Risks	+R2+R4+R6+R8 (35-45%)	-35%	$88B
L3: +Low Probability Risks	+R5+R7+R10 (<35%)	-42%	$79B
L4: +Cluster Effect	+Synergy Amplification	-45%	$75B

Key Insights:

1. Market pricing is near L0 — $136B implies the market has priced in almost no risk (or believes these risks are offset by growth options)
2. Our fair value is between L2-L3 — $80-88B, consistent with the five-engine aggregate of $80-85B
3. The largest jump is from L0 to L1 (-$27B) — Considering only risks with >50% probability reduces the valuation by 20%
4. This explains why ARM's volatility is high — If any high probability risk is re-focused on by the market, a move from $136B to $109B (-20%) would be a reasonable short-term adjustment

23.11 Cross-Validation of Risk Scenarios and Valuation

Cross-validating the risk analysis from Chapter 21 with the valuation conclusions from Chapter 20:

Validation Dimension	Risk Analysis (Ch21)	Valuation Analysis (Ch20)	Consistency
Core Downside Risk	RISC-V (R1, 60%)	E3 CDS-driven/Bearish (35%)	Consistent
Median Valuation	L2-L3: $79-88B	E2/E4: $83-86B	Highly Consistent
Tail Downside	Cluster 2: -50~70%	E3 Bearish: $7B (-95%)	Analysis more extreme
Tail Upside	—	E4 Bull Case: $175B (+29%)	Risk analysis does not cover upside
SoftBank Impact	R2: -22% (Median)	E1/E2: Governance discount already included	Consistent
Core Disagreement Point	R1: Whether pricing power is lost	E3 vs E5: GAAP/Non-GAAP	Two sides of the same issue

Most important cross-validation: The risk analysis (R1 RISC-V pricing ceiling, 60% probability) and valuation analysis (GAAP vs Non-GAAP divergence = 40% valuation difference) point to the same core issue—whether ARM's pricing power is sufficient to support a valuation upgrade from an "IP licensing company" to a "payment network company". If the answer is "yes," ARM could be worth $100-120B (Non-GAAP framework); if the answer is "no," ARM could be worth $60-85B (GAAP framework + risk discount). The current $136B would require the answer to be a "strong yes."

23.12 ARM's "Valuation Regimes": What Stage Are We Currently In?

Semiconductor companies' valuations typically undergo four "regimes":

Regime	Characteristics	P/E Range	Typical Companies	ARM Currently Applicable?
Growth Discovery	Market first recognizes growth potential	30-60x	2020-2021 CDNS	Passed
Narrative Euphoria	Growth narrative amplified into a "paradigm"	60-150x	Current ARM	Yes
Evidence Scrutiny	Market begins to demand evidence	40-80x	—	Next 2-3 Years
Steady-State Valuation	Growth is proven or disproven	20-40x	Current SNPS	Long-term

ARM is currently in a transition period from "Narrative Euphoria" to "Evidence Scrutiny":

• The "Chip Visa" narrative (H3) has not yet been proven or disproven
• There is no quantitative evidence yet for the RISC-V threat (H2)
• Phoenix (H4) has not yet contributed revenue
• 142x P/E = Market at the tail end of "Narrative Euphoria"

Historical Analogy: Speed of Regime Transition

Company	From "Euphoria" to "Scrutiny"	P/E Compression	Trigger Event
Cisco (2000)	~6 months	110x→35x (-68%)	.com Bubble Burst
Netflix (2021-2022)	~9 months	55x→18x (-67%)	Growth Deceleration Exposed
CDNS (2021-2022)	~12 months	90x→45x (-50%)	Rising Interest Rates + Valuation Normalization
ARM (Expected)	12-24 months	142x→?	v10 Pricing / RISC-V Progress

If ARM enters the "Evidence Scrutiny" regime:

• Conservative P/E: 60-80x (CDNS level + ARM growth premium)
• Corresponding EPS (FY2027 consensus $2.14): 60x×$2.14 = $128 to 80x×$2.14 = $171
• Regime transition could potentially bring ARM's valuation down to $80-128 (depending on the extent of P/E compression and whether EPS targets are met)
• Key: If EPS beats expectations + moderate P/E compression, ARM might "soft land" at $100-130; If EPS misses + significant P/E compression, ARM might "hard land" at $65-80

---

Chapter 24: Kill Switch Registry

24.1 Kill Switch Definition and Framework

A Kill Switch is a predefined event trigger—if triggered, the entire investment thesis requires re-evaluation. This chapter contains 8 core triggers.

KS-1: RISC-V Server Share Exceeds 10%

Field	Value
Trigger Condition	Independent third-party data shows RISC-V share of DC CPU shipments >10%
Current Value	<2% (2025)
Trigger Time Window	2027-2029
Data Source	Mercury Research / IDC QST
Verification Frequency	Quarterly
Action Upon Trigger	CQ-4 Confidence→>80% → H2 Confirmed → Rating Downgrade
Impact Path	RISC-V 10% = Software Ecosystem No Longer a Barrier → Pricing Power Erosion
Related CQ	CQ-4 (15%) + CQ-1 (25%)
Current Probability	25% (Before 2029)

KS-2: SoftBank Major Stake Reduction >5%

Field	Value
Trigger Condition	SoftBank reduces ARM stake by >5% (approx. 53M shares) within 6 months
Current Value	0%
Trigger Time Window	Anytime
Data Source	SEC 13D/13F + SoftBank Quarterly Reports
Verification Frequency	Monthly
Action Upon Trigger	Re-evaluate free float structure → Supply shock → Valuation discount 10-20%
Related CQ	CQ-3 (18%)
Current Probability	15% (Within 1 year), 35% (Within 3 years)

KS-3: GAAP/Non-GAAP Gap Widens to >3x

Field	Value
Trigger Condition	Non-GAAP OP / GAAP OP >3.0x for 2 consecutive quarters
Current Value	~2.6x (Q3 FY2026)
Trigger Time Window	Anytime
Data Source	ARM 6-K/20-F + Quarterly Earnings
Verification Frequency	Quarterly
Action Upon Trigger	H1 Confirmed → Rebuild valuation using adjusted metrics → 20-30% Downside
Related CQ	CQ-1 (25%)
Current Probability	30% (Next 4 quarters)

KS-4: Hyperscale Customer Announces RISC-V Server Replacement

Field	Value
Trigger Condition	Any of AWS/Google/MSFT/Meta announces commercial RISC-V server CPU
Current Value	None (Meta RISC-V limited to AI co-processors)
Trigger Time Window	2027-2030
Data Source	Public News + Earnings Calls
Verification Frequency	Quarterly
Action Upon Trigger	CQ-2 and CQ-4 simultaneously downgraded → Significant valuation adjustment
Related CQ	CQ-2 (20%) + CQ-4 (15%)
Current Probability	10% (Before 2028), 25% (Before 2030)

KS-5: ARM Licensing Pipeline (ATA) Declines for 3 Consecutive Quarters

Field	Value
Trigger Condition	ARM's disclosed number of newly signed license agreements (ALA+TLA+CSS) declines for 3 consecutive quarters
Current Value	— (Not detailed by ARM, needs to be inferred from 6-K)
Trigger Time Window	Anytime
Data Source	ARM 6-K/20-F + Management guidance
Verification Frequency	Semi-annually (6-K cycle)
Action Upon Trigger	License revenue growth outlook lowered → Total revenue growth forecast -5~10pp
Impact Path	Decline in new signings = narrowing of royalty pipeline for the next 3-5 years → discount on long-term growth
Related CQ	CQ-1 (25%)
Current Probability	20% (within 2 years)

KS-6: China's Mandatory RISC-V Procurement Policy

Field	Value
Trigger Condition	China's central government issues a mandatory (non-encouraging) RISC-V procurement policy
Current Value	Encouraging policy (promoted by 8 government agencies but not mandatory)
Trigger Time Window	2026-2029
Data Source	China State Council/MIIT policy documents
Verification Frequency	Quarterly
Action Upon Trigger	Arm China revenue path → most pessimistic scenario → CQ-5 confidence → 80%+
Impact Path	Mandatory procurement → rapid decline in ARM's China market share → Arm China revenue halved
Related CQ	CQ-5 (10%) + CQ-4 (15%)
Current Probability	15% (within 3 years)

KS-7: DSO continuously exceeding 200 days

Field	Value
Trigger Condition	ARM DSO exceeds 200 days for 2 consecutive quarters (currently 173 days)
Current Value	173 days (FY2025)
Trigger Time Window	Anytime
Data Source	ARM 20-F + Quarterly estimation
Verification Frequency	Quarterly
Action Upon Trigger	H1 (Earnings Quality) confidence → 80%+ → Revenue Recognition Review
Impact Path	DSO > 200 days = potentially aggressive revenue recognition → significant concerns about earnings quality
Related CQ	CQ-1 (25%) + H1
Current Probability	25% (within 2 years)

KS-8: Rene Haas Departure or SoftBank Management Change

Field	Value
Trigger Condition	CEO Rene Haas departs/is replaced, or SoftBank appoints new management
Current Value	Haas in office (since 2022)
Trigger Time Window	Anytime
Data Source	Public news + SEC filing
Verification Frequency	Continuous
Action Upon Trigger	Evaluate new management's strategic direction → Potential changes to Phoenix/CSS strategy
Impact Path	Management change + strategic uncertainty → multiple valuation discounts
Related CQ	All CQs
Current Probability	10% (within 2 years)

24.2 Kill Switch Monitoring Dashboard

KS#	Trigger	Current Value	Trigger Threshold	Proximity to Trigger	Probability	Priority
KS-1	RISC-V DC Share	<2%	>10%	Far	25%	P0
KS-2	SB Divestment	0%	>5%/6mo	Far	35%/3yr	P0
KS-3	GAAP/Non-GAAP	2.6x	>3.0x	Close	30%	P1
KS-4	Hyperscaler RISC-V	None	First Announcement	Medium	25%/5yr	P0
KS-5	Licensing Pipeline	—	3Q Consec. ↓	?	20%	P1
KS-6	China Mandatory RISC-V	Encouraging	Mandatory	Medium	15%/3yr	P2
KS-7	DSO	173 days	>200 days	Close	25%	P1
KS-8	CEO Change	In Office	Departure/Replacement	Far	10%	P2

Most concerning: KS-3 and KS-7 are closest to triggering (GAAP/Non-GAAP 2.6x → 3.0x is a difference of only 0.4x; DSO 173 days → 200 days is a difference of only 27 days). The FY2026 annual report (May 2026) will be a critical data point.

24.3 Stress Test Preview

The Bearing Wall Test stems from a belief inversion. The ARM valuation 'edifice' is built upon 6 bearing walls (B1-B6):

Wall	Belief	Independent Strength	Most Vulnerable Link
B1	Revenue CAGR >14% (10 years)	Medium	Highly correlated with B3/B4
B2	Net Margin 17% → 25%+	Low	SBC normalization timeline
B3	Pricing power unconstrained by RISC-V	Very Low	B3 = Keystone Pillar
B4	DC penetration 10% → 40%+	Medium	Highly correlated with B3
B5	Arm China sustained growth	Medium	Geopolitics
B6	SoftBank does not reduce stake	Medium	Independent (event)

Joint Probability: P(all 6 walls hold true) ≈ 0.40×0.60×0.35×0.55×0.65×0.70 = 3.4%

Key Finding: B3 (pricing power) is the sole 'keystone pillar'—if B3 fails:

• B1 is also likely to fail (no pricing power → revenue growth constrained)
• B4 is also affected (RISC-V directly competes in DC)
• The entire valuation framework collapses (B3 failure → chain reaction → $136B → $75-90B)

Bearing Wall Sensitivity: Each Wall's 'Tolerance Space'

Wall	Assumption	Minimum Acceptable Value	EV Impact per 10pp Deviation
B1	Rev CAGR >14%	>10% (otherwise growth narrative collapses)	-$8-12B
B2	NM 17% → 25%+	>20% (otherwise P/E cannot narrow)	-$10-15B
B3	Pricing power intact	v10 ≥ 1.2x (otherwise H2 confirmation)	-$15-25B
B4	DC penetration → 40%	>15% (otherwise DC narrative fails)	-$5-10B
B5	Arm China sustained growth	>0% (otherwise structural contraction)	-$3-5B
B6	SB does not reduce stake	<5%/year (otherwise free float impact)	-$5-10B

B3 is the only 'zero-tolerance' wall: B1/B2/B4/B5/B6 all have acceptable lower bounds (even if lower, valuation still has support). But B3 (pricing power) has no 'acceptable minimum value'—if RISC-V genuinely constrains ARM pricing (v10 < 1.2x v9), the entire 'ARM is Visa' narrative is debunked, and P/E could instantly compress from 142x to 60-80x (aligning with EDA peers).

A full RT-1b joint probability in-depth test will be conducted.

24.4 Kill Switch Interaction Matrix

KSs are not independent—the triggering of certain KSs significantly increases the probability of others. This interaction matrix is the investor's 'domino map':

KS Trigger → Impact	KS-1	KS-2	KS-3	KS-4	KS-5	KS-6	KS-7	KS-8
KS-1 (RISC-V DC)	—	+15%	0	+40%	+10%	+10%	0	0
KS-2 (SB Stake Reduction)	0	—	+20%	0	0	0	+15%	+20%
KS-3 (GAAP Gap)	0	+10%	—	0	+15%	0	+30%	+10%
KS-4 (Hyperscaler RISC-V)	+40%	0	0	—	+20%	+15%	0	0
KS-5 (Licensing Pipeline)	+5%	0	+10%	+10%	—	0	+10%	+15%
KS-6 (China Mandate)	+15%	0	0	+10%	+5%	—	0	0
KS-7 (DSO)	0	+10%	+30%	0	+15%	0	—	+10%
KS-8 (CEO Change)	0	+5%	0	0	+20%	0	0	—

Interpretation: After KS-1 is triggered, the conditional probability of KS-4 increases by 40% (=after RISC-V DC penetration exceeds 10%, the probability of a hyperscaler publicly announcing RISC-V increases from 25% to ~35%).

Most Dangerous Domino Chain:

Two Independent Domino Chains:

1. Competitive Chain (Left): KS-4→KS-1→KS-5→KS-8 — "RISC-V Breakthrough→Loss of Pricing Power→Pipeline Contraction→Management Reshuffle"
2. Financial Chain (Right): KS-7→KS-3→KS-2→KS-8 — "Revenue Quality Deterioration→Accounting Exposure→SoftBank Stake Reduction→Management Change"

Convergence Point: KS-8 (CEO Change) — The outcome from either chain is maximized strategic uncertainty.

Chain Reaction Probability Estimates:

• Competitive Chain Fully Triggered (5 years): KS-4(25%) × KS-1|4(35%) × KS-5|1(30%) × KS-8|5(25%) = 0.66%
• Financial Chain Fully Triggered (5 years): KS-7(25%) × KS-3|7(55%) × KS-2|3(45%) × KS-8|2(30%) = 1.86%
• At least one chain fully triggered: ~2.5%
• While the probability is low, either chain fully triggered = significant ARM valuation repricing (-50%+)

24.5 Kill Switch Time Series: FY2026-FY2028 Monitoring Calendar

Time Point	KS to Monitor	Data Source	Decision Impact
May 2026 (FY2026 Annual Report)	KS-3, KS-7	20-F	GAAP/Non-GAAP Gap + DSO Key Updates
August 2026 (Q1 FY2027)	KS-5	6-K	First Data Point for Licensing Pipeline
September 2026	KS-1	Industry	Tenstorrent RISC-V DC Chip Mass Production Assessment
November 2026	KS-6	China Policy	China's 15th Five-Year Plan Details (RISC-V Clauses)
February 2027 (Q2 FY2027)	KS-3, KS-5, KS-7	6-K	Simultaneous check on three KS
May 2027 (FY2027 Annual Report)	All KS	20-F	Annual Comprehensive Review
Mid-2027	KS-4	Industry	Hyperscale Client RISC-V Roadmap (If Any)
Q3 2027	KS-2	SoftBank	SoftBank Debt Refinancing Cycle + ARM Equity Disposal
Q1 2028	KS-1, KS-4	Industry	RISC-V DC Share Reaches 5% for the First Time? (If Trend Accelerates)

Investor Action Triggers:

• Yellow Light (Close Monitoring): Any single KS triggered
• Orange Light (Consider Reducing Position): 2 KS triggered in the same chain
• Red Light (Significant Reduction/Exit Position): 3+ KS triggered in the same chain, or 1 KS triggered in each of 2 different chains

---

Chapter 25: Deep Dive into Arm China and Geopolitical Risks

25.1 Arm China Overview

Arm China is ARM's exclusive operating entity in mainland China. In 2022, ARM regained control after the Allen Wu governance crisis (see Ch17A.5 for details):

Aspect	Details
Legal Entity	Arm Technology (China) Co., Ltd.
Chinese Control	51% operational control (Chinese shareholders)
ARM Economic Interest	~4.8% direct economic interest (indirectly via SoftBank)
FY2025 Revenue Contribution	~$800-900M (20-22% of ARM's total revenue)
Growth Rate	+7.5% YoY (vs ARM Group +24%)
Key Customers	Huawei/OPPO/vivo/Xiaomi (Mobile) + Unisoc/Allwinner/GigaDevice (IoT)

25.2 China Revenue Trends: Structural Deceleration Confirmed

Metric	FY2023	FY2024	FY2025	FY2026E	Trend
China Revenue ($M)	~$700	~$697	~$749	~$780	Low Growth
% of Total Revenue	~26%	~22%	19%	~16%	Continuous Decline
YoY Growth	—	~0%	+7.5%	+4%	Significantly Below Group
vs Group Growth	—	-21pp	-16pp	-18ppE	Gap Not Narrowing

Structural vs. Cyclical Diagnosis: China revenue's share dropping from 26% to 16% (a 10pp decline over 3 years) is not a cyclical fluctuation—it's a structural deceleration. Among the three driving factors (RISC-V substitution, export controls, cyclicality), RISC-V substitution and export controls are structural (accounting for 70% of explanatory power), while cyclicality accounts for only 30%. Even if the Chinese economy recovers, it is unlikely to reverse this trend.

25.3 Quantifying Three Geopolitical Scenarios

Path 1: De-escalation (Probability 20%)

Assumption: US-China relations significantly improve in 2027-2028, technology controls ease, and Huawei regains access to ARM's advanced cores.

• Arm China's growth recovers to ~15% CAGR (aligning with ARM Group)
• FY2030 revenue of $1.5-1.8B (vs. baseline $1.1B)
• Incremental $400-700M/year × 30x → +$12-21B market cap
• However, probability is low: US-China tech competition shows a trend of structural intensification.

Path 2: Stalemate (Probability 55%)

Assumption: The current state persists—controls neither intensify nor relax, and Arm China operates independently.

• Arm China's growth ~5-8% CAGR
• FY2030 revenue of $1.0-1.3B
• Largely priced in by the market
• China RISC-V Penetration: China could become the fastest-penetrating RISC-V market globally (driven by both policy and economy)
• Arm China's market share in China will slowly yield from ~80% to 65-70%

Path 3: Decoupling (Probability 25%)

Assumption: US-China technological decoupling intensifies—new export controls restrict ARM's advanced cores, or China massively shifts to RISC-V.

• Arm China revenue stagnation or decline (-5%~0% CAGR)
• FY2030 revenue of $0.5-0.8B
• Difference of $250-500M/year × 30x → -$7.5-15B market cap
• Extreme sub-scenario: China fully shifts to RISC-V (probability <10%) → Arm China shrinks to $300-500M

Probability-weighted Arm China FY2030 Revenue: 20%×$1.65B + 55%×$1.15B + 25%×$0.65B = $1.13B

25.4 Arm China Standalone Valuation

Method	Valuation Range	Assumptions
Revenue Multiple (P/S 10-20x)	$8.5-17B	IP Company P/S
EV/EBIT (25x)	$5-10B	EBIT $200-400M
DCF (Baseline Path)	$7-12B	8% CAGR + 25% margin
Composite	$7-13B	5-10% of ARM's Total Market Cap

Implication: Arm China's share of ARM's total market cap (5-10%) is significantly lower than its revenue contribution (20-22%)—the market has already applied a "geopolitical discount." However, if Path 3 materializes, the discount might be insufficient:

• Path 3 Valuation: $3-7B (vs. baseline $7-13B → additional loss of $4-10B)
• Impact on ARM's Total Market Cap: -3~7%

25.5 Taiwan Strait Conflict Scenario (Extreme Tail Risk)

Metric	Normal	Crisis (6 months)	Crisis (>1 year)
ARM Revenue Impact	—	-30~40%	-50~60%
ARM Market Cap Impact	—	-50~70%	-70~80%
5-Year Probability	—	5%	2%

Probability-weighted Impact: 5%×(-60%) + 2%×(-75%) = -4.5% EV

Due to its reliance on TSMC and exposure to China revenue, ARM is one of the most vulnerable semiconductor companies to a Taiwan Strait conflict (second only to TSMC).

25.6 Deep Dive into Taiwan Strait Conflict Scenario: ARM's Special Vulnerability

Why is ARM particularly vulnerable to a Taiwan Strait conflict?

Unlike other semiconductor companies, ARM faces a two-pronged strike from a Taiwan Strait conflict—not only supply chain disruptions (common to all semiconductor companies), but also simultaneous damage to its two major revenue streams:

Impact Channel	Mechanism of Impact	Share of ARM Revenue	Degree of Impact
TSMC Foundry Disruption	ARM licensees unable to produce → royalty stagnation	~60% (Indirect)	-40~60%
Direct Interruption of China Revenue	Sanctions/Bans/Arm China Operations Halt	~18% (Direct)	-80~100%
Global Demand Shock	Plummeting Consumer Electronics Demand	~100%	-20~30%
Combined Impact	ARM is the only major semiconductor simultaneously impacted across all three channels	—	-50~70%

Vulnerability Ranking Compared to Other Semiconductor Companies:

Company	TSMC Dependence	China Revenue	Demand Sensitivity	Overall Vulnerability
TSMC	100%	~10%	High	Extreme
ARM	~60% (Indirect)	~18%	High	Very High
NVDA	~80% (Indirect)	~12%	Very High	Very High
SNPS/CDNS	~30% (Indirect)	~15%	Medium	Medium
INTC	~10%	~28%	High	Medium-High (due to China exposure)

ARM Revenue Recovery Curve Under a Taiwan Strait Crisis:

Crisis Phase	Time	ARM Revenue (Annualized)	Recovery %
Normal	T-0	$4.9B (FY2026)	100%
Acute Shock	T+0~3 Months	$1.5-2.0B	30-40%
Adjustment Period	T+3~12 Months	$2.5-3.0B	50-60%
Slow Recovery	T+1~2 Years	$3.0-3.5B	60-70%
New Normal	T+2 Years+	$3.5-4.5B	70-90%

Permanent Loss: Even if the crisis is fully resolved, ARM may permanently lose ~10-20% of its revenue (due to a permanent shift in the Chinese market to RISC-V, and TSMC capacity reorganization leading some customers to non-ARM architectures).

25.7 Geopolitical Risk Hedging Strategies—What Can ARM Do?

Hedging Strategy	Feasibility	Effectiveness	Is ARM Executing?
Foundry Diversification (Samsung/Intel Foundry)	Medium	Medium	Indirect (customer decision)
Accelerate Non-China Market Growth	High	Medium	Yes (DC+Automotive)
Establish Geopolitical Insurance Mechanism	Low	Low	No
Reduce Dependence on a Single Market	Medium	High	Occurring Naturally
Sign "Security Agreements" with Various Governments	Medium	Medium	UK Headquarters has an advantage

ARM's Geopolitical Positioning: As a UK company (non-US), ARM holds a unique position between the US and China—not directly subject to US export controls (though still impacted), while maintaining relatively neutral relations with the Chinese government. This gives ARM greater geopolitical maneuvering room than NVDA/QCOM. However, SoftBank (Japan) control + NASDAQ listing still deeply embeds ARM within the US-Japan technology alliance framework.

25.8 Arm China IPO/Spin-off Scenario Analysis

An independent IPO or spin-off of Arm China is a low-probability but high-impact event. SoftBank considered an independent IPO for Arm China on Shanghai's STAR Market in 2021:

Four Paths for an IPO/Spin-off:

Path	Description	Probability (3 Years)	Impact on ARM
A: STAR Market IPO	Arm China independently listed on Shanghai's STAR Market	10%	+$5-10B (Control Premium)
B: Privatization + A-share Listing	Chinese shareholders privatize Arm China → Backdoor listing on A-share market	5%	-$3-5B (Loss of Control)
C: ARM Increases Stake	ARM increases economic interest in Arm China to >50%	10%	+$2-5B (Integration Premium)
D: Status Quo	Maintain current 4.8% economic interest + operational control	75%	0 (Already priced in)

Detailed Analysis of Path A (STAR Market IPO):

If Arm China is listed on the STAR Market with a valuation of $10-15B:

• ARM holds 4.8% economic interest → $0.5-0.7B direct equity value (minor)
• But ARM retains exclusive licensing relationship → can recognize $800M+/year in royalty/licensing revenue (high value)
• A STAR Market listing means Arm China becomes more independent → potentially developing its own IP in the long term (competing with ARM?)
• Net Effect: Short-term neutral to positive (valuation transparency + capital injection), long-term risk (independence → competition)

Risks of Path B (Privatization):

Worst-case scenario—Chinese shareholders use their 51% controlling stake to push for privatization:

• ARM loses operational influence over Arm China
• Royalty agreements may be renegotiated upon expiry (on unfavorable terms)
• The Chinese government may pressure for lower royalty rates
• Valuation Impact: Arm China from $7-13B (ARM consolidated) → $3-5B (discount due to loss of control)

Possibility of Path C (ARM Increased Stake):

ARM may wish to increase its stake in Arm China to strengthen control:

• Requires Chinese regulatory approval (SAMR) — uncertain
• Requires SoftBank funding — tight on capital
• Estimated acquisition cost: $5-8B (50%+ of Arm China's valuation of $10-15B)
• If successful: ARM revenue: +$500M in direct profit (currently only collects royalties), market assigns integration premium

Combined Probability Weighted: 0.10×$7.5B + 0.05×(-$4B) + 0.10×$3.5B + 0.75×$0 = +$0.9B — small but positive expected value. The key is to monitor Path B (privatization risk).

25.9 Summary of Arm China's Comprehensive Impact on ARM Valuation

Factor	EV Impact	Certainty	Time Horizon
Structural slowdown (decreasing China share)	Priced in	High	Ongoing
RISC-V Substitution (Probability Weighted)	-$4.8B (-3.5%)	Medium	3-5 years
Geopolitical Decoupling (Path 3)	-$7.5-15B (25% Probability)	Medium	2-5 years
Taiwan Strait Conflict (Tail Risk)	-$6.1B (Probability Weighted -4.5%)	Low	5+ years
IPO/Spin-off (Expected)	+$0.9B	Low	3+ years
Combined Probability Weighted	-$3-5B (-2~4% EV)	—	—

Conclusion: The probability-weighted net impact of Arm China on ARM's valuation is approximately -2~4% — not a decisive factor, but an ongoing structural drag. The real risk is not in the baseline path (already priced in), but in Path 3 (decoupling) and Taiwan Strait tail risk — these two low-probability events, if they occur, could lead to an additional loss of -15~20%.

25.10 China Semiconductor Self-Sufficiency Timeline and ARM Impact

Year	China Self-Sufficiency Milestone	ARM Impact	Probability
2026	China RISC-V IoT chips fully commercialized	Arm China IoT royalties -10~15%	70%
2027	Huawei HiSilicon releases RISC-V hybrid SoC	Significance > Actual Impact	40%
2028	First commercial China RISC-V mobile chip	Arm China mobile royalties begin to erode	25%
2029	China RISC-V servers enter government procurement	Arm China DC revenue stagnates	35%
2030	China RISC-V ecosystem becomes independently mature	Arm China market share <50%	20%

---

Chapter 26: Red Team Seven Questions (RT-1~RT-7)

Red Team Principles and Methodology

The purpose of the Red Team is to challenge our own analysis — not to overturn conclusions, but to test their robustness. The Red Team adheres to three iron rules:

1. Truly challenge core assumptions (not performative rhetoric) — each RT must have a quantifiable impact
2. Validity Gate: If all 7 RTs point in the same direction or the net effect is <5pp, mark as "performative" and redo
3. Two-way Calibration: RT not only examines "if we are overly pessimistic" but also "if we are overly optimistic"

Red Team Coverage Matrix:

RT#	Challenge Objective	Associated CQ	Associated Hypothesis	Direction	Impact Level
RT-1	RISC-V Ecosystem Never Catches Up	CQ-4	H2	↑	High
RT-2	45% Net Margin Achievable	CQ-1	H1	↑	High
RT-3	Phoenix Achieves Great Success	CQ-6	H4	↑	Medium
RT-4	Ecosystem Lock-in Underestimated	CQ-4	H2	↑	Medium
RT-5	SoftBank is a Protector	CQ-3	H5	↑	Medium
RT-6	DC Growth Exceeds Expectations	CQ-2	—	↑	High
RT-7	WACC is Too High	—	—	↑	Medium
RT-8	AI Democratization Weakens Pricing Power	CQ-1	H2	↓	High
RT-9	ARM Chiplet Ecosystem Valuation	CQ-6	—	↑	Medium
RT-10	v10 Delayed by 3 Years	CQ-4	H2	↓	High

Key Improvement: Added RT-8/RT-9/RT-10 to achieve two-way Red Team — the initial 7 RTs all had an "upward" direction (= systematic bias), adding 3 "downward" directions ensures the Red Team is truly bidirectional.

Red Team Methodology Description:

Why Red Team Does Not Equal "Bearish"? The purpose of the Red Team is to **test our analytical framework** – if our analysis is bearish (prior valuation: 37-41% overvalued), the Red Team should challenge from a bullish perspective. This is why RT-1~RT-7 (bullish challenges) are reasonable. Meanwhile, RT-8~RT-10 examine from a bearish perspective "whether we are not bearish enough" – ensuring the Red Team is not just "speaking for the bulls".

"Necessary but Not Sufficient" Criteria for Red Team Effectiveness:

• ✅ Directional Balance: At least 2 bearish (↓) directions (present)
• ✅ Quantitative Impact: Each RT has Probability × Impact (present)
• ✅ Genuine Challenge: RT changed at least one key assumption (RT-5 SoftBank revision)
• ✅ Core Robustness: Rating maintained after Red Team (present)
• ⚠️ Boundary Conditions: "If ARM's business model shifts from IP → Chiplet" paradigm change not fully tested – subsequent analysis needed

RT Execution Standard Comparison (EVO Accumulated Experience):

Standard	AMAT Lesson	MSFT Lesson	ARM Initial Version	ARM Revised Version
Directional Balance	All ↑ (Failed)	All ↑ (Performative)	All ↑ (Failed)	7↑+2↓+1≈(✓)
Net Effect	+$15B (0pp)	+$50B (0pp)	+$43B (0pp)	+$27B (+12pp)
Core Assumption Change	0	0	0	1 (RT-5)
Rating Change	None	None	None	None (but magnitude narrowed)
Effectiveness	Failed	Performative	Failed	7.8/10 (Passed)

---

RT-1: What if RISC-V can never catch up to ARM in software ecosystem?

Challenge Objective: H2 (RISC-V=CDS) hypothesis and CQ-4 (RISC-V pricing ceiling)

Core Argument: We assume RISC-V's software ecosystem will catch up to ARM by 2027-2028, but this may be overly optimistic. Establishing a software ecosystem is not a linear process – there's a "critical mass" effect:

Software Ecosystem Maturity Comparison (6 Dimensions):

Dimension	ARM	RISC-V	Gap (Years)	Catch-up Difficulty
Operating Systems	Linux/Windows/macOS/Android	Linux (complete) + Android (GKI in progress)	2-3 years	Medium
Compiler Toolchains	GCC/LLVM/Proprietary (Keil, etc.)	GCC/LLVM (basic complete)	1-2 years	Low
Debugging/Analysis	DS-5/Trace/CoreSight complete	Basic GDB/OpenOCD	3-5 years	High
Safety Certification	ISO 26262/IEC 61508/DO-178C	Virtually blank	5-10 years	Extremely High
Enterprise Middleware	Java/Oracle/SAP fully optimized	Basic support	3-5 years	High
AI/ML Frameworks	TensorFlow/PyTorch/ONNX optimized	Basic CPU support	2-3 years	Medium

Key Finding: Safety certification is the **biggest bottleneck** for RISC-V to catch up with ARM. Automotive (ISO 26262 ASIL-D) certification costs $10-50M + 2-3 years per instance. Aerospace (DO-178C Level A) is even more stringent. Even if RISC-V hardware matures, the lack of safety certification will set it back 5-10 years behind ARM in the automotive/industrial/aerospace sectors.

Quantified Impact Pathway:

Detailed Valuation Impact Calculation:

• B3+B1 Enhancement → FY2030 Royalties $7-9B (vs probability-weighted $4.95B)
• Net Margin 30% (after calibration) → NI $2.1-2.7B
• Terminal P/E 35x (full pricing power = sustained high growth) → Market Cap $73.5-94.5B (2030)
• Discounted (10%, 4 years): $50-65B
• Plus recent cash flow PV + licensing revenue: $105-135B
• Uplift: +$25-50B (+18-37%)

Rebuttal:

1. RISC-V doesn't need to "catch up" in the entire ecosystem – it only needs to establish a "good enough" ecosystem in IoT/DC/automotive to limit ARM's pricing power. Historical analogy: Linux never caught up to Windows but was sufficient to limit Microsoft's OEM pricing power.
2. While ARM's 5-10 year lead in safety certification is strong, it only protects ARM's royalties in automotive/industrial ($200-500M/year) – it does not protect the 60% share in mobile.
3. Google and the Chinese government are separately driving RISC-V adoption from the consumer and policy ends – this "dual-engine" impetus is historically rare.

Probability Assessment: RISC-V software never catches up to ARM (all dimensions) = **15%**
Net Effect: 15% × $37.5B (median) = **+$5.6B**

RT-1 Sensitivity: If only "safety certification area does not catch up" (more reasonable intermediate scenario, 40% probability):

• Protects ARM automotive + industrial royalties $300-500M/year
• Valuation impact: +$9-15B × 40% = +$3.6-6.0B
• This intermediate scenario is more likely to occur and the impact remains significant.

Historical Analogy: x86 vs ARM Ecosystem Catch-up Lessons

ARM took 15 years to move from embedded to server (2012 Calxeda → 2027 estimated scale-up) – the actual path was slower than anyone predicted:

Milestone	Expected Time	Actual Time	Delay
First ARM server chip	2012	2012	0 years
First hyperscaler adoption	2014-2015	2018 (AWS Graviton 1)	3-4 years
ARM DC Revenue >$100M	2018	2024	6 years
ARM DC Share >10%	2020	2027E?	~7 years

If RISC-V follows a similar trajectory (first RISC-V server chip 2024 → scaling 2035-2040), ARM might have an 8-10 year first-mover advantage window in DC—much longer than our 2-3 year assumption.

However, the key difference: RISC-V's starting point is much better than ARM's was back then:

• ARM had no open-source community (2012) vs. RISC-V has a global open-source ecosystem (2024)
• ARM had no hyperscaler support (initial stages) vs. RISC-V has Google + Alibaba + SiFive
• ARM faced a mature x86 ecosystem vs. RISC-V faces ARM, which is also catching up
• Conclusion: RISC-V's catch-up speed might be 2-3x faster than ARM's was back then → but it will still require 5-8 years, not 2-3 years

Counter-counterargument (three-layered dialectic):

1. Layer 1: "RISC-V doesn't need to catch up completely—it just needs to be 'good enough' to cap pricing"
2. Counter: "However, 'good enough' is defined differently in different markets—'good enough' for IoT (already achieved) vs. 'good enough' for DC (5 years away) vs. automotive certification (10 years away)"
3. Synthesis: RISC-V will reach "good enough" for IoT/MCU in 2027-2028, "good enough" for DC in 2029-2030, and automotive certification level in 2033+. ARM's pricing power window will narrow market by market, not "all suddenly lost"

---

RT-2: What if ARM's 35-45% Net Margin Assumption is Achievable?

Challenge Goal: We use a 25% net margin in E3/E4 (vs. consensus implied 37-45%)

Core Argument: Our GAAP framework for net margin might be systematically understated. Requires serious examination of Non-GAAP margin paths:

Margin Breakdown: Three Paths from FY2026 to FY2030

Item	FY2026 (Actual)	Path A: Conservative (GAAP)	Path B: Baseline (adj GAAP)	Path C: Optimistic (Non-GAAP)
Revenue	$4.9B	$7.5B	$8.0B	$9.0B
Gross Margin	95%	94%	95%	95.5%
R&D/Rev	56%	45%	40%	36%
SG&A/Rev	19%	15%	13%	12%
SBC/Rev	15%	12%	8%	5%
GAAP OM	5%	22%	34%	42.5%
Tax Rate	15%	16%	15%	14%
GAAP NM	12%	18.5%	28.9%	36.6%

Path Probability Assessment:

• Path A (Conservative GAAP): SBC remains high + R&D does not decrease → Probability 40%
• Path B (Baseline adj): SBC gradually normalizes + R&D moderately decreases → Probability 40%
• Path C (Optimistic Non-GAAP): SBC rapidly normalizes + R&D significantly optimizes → Probability 20%
• Probability-Weighted NM: 40%×18.5% + 40%×28.9% + 20%×36.6% = 26.3%

Key Variables for SBC Normalization:

The peak in ARM's SBC post-IPO resulted from concentrated RSU grants. Historical SBC evolution for semiconductor companies post-IPO:

Company	IPO Year	SBC/Rev IPO+1 Year	IPO+3 Years	IPO+5 Years	Steady State
Arm	2023	17%	15%(current)	?	?
Snowflake	2020	30%	22%	18%	~15%
Palantir	2020	50%	25%	15%	~12%
Cloudflare	2019	18%	14%	11%	~10%
CrowdStrike	2019	15%	12%	9%	~8%
Median Trend	—	~20%	~15%	~12%	~10%

Based on Peers, ARM SBC Path Forecast: FY2026: 15% → FY2028: 10-12% → FY2030: 8-10%

If SBC Reaches 8% by FY2030 (Peer Median Steady State):

• GAAP NM increases from 25% (our baseline) to ~29-30%
• This makes Path B the most likely steady state
• E3/E4 valuation revised upwards accordingly: $80-85B → $95-105B

Valuation Impact:

• Probability-weighted NM 26.3% (vs. our 25%): Minor adjustment +$3-5B
• If Path B is realized (40% probability, NM 29%): +$15-20B
• If Path C is realized (20% probability, NM 37%): +$30-40B
• Probability-weighted Upside Revision: 40%×$17.5B + 20%×$35B = +$14B

Rebuttals:

1. SBC "normalization" ≠ SBC reduction—ARM may continue to grant high-value new RSUs to retain talent (AI talent war)
2. A reduction in R&D from 56% to 36% means ARM is cutting over $1.5B in R&D spending—is this wise given intensifying RISC-V competition?
3. Path C (Non-GAAP 37%) requires simultaneously achieving an SBC reduction to 5% + R&D reduction to 36%—historically, only integrated companies like Broadcom have achieved this, with no precedent for pure-play IP companies

ARM vs. Synopsys/Cadence Margin Comparison (Terminal State Reference):

ARM is frequently benchmarked against SNPS/CDNS—but the profit margin structure differs significantly:

Metric	ARM FY2026	SNPS	CDNS	ARM Gap	Likelihood of Convergence
Gross Margin	95.4%	80%	89%	+6-15pp	ARM already superior
R&D/Rev	56%	30%	29%	+26pp	Low (ARM's R&D is its product)
SBC/Rev	15%	8%	7%	+8pp	Medium (IPO effect subsides)
GAAP OM	5%	30%	38%	-25~33pp	Difficult to converge within 5 years
Non-GAAP OM	40.7%	36%	42%	Already comparable	—

Key Insight: ARM's Non-GAAP OM (40.7%) is already comparable to CDNS (42%)—the gap is almost entirely due to SBC. If investors use a Non-GAAP framework, ARM's profit margins are already "normal." If using a GAAP framework, ARM's profit margins significantly lag behind peers.

This framework choice is "you buy what you see"—Non-GAAP investors see a high-margin IP company with 40%+ OM → fair valuation $110-130B. GAAP investors see a loss-making company with 5% OM → fair valuation $50-70B. The true answer lies somewhere in between—but the market pricing ($136B) is only justifiable under a Non-GAAP framework.

Probability Assessment: FY2030 GAAP NM >30% = 35% (Upgraded from 25% in the initial version, supported by SBC peer benchmarking)
Net Effect: +$14B × probability calibration ≈ +$9.8B

The "Hidden Minefield" of SBC Normalization: RSU Grant Pace Analysis

SBC "normalization" has an overlooked prerequisite—ARM must actively reduce new RSU grants. However, in the AI talent war:

Year	New RSU Grants (Est.)	Vesting/Expiration	Net SBC Change	Driver
FY2025	~$700M	~$400M	+$300M inventory	Significant grants post-IPO
FY2026E	~$600M	~$500M	+$100M	Slowing grants but accelerating vesting
FY2027E	~$500M	~$550M	-$50M	First net reduction (if grants are moderated)
FY2028E	~$450M	~$500M	-$50M	Gradual normalization

Key Risk: If ARM initiates large-scale DC/Phoenix hiring in 2026-2027 (estimated +2000 people required), a $300K RSU package per person = +$600M new grants → SBC normalization delayed by 2-3 years. The talent requirements of the Phoenix initiative could be the biggest obstacle to SBC normalization.

The Chicken-and-Egg Dilemma of SBC and Employee Growth:

• ARM needs to expand (DC/Phoenix) → requires top talent → requires significant RSU grants → SBC remains elevated
• ARM needs profit margin expansion (Wall Street expectation) → requires SBC reduction → limits hiring → slows expansion
• These two goals are mutually exclusive in FY2027-2029—this is a contradiction not yet fully understood by investors

Counter-argument: Broadcom (AVGO) achieved NM of 35%+ through an acquisition + layoff path—is it possible for ARM to take a similar path?

• Extremely low probability: ARM is a pure IP company with no redundant product lines to cut; ARM's culture is engineering R&D-driven, not cost-optimization-driven
• Only path: Revenue growth to depress SBC/Revenue (expanding the denominator) rather than an absolute decline in SBC → requires revenue CAGR >20% for 3 consecutive years

---

RT-3: What if Phoenix successfully transforms ARM from an IP licensor into a chip company?

Challenging Premise: H4 (Phoenix Royalty Cannibalization) Hypothesis—we position Phoenix as a "risk" but also a potential "opportunity"

Conditions Chain for Phoenix Success:

Three Potential Outcomes for Phoenix:

Outcome	Description	FY2030 Phoenix Revenue	Royalty Cannibalization	Net EV Impact	Probability
Major Success	DC CPU Share >10%	$3-5B	-$0.5-1B	+$25-50B	10%
Moderate Success	DC CPU Share 3-5%	$1-2B	-$0.3-0.5B	+$5-15B	20%
Failure	No volume production/customers	<$0.3B	-$0.1B	-$3-5B(R&D waste)	35%
Backfire	Customers accelerate RISC-V adoption	<$0.5B	-$1-2B	-$15-25B	15%
Cancellation	ARM cancels Phoenix	$0	$0	+$1-3B(R&D savings)	20%

Probability-Weighted Phoenix Net Effect:

The Core Paradox of Phoenix: The probability-weighted net effect is only +$1.75B—almost zero. This means Phoenix is almost neutral to ARM in terms of expected value. Bulls are betting on "Major Success" (10% probability), while bears are betting on "Backlash" (15% probability). Current market pricing might already imply "Moderate Success" (20% probability).

Key Observation: Phoenix's biggest risk is not "Failure" (losses are controllable), but "Backlash"—ARM successfully developing chips but accelerating licensees' migration to RISC-V. This is a nonlinear risk: the greater the success, the stronger the backlash.

The Transmission Mechanism of Phoenix's "Backlash":

Each step in this transmission chain has a reasonable probability (60-80%). However, the cumulative probability of 7 sequential steps = (0.7)^7 ≈ 8%—which is close to our "Backlash" probability estimate (15%), indicating that while the backlash path is complex, the credibility of each link is high.

Analogy between Phoenix and Intel Foundry Services (IFS):
Intel announced IFS in 2021, attempting to transition from a chip designer to a foundry → Results:

• Angered TSMC (reduced Intel's priority)
• Made Apple/QCOM more steadfast in using TSMC (no designs given to Intel)
• IFS revenue remains negligible to date (external customers <$200M out of FY2024 ~$900M)
• Intel stock price fell from 2021 high of $68 → 2025 low of $18 (-74%)

Limitations of the Analogy: ARM is not "transforming into a competitor" (Intel as a foundry = competing with TSMC), but rather "expanding downstream" (ARM making chips = competing with its customers). The backlash from the latter is more direct—licensees are your customers, and the competitive relationship is irreconcilable.

Counter-argument: However, ARM can avoid direct competition through "differentiated positioning":

• Phoenix focuses on SoftBank's internal demand (not the open market) → reduced customer resentment
• Phoenix as a "reference design" rather than a competing product → similar to Google Pixel's positioning
• Counter: SoftBank's internal demand ~100,000 units/year → insufficient to support Phoenix's scale → must enter the open market → backlash inevitable

---

RT-4: Have We Underestimated ARM's Ecosystem Lock-in?

Challenge Target: A-Score 7.17 (moat decline amidst migration) and RISC-V penetration expectations

Five Lines of Defense for Ecosystem Lock-in:

Defense Line	Description	Difficulty to Break Through	RISC-V Progress	Estimated Duration
L1: Instruction Set	ARM ISA Compatibility	Low (RISC-V already has it)	Already Broken Through	0 years
L2: Development Tools	Compilers + IDE + Debugging	Medium	Basic Completion	1-2 years
L3: Operating Systems	Linux/Android/RTOS	Medium	In Progress	2-3 years
L4: Middleware/Applications	Enterprise Software + AI Frameworks	High	Just Starting	3-5 years
L5: Certification Ecosystem	Security/Automotive/Aerospace	Extremely High	Almost Blank	5-10 years

Economic Quantification of Ecosystem Lock-in:

For a medium-sized chip design company (annual revenue $1-3B), the cost of migrating from ARM to RISC-V:

Cost Item	Amount	Duration	Can be Gradual?
Toolchain Migration	$5-10M	1 year	Yes
Software Re-validation	$10-30M	1-2 years	Partially
Security Certification (if required)	$10-50M	2-3 years	No
Talent Training/Recruitment	$5-15M	2-3 years	Yes
Customer Validation/Certification	$5-20M	1-2 years	Partially
Total Migration Cost	$35-125M	2-4 years	—

Compared to ARM Royalty Costs: Annual royalties are $5-20M (depending on shipment volume). Even if ARM royalties double ($10-40M/year), migration costs would still take 3-6 years to pay back. This is ARM's "CDS pricing anchor"—ARM can raise prices below the "annualized migration cost" without fear of customer churn.

Annualized Migration Cost vs. ARM Royalties:

Company Type	Total Migration Cost	Annualized (5 years)	Current ARM Royalties/Year	ARM Safe Price Increase Headroom
Large (QCOM/MTK)	$80-125M	$16-25M	$200-400M	Low (already major customers, strong bargaining power)
Medium (UNISOC/MediaTek)	$50-80M	$10-16M	$20-50M	Medium (price increase headroom $10-16M)
Small (IoT Design)	$35-50M	$7-10M	$2-5M	Negative (royalties < migration cost → already locked-in)

Key Finding: Small customers are deeply locked into the ARM ecosystem (royalties are far below migration costs), but large customers have bargaining power. ARM's pricing strategy should be "maintain/slightly increase for small customers, conservative for large customers"—but v10 might attempt to uniformly raise prices for all customers, which would trigger RISC-V evaluations by large customers.

Migration Cost "Decay Curve" — Lock-in is Not Forever:

As the RISC-V ecosystem matures, migration costs decrease year by year:

Year	RISC-V Toolchain Cost	Software Validation Cost	Total Migration Cost (Mid-sized)	vs. ARM Royalty Ratio
2024	$10M	$30M	$80M	4x(Strong Lock-in)
2026	$7M	$20M	$55M	2.8x(Medium Lock-in)
2028	$5M	$12M	$35M	1.8x(Weak Lock-in)
2030	$3M	$8M	$22M	1.1x(Tipping Point)
2032	$2M	$5M	$15M	0.75x(Reversal → RISC-V Cheaper)

Tipping Point: Around 2030, the annualized total RISC-V migration cost for mid-sized customers ≈ ARM royalties → ARM completely loses pricing power over mid-sized customers → retains pricing power only for small customers (royalties < migration cost).

This timeline is consistent with our RISC-V timeline assumption (2028-2029 "good enough") → but the loss of pricing power will lag by 1-2 years (contract cycle). The actual impact will become apparent in 2030-2032.

Double Check: This migration cost decay model complements the RT-1 security certification analysis—

• RT-1: Analyzes from a "technical feasibility" perspective → security certification sector has the longest lock-in (5-10 years)
• RT-4: Analyzes from an "economic feasibility" perspective → total migration cost reaches a tipping point in 2030
• Both analyses converge to the same conclusion: ARM faces a pricing power inflection point in 2028-2030

Valuation Impact:

• Ecosystem lock-in maintained >10 years (30% probability): DCF places higher weight on near-term cash flows → +$15-25B
• Net Effect: 30% × $20B = +$6.0B

The "Boiling Frog" Erosion Model for Ecosystem Lock-in:

Even if ARM's ecosystem lock-in is extremely strong today, it may be subject to gradual erosion rather than a sudden breakthrough:

Time	Erosion Event	Lock-in Strength (0-10)	ARM Pricing Power Impact
2024	RISC-V IoT Shipments >1B	10→9	Minimal (IoT royalties are thin)
2025	Google Android GKI supports RISC-V	9→8	Small (Android developers begin to pay attention)
2026	SiFive/Ventana first DC chip mass production	8→7	Medium (DC customers have alternatives)
2027	RISC-V Linux enterprise certification	7→6	Medium (Enterprise IT begins evaluation)
2028	First RISC-V Android Flagship Phone	6→5	Significant (Mobile signal effect)
2029	RISC-V automotive chip ISO 26262 ASIL-B certification	5→4.5	Medium (First step in automotive)
2030	RISC-V DC market share >5%	4.5→4	Significant (Pricing ceiling forms)

Key Insight: Following the "boiling frog" path, ARM's ecosystem lock-in will not collapse suddenly—instead, it gradually decreases from 10/10 in 2024 to 4/10 in 2030. Each 1-point drop has little impact, but the cumulative effect becomes undeniable at a certain "tipping point" in 2028-2029.

"Tipping Point" Criteria:

• RISC-V reaches 10% market share in any single end market = ARM loses pricing power in that market
• IoT is expected first (2025-2026), followed by DC (2029-2030), then mobile (2030+), and automotive furthest (2033+).
• When ≥3 end markets reach the tipping point → ARM's overall pricing power is impaired → expected to occur in 2030-2032.

Counter-Rebuttal: The "boiling frog" scenario could be interrupted by a "major leap" in ARM v10:

• If v10 introduces revolutionary features in 2027-2028 (e.g., native CXL support, secure enclave standardization)
• → Ecosystem lock-in could rebound to 8-9/10 (customers stay in the ARM ecosystem for new features)
• → But this presumes v10 truly offers "revolutionary" features—otherwise, it's merely an incremental upgrade.

---

RT-5: SoftBank a Protector, Not a Threat?

Challenge Objective: The "threat" aspect of CQ-3 (SoftBank's Double-Edged Sword)

Arguments for SoftBank as a Protector:

Argument	Quantification	Durability	Credibility
Strategic Patience (Long-term Investment)	Son has held ARM for 8 years without divestment	Medium (depends on funding needs)	Medium
Acquisition Defense (90% Stake)	Blocked NVIDIA's $40B acquisition	High	High
Affiliated Purchasing (Stimulates Revenue)	+$450-730M/year in licensing	Low (SoftBank is cash-strapped)	Medium
AI Infrastructure Synergy (Izanagi/Stargate)	Indirect ARM royalties +$200-400M	Medium	Medium
Float Scarcity (Supply-Demand Imbalance)	9.7% float → partially justifies 142x P/E	Low (disappears upon divestment)	Low

SoftBank Protector/Threat Probability Evolution:

Time Window	Protector Probability	Threat Probability	Driving Factor
0-1 year	70%	30%	No urgent funding needs
1-3 years	50%	50%	Increased Izanagi funding requirements
3-5 years	35%	65%	Debt maturities + accumulated divestment pressure
5-10 years	20%	80%	Long-term divestment inevitable

Probability-Weighted Net Effect (5-year perspective): ~40% of the time as protector × (+$10B) + ~60% of the time as threat × (-$15B) = -$5B

Rebuttal:

• Float scarcity premium is "borrowed valuation"—not ARM's intrinsic value
• Related-party procurement is "circular credit"—SoftBank's procurement for ARM is financed by ARM's stock appreciation → unsustainable
• Acquisition defense decreases in value after ARM's IPO—ARM can independently raise capital

Probability Assessment: SoftBank's net impact within a 5-year period is positive = 30% (downgraded from initial 35% due to increased Izanagi financing pressure)
Net Effect: 30% × $10B - 70% × (-$5B negative impact) = +$3B - $3.5B = -$0.5B

Important Correction: The initial version calculated the net effect of RT-5 as +$3.5-7.0B, but after two-way probability weighting, it is actually closer to neutral to negative. This is a key correction from the Red Team—SoftBank is more likely to experience a net negative impact over a 5-year period (sell-down pressure > protection value).

Game Theory Analysis of SoftBank's Sell-Down Paths:

SoftBank faces a multi-stage game—not a binary "sell-down/no sell-down" choice:

Strategy	Description	Impact on ARM Share Price	Impact on SoftBank NAV	Probability
S1: No Sell-Down	Maintain ~90% stake	Neutral (low float scarcity persists)	Paper gains but poor liquidity	25%
S2: Slow Sell-Down	1-2% annually, reduced to 80% in 5 years	Moderate downward pressure (supply increase)	Gradual monetization, manageable impact	35%
S3: Large Sell-Down	One-time 5-10% (secondary offering)	Severe downward pressure (-15~25%)	Significant monetization but price discount	20%
S4: Collateralized Financing	No sale but pledge more shares	Indirect downward pressure (leverage risk)	Maintain stake but increase vulnerability	15%
S5: Forced Liquidation	Margin call → forced large-scale selling	Devastating (-30~50%)	Passive sell-down + massive losses	5%

Probability-Weighted Impact:

vs. Protector Value: Strategic value provided by SoftBank during all strategies (acquisition defense + related-party procurement) ≈ +$5-8B

Net Effect = +$6.5B (protection) - $9.7B (sell-down risk) = -$3.2B → more pessimistic than our previous -$0.5B

Adjustment: However, this calculation assumes all sell-down occurs within 5 years. If spread over 10 years, the NPV impact is approximately half → -$1.6B

Taking the midpoint between conservative and previous estimates: -$0.5B ↔ -$1.6B → approx. -$1.0B

SoftBank's ARM Holding "Prisoner's Dilemma":

• SoftBank holds ARM because ARM's share price is high (=NAV support)
• ARM's high share price is partly due to SoftBank not selling down (=low float scarcity)
• If SoftBank sells down → ARM's share price falls → SoftBank's NAV decreases → potentially triggering debt covenants → leading to more forced sell-down
• This is a positive feedback loop—once started, it's hard to stop
• Trigger condition: ARM's share price falls to SoftBank's margin call threshold (estimated $80-90/share)
• Current buffer: $128→$85 ≈ -34% → still a significant safety cushion

---

RT-6: Data Center Growth May Exceed Expectations?

Challenge Target: Baseline projection for CQ-2 (Data Center Quest)

Quantitative Validation of AI CapEx Supercycle:

Year	Hyperscale AI CapEx ($B)	YoY	ARM DC Royalties (Est.)	DC % of ARM Royalties
2023	~$150	—	~$100M	~5%
2024	~$200	+33%	~$200M	~8%
2025	~$280	+40%	~$350M	~12%
2026E	~$350	+25%	~$500M	~15%
2027E	~$400	+14%	~$700M	~18%
2028E	~$430	+8%	~$900M	~20%

CSS (Compute Sub-System) Growth Flywheel:

CSS Customer-by-Customer Analysis (19 known licenses):

Customer	CSS Type	Status	Estimated Mass Production	Annual Shipments (Est.)	Royalty/Chip (Est.)	Annual Royalties (Est.)
AWS	Graviton 4/5	In Mass Production	2024	1-2M	$50-100	$50-200M
Microsoft	Cobalt/In-house	In Mass Production	2025	0.5-1M	$50-80	$25-80M
Google	Axion	In Mass Production	2024	0.5-1M	$50-80	$25-80M
NVIDIA	Grace	In Mass Production	2024	0.3-0.5M	$80-120	$24-60M
Ampere	Altra Max/AmpereOne	In Mass Production	2022	0.2-0.4M	$30-60	$6-24M
Oracle	In-house Arm DC	In Design	2026-2027	0.2-0.5M	$50-80	$10-40M
Meta	In-house Chip	In Design	2026	0.3-1M	$50-80	$15-80M
SoftBank	In-house (Izanagi)	In Design	2027	0.1-0.3M	$50-80	$5-24M
Other 11 Companies	Various	In Design	2027-2029	0.1-0.5M each	$30-80	$3-40M each
Total (5 in Mass Production)		Mass Production		~3-5M		$130-444M
Total (All 19)		Including In Design		~5-12M		$250-900M

Current DC Royalty Anchor: FY2026 ARM DC royalties ~$350M → Consistent with the estimated $130-444M range for "5 mass production customers". This indicates that current DC royalties are primarily contributed by the four major customers: AWS/Google/NVIDIA/Microsoft.

Key Bottlenecks from 5 → 15 in Mass Production:

1. TSMC Capacity Allocation: N3/N2 capacity is limited → New customers may receive secondary capacity → Performance substandard → Cancellation
2. Software Ecosystem Readiness: Middleware porting for enterprise customers (Oracle/SAP) requires 1-2 years → DC ARM adoption curve is S-type, not linear
3. ROI Validation: First batch of non-hyperscaler customers need to demonstrate ARM server TCO superiority over x86 → Validation requires 6-12 months
4. Design Cancellation Rate: Historically, the mass production cancellation rate for chip designs is approximately 30-40% → Only 11-13 out of 19 licenses may ultimately reach mass production

Quantified Impact of Accelerated CSS Mass Production:

Number of CSS in Mass Production	ARM DC Royalties ($B)	% of Total Royalties	Probability (FY2030)
5 (Current)	$0.35B	12%	Achieved
10	$0.8-1.2B	20-25%	50%
15	$1.5-2.5B	30-40%	30%
19 (All)	$3.0-4.0B	45-55%	10%

Historical Reference for CSS Success Rate (ARM Itself):

ARM's historical success rate for CSS (Cortex-M/A series) from licensing to mass production in the mobile sector:

• Cortex-A9 (released 2007): 45+ licenses → 30+ mass production = ~67% success rate
• Cortex-A15 (2010): 20+ licenses → 12+ mass production = ~60%
• Cortex-A72 (2015): 30+ licenses → 22+ mass production = ~73%

DC CSS Success Rate May Be Lower:

• DC chip design cycles are longer (2-3 years vs mobile 1-2 years)
• DC customers are hyperscalers (with ability to develop in-house alternatives)
• TSMC N3/N2 capacity bottlenecks increase cancellation risk
• Estimated DC CSS Success Rate: 55-65% → 19 licenses × 60% = 11-12 ultimately reaching mass production

If DC Significantly Exceeds Expectations (15+ CSS in Mass Production):

• FY2030 DC royalties $2.5-4.0B (vs. baseline $1.2B)
• Incremental $1.3-2.8B × 30x P/E = +$39-84B
• Probability-weighted: 30% × $60B (midpoint) = +$18B

Rebuttal:

1. Hyperscale customers are "smart buyers" – Google/Meta simultaneously invest in ARM and RISC-V, unwilling to bet on only one.
2. AI CapEx cycles may slow down in 2027-2028 (historically, +30% CapEx growth has never sustained for >3 years)
3. 14 out of 19 CSS licenses not in mass production – customer cancellations/delays are common (typically 30-40% of design wins do not ultimately reach mass production)
4. CSS chip shipments are constrained by TSMC capacity – ARM does not control capacity

Probability Assessment: DC revenue significantly exceeds expectations (FY2030 >$2.5B) = 25%
Net Effect: 25% × $18B = +$4.5B (Adjusted down from initial +$7.5-15B, due to consideration of CapEx cycle pullback)

Historical Anchoring of AI CapEx Cycle:

All historical large-scale IT CapEx cycles have come to an end:

Cycle	Start	Peak	Duration	Peak→Trough Decline
Dot-com Bubble (Telecom Equipment)	1997	2000	3 years	-70%
Cloud Computing First Wave	2011	2014	3 years	-20% (Moderate)
5G Buildout	2019	2022	3 years	-30%
AI CapEx	2023	2026-2027?	3-4 years?	?

Pattern: IT CapEx cycles typically last 3-4 years. AI CapEx, starting from 2023, may peak in 2026-2027.

Three Arguments Why AI Might Be an Exception:

1. The ROI of AI investment is already quantifiable (ad targeting, programming assistance, customer service automation) → Unlike the dot-com bubble, which was "spend first, find business model later"
2. Inference demand is continuous (vs. 5G buildout, which is one-time) → CapEx will not drop sharply
3. The US-China AI arms race adds non-commercial drivers → Government investment provides bottom support

Probability-Weighted DC Path (More Detailed):

Scenario	FY2030 ARM DC Royalty	Probability	Weighted
AI CapEx Sustained Growth (Optimistic)	$2.5-4.0B	25%	$0.8B
AI CapEx Moderate Slowdown (Base Case)	$1.0-1.5B	45%	$0.56B
AI CapEx Cycle Pullback (Pessimistic)	$0.5-0.8B	20%	$0.13B
AI CapEx Collapse (Extreme Bear)	$0.2-0.5B	10%	$0.04B
Probability-Weighted		100%	$1.53B

$1.53B vs. our previous estimate of $1.2B → Difference +$0.33B → Valuation Impact +$0.33B×30x = +$10B → Probability weighting does not change the base case conclusion. The possibility of DC exceeding expectations has largely been captured by the previous valuation base case.

---

RT-7: Is Our WACC/Discount Rate Too High?

Challenge Target: E4 (Scenario-Weighted DCF) uses 10% WACC

WACC Breakdown:

Arguments Why 10% Might Be Too High:

1. ARM's beta = 2.0 primarily reflects post-IPO volatility, not long-term risk
2. ARM's business model (royalties) has much lower revenue elasticity than typical semiconductors (revenue has never declined YoY >10%)
3. Payment networks (V/MA) have an implied WACC of approximately 7-8% — if ARM's business model is closer to V/MA, its WACC should be lower

Payment Network vs. Semiconductor WACC Comparison:

Company	β	Implied WACC	Business Model
Visa	0.9	7.3%	Transaction Fees
Mastercard	1.0	7.8%	Transaction Fees
SNPS	1.2	9.0%	IP Licensing + Tools
CDNS	1.3	9.5%	IP Licensing + Tools
ARM	2.0	15.3%	IP Licensing + Royalties
ARM (Adjusted)	1.2-1.5	9-12%	Based on Business Model

If WACC from 10%→8%: E4 from $85.5B→$107B(+25%)
If WACC from 10%→9%: E4 from $85.5B→$96B(+12%)

Rebuttals:

• ARM beta=2.0 is not just "IPO volatility" — it reflects exposure to semiconductor cycles + valuation risk (142x P/E de-rating risk)
• V/MA's low beta is built on 40 years of stable operations — ARM has only been listed for 2 years, and beta has not yet converged
• If the RISC-V threat materializes, ARM's beta could rise from 2.0 to 2.5+ (instead of falling)

Probability Assessment: ARM's true long-term WACC <9% = 15% (Adjusted down from initial 20%)
Net Effect: 15% × (+$10.5B) = +$1.6B

WACC Sensitivity Matrix (E4 Engine):

WACC \ Terminal P/E	20x	25x (Baseline)	28x (Calibrated)	35x
8%	$78B	$97B	$107B	$131B
9%	$69B	$86B	$96B	$117B
10% (Baseline)	$61B	$76B	$85B	$104B
11%	$55B	$68B	$76B	$93B
12%	$49B	$61B	$68B	$83B

Assumptions Required for Current Market Cap of $136B:

• Requires WACC 8% + P/E 35x (=Payment Network + Low Risk)
• Or WACC 9% + P/E 40x+ (=Ultra-high Growth Perpetuity)
• Even the most optimistic WACC/P/E combination (8%/35x) only yields $131B → Below $136B
• The set of assumptions implied by market pricing is outside this matrix's range

---

RT Summary: Red Team Adjustments

RT#	Challenge	Direction	Probability-Weighted Net Impact	Change vs. Initial Version
RT-1	RISC-V Never Catches Up	↑	+$5.6B	No Change
RT-2	High Net Margin Achievable	↑	+$9.8B	+$3.0B (SBC Peer Validation)
RT-3	Phoenix Success	↑	+$1.75B	-$2.5B (Increased Probability of Backlash)
RT-4	Ecosystem Lock-in Underestimated	↑	+$6.0B	No Change
RT-5	SoftBank Positive	↑→Neutral	-$0.5B	-$5.8B (Two-way Correction)
RT-6	DC Outperforms Expectations	↑	+$4.5B	-$6.8B (CapEx Cycle)
RT-7	WACC Too High	↑	+$1.6B	-$2.4B
RT-1~7 Total	—	↑	+$28.8B	-$14.4B

Key Revision: RT net impact lowered from initial +$43.2B to **+$28.8B** — primarily due to RT-5 two-way correction (-$5.8B) and RT-6 CapEx cycle calibration (-$6.8B). The initial red team was too "bullish-sympathetic."

---

Chapter 27: RT-8~RT-10 Supplementary Red Team — Two-way Balancing

RT-8: Will AI Compute Democratization Undermine ARM's Pricing Power?

Challenge Direction: ↓ (Downgrade) — This is the first "downgrade" direction challenge for the red team.

Core Argument: The efficiency improvements in AI models (e.g., DeepSeek-R1) may reduce demand for the most advanced chips, thereby depressing ARM's royalty growth in DC:

Quantified Impact of AI Efficiency Improvements:

Dimension	2024	2026E	2028E	Impact on ARM
Training Cost/Performance	$100M/GPT-4 level	$10M	$1-3M	Small players won't need top-tier chips
Inference Efficiency (tokens/$)	1M/$10	1M/$1	1M/$0.10	Chip demand per $1 AI spend ↓
Model Distillation	Research Phase	Early Commercialization	Widespread Adoption	Small models replace large models → Chip demand ↓

If AI Democratization Weakens DC Chip Demand Growth:

• AI CapEx shifts from continuous growth to "efficiency-driven replacement growth" → Chip shipment growth rate drops from +25% to +10%
• ARM DC royalty growth drops from +40% CAGR to +20% CAGR
• FY2030 DC royalties drop from baseline $1.2B to $0.8B → -$0.4B × 30x = -$12B

Rebuttal:

1. Efficiency improvements are often offset by demand elasticity (Jevons Paradox) — AI becomes cheaper → more people use AI → total chip demand increases instead
2. Inference demand (unlike training) is continuous and grows linearly with the number of users — even if unit efficiency improves, total volume still rises
3. ARM's royalty rates for inference chips (Cortex-A/Neoverse N series) may be higher than for training chips

Quantified Test of Jevons Paradox:

Dimension	Jevons Effect (Demand Elasticity)	Anti-Jevons Factors	Net Effect
Training	Efficiency ↑ → More Model Training	Training compute demand has a ceiling (per model)	Partially Offset (+30%/-50%)
Inference	Cheaper → More Deployment → More Tokens	AI Penetration Increases but Per-Chip Demand ↓	Completely Offset (+100%/-80%)
Edge Inference	AI deployment to edge devices	Edge chips ≠ DC chips (lower ARM royalties)	Unfavorable for DC Royalties

Key Distinction: The Jevons paradox may hold true for inference (the cheaper, the more it's used) but not for training (training has an upper limit). ARM DC royalties primarily come from inference chips → the Jevons effect may partially protect ARM.

However: Even if Jevons protects total demand, "democratization" means computing power is decentralized from hyperscalers (high ASP ARM chips) to more smaller players (potentially using RISC-V or low-end ARM) → ARM's unit royalty may decrease even if total shipments increase.

Probability Assessment: AI democratization significantly weakens ARM DC royalty growth = 20%
Net Effect: 20% × (-$12B) = -$2.4B

---

RT-9: Does the ARM Chiplet/SOAFEE Ecosystem Create Unpriced Value?

Challenge Direction: ↑(Upward adjustment)

Core Argument: The ecosystem ARM is building (Total Design, SOAFEE automotive framework, Chiplet standards) may create "platform value" beyond royalties — similar to AWS evolving from storage services to a cloud platform:

ARM Ecosystem Value Tiers:

Tier	Current Status	Potential Value	Already Priced?
L1: IP Licensing (Core)	Mature	$80-100B	Yes
L2: CSS Design Platform	Rapid Growth	$10-20B	Partially
L3: Total Design Ecosystem	Early Stage	$5-15B	No
L4: SOAFEE Automotive Standard	Early Stage	$3-8B	No
L5: Chiplet Interoperability	Conceptual	$2-5B	No

Unpriced Ecosystem Value: L3+L4+L5 = $10-28B (Probability-Weighted)

Rebuttal:

• Total Design/SOAFEE currently have no independent revenue — purely "ecosystem investment"
• Chiplet standards are open standards; ARM may not be able to monopolize them
• Automotive software standards (AUTOSAR) already have a mature ecosystem; can SOAFEE replace them?

Quantification of "Standard-Setter Premium" for Chiplet Ecosystem:

Historical valuation premium for standard setters:

Company	Standard	Standard-Setter Premium (P/E vs Industry)	Premium Duration
Intel	x86 ISA	+30-50%	30 years (1990-2020)
Qualcomm	CDMA/4G/5G SEP	+20-40%	20 years (2000-2020)
Dolby	Audio Encoding Standard	+40-60%	15+ years
ARM	Chiplet/SOAFEE?	+?%	?

Assessment of ARM's Credibility as a Chiplet Standard Setter:

• UCIe (Universal Chiplet Interconnect Express): ARM is one of the founding members, but Intel/AMD/TSMC are also → ARM is not the sole standard setter
• AMBA (Advanced Microcontroller Bus Architecture): ARM is indeed a standard setter → but this standard has existed for 20 years and is already priced in
• SOAFEE: ARM-led → but automotive software standards are fragmented (competition from AUTOSAR/Adaptive AUTOSAR/Android Automotive)
• Conclusion: ARM's position as a Chiplet standard setter is not strong enough (non-exclusive) → limited premium potential

Probability Assessment: ARM ecosystem creates >$10B in unpriced value = 20%
Net Effect: 20% × $15B (median) = +$3.0B

---

RT-10: What if v10 is delayed by 3 years (2030+)?

Challenge Direction: ↓(Downward adjustment) — v10 is the most important single variable in our model

Core Argument: v10 is expected to be released in 2027-2028, but if delayed until 2030+:

Possible Reasons for v10 Delay:

1. RISC-V competition forces ARM to redesign v10 (adding more features → delay)
2. Technical complexity (e.g., native Chiplet support) leads to postponement
3. ARM's internal resources diverted to Phoenix → v10 development slows down
4. Customers (Apple/QCOM) object to v10 pricing → ARM forced to renegotiate

Valuation Impact of v10 Delay:

v10 Timeline	Royalty Rate Increase Starts	FY2030 Impact	Valuation Impact
2027 (On Schedule)	FY2028	+$500M-1B	Baseline (Already priced in)
2028 (1-Year Delay)	FY2029	+$300-600M	-$6-12B
2030 (3-Year Delay)	FY2031	+$0 (No FY2030 impact)	-$15-25B

If v10 is delayed by 3 years:

• FY2028-2030 still use v9 royalty rates → royalty CAGR drops from 14.5% to 10-12%
• Market narrative shifts from "v10 price increase drives growth" → "ARM fails to launch next-gen architecture"
• P/E could compress from 142x to 80-100x (narrative damaged)
• EV Impact: -$15-25B

Rebuttal:

• ARM has ample resources (annual R&D $2.3B+) to complete v10
• v10 is not an entirely new architecture; it's an evolution of v9 — 3-year delay is unlikely
• Even if v10 is delayed, minor v9.x updates could provide modest price increases

Second-Order Effects of v10 Delay (Beyond Direct Revenue Impact):

Impact of P/E Compression Underestimated: The biggest risk of a v10 delay is not revenue reduction ($8-12B), but rather P/E compression due to a narrative shift ($35-45B). If the market perceives "ARM is unable to launch its next-generation architecture" → ARM is repriced from a "growth stock" to a "mature IP company" → P/E falls from 142x to 50-70x → market cap of $45-65B (1/3-1/2 of current).

This is why v10's progress is ARM's most important single tracking signal (TS-14).

Early Warning Signals for v10 Delay:

How to identify a potential v10 delay before it's officially announced?

Early Warning Signal	Observable Time	Reliability	Source
Management stops mentioning v10 timeline in earnings reports	-12 months	High	6-K/20-F
ARM TechCon does not showcase v10 technical details	-6~12 months	High	Public Events
ARM significantly increases v10-related hiring	-18 months	Medium	LinkedIn/Job Postings
Key customer (Apple) delays next-gen chip design	-6 months	Medium (Indirect)	Supply Chain Reports
Increased ARM insider departures (especially architecture team)	-6~12 months	Medium	LinkedIn

Most Reliable Signal: ARM management avoids v10 timeline questions in two consecutive quarterly earnings reports. If, during the earnings reports/conference calls for FY2026Q4 (Feb 2027) and FY2027Q1 (May 2027), management is asked about v10 but only provides vague answers → probability of delay increases from 10% to 25-30%.

Probability Assessment: v10 delay ≥3 years = 10%
Net Effect: 10% × (-$20B) = -$2.0B (direct revenue impact only; the impact is far greater including the P/E compression risk)

---

Full Summary of RT-1~RT-10

RT#	Challenge	Direction	Net Effect
RT-1	RISC-V Never Catches Up	↑	+$5.6B
RT-2	High Net Margin Achievable	↑	+$9.8B
RT-3	Major Success for Phoenix	↑	+$1.75B
RT-4	Ecosystem Lock-in Underestimated	↑	+$6.0B
RT-5	SoftBank Positive	≈	-$0.5B
RT-6	Data Center Exceeds Expectations	↑	+$4.5B
RT-7	WACC Too High	↑	+$1.6B
RT-8	Weakening from AI Democratization	↓	-$2.4B
RT-9	Chiplet Ecosystem Value	↑	+$3.0B
RT-10	v10 Delay of 3 Years	↓	-$2.0B
Total	—	↑	+$27.35B

Red Team Net Adjustment: +$27.4B

Two-Way Balance Check: 7 upward adjustments ($32.25B) vs. 2 downward adjustments (-$4.4B) + 1 neutral (≈) (-$0.5B). Upward/Downward Adjustment Ratio = 7.3:1 — still biased towards upward adjustments, but the downward adjustments of the revised version ($4.9B) are a significant improvement over the initial version (0). The remaining bias may reflect a conservative bias in the preceding valuation (reasonable).

Red Team Bias Diagnosis: Why More Upward RT Adjustments?

Three possible explanations:

1. Preceding valuation was indeed conservative (reasonable bias) — Our NM assumption (25%) was indeed below consensus (37%), and the RISC-V timeline was indeed aggressive. It is normal for the Red Team to find more reasons for upward adjustments.
2. Red Team Methodological Bias (Cognitive Bias) — Red Team members (= ourselves) might have unconsciously designed "easier to justify" upward RT adjustments. Self-check: RT-8 (Weakening from AI Democratization) and RT-10 (v10 Delay) are both real risks, not just placeholders.
3. Red Team Coverage Gap (Structural Omissions) — Certain downside risks may not have been covered by the RT:

• Unexamined: Is the "liquidity premium" (9.7% free float) in ARM's valuation reasonable?
• Unexamined: If the semiconductor industry enters a deep cyclical downturn (2008-level), what is ARM's royalty elasticity?
• Unexamined: Could ESG/AI regulations potentially limit ARM's AI chip royalties?

Supplemental Downside Risk Assessment (RT-11/RT-12 Informal):

RT-11 (Informal): Disappearance of Liquidity Premium

• Current 9.7% free float → P/E has a 20-30% "scarcity premium"
• If SoftBank reduces its stake, increasing free float to 25% → scarcity disappears → P/E could compress by 20-30%
• Probability: 30% (within 3 years) × Impact -$25-35B = -$8-10B probability-weighted
• This is the largest downside risk not formally included in the RT table

RT-12 (Informal): Deep Semiconductor Downturn

• In 2008, semiconductor revenue declined -25%, while ARM royalties declined ~15% (relatively counter-cyclical)
• If a similar downturn occurs in 2027-2028: ARM royalties -10-15% → revenue $4.0-4.3B → market panic
• P/E typically compresses to 30-50x during a downturn (from 142x) → market cap $30-50B (temporary)
• Probability: 15% (any 3-year window) × Impact -$80-100B (temporary) = -$12-15B probability-weighted
• However, this is a "temporary" impact → long-term investors may not be affected

Complete Net Effect Including Informal RTs:

This reduces the Red Team's net impact from +$27.4B to **+$11.7B**—a more balanced outcome. However, RT-11/RT-12 fall under "macro/structural" risks rather than "ARM-specific" risks, so they are not included in the official RT summary, but are covered in the weight-bearing wall and scenario analyses.

---

Chapter 28: RT-1b Weight-Bearing Wall Joint Probability Test (Full Version)

28.1 Red Team Adjusted Weight-Bearing Walls

Based on challenges from RT-1 to RT-10, updated probabilities for B1-B6:

Wall	Belief	Prior Estimate	RT Adjustment ↑	RT Adjustment ↓	Final Probability	Change
B1	Rev CAGR >14.5%	45%	RT-6(+3pp)	RT-8(-2pp) RT-10(-1pp)	45%	0pp
B2	Net Margin → 25%+	35%	RT-2(+8pp)	—	43%	+8pp
B3	Pricing Power Unconstrained by RISC-V	40%	RT-1(+3pp) RT-4(+2pp)	RT-8(-1pp) RT-10(-2pp)	42%	+2pp
B4	DC Share → 40%+	50%	RT-6(+3pp) RT-9(+1pp)	RT-8(-2pp)	52%	+2pp
B5	Arm China Growth	55%	—	—	55%	0pp
B6	SoftBank Not Divesting	60%	RT-5(+0pp)	—	60%	0pp

28.2 "If Only One Wall Collapses" — Single-Wall Scenario

In a single-wall collapse scenario, ARM could still be reasonably valued:

Collapsed Wall	Scenario Description	Valuation Impact	Adjusted Market Cap	vs $136B
B1	Revenue Growth Slows to 10% CAGR	-$15-20B	$116-121B	-11~15%
B2	NM Maintained at 15% (SBC Not Normalized)	-$10-15B	$121-126B	-7~11%
B3	RISC-V Weakens Pricing Power	-$30-40B	$96-106B	-22~29%
B4	DC Share Stagnates at 15%	-$8-12B	$124-128B	-6~9%
B5	Arm China Revenue Declines by 30%	-$5-8B	$128-131B	-4~6%
B6	SoftBank Divests 20% Stake	-$15-25B	$111-121B	-11~18%

Key Finding: The collapse of B3 (Pricing Power) is the only wall that can singularly drive ARM's valuation down by >20%—making B3 the "keystone". No other single-wall collapse is sufficient to change a "Cautious Watch" rating to "Deeply Cautious".

28.3 Joint Probability Calculation (Including Conditional Probabilities)

Cannot Simply Multiply—Strong conditional dependencies exist between B1/B3/B4:

• B3 (Pricing Power) holds true → B1 probability increases to 55% (revenue growth becomes easier)
• B3 (Pricing Power) holds true → B4 probability increases to 60% (DC customers are more willing to use ARM)
• B3 fails → B1 drops to 35%, B4 drops to 35%

Exact Calculation:

Wait—Path B calculation is incorrect (It's contradictory for all to hold true when B3 fails). Correct Calculation:

Note: This is lower than a simple multiplication (5.2%)—because conditional probabilities reflect B3's "keystone" status (B1/B4 only receive high probabilities if B3 holds true).

28.4 Sensitivity Analysis: Changing Single Probability by ±10pp

Wall	Probability From→To	P(All) Change	Elasticity
B3	42%→52%	2.0%→2.5%	+0.5pp
B3	42%→32%	2.0%→1.5%	-0.5pp
B2	43%→53%	2.0%→2.4%	+0.4pp
B1	55%→65%(Conditional on B3)	2.0%→2.4%	+0.4pp
B6	60%→70%	2.0%→2.3%	+0.3pp
B5	55%→65%	2.0%→2.4%	+0.4pp
B4	60%→70%(Conditional on B3)	2.0%→2.3%	+0.3pp

Highest Elasticity: B3 (Pricing Power) — Changing B3 by ±10pp has the largest impact (±0.5pp), because B3 simultaneously affects the conditional probabilities of B1 and B4. B3 is the only "linchpin" of the entire valuation edifice.

28.5 Correlation Structure Among Bearing Walls

Bearing walls are not independent events. B3 (Pricing Power) is the "main pillar"—its success or failure affects B1 and B4:

Correlation Matrix (Qualitative):

	B1	B2	B3	B4	B5	B6
B1(Rev)	1.0	0.3	0.6	0.4	0.2	0.0
B2(NM)	0.3	1.0	0.2	0.1	0.1	0.0
B3(Pricing Power)	0.6	0.2	1.0	0.5	0.3	0.0
B4(DC)	0.4	0.1	0.5	1.0	0.1	0.0
B5(China)	0.2	0.1	0.3	0.1	1.0	0.1
B6(SoftBank)	0.0	0.0	0.0	0.0	0.1	1.0

Key Observations:

• B6 (SoftBank) is almost uncorrelated with other walls → SoftBank risk is an "independent shock", not driven by ARM's business fundamentals
• B3 is highly correlated with B1/B4 (0.5-0.6) → Pricing power determines revenue growth and DC share
• B5 (China) is moderately correlated with B3 (0.3) → China's policy forcing RISC-V would indirectly weaken global pricing power
• B2 (NM) is relatively independent (highest correlation only 0.3) → Profit margins are mainly determined by internal factors like SBC/R&D

Impact on Joint Probability: High correlation means the probability of "all succeeding" is higher than under an independence assumption, but the probability of "all failing" is also higher than under an independence assumption → The distribution is bimodal (either extremely good or extremely bad), not normal → ARM is a "binary bet" stock (binary outcome) rather than a "moderate return" stock

28.6 P(≥K/6) Distribution

K (Number of Walls Holding)	Probability	Valuation Implication
6/6	2.0%	$136B reasonable (required by current price)
≥5/6	~14%	$110-130B (Moderately Overvalued)
≥4/6	~32%	$90-110B (Calibration Range)
≥3/6	~55%	$70-90B (Our Baseline)
≤2/6	~45%	$40-70B (Severely Overvalued)

Key Conclusion: The market price ($136B) implies all 6/6 walls hold—but with only a 2% probability. Our valuation (calibrated to $94B) corresponds to 4/6 walls holding (32% probability)—more reasonable but still requiring "most assumptions to be correct."

28.7 "Vulnerability Ranking" of Bearing Walls (Which Wall is Most Likely to Fall First?)

Based on the preceding comprehensive analysis, here is the ranking of the impact of a single wall's collapse for the 6 walls:

Rank	Wall	Probability of Collapse	EV Impact upon Collapse	Risk Product (P×I)	Observability
1	B3 (Pricing Power)	58%	-$35-45B	$23B	Low (3-year lag)
2	B1 (Rev CAGR)	55% (unconditional)	-$20-30B	$14B	High (quarterly earnings)
3	B2 (Net Margin)	57%	-$15-25B	$11B	High (quarterly earnings)
4	B6 (SoftBank)	40%	-$20-40B	$12B	Medium (13F delay)
5	B4 (DC Contribution)	48% (unconditional)	-$10-20B	$7B	High (ARM disclosure)
6	B5 (China)	45%	-$5-15B	$5B	Medium (geopolitical events)

Most Dangerous Wall: B3 (Pricing Power) – not only because its risk product is the highest ($23B), but also because its observability is the lowest. As ARM loses pricing power, royalty revenue may not immediately decline (due to contract lock-in) → investors might only realize it 2-3 years later, at which point it would be too late.

"Hidden Collapse" Scenario:

In this scenario, B3 "confirms collapse" only in 2030, but the seeds of collapse were planted in 2027 (excessive v10 price hike spurred RISC-V evaluation). If investors only look at short-term financial data, they would remain bullish on ARM from 2026-2028 → only to discover the issue in 2030.

28.8 "Scenario Mapping" of Joint Probabilities

Mapping the load-bearing wall outcomes to valuation ranges:

Scenario	Number of Intact Walls	Representative Combination	Valuation Range	Probability
Extreme Bull	6/6	All Intact	$130-150B	2%
Strong Bull	5/6	B5 Collapse (China slowdown but all others good)	$110-130B	12%
Moderate Bull	4/6	B5+B6 Collapse (China+SoftBank double pressure)	$90-110B	20%
Base Case	3/6	B3+B5+B6 Collapse (Pricing Power+China+SB)	$70-90B	23%
Moderate Bear	2/6	B1+B3+B5+B6 Collapse	$50-70B	20%
Strong Bear	1/6	Only B4 Intact (DC is the only highlight)	$35-50B	15%
Extreme Bear	0/6	All Collapse	$25-35B	8%

Probability-Weighted Valuation = 2%×$140B + 12%×$120B + 20%×$100B + 23%×$80B + 20%×$60B + 15%×$42.5B + 8%×$30B
= $2.8B + $14.4B + $20B + $18.4B + $12B + $6.4B + $2.4B = $76.4B

The valuation derived from this underlying probability calculation ($76B) is lower than our five-engine composite ($80-85B) but within the same order of magnitude – cross-validation passed. The difference stems from the five-engine model giving additional weight to "option value".

---

Chapter 29: Possibility Space Mapping — Scenario Tree

29.1 Five Futures for ARM

Based on the full analysis of Parts I-IV (314K chars), ARM could be in one of the following five states in 5 years (FY2030):

v2.0 vs v1.0 Key Adjustments:

• S1 from $175B→$160B (more cautious after red team calibration)
• S2 from $110B→$105B (NM from 30%→28%, GAAP calibration)
• S4 probability from 15%→17% (risk adjusted upward after RT-5 SoftBank two-way correction)
• S3 probability from 35%→33% (RT-1 security certification barrier delays RISC-V)
• Added detailed 5-year P&L Build-out (v1.0 only had terminal figures)

29.2 Scenario Details — Five-Year P&L Build-out

Scenario 1: "Chip Royalty Empire" (Probability 12%, Market Cap $160B, +18%)

Trigger Condition Chain:

• RISC-V software ecosystem still significantly lags ARM by 2030 (security certification + enterprise middleware) → v10 successfully achieves 1.8-2.0x price increase
• CSS achieves full penetration (≥15 customers in mass production) → Royalty CAGR 18-20%
• DC share reaches 35-40% → ARM becomes the standard architecture for AI inference
• SBC normalizes to 8-10% (peer IPO + 5-year median)
• Phoenix moderate success (SoftBank internal + 1-2 external customers)

Five-Year P&L Path:

Metric	FY2026	FY2027E	FY2028E	FY2029E	FY2030E
Revenue ($B)	4.9	6.0	7.5	9.5	12.0
Revenue YoY	+26%	+22%	+25%	+27%	+26%
Royalties ($B)	2.5	3.2	4.2	5.5	7.2
Licensing ($B)	2.4	2.8	3.3	4.0	4.8
Gross Margin	95%	95%	95%	95.5%	95.5%
R&D/Rev	56%	48%	42%	38%	36%
SBC/Rev	15%	12%	10%	8%	7%
GAAP OM	5%	22%	32%	40%	44%
GAAP NM	12%	18%	27%	34%	38%
Net Income ($B)	0.59	1.08	2.03	3.23	4.56
FCFF ($B)	0.3	0.8	1.5	2.5	3.8
DC Royalty Share	12%	16%	22%	30%	38%

Terminal Valuation: NI $4.56B × P/E 33x = $150B. Includes option premium (+$10B) → $160B
5-Year CAGR: ($160B/$136B)^(1/5) - 1 = +3.3%/year — a modest return even in the most optimistic scenario

S1's 'Possible but Uninvestable' Paradox: A +3.3%/year return is far from sufficient for a stock with a beta of ~2.0 — during the same period, SPY's expected return is ~8-10%/year. Even if S1 (best-case scenario) is realized, ARM's risk-adjusted return would be negative (3.3% return / 2.0 beta = 1.65% → significantly below the risk-free rate of 4.3%). This implies that ARM is not a good risk-adjusted investment in any scenario — which is the deepest reason for 'prudent caution'.

S1 FCFF → Fair Value Bridge:

S1 Time Dependence: S1 is not "if ARM achieves this, it will happen" — it also requires:

• AI CapEx does not decline (at least extends to FY2029): Probability ~55%
• SoftBank maintains stable holding (not forced to reduce stake): Probability ~70%
• RISC-V DC <3% (security certification barrier remains): Probability ~45%
• All three conditions hold simultaneously: 55%×70%×45% = 17% ← Higher than S1 probability (12%) because S1 also requires v10 major success

Key Milestones in Scenario 1 Path:

• FY2027 Net Margin >18%: First validation of SBC normalization
• FY2028 DC Royalties >$1B: Confirmation of DC scaling
• FY2029 v10 Royalties >1.5x v9: Full confirmation of pricing power

Scenario 2: "Steady Growth IP" (Probability 28%, Market Cap $105B, -23%)

Trigger Conditions Chain:

• RISC-V gradual penetration but not disruptive (IoT share drops to 45%, DC <5%)
• v10 modest price increase (1.3x vs v9) → Royalty CAGR 13-15%
• CSS steady expansion (10-12 customers in mass production)
• Phoenix delayed or scaled back → neither cannibalizes nor contributes
• SBC normalizes to 10-12% (moderate path)

Five-Year P&L Path:

Metric	FY2026	FY2027E	FY2028E	FY2029E	FY2030E
Revenue ($B)	4.9	5.7	6.5	7.3	8.2
Revenue YoY	+26%	+16%	+14%	+12%	+12%
Royalties ($B)	2.5	3.0	3.5	4.0	4.6
Licensing ($B)	2.4	2.7	3.0	3.3	3.6
Gross Margin	95%	95%	94.5%	94.5%	94%
R&D/Rev	56%	50%	46%	43%	41%
SBC/Rev	15%	13%	12%	11%	10%
GAAP OM	5%	18%	25%	30%	33%
GAAP NM	12%	15%	21%	25%	28%
Net Income ($B)	0.59	0.86	1.37	1.83	2.30
FCFF ($B)	0.3	0.6	1.0	1.4	1.8
DC Royalty Contribution	12%	15%	18%	21%	24%

Terminal Valuation: NI $2.30B × P/E 28x = $64B. Including recent FCFF PV (~$4B) + Options (+$10B) + Net Cash (+$2B) → $80B
Using Market Cap Method: NI × P/E + Options → $64B + $15B = $79B
Taking the average → Scenario 2 Market Cap **~$105B** (v1.0 method: direct NI×P/E yielded a lower value; adjusted upwards after including recent FCFF and options)

Let me actually recalibrate – using the Red Team's calibrated unified framework:

Revised Scenario 2: FY2030 Market Cap $70B → Return for today's holders = -49% (5-year cumulative)

This is more pessimistic than v1.0's -19% – because v1.0 used $110B without considering discounting. But investors care about "how much ARM will be worth in 5 years," and the answer is an FY2030 value of $70B or $55B discounted to today.

For consistency: The scenario table uses "FY2030 Nominal Market Cap" (undiscounted), but the return rate is calculated as "Buy Today → Sell FY2030".

Scenario	FY2030 Market Cap	Return (5 years)	Annualized
S1	$160B	+18%	+3.3%
S2	$70B	-49%	-12.5%

Wait – S2 yielding -49% means a 49% loss in a "steady growth" scenario? This is because the P/E compression from 141x→28x (-80%) swallowed up the earnings growth (+290%).

This is ARM's core issue: even with decent business performance (NM 28%, Revenue CAGR 13%), P/E compression alone is enough to cause significant losses.

Let me correct the scenario table – v1.0's S2 using $110B (without discounting or P/E compression) was misleading:

Scenario 2 Addendum: Quantifying the P/E Compression Effect

Earnings growth of +137% is completely offset and reversed by P/E compression of -80% → This is the mathematical essence of a "valuation trap"

Scenario 3: "RISC-V Erosion" (33% Probability, Market Cap $43B, -68%)

Trigger Condition Chain:

• RISC-V software ecosystem reaches "good enough" by 2028 → ARM's pricing power erodes
• v10 unable to effectively increase prices (only 1.1x vs v9) → Royalty CAGR drops to 8-10%
• 2-3 hyperscale customers announce RISC-V evaluation/migration plans
• IoT ARM share drops to 35%, automotive RISC-V begins initial mass production
• Net margin stagnates at 22% (increased R&D for RISC-V defense)

Five-year P&L Path:

Metric	FY2026	FY2027E	FY2028E	FY2029E	FY2030E
Revenue ($B)	4.9	5.5	5.9	6.2	6.5
Revenue YoY	+26%	+12%	+7%	+5%	+5%
Royalties ($B)	2.5	2.8	3.1	3.3	3.5
Licensing ($B)	2.4	2.7	2.8	2.9	3.0
GAAP NM	12%	14%	18%	20%	22%
Net Income ($B)	0.59	0.77	1.06	1.24	1.43
FCFF ($B)	0.3	0.5	0.7	0.9	1.1
DC Royalties Share	12%	14%	16%	17%	18%
RISC-V DC Market Share	<2%	3%	5%	7%	10%

Terminal Valuation: NI $1.43B × P/E 22x = $31B. Including FCFF + Options → $43B

S3 FCFF → Fair Value Bridge:

S3's P/E Compression Math:

Scenario 3's "Boiled Frog" Path: Revenue never declines (still grows 5-12% annually) → Investors may continue to hold → but P/E compresses from 141x → 22x (85% compression) → 5-year loss of -68%. This is the most likely scenario (33%) and the most deceptive — because ARM's revenue continues to grow, investors might mistakenly believe "everything is normal."

S3's "Deceptive" Timeline:

Year	Revenue Growth	ARM Narrative	Investor Sentiment	Actual P/E Compression
FY2027	+12%	"Solid Growth"	Optimistic	141→120x(-15%)
FY2028	+7%	"Temporary Slowdown"	Hesitant	120→80x(-33%)
FY2029	+5%	"Transition to Maturity"	Uneasy	80→40x(-50%)
FY2030	+5%	"Stabilization Phase"	Panic	40→22x(-45%)

Revenue grows every year, but P/E compresses by 20-50% annually → cumulative effect is devastating. Investors in FY2027-2028 still harbor the illusion of "growth recovery," only realizing by FY2029 that the growth rate has permanently shifted down — but by then, P/E has fallen from 141x to 40x (a 72% loss). S3's harm lies in "chronic bleeding," not "acute shock." []

Scenario 4: "SoftBank Spiral" (Probability 17%, Market Cap $35B, -74%)

Trigger Condition Chain:

• SoftBank Izanagi project requires $200B+ financing → ARM equity pledged to its full extent
• ARM share price drops to $90 for any reason → triggers margin call
• SoftBank forced to divest 10-15% → supply shock → ARM share price declines further
• Market re-evaluates related party transactions + earnings quality → panic selling
• "SoftBank-ARM Positive Feedback Loop" forms (RT-5 Prisoner's Dilemma)

Five-year P&L Path:

Metric	FY2026	FY2027E	FY2028E	FY2029E	FY2030E
Revenue ($B)	4.9	5.5	6.0	6.5	7.0
Revenue YoY	+26%	+12%	+9%	+8%	+8%
GAAP NM	12%	14%	18%	20%	22%
Net Income ($B)	0.59	0.77	1.08	1.30	1.54

Note: S4's P&L is similar to S3 – fundamentals are largely unchanged, the difference lies in governance discount + supply shock:

• Governance Discount: P/E further compressed from 22x to 16x (-27%)
• Supply Shock: Free float from 9.7% → 20%+ → scarcity premium disappears (-15%)
• Terminal Valuation: NI $1.54B × P/E 16x = $24.6B + FCFF/Options = $35B

S4 FCFF → Fair Value Bridge:

S4's Liquidity Shock Model:

Phase	SoftBank Action	Free Float Change	P/E Impact	Market Cap Impact
T0	Initial 5% Divestment (Financial Report Disclosure)	9.7%→14.7%	-15%	-$20B
T0+3 Months	Forced Additional 5% Divestment (Margin Call)	14.7%→19.7%	-10%	-$12B
T0+6 Months	Market Panic + Short Sellers Influx	Short Interest Doubles	-20%	-$18B
T0+12 Months	Price Stabilizes at New Equilibrium	Free Float ~20%	+5% (Rebound)	+$4B
Net Effect	—	—	—	-$46B

Chained Probability: T0 Trigger (35%/5 years) × T0 leading to T0+3 Months (40%) × Panic (60%) = 8.4% ← approaches the lower bound of S4 probability (17%). S4's 17% includes other trigger paths (divestments not triggered by margin calls).

Scenario 4 Key Time Window: SoftBank's Izanagi project financing peak is in 2027-2028. If ARM's stock price falls below $90 during this period (regardless of reason) → the spiral trigger probability jumps from 17% to 40%. ARM's stock price is SoftBank's "deadman's switch".

Scenario 5: "Total Collapse" (Probability 10%, Market Cap $18B, -87%)

Trigger Conditions: Scenario 3 + Scenario 4 + Geopolitical Shock occur simultaneously

• RISC-V Acceleration (China policy + Google push) + SoftBank Crisis + Taiwan Strait conflict leads to supply chain disruption
• ARM reclassified from "growth stock" to "declining cyclical stock"
• Arm China completely out of control (revenue zeroed out)

Five-year P&L Path:

Metric	FY2026	FY2027E	FY2028E	FY2029E	FY2030E
Revenue ($B)	4.9	5.0	4.5	4.0	4.0
Revenue YoY	+26%	+2%	-10%	-11%	0%
GAAP NM	12%	10%	8%	8%	10%
Net Income ($B)	0.59	0.50	0.36	0.32	0.40

Terminal Valuation: NI $0.40B × P/E 10x = $4B + Residual Assets (IP+Cash ~$14B) = $18B

S5 FCFF → Fair Value Bridge:

S5's Triple Blow Path (requires three catalysts to occur simultaneously):

Catalyst	Independent Probability	Impact	Time Window
RISC-V DC>10%	15%	Royalty CAGR drops to 3-5%	FY2028-2030
SoftBank forced to fully divest	8%	Float shock + governance crisis	2027-2028
Arm China revenue zeroes out	5%	Revenue reduced by 19% + geopolitical ripple effects	Unpredictable
All three simultaneously	~0.6%	Complete collapse	—

0.6% vs S5's 10% Probability Difference: S5's 10% probability is higher than all three simultaneously (0.6%) because S5 does not require the three events to happen "simultaneously"—any one catalyst can trigger a chain reaction. For example: SoftBank crisis → ARM stock plummets → talent drain → RISC-V accelerates → Arm China impacted. The transmission chain gives seemingly independent events an outsized correlation.

S5's "Residual Value" Analysis: Even if S5 occurs, ARM's IP asset value remains at $12-15B:

• Patent Portfolio (~6,000 core patents): $5-8B (compared to Motorola patents acquired by Google for $12.5B)
• Customer Contracts (long-term licenses): $3-5B (present value of future royalties)
• Brand + Ecosystem Residual Value: $2-3B
• Implies ARM's "floor price" is approximately $12-15/share → S5's $21/share already includes a reasonable margin of safety

29.3 Scenario Probability-Weighted Returns (v2.0 Calibrated Version)

Scenario	Probability	FY2030 Market Cap	vs $136B	Annualized Return	Contribution
S1 Chip Royalty Empire	12%	$160B	+18%	+3.3%	+2.2%
S2 Steady Growth	28%	$70B	-49%	-12.5%	-13.7%
S3 RISC-V Erosion	33%	$43B	-68%	-20.8%	-22.6%
S4 SoftBank Spiral	17%	$35B	-74%	-23.2%	-12.6%
S5 Complete Collapse	10%	$18B	-87%	-33.2%	-8.7%
Expected Value	100%	$60.7B	-55%	—	-55.4%

Probability-Weighted Expected Market Cap: $60.7B → Expected Return -55%

vs v1.0 Differences: v1.0 expected market cap $80.4B (-41%), v2.0 $60.7B (-55%). Differences arise from:

1. Lower S2/S3/S4 valuations after including detailed P&L (P/E compression effect was underestimated in v1.0)
2. S3 has the highest probability (33%) and a terminal value of $43B (vs v1.0's $65B)
3. P/E compression is the biggest correction factor in v2.0—v1.0 intuitively thought "FY2030 should be worth $XX", while v2.0, by thinking "today 141x → FY2030 terminal P/E", found the compression magnitude far exceeded earnings growth

vs Calibrated Fair Value: Calibrated $100B, Scenario Tree $61B—a difference of $39B. Differences arise from:

• Calibrated fair value from a "reasonable valuation" perspective (giving ARM a reasonable valuation)
• Scenario tree from an "expected value" perspective (probability-weighted across all scenarios, including tail risks)
• Tail risks (S4+S5, 27% probability) dragged down the expected value by ~$25B
• Taking the midpoint of both as the final valuation anchor: ($100B + $61B) / 2 = $80.4B

29.4 Return Distribution Characteristics

Metric	Value
Expected Return	-41% (based on the final anchor of $80.4B)
Upside Probability (> current price)	12% (S1 only)
Downside Probability (< current price)	88% (S2-S5)
Max Upside	+18% (S1)
Max Downside	-87% (S5)
Median Return	-68% (S3 is the probability-weighted median)
Upside/Downside Ratio	1:7.3 (highly asymmetric)
P(Loss > 50%)	60% (S3+S4+S5)

Summary of Return Distribution: 88% probability of loss, 60% probability of losing more than half, limited upside (+18% cap)—ARM's current price exhibits severe risk/reward asymmetry. []

29.5 Transition Paths Between Scenarios

Scenarios are not static—ARM may "slide" from one scenario into another:

The most likely "downward spiral" path: S2→S3 (Progressive RISC-V Erosion)→S4 (SoftBank Forced Divestment). The cumulative probability of this path is approximately 12-15%—once triggered, ARM's slide from $105B/share to $35B could occur within 12-24 months.

29.6 Mapping Scenarios to CQ/Hypotheses

Each scenario corresponds to different CQ/hypothesis outcomes:

Scenario	CQ-1 Royalties	CQ-2 DC	CQ-3 SB	CQ-4 RISC-V	CQ-5 China	CQ-6 Phoenix	H1 Profitability	H2 CDS
S1	Excellent	Surge	Protection	Not Materialized	Stable	Highly Successful	Not Materialized	Not Materialized
S2	Good	Steady Growth	Neutral	Weakly Materialized	Stable	Neutral	Weakly Materialized	Weakly Materialized
S3	Average	Stagnant	Neutral	Materialized	Slowdown	Failure	Materialized	Materialized
S4	Average	Stagnant	Spiral	Weakly Materialized	Slowdown	Neutral	Materialized	Weakly Materialized
S5	Deterioration	Collapse	Spiral	Materialized	Out of Control	Backfire	Materialized	Materialized

Key Drivers for S3 (Most Likely Scenario, 33%): CQ-4 (RISC-V) and H2 (CDS) – these two mutually reinforce each other. If RISC-V is "good enough" (CQ-4 materializes) → ARM pricing power is limited (H2 materializes) → Royalties CAGR decreases to 8-10% → P/E compression → S3 realized.

Simplification for Investors: "One Metric": If investors can only track one metric to determine which scenario ARM is in – they should track TS-7 (RISC-V DC Shipment Share):

• TS-7 < 2% (Current): S1 or S2 (Optimistic)
• TS-7 3-7%: S3 (Most Likely)
• TS-7 > 7%: S3 → S4/S5 (Pessimistic)

29.7 Scenario Probability Sensitivity

Scenario probabilities themselves are uncertain. If a single assumption is changed:

Assumption Change	S1	S2	S3	S4	S5	Expected Market Cap
Baseline	12%	28%	33%	17%	10%	$61B
RISC-V delayed by 2 years	18%	35%	25%	12%	10%	$66B
RISC-V accelerated by 2 years	5%	20%	40%	22%	13%	$44B
SoftBank does not divest	15%	32%	33%	10%	10%	$63B
SoftBank significantly divests	8%	22%	33%	25%	12%	$47B
AI CapEx does not decline	18%	35%	27%	12%	8%	$68B

Maximum Sensitivity: "RISC-V accelerated by 2 years" reduces the expected market cap from $61B to $44B (-$11B);
"AI CapEx does not decline" increases the expected market cap from $61B to $68B (+$13B).
RISC-V timeline and AI CapEx cycle are the two largest exogenous variables driving scenario probabilities.

29.8 Risk-Adjusted Return Analysis (CAPM Framework)

Risk-adjusted returns for different scenarios – investors are not only concerned with "how much return," but also with "how much return for each unit of risk":

Scenario	Nominal Annualized Return	ARM β (Est)	Risk-Adjusted Return	Rf (4.3%)	Sharpe Ratio (Est)	Investment Attractiveness
S1	+3.3%	1.5	+2.2%	4.3%	-0.08	Unattractive (Below Rf)
S2	-12.5%	1.8	-6.9%	4.3%	-0.93	Very Poor
S3	-20.8%	2.0	-10.4%	4.3%	-1.25	Extremely Poor
S4	-23.2%	2.5	-9.3%	4.3%	-1.10	Extremely Poor
S5	-33.2%	3.0	-11.1%	4.3%	-1.25	Disastrous
Weighted	-15.6%	~2.0	-7.8%	4.3%	-1.00	Uninvestable

Key Findings:

1. Even S1 is Unattractive: Even if the best-case scenario (+3.3%/year) is achieved, ARM's β≈1.5 implies a CAPM required return = 4.3% + 1.5 × (8% - 4.3%) = 9.9%. 3.3% is far below 9.9% → alpha is -6.6%/year.
2. Probability-Weighted Sharpe Approx. -1.0: In portfolio construction, this means "should short rather than go long."
3. "Under what conditions does ARM become investable?": Requires a nominal return ≥ 10% (CAPM minimum requirement) → requires FY2030 market cap ≥ $220B (1.10⁵ times the current $136B). This necessitates NI $5.5B + P/E 40x (exceeding S1) → probability < 5%.
4. Conclusion: From a risk-adjusted return perspective, ARM at its current price is not a good investment under any probability assumption. This is a stronger statement than "overvalued by 27-41%" — even if ARM is not overvalued (P/E is reasonable), its β and expected return do not match.

---

Chapter 30: Investment Thermometer Final Version

30.1 Thermometer Calculation (10 Dimensions, v2.0)

Dimension 1: Macro Environment (Weight 10%)

Metric	Value	Score (-5 to +5)
Shiller P/E (CAPE)	40.08 (98th percentile)	-2.0
Buffett Indicator	219% (100th percentile)	-2.5
ERP	4.5% (66th percentile)	-1.0
Macro Sub-score	—	-1.8

Dimension 2: Absolute Valuation (Weight 15%)

Metric	Value	Score (-5 to +5)
GAAP P/E	141.6x (Extremely High)	-4.0
Non-GAAP P/E	~55-65x (High)	-2.0
FCF Yield	0.87% (Extremely Low)	-4.0
EV/EBITDA	~95x (Extremely High)	-3.5
Absolute Valuation Sub-score	—	-3.4

Dimension 3: Relative Valuation (Weight 15%)

Metric	Value	Score (-5 to +5)
P/E vs EDA (SNPS/CDNS)	141x vs 65x (2.2x Premium)	-3.0
EV/Sales vs EDA	23.4x vs 14.5x (1.6x Premium)	-2.5
P/E vs Payment Networks (V/MA)	141x vs 30-35x (4x Premium)	-4.0
Five-Engine Fair Value vs Market Price	$100B vs $136B (-27%)	-2.0
Relative Valuation Sub-score	—	-2.9

Dimension 4: Revenue Growth (Weight 10%)

Metric	Value	Score (-5 to +5)
Revenue YoY	+26% (Strong)	+3.0
Royalty CAGR (3-year)	~18% (Excellent)	+3.5
Revenue Diversification	DC from 5% → 12% (Improving)	+2.0
Revenue Growth Sub-score	—	+2.8

Dimension 5: Earnings Quality (Weight 10%)

Metric	Value	Score (-5 to +5)
GAAP/Non-GAAP Gap	2.6x (High)	-2.5
DSO	173 Days (Industry High)	-3.0
OCF/NI	1.90 (Incl. SBC, inflated)	-1.5
Related-Party Transactions % of Licensing	~30% (Significant)	-2.5
Earnings Quality Sub-Score	—	-2.4

Dimension 6: Moat Strength (Weight 10%)

Metric	Value	Score (-5 to +5)
A-Score	7.17/10 (High)	+2.0
Moat Migration Direction	T→N Downward (Network effect decay)	-1.0
Ecosystem Lock-in Strength	8/10 → Declining Annually	+1.0
Moat Sub-Score	—	+0.7

Dimension 7: Management and Governance (Weight 5%)

Metric	Value	Score (-5 to +5)
CEO Competency (Rene Haas)	Professional + Strategic Clarity	+2.0
Governance Structure	Dual Exemption + SoftBank 90% Holding	-3.0
Capital Allocation	No Dividends/Buybacks (Rational)	+0.5
Management & Governance Sub-Score	—	-0.2

Dimension 8: Catalysts (Weight 10%)

Metric	Value	Score (-5 to +5)
v10 Architecture (Upside)	Expected 2027-2028	+2.0
CSS Acceleration (Upside)	19 Licenses / 5 Mass Production	+2.0
Phoenix (Two-Way)	Upside Option + Backlash Risk	+0.5
RISC-V (Downside)	Accelerating Penetration	-2.0
SoftBank Stake Reduction (Downside)	35%/3 Year Probability	-1.5
Catalysts Sub-Score	—	+0.2

Dimension 9: Technicals (Weight 5%)

Metric	Value	Score (-5 to +5)
Short-Term Trend (RSI/MA)	Neutral (Near MA50)	0.0
Short Interest %	~15% (High)	-1.0
Float Structure	9.7% (Extremely Low → High Volatility)	-2.0
Technicals Sub-Score	—	-1.0

Dimension 10: Industry Cycle (Weight 10%)

Metric	Value	Score (-5 to +5)
Semiconductor Cycle Position	Upturn Phase (but AI CapEx may be peaking)	+1.0
AI CapEx Trend	+30-40% YoY (May peak 2026-2027)	+1.5
Inventory Cycle	Normal (ARM has no inventory)	+1.0
Industry Cycle Sub-Score	—	+1.2

30.2 Composite Thermometer

Composite Temperature: -0.94 (Range -5 to +5)

Temperature Range	Meaning	ARM Position
+3 ~ +5	Extremely Cold (Severely Undervalued, Deep Focus)	—
+1 ~ +3	Cold (Undervalued, Focus)	—
-1 ~ +1	Neutral (Fair Range)	← ARM: -0.94
-3 ~ -1	Hot (Overvalued, Cautious Focus)	—
-5 ~ -3	Extremely Hot (Severely Overvalued, Strong Caution)	—

30.3 Historical Comparison: Fate of High P/E Stocks at Similar Temperatures

Where does ARM's thermometer score of -0.94 stand among historical high P/E stocks?

Company	Date	P/E	Temperature (Est.)	Subsequent 3-Year Return	Reason
Cisco(CSCO)	March 2000	196x	-3.5	-80%	Dot-com bubble + Growth deceleration
Intel(INTC)	March 2000	52x	-2.0	-75%	Cycle peak + Market bubble
Qualcomm(QCOM)	January 2000	283x	-4.0	-82%	Bubble + Competition + Growth steep decline
ASML	November 2021	55x	-1.5	-28%	Valuation mean reversion + Cycle
NVIDIA(NVDA)	November 2021	90x	-2.0	-60%(to 2022 low)	Crypto crash + Inventory
Snowflake(SNOW)	November 2021	N/A (Loss-making)	-3.0	-70%	Growth deceleration + SBC + Valuation
ARM	February 2026	141x	-0.94	?	—

Key Observations:

1. Stocks with P/E > 100x: Cisco (196x → 80%) and QCOM (283x → 82%) – both lost 75-80% within 3 years. ARM's 141x falls within this range.
2. Temperature of -0.94 vs. Bubble Stocks: ARM's temperature is milder than 2000 dot-com bubble stocks (-2.5~-4.0) – because ARM is supported by real high growth (+26% revenue) and high gross margin (95%). Growth for bubble stocks was often artificial.
3. Most Similar Case: ASML 2021 (-1.5, subsequent -28%) – High-quality semiconductor IP monopolist, overvalued but with solid fundamentals. ARM's temperature (-0.94) is milder than ASML's at that time, but its P/E is higher (141x vs 55x).
4. ARM's Potential Paths: If ARM follows the ASML path (mild reversion) → -20~30% (S2 zone). If it follows the Cisco path (bubble burst) → -70~80% (S4/S5 zone).

Conclusion: Historically, high-quality companies with temperatures in the -0.5~-1.5 range typically see subsequent 3-year returns of -15%~-40% – largely consistent with our S2/S3 scenarios (-49%/-68%). ARM's temperature (-0.94) suggests a **median return of approximately -30%**, which aligns with the -41% implied by the $80.4B valuation anchor (though the thermometer is slightly more optimistic than the scenario tree).

Reconciling Differences: Thermometer vs. Scenario Tree: The thermometer (-0.94) is milder than the scenario tree (-55%) because it includes positive factors such as "revenue growth (+2.8)" and "industry cycle (+1.2)" – these factors indeed provide a buffer in the short term (1-2 years). However, in the medium to long term (3-5 years), P/E compression effects dominate → the scenario tree is more accurate. **The thermometer is suitable for judging the "present moment," while the scenario tree is suitable for judging the "terminal state."**

30.4 Thermometer Interpretation and Rating Reconciliation

Reconciling Thermometer -0.94 (warm side of the spectrum) with "Cautious Watch" Rating (Overvalued -27~41%):

The thermometer is milder than a pure valuation analysis because:

1. Revenue Growth (+2.8) Boost: ARM's growth rate is indeed very strong – this is real.
2. Industry Cycle (+1.2) Boost: The AI CapEx supercycle provides a tailwind for ARM.
3. Absolute Valuation (-3.4) Diluted by Other Dimensions: The thermometer's weighting is spread across 10 dimensions.

Statement on Thermometer Limitations: The thermometer reflects the "risk/reward ratio under current trends," not the "absolute valuation level." ARM's fundamentals are indeed strong (revenue growth + gross margin + DC growth), but strong fundamentals ≠ cheap valuation. A thermometer reading of -0.94 means "ARM is not an extreme bubble (not -3~-5)," not "ARM is reasonably valued."

Final Reconciliation with Rating: The thermometer is in the warm side of the spectrum (-0.94), while the Five Engines + Scenario Tree indicate "overvalued by 27-41%" → the intersection is **"Cautious Watch"** – not "Strong Cautious Watch" (because of fundamental support), but also not "Neutral" (because valuation is clearly elevated).

30.5 Thermometer Dimension Contribution Analysis

Which dimensions pull the thermometer down/up the most?

Structure: Bearish forces (-1.37) are significantly stronger than bullish forces (+0.47). The "components" of the net temperature of -0.90 are:

• Valuation dimensions contribute 70% of bearish forces (-0.95/-1.37) – ARM's "overvaluation" is almost entirely driven by valuation dimensions.
• Revenue growth contributes 60% of bullish forces (+0.28/+0.47) – ARM's "fundamental support" primarily comes from revenue growth rate.

Thermometer "Flip Condition": If ARM's P/E falls from 141x to 60x (assuming stock price drops to $55):

• Absolute valuation sub-score from -3.4 → -0.5 (significant improvement)
• Relative valuation sub-score from -2.9 → -0.5 (comparable to peers)
• Thermometer from -0.94 → +1.5 (enters "cold" zone = undervalued)
• Conclusion: ARM would only become "thermometer neutral to cold" around $55/share – this means a 57% drop from the current $128 is needed to reach "undervalued" []

---

Chapter 31: Conditional Rating Matrix — 6x6 Matrix

31.1 Why Conditional Ratings Are Needed

PW=6 (hybrid mode) means ARM's valuation is highly dependent on several key conditions. A single rating cannot reflect this conditionality. The 6x6 matrix provides "ARM's rating under different P/E and NM combinations."

31.2 Core Variable Selection

ARM's two largest valuation drivers (Red Team confirmed):

• X-axis: FY2030 Terminal P/E — Reflects market belief in ARM's growth sustainability
• Y-axis: FY2030 GAAP Net Margin — Reflects ARM's earnings quality conversion

31.3 6x6 Conditional Rating Matrix

FY2030 NI = Revenue($B) × NM%, Market Cap = NI × P/E. Assuming FY2030 revenue of $8B (scenario-weighted median).

NM \ P/E	15x	20x	25x	28x	33x	40x
40%	$48B	$64B	$80B	$90B	$106B	$128B
	Cautious	Cautious	Cautious	Cautious	Neutral	Neutral
35%	$42B	$56B	$70B	$78B	$92B	$112B
	Cautious	Cautious	Cautious	Cautious	Cautious	Neutral
28%	$34B	$45B	$56B	$63B	$74B	$90B
	Cautious	Cautious	Cautious	Cautious	Cautious	Cautious
25%	$30B	$40B	$50B	$56B	$66B	$80B
	Cautious	Cautious	Cautious	Cautious	Cautious	Cautious
20%	$24B	$32B	$40B	$45B	$53B	$64B
	Cautious	Cautious	Cautious	Cautious	Cautious	Cautious
15%	$18B	$24B	$30B	$34B	$40B	$48B
	Cautious	Cautious	Cautious	Cautious	Cautious	Cautious

Matrix Guide: First row in cell = FY2030 Market Cap, Second row = Rating (vs. current $136B).

• "Neutral Attention": FY2030 Market Cap ≥ $100B (vs. $136B overvaluation < 27%)
• "Cautious Attention": FY2030 Market Cap < $100B

Key Findings: Out of 36 cells, only 3 reached "Neutral Attention" — requiring NM ≥ 35% + P/E ≥ 33x or NM 40% + P/E 28x. Under 93% of NM/P/E combinations, ARM remains under "Cautious Attention".

Cell corresponding to current market cap of $136B: requires NM 40% + P/E 40x (the $128B in the top-right corner of the matrix is still slightly below $136B) → Current pricing requires extreme assumptions "beyond the top-right corner" of the matrix.

31.4 Matrix "Isotherms" and Revenue Sensitivity

In the 6x6 matrix, the "$100B isotherm" (Cautious → Neutral boundary):

Revenue Sensitivity (Supplementary, holding P/E constant at 28x):

Revenue ($B) \ NM	22%	28%	35%	40%
$6B	$37B	$47B	$59B	$67B
$8B (Base Case)	$49B	$63B	$78B	$90B
$10B	$62B	$78B	$98B	$112B
$12B	$74B	$94B	$118B	$134B

Implied by $136B Market Cap: requires revenue of $12B + NM 40% + P/E 28x = $134B (approximate) → Market implies "2.4x revenue growth + 3.3x margin expansion + maintenance of high P/E" all simultaneously achieved.

31.5 Condition Trigger Timeline and Decision Tree

Time	Verifiable Condition	Optimistic Scenario	Pessimistic Scenario
FY2026 Annual Report (May 2026)	Full-year GAAP NM + SBC Disclosure	NM>18%→NM path validated	NM<15%→Path deteriorates
Q1 FY2027 (August 2026)	Revenue Growth Rate + DC Share	>20% growth rate→Growth sustained	<12%→Deceleration confirmed
FY2027 Annual Report (May 2027)	Key Validation Window	NM>22%→Path better than expected	NM<18%→SBC not normalized
H1 FY2028 (Mid-2027)	v10 Release + RISC-V DC Data	v10 confirmed + RISC-V<3%	v10 delayed + RISC-V>5%
FY2029 (March 2029)	RISC-V DC Share + ARM NM	Terminal Rating Framework can be established	Terminal Rating Framework can be established

---

Chapter 32: Integrated Rating and Core Arguments

32.1 Final Rating

Metric	Previous Chapter Value	Current Chapter Update
Rating	Cautious Watch	Cautious Watch (Confirmed)
Calibrated Fair Value	$100B	$100B (Unchanged)
Scenario Tree Expected Value	—	$60.7B
Final Valuation Anchor	—	$80.4B (~$76/share)
Current Market Cap	$136.1B	$136.1B
Implied Downside	-27%	-41%
Thermometer	—	-0.94 (Zone Too Hot)
CQ Weighted Confidence	56.4%	56.4%
Upside Probability	—	12%

Why is the implied downside (-41%) in this chapter greater than the Red Team's calibrated value (-27%)?
The calibrated $100B is the "calibrated fair value" (a reasonable valuation for ARM). The $80.4B in this chapter is the "final valuation anchor" (the median of fair value and scenario tree expected value). The difference stems from:

• Tail risks in the scenario tree (S4+S5, 27% probability) significantly pull down the expected value
• P/E compression effect is more pronounced after the scenario P&L build-out
• $100B (What ARM is worth under reasonable conditions) vs $80.4B (What ARM might be worth probability-weighted)

32.2 Core Arguments – Three-Part Structure (Evidence → Reasoning → Conclusion)

Argument 1: ARM's P/E is a "forward check pre-paying for 10 years of growth"

Section	Content
Evidence	ARM P/E 141x, NI $0.97B; peers SNPS 65x/CDNS 70x; payment networks V/MA 30-35x. Even under Non-GAAP, ARM's P/E ~55x remains higher than all benchmarks. The market implies that FY2030 NI needs to reach $3-4B (currently 4-7x) for P/E to fall back to 30x.
Reasoning	A 141x P/E mathematically requires revenue CAGR of 14.5% + NM to increase from 17% to 30% + simultaneously maintain a 30x terminal P/E. Load-bearing wall analysis shows only a 2% probability of all 6 walls holding. In a 6x6 matrix, ARM is overvalued under 93% of NM/P/E combinations.
Conclusion	ARM is not without value—it is the "invisible infrastructure" of global semiconductors. However, the $136B price has priced in all optimistic assumptions, leaving minimal return potential for new investors (S1 only +18%) and significant downside (S3-S5: -68~87%).

Argument 2: RISC-V is a "Credit Default Swap," not a "substitute"

Section	Content
Evidence	QCOM spent $2.4B acquiring Ventana; Google is advancing Android RISC-V GKI; China's policies mandate domestic substitution; SiFive/Tenstorrent DC chips are in mass production. However, RISC-V still significantly lags in security certification (5-10 years) and enterprise middleware (3-5 years).
Reasoning	RISC-V does not need to "replace" ARM—its mere existence permanently limits ARM's pricing power. ARM migration costs are annualized, reaching a tipping point around 2030 (=ARM royalties). If v10 over-prices (>1.5x)→accelerates RISC-V evaluation→self-fulfilling prophecy. The "CDS pricing anchor" model and migration cost decay curve (Chapter 14A) quantitatively validate this mechanism.
Conclusion	H2 (RISC-V=CDS) holds with 50% confidence. ARM's golden window for pricing power is closing market by market from 2028-2032. Long-term investors need to assume ARM cannot sustain royalty CAGR >10% post-2030.

Argument 3: Earnings Quality is Worse Than Appears – But Improving

Section	Content
Evidence	SBC/Rev 15% (high for the industry, but IPO+3 years peer benchmark median ~15%); DSO 173 days (peers 70-80 days); related-party transactions account for ~30% of licensing; GAAP/Non-GAAP gap 2.6x. However, the SBC normalization path is clear (FY2028 → 10-12% estimated).
Reasoning	RT-2's SBC peer benchmark indicates ARM may reach SBC 10-12% by IPO+5 years (FY2028). If achieved, GAAP NM would increase from 17% to 22-25%. However, there's also the "SBC chicken-and-egg dilemma": expansion (Phoenix/DC) requires talent→requires RSU→SBC doesn't decrease→NM doesn't increase.
Conclusion	H1 (Triple Distortion of Earnings Quality) confidence decreased from 70% to 58%—SBC normalization is more likely than we initially expected. However, DSO and related-party transaction issues have not improved. Overall, earnings quality is "30-40% worse than appears" (revised from the initial 50%).

32.3 "If I'm Wrong" – A Systematic Review

Three scenarios most likely to invalidate the "Cautious Watch" rating:

Scenario A: ARM becomes "The ASML of the AI Era" (Probability ~8%)

During the EUV era, ASML transformed from an "equipment supplier" into an "irreplaceable monopolist"→P/E increased from 20x→40x. If ARM achieves a similar transformation in the AI era:

• ARM CSS becomes the de facto standard for AI inference (not just IP, but a complete solution)
• AI inference demand far exceeds expectations (Jevons Paradox fully effective) → chip shipments ×10
• ARM shifts from "tax per chip" → "tax per AI inference" (business model upgrade)
• Result: FY2030 revenue $15B+, NM 40%, P/E 35x → market cap $210B (+54%)
• In this scenario, my rating is not just wrong, but seriously wrong

Scenario B: Non-GAAP framework ultimately "prevails" (probability ~25%)

If the market confirms in FY2027-2028:

• SBC normalizes to 8% (peer median) → GAAP NM from 17% → 25%+
• Market determines "Non-GAAP reflects true earning power" (similar to early SaaS companies)
• P/E re-anchors on a Non-GAAP basis (~55x → 35x terminal state)
• Result: ARM transforms from a "GAAP loss stock" → "high-profit IP platform" → $110-130B reasonable
• In this scenario, my rating is only "slightly" wrong (overestimation narrows to 0-10%)

Scenario C: RISC-V encounters the "Linux on Desktop" dilemma (probability ~15%)

Linux Desktop has been "about to surpass Windows" every year since 1999 — 25 years later it still only holds a 3% share. If RISC-V faces the same fate:

• Software ecosystem never reaches "good enough"
• Open-source community fragmentation (incompatible RISC-V extensions from different vendors)
• Enterprise clients abandon RISC-V evaluation → ARM's pricing power fully preserved
• Result: ARM royalty CAGR 18%+ → FY2030 revenue $12B → $140-160B reasonable
• In this scenario, my rating is seriously wrong

Combined Probability: At least one of Scenario A (8%) + B (25%) + C (15%) materializes ≈ 40% (considering overlap)

Implications for Rating: 40% probability my rating is "wrong" (but mostly "slightly wrong" in Scenario B, not "seriously wrong" in A/C). 60% probability the rating is correct or not prudent enough. "Prudent Watch" is a reasonable risk-adjusted choice under this probability distribution.

Symmetry Check of "Error Probability": We also need to check the probability of the "Prudent Watch rating being insufficiently prudent":

Scenario D: RISC-V acceleration + P/E collapse (probability ~15%)
If RISC-V reaches a 5% share in DC by 2028 (2 years early) + accelerated China RISC-V policy:

• ARM's pricing power starts to erode in 2028 (instead of 2030) → royalty CAGR drops to 5%
• Market prices in early → P/E drops from 141x → 30x within 2 years (instead of 5 years)
• Result: S3/S4 deeper and faster than our expectations → FY2028 market cap $40-50B (-63%~-70%)
• In this scenario, the rating is insufficiently prudent

Scenario E: Multiple black swans simultaneously (probability ~3%)
Taiwan Strait conflict + SoftBank crisis + AI CapEx plummeting occur in the same year:

• ARM revenue -30%+ floating shares impact + industry cycle downturn
• Result: Market cap $15-25B (-82%~-89%)
• In this scenario, the rating is seriously insufficiently prudent

Two-Way Probability Summary:

Direction	Scenario	Probability	Deviation Magnitude
Rating too pessimistic	A (AI ASML) + B (Non-GAAP) + C (RISC-V Failure)	~40%	My error 0~54%
Rating appropriate	Base Case (S2/S3)	~42%	—
Rating too optimistic	D (RISC-V Acceleration) + E (Multiple Black Swans)	~18%	Insufficiently prudent 10~40%

Conclusion: Probability of "too pessimistic" rating (40%) > probability of "too optimistic" rating (18%) → our rating is indeed conservative — but this is intentional (risk management principle under PW=6: better to make the mistake of "missing an opportunity" than "being overly optimistic").

32.4 Position Recommendation Framework (Not a Recommendation)

This section provides a decision framework, not investment advice. Investors should make judgments based on their own risk appetite and current holdings.

Investor Type	Current Recommendation	Trigger Conditions (Change in Recommendation)
Existing Holder (Long-term)	Consider partial reduction (retain ≤50% position)	NM > 22% + DC > 20% + RISC-V < 3% → Hold full position
Existing Holder (Short-term)	Set stop-loss at $100 (-22%)	P/E < 80x → Stop-loss triggered
New Holder (Bullish on ARM business)	Wait for below $80 (thermometer ~0, neutral zone)	FY2027 NM > 25% → Consider entry at $90 ahead of schedule
New Holder (Value Investor)	Wait for below $55 (thermometer +1.5, undervalued zone)	Any KS triggered → May reach this price point sooner
Short Seller	Shorting not recommended (float 9.7%, high short squeeze risk)	Float > 15% AND SoftBank divestment → Short seller safety cushion increases

"Staggered Accumulation" Approach (if deciding to wait for entry):

Why Shorting is Not Recommended: ARM's free float is only 9.7% (SoftBank holds 90%) → extremely high shorting costs (high borrow fees) + significant short squeeze risk. Even if the direction is correct, short sellers might be forced out during the process. Shorting ARM requires waiting until SoftBank reduces its stake and the free float increases to >15%.

32.5 One-Sentence Investment Thesis

Bull Case: "ARM is the invisible infrastructure of the AI era, its royalty model equals a chip tax empire, DC growth is just beginning, 95% gross margin + network effect = pricing power, SBC will normalize, 141x P/E is only 55x under Non-GAAP."

Bear Case: "ARM is a $136B stock with only a 12% probability of positive returns, an 88% probability of loss, and a 60% probability of losing more than half. RISC-V doesn't need to replace ARM, merely limiting its pricing is sufficient. The SBC chicken-and-egg dilemma means growth and margins cannot be achieved simultaneously. The probability of all six supporting walls holding is only 2%."

Our Version: "ARM's business is good, but the price is not. Buying ARM at $128 requires an ALL-IN bet on 'everything perfect' (2% probability). Below $80, ARM becomes interesting. Below $55, ARM is a bargain. Patience is the optimal strategy."

32.6 Summary of Key Report Findings (Top 10)

Reviewing the complete analysis (374K+ chars), here are the 10 most valuable findings:

Rank	Finding	Source	Impact (Valuation/Strategy)	Type
1	P/E Compression > Earnings Growth: Even in S2 (Robust), P/E from 141x to 28x leads to -49%	P5 Ch29	Fundamental - illustrates ARM's losses in any "reasonable" scenario	CI-04
2	RISC-V = CDS, not a substitute: Price Cap > Substitution	P2 Ch14	±$30-40B Fair Value	CI-01
3	B3 (Pricing Power) is the only Jack: One wall determines $23B	P4 Ch25	Focus 80% of research resources on B3	CI-03
4	Joint Probability of Load-Bearing Walls is only 2%: Market is betting all 6 walls stand	P3 Ch21	Current $136B price requires "everything perfect"	Analysis
5	SBC Chicken-and-Egg Dilemma: Growth and Net Margin expansion cannot coexist	P4 RT-2	FY2028 Net Margin 22-25% (below consensus 28-30%)	CI-02
6	Risk-Adjusted Return is Negative: S1 (+3.3%/year) < CAPM Requirement (9.9%)	P5 Ch29.8	ARM is a poor risk-adjusted investment in all scenarios	Analysis
7	SoftBank 5-Strategy Game Theory: Slow divestment of 35% most likely	P4 RT-5	Free Float from 9.7% → 30% → P/E compression 20-30%	Analysis
8	93% Matrix Cells = Caution Warranted: Requires Net Margin ≥ 35% + P/E ≥ 33x for neutrality	P5 Ch31	Current pricing requires "beyond the upper-right corner of the matrix"	Analysis
9	Thermometer Flip requires $55/share: From -0.94 → +1.5 (cold zone) requires a 57% drop	P5 Ch30.4	"Buy-the-dip" threshold is far above market expectations	CI-07
10	Scenario Slippage Path: S2 → S3 → S4 Series Probability 12-15%	P5 Ch29.5	Downside tail is thicker than implied by independent probabilities	CI-06

Distribution of Findings: 6 out of 10 core findings emerged in the latter half (after deeper analysis) → illustrating that the marginal value of in-depth analysis is higher in the later stages (Reverse DCF → Red Team → Scenario Tree's "Progressive Insight Chain").

Direction of Findings: 8 out of 10 findings are bearish, 1 is neutral (SBC normalization path), and 1 is structural (conditional framework). This reflects that at $136B, the density of bearish evidence for ARM significantly outweighs bullish evidence – but this does not equate to "ARM is a bad company," only to "ARM is an expensive company."

32.7 Final Note to Investors

But a great company ≠ a great investment.

At $128/share (P/E 141x), ARM's price implies "all load-bearing walls standing simultaneously" (2% probability). Even if ARM executes perfectly (S1 scenario), the 5-year annualized return is only +3.3% - less than 80% of the risk-free rate. This is not a repudiation of ARM's business, but a repudiation of ARM's valuation.

Our advice is not "do not touch ARM," but rather "wait until the price is reasonable". Below $80, it starts to get interesting; below $55 is the real opportunity. Mr. Market will eventually offer such a price – either through P/E normalization or a black swan event. Patience is the best strategy.

32.8 CQ Closed Loop – 5-Element Format

CQ-1: Royalty Economics End State (Weight 25%, Confidence 57%)

Element	Content
Question	What level can ARM royalties reach by FY2030? Can v10 price increases be sustained?
Finding	v9→v10 price increase 1.3-2.0x (range); FY2030 royalties $3.5-7.2B (Scenarios 1-3); CSS single-chip $50-100+ royalties (DC) vs $0.10-0.50 (Mobile); Royalty CAGR 5-20% (scenario dependent)
Impact	Royalty path determines 60% of ARM's revenue → directly determines Fair Value ±$30-50B
Remaining Uncertainty	v10 royalty rate unconfirmed; CSS mass production cancellation rate unknown; When RISC-V's royalty pricing ceiling becomes effective
Monitoring Indicators	TS-1 (v9/v10 proportion), TS-3 (DC royalty share), TS-14 (v10 progress)

CQ-2: Data Center Journey (Weight 20%, Confidence 52%)

Element	Content
Question	How much can ARM's share in DC grow from 12%? Can the CSS model be sustained?
Finding	19 CSS licenses, 5 in mass production; Estimated 11-15 in mass production by FY2030 (60% success rate); DC royalties $0.8-4.0B (scenario dependent); AI CapEx may peak in 2026-2027; Jevons Paradox partially protects inference demand
Impact	DC is ARM's fastest-growing segment → from 12% → 18-38% (scenario dependent) → impacts Fair Value ±$20-40B
Remaining Uncertainty	When will AI CapEx cycle peak; CSS mass production cancellation rate; RISC-V DC penetration speed
Monitoring Indicators	TS-2 (Number of CSS mass productions), TS-3 (DC royalty share), TS-7 (RISC-V DC share)

CQ-3: SoftBank Double-Edged Sword (Weight 18%, Confidence 58%)

Element	Content
Question	Is SoftBank's 90% stake a protection or a threat? When/how will it divest?
Finding	RT-5 two-way correction → 5-year net effect approximately -$1B (neutral to slightly negative); 5-strategy game theory (S2 slow divestment of 35% most likely); Margin call trigger line ~$85/share; Positive feedback spiral risk (unstoppable once triggered); Prisoner's Dilemma structure
Impact	SoftBank's actions could lead to -$30~50B stock price impact (S4/S5 scenarios); Free float scarcity disappears → P/E compression 20-30%
Remaining Uncertainty	Masayoshi Son's personal decisions are unpredictable; Scale of Izanagi's financing needs; Precise margin terms undisclosed
Monitoring Indicators	TS-6 (SoftBank Margin), KS-2 (Divestment > 5%), TS-13 (Short interest proportion)

CQ-4: RISC-V Pricing Ceiling (Weight 15%, Confidence 63%)

Element	Content
Question	How much does RISC-V's existence long-term limit ARM's pricing power?
Finding	CDS pricing anchor model confirmed: ARM can raise prices up to the "annualized migration cost" but no higher; Migration costs decay to a tipping point by 2030; Security certification barrier protects automotive/industrial for 5-10 years; Boiling frog erosion path: lock-in strength from 10/10 → 4/10 (2024→2030)
Impact	Pricing power is the "jack" (B3) among the 6 load-bearing walls – with the highest risk impact at $23B; B3 collapse is the only single wall that can independently change valuation >20%
Remaining Uncertainty	RISC-V software ecosystem "critical mass" tipping point; Whether v10 features are sufficient to "reset" lock-in; Actual effect of the Google/China dual-engine push
Monitoring Indicators	TS-7 (RISC-V DC share), TS-14 (v10 progress), KS-1/KS-4 (RISC-V milestones)

CQ-5: Arm China Deceleration (Weight 10%, Confidence 65%)

Element	Content
Problem	What are the revenue and governance risks associated with Arm China (Anmou Technology)?
Findings	Arm China accounts for 19% of revenue ($900M+); Governance crisis with CEO dismissal in 2022; Chinese policy driving domestic substitution → long-term revenue facing policy risk; Taiwan Strait vulnerability analysis + hedging strategies + IPO scenarios (Chapter 23)
Impact	Arm China revenue zeroed out (extreme) → EV impact -$15-25B; Revenue decline 30% (moderate) → -$5-8B
Remaining Uncertainty	Intensity of Chinese policy execution; Likelihood of Arm China independent IPO; Taiwan Strait geopolitical risk (Unknowability 9/10)
Monitoring Metrics	TS-8 (China Revenue Share), KS-6 (China Mandating RISC-V)

CQ-6: The Phoenix Transformation Paradox (Weight 12%, Confidence 45%)

Element	Content
Problem	Is Arm's transformation from an IP company to a chip company an opportunity or a risk?
Findings	Five end-state models: Major Success (10%), Moderate Success (20%), Failure (35%), Backlash (15%), Cancellation (20%); Probability-weighted net effect +$1.75B (≈Neutral); Backlash transmission chain 7-step series ~8%; IFS analogy: Intel's failed transformation precedent
Impact	Phoenix is nearly neutral to Arm in terms of expected value → Should not be a reason to buy or sell
Remaining Uncertainty	TSMC capacity allocation; SoftBank internal demand scale; Customer reaction (whether to initiate RISC-V evaluation)
Monitoring Metrics	TS-12 (Phoenix Pipeline/Revenue), TS-2 (CSS Customer Indirectly Reflecting Competition)

---

Chapter 33: Non-Consensus Insight Registry (CI Registry)

CI Registry Definition

Non-Consensus Insights (CIs) are unique findings in this report that are different from mainstream market views and supported by quantitative data. The value of CIs lies in providing investors with information advantages not yet widely recognized.

CI-01: Quantifying RISC-V CDS Pricing Power – "Insurance Premium" More Important Than "Substitution"

Field	Content
Insight	The real threat of RISC-V to Arm is not "substitution" (low probability) but "price capping" – Arm's royalty cap is anchored by the annualized cost of RISC-V migration. Arm can raise prices to a level where "it's not worth clients spending $35-125M to migrate," but no higher.
Degree of Non-Consensus	High – Market discussions focus on "Can RISC-V replace Arm?" (binary), ignoring "RISC-V's existence implies price capping" (continuous)
Quantitative Support	Annualized migration cost $7-25M (by company type); Arm royalties $5-400M/year (by client size); Tipping point around 2030 (migration cost declines to royalty level)
Transferability	All IP companies facing open-source alternatives (Synopsys vs OpenROAD? Dolby vs open-source codecs?)
Impact	Terminal royalty CAGR ceiling ~10% (vs market implied 14.5%+)
Source	Part II(Ch14)+Part IV(RT-4)

CI-02: SBC Chicken-and-Egg Dilemma – Growth and Margins Mutually Exclusive

Field	Content
Insight	Arm needs high SBC to attract talent → expansion (DC/Phoenix), while simultaneously needing SBC to decline → margin expansion (Wall Street expectations). These two goals are mutually exclusive in FY2027-2029. The market implies both are achieved simultaneously.
Degree of Non-Consensus	Medium – Analysts forecast "revenue growth" and "margin expansion" separately but rarely examine the inherent conflict in talent demand between the two
Quantitative Support	DC/Phoenix hiring ~2,000 people × $300K RSU = $600M SBC increment; SBC normalization requires grants < vesting; these two are contradictory
Transferability	All high-SBC post-IPO companies (SNOW/PLTR/CRWD)
Impact	FY2028 NM may be 22-25% instead of the consensus 28-30%
Source	Part IV(RT-2)

CI-03: Load-Bearing Wall B3 is the Only "Leverage Point" – Pricing Power Determines Everything

Field	Content
Insight	Among the 6 load-bearing walls, B3 (pricing power) is the only wall that can individually change valuation >20%. B3's success or failure simultaneously impacts the conditional probabilities of B1 (revenue) and B4 (DC share). The collapse of any other single wall is not sufficient to change the rating.
Degree of Non-Consensus	Medium – The market focuses on multiple risks but rarely prioritizes "which risk is most important"
Quantitative Support	B3 risk product $23B (highest); B3 collapse → EV impact -$35-45B (largest single wall); B3 sensitivity ±10pp → largest change in joint probability (±0.5pp)
Transferability	Any company with a "moat core" – identifying which wall is the leverage point
Impact	Investors should focus 80% of their attention on tracking B3 (RISC-V pricing power)
Source	Part IV(Ch25)

CI-04: P/E Compression > Earnings Growth – The Math of the "Valuation Trap"

Field	Content
Insight	In scenario S2 (stable growth), Arm's earnings grew +137% (NI from $0.97B → $2.30B), but P/E compressed -80% (from 141x → 28x) → net return -49%. Even with strong business performance, valuation compression alone can lead to substantial losses.
Degree of Non-Consensus	High – Bullish arguments that "Arm earnings are growing" overlook the mathematical effect of P/E compression
Quantitative Support	S2: Earnings contribution +$186B, P/E compression -$258B, Net -$72B (-49%); S3 even more extreme: Net -$93B (-68%)
Transferability	All high P/E growth stocks (SNOW 80x, PLTR 100x+, ARM 141x)
Impact	Arm "must" be in S1 (Chip Tax Empire, 12% probability) to deliver positive returns to investors
Source	Part V(Ch29)

CI-05: Arm is Unsuited for "Precise Valuation" – Apply a "Conditional Framework"

Field	Content
Insight	Among Arm's 5 largest uncertainties, 2 have unknowability ≥8 (SoftBank behavior 8/10, China geopolitics 9/10). Even with perfectly correct business analysis, exogenous shocks could cause valuation deviations of ±30%. The correct question is not "What is Arm worth?" but "Under what conditions am I willing to hold Arm?"
Degree of Non-Consensus	Medium – Sell-side analysts are still providing precise price targets for Arm ($130-$200), but this precision is false
Quantitative Support	Unknowability ratings: China geopolitics 9/10, SoftBank 8/10, Business model end-state 7/10; Exogenous variables could affect valuation by ±30%
Transferability	All companies with PW≥5 – "Discovery System" superior to "Precise Valuation"
Impact	Investors should use a conditional rating matrix (Ch31) rather than a single price target
Source	Part IV(Ch28)+Part V(Ch31)

CI-06: Scenario Erosion More Dangerous Than Scenario Jumps – "Path Dependence" Underestimated

Field	Content
Insight	ARM's scenarios are not isolated events—there is a clear 'slippery slope path': S2→S3 (RISC-V gradual adoption)→S4 (SoftBank forced action). A serial probability of 12-15% implies that once S3 is entered, ARM has a 35-40% probability of further sliding towards S4. The market prices each scenario independently, ignoring path dependency.
Degree of Non-Consensus	Medium—Analysts assign independent probabilities to scenarios, ignoring conditional probabilities between scenarios (e.g., 'higher probability of S4 once S3 is entered').
Quantitative Support	S2→S3 Serial Probability: 28%×40%=11.2%; S3→S4 Serial Probability: 33%×35%=11.6%; S2→S3→S4 Full Chain: 28%×40%×35%=3.9%
Transferability	All companies facing multiple downside catalysts—Examine conditional probabilities between scenarios rather than independent probabilities.
Impact	The downside tail is 'thicker' than implied by independent probability models—the actual probability of significant losses is higher than what the numbers suggest.
Source	Part V (Ch29.5 Scenario Transition Paths)

CI-07: $55/share is "Thermometer Neutral"—The True Decline Needed for Undervaluation Far Exceeds Bullish Expectations

Field	Content
Insight	At $128/share, ARM's thermometer reading is -0.94 (warm zone, tending hot). Through thermometer reversal analysis, ARM needs to fall to approximately $55/share (P/E ~60x) to enter the 'cold zone' (undervaluation). This means ARM needs to drop 57% from its current price to reach an 'attractive' area. The market's threshold for 'buy the dip' is set too high—$100 or even $80 is still 'overpriced'.
Degree of Non-Consensus	High—Bulls view any 5-10% pullback as a 'buying opportunity', but the thermometer shows that even a 20% pullback is still in the 'warm zone'.
Quantitative Support	$128→Thermometer -0.94; $100→Thermometer ~-0.2 (still in warm zone); $80→Thermometer ~+0.5 (warm zone, tending cold); $55→Thermometer ~+1.5 (cold zone)
Transferability	All high P/E stocks—Use the thermometer, not intuition, to determine 'what price constitutes undervaluation'.
Impact	The advice to 'wait until below $80 before considering' is more disciplined than 'buy the dip at $100'.
Source	Part V (Ch30.4 Thermometer Reversal Analysis)

---

Chapter 34: Tracking Signal Dashboard

34.1 Core Dashboard (14 Signals, 6-Field Format)

TS#	Signal	Current Value	Optimistic Threshold	Warning Threshold	CQ/Hypothesis	Frequency	Current Signal
1	v9/v10 Royalty Share	>50%	>70%	<40%	CQ-1	Semi-annual	🟡
2	Number of CSS Mass Production Customers	5	>10	<4	CQ-2	Quarterly	🔴
3	DC Revenue Royalty Share	~12%	>20%	<8%	CQ-2	Quarterly	🔴
4	GAAP/Non-GAAP Gap	2.6x	<2.0x	>3.0x	H1	Quarterly	🔴
5	DSO (Days)	173	<120	>200	H1	Quarterly	🔴
6	SoftBank Margin Withdrawal	$8.5B	<$6B	>$15B	CQ-3	Quarterly	🟡
7	RISC-V DC Shipment Share	<2%	<2%	>5%	CQ-4	Quarterly	🟢
8	Arm China/Total Revenue	19%	>22%	<15%	CQ-5	Semi-annual	🟡
9	FCF Margin	4.4%	>15%	<5%	H1	Quarterly	🔴
10	R&D/Revenue	56.3%	<45%	>60%	RT-2	Quarterly	🟡
11	SBC/Revenue	~15%	<10%	>18%	H1	Semi-annual	🟡
12	Phoenix Pipeline/Revenue	$0	>$500M	Canceled	CQ-6	Semi-annual	🔴
13	Short Interest Ratio	~15%	<10%	>25%	—	Monthly	🟡
14	v10 Progress (Management Commentary)	In Development	Roadmap Confirmed	Delay Signal	CQ-4	Quarterly	🟡

Current Signal Summary: 🟢1 / 🟡7 / 🔴6 → Predominantly Red (Consistent with "Cautious Watch")

34.2 TS Deep Dive (Detailed Key Signals)

TS-7 (RISC-V DC Share) — "One Signal to Rule Them All":

This is the most important single tracking signal confirmed by the Red Team (A directly observable proxy for the B3 Key Driver):

RISC-V DC Share	Implication	Impact on ARM Rating	Corresponding Scenario
<1%	RISC-V DC progress is slow	Cautious → Potential Upgrade	S1/S2
1-3%	Initial penetration, not yet established	Maintain Cautious	S2
3-5%	"Good enough" signal appears	Maintain Cautious	S2→S3
5-8%	Pricing power begins to erode	Maintain/Deepen Cautious Stance	S3
>10%	ARM's pricing power substantially eroded	Deepen Cautious Stance	S3→S4

TS-4 (GAAP/Non-GAAP Gap) — Thermometer of the "Battle of Accounting Frameworks":

Gap Multiple	Implication	Impact on ARM Rating
<1.5x	SBC fully normalized, GAAP=Non-GAAP	Strong positive → Potential upgrade
1.5-2.0x	SBC is normalizing	Positive → Reduces overvaluation
2.0-2.5x	SBC remains high but is improving	Neutral
2.5-3.0x(Current)	SBC remains high, no improvement	Negative → Maintain cautious stance
>3.0x	SBC deteriorating	Strong negative → Potentially deepen cautious stance

TS-6 (SoftBank Margin Call) — "Black Swan Detector":

SoftBank margin loan draw is a "real-time leverage indicator" — if it moves from $8.5 billion (current) → $15 billion+:

• Implies SoftBank is significantly increasing leverage (Izanagi financing?)
• ARM, as the primary collateral, faces higher forced liquidation risk
• Trigger condition may shift from "ARM falls to $85" → potentially to "ARM falls to $100" (lower safety cushion)

34.3 TS Correlation and Interaction Effect Analysis

Tracking signals are not independent — changes in certain signals can drive others:

Positively correlated signal pairs (one deteriorates → the other may follow):

Signal A	Signal B	Correlation	Transmission Mechanism
TS-7 (RISC-V DC)	TS-1 (v10 Share)	Strong Positive Correlation	RISC-V acceleration → ARM forced to accelerate v10 launch + price reductions
TS-6 (SoftBank Margin Call)	TS-13 (Short Interest)	Strong Positive Correlation	SoftBank leverage ↑ → Market risk perception ↑ → Short sellers flock in
TS-4 (GAAP/Non-GAAP Gap)	TS-9 (FCF Margin)	Strong Positive Correlation	SBC does not decrease → Large GAAP gap → Low FCF margin
TS-2 (CSS Customer Count)	TS-3 (DC Share)	Moderate Positive Correlation	CSS mass production → DC royalty growth (6-12 month lag)
TS-12 (Phoenix)	TS-2 (CSS Customer Count)	Weak Negative Correlation	If Phoenix competes with customers → Slower CSS new licensing

Signal Cluster Identification:

"Domino Warning": If Cluster B (Competitive Threat) begins to deteriorate (TS-7 > 5% + TS-14 delay) → Cluster A (Profitability Quality) may follow suit 6-12 months later (SBC↑ → TS-4↑ → TS-9↓). This is the "RISC-V → Profitability Quality → Valuation" transmission chain.

Earliest Warning Signal: TS-7 (RISC-V DC share) is the "upstream" of the entire signal network — if TS-7 deteriorates, investors should expect multiple downstream signals to deteriorate concurrently 6-18 months later.

"Signal Cascade" Scenario Simulation: If TS-7 jumps from <2% to 5% (in a certain quarter of FY2028):

Time T	Event	TS Change	Rating Impact
T+0	RISC-V DC quarterly report released	TS-7: 🟢→🔴	Observe
T+1Q	ARM management questioned on response	TS-14: 🟡→🔴(v10 acceleration)	Observe
T+2Q	ARM announces DC customer RISC-V evaluation	TS-2: 🔴(CSS new licensing slows)	Cautious (Strong)
T+3Q	ARM accelerates R&D hiring in response	TS-11: 🟡→🔴(SBC↑)	Maintain
T+4Q	FY annual report confirms Net Margin deterioration	TS-4: 🔴(GAAP gap widens)	Maintain
T+5Q	P/E begins to compress	TS-13: 🟡→🔴(Short sellers flock in)	Possible downgrade
T+6Q	Value of SoftBank's stake shrinks	TS-6: 🟡→🔴(Margin call pressure)	Emergency review

From TS-7 turning red to TS-6 turning red (full chain transmission): approximately 18 months. This means investors have a window of ~4 quarters to reduce positions — if they paid attention to TS-7 at T+0. Most investors will not react at T+0 (because "RISC-V share is still small"), but will only react at T+4Q (Net Margin deterioration) → at which point P/E has already compressed by 30-40%.

Signal Weight Calibration:

Signal	Current Weight (Implied)	Recommended Weight	Reason
TS-7	Equal	25%	B3 Jack + Upstream Signal + Scenario Splitter
TS-4	Equal	15%	Earnings Quality Core Proxy
TS-6	Equal	15%	Black Swan Detector
TS-2	Equal	10%	DC Growth Validation
Other 10	Equal	35% Total	Supplementary Information

34.4 Signal Interpretation Guide

How to use the TS Dashboard:

1. Quarterly Update: Update TS-2/3/4/5/9/10/14 after earnings release
2. Semi-Annual Update: Update TS-1/8/11/12 after 20-F/Annual Report
3. Continuous Monitoring: TS-6 (SoftBank) and TS-13 (Shorts) may change at any time

Rating Change Trigger Rules:

• Upgrade to "Neutral": 🔴 drops to ≤2 AND 🟢≥4 AND no KS triggered
• Downgrade to "Prudent (Strong)": Any KS triggered OR 🔴≥8
• Emergency Review: Any single TS movement exceeding 2 levels (🟢→🔴 or 🔴→🟢)

34.5 Investment Calendar (FY2027-FY2028)

34.6 90-Day Action List

Priority	Action	Target Date	Trigger Condition	Expected Output
P0	Monitor ARM FY2026 Annual Report (20-F)	May 2026	Upon release	Update all TS + NM path validations
P0	Update SBC precise data	May 2026	Full 20-F disclosure	Resolve OQ-1 (SBC precise value)
P1	Track SoftBank Quarterly Report	May 2026	SB earnings release	Update TS-6 + KS-2
P1	RISC-V DC Shipment Tracking	Ongoing	Any outperformance	Update TS-7 + KS-1
P2	Confirm CSS Mass Production Customer Count	August 2026	Q1 FY2027	Update TS-2
P2	Monitor v10 Architecture Pre-release	H2 2026	ARM Tech Day	Update TS-14
P3	QCOM/MTK RISC-V Announcement	Ongoing	Any official release	Update KS-4

---

Chapter 35: Open Question List

35.1 High-Priority Open Questions

OQ#	Question	Importance	Obstacle	Research Path	Data Source	Expected Resolution Time
OQ-1	What is ARM's true and precise SBC value?	Very High	FPI exemption, FY2025 20-F not fully disclosed	Wait for full SBC disclosure in FY2026 Annual Report (20-F); simultaneously estimate using RSU grants × fair value	SEC EDGAR 20-F	May 2026
OQ-2	Terms of SoftBank's related-party licensing agreement?	High	Limited disclosure in 20-F	Analyze FY2026 20-F "Related Party" notes; compare SB subsidiary purchase amounts with ARM licensing revenue	SEC EDGAR + SB Financial Reports	May 2026
OQ-3	When will v10 architecture be released + royalty rate?	Very High	ARM has not yet announced roadmap	Track ARM Tech Day/Techcon; monitor patent applications (USPTO); analyze changes in management's wording	ARM IR + USPTO	2026H2-2027H1
OQ-4	What is the precise royalty rate range for CSS?	High	ARM does not publicly disclose	Reverse engineer via clients: NVIDIA Grace BOM cost - TSMC foundry fee - ARM license fee → residual value = royalty; similar method for AWS Graviton	Teardown Reports + Supply Chain	Ongoing
OQ-5	What is the gross margin structure for the Phoenix product?	High	Product not yet shipped	Wait for financial breakdown after first shipments; benchmark against Ampere AmpereOne gross margin (estimated 50-60%)	ARM IR + Ampere Benchmark	2027H2+

35.2 Medium-Priority Open Questions

OQ#	Question	Importance	Research Path	Expected Resolution Time
OQ-6	Arm China Independent Listing Plan?	Medium	Monitor Chinese regulations + Arm China personnel changes; Pay attention to Shanghai Stock Exchange STAR Market	Unpredictable
OQ-7	SoftBank Margin Loan Trigger Clause?	Medium	SoftBank semi-annual report disclosure (Japan GAAP requirements); Analyze SB debt maturity schedule	H2 2026
OQ-8	ARM Royalty Rate in NVIDIA Grace?	Medium	GB200/GB300 teardown; Compare ARM financial report DC royalty total / NVIDIA shipments	H2 2026 (after teardown)
OQ-9	Windows on ARM Market Share Trajectory?	Medium	Mercury Research quarterly PC report; Qualcomm Snapdragon X shipment data	Quarterly Tracking
OQ-10	ARM Dividend/Buyback Possibility?	Low	Track capital allocation policy statements; Increased free float after SoftBank's stake reduction → buybacks become more meaningful	2027+

35.3 OQ Prioritization and Research Resource Allocation

Priority	OQ	Estimated Research Time	Expected Information Gain	ROI (Gain/Time)
1	OQ-3 (v10 Royalty Rate)	Continuous Tracking	Very High (Directly impacts B3)	Very High
2	OQ-1 (SBC Precise Value)	1 Week After FY2026 Annual Report	High (Resolves H1 Framework)	High
3	OQ-4 (CSS Royalty Rate)	After Teardown Report Release	High (Validates DC Model)	High
4	OQ-2 (SB Related Contracts)	After FY2026 Annual Report	Medium (Quantifies Related Impact)	Medium
5	OQ-5 (Phoenix Gross Margin)	H2 2027+	Medium (Long-term Impact)	Low (Distant Horizon)

Highest ROI Research Action: OQ-3 (v10 Royalty Rate) — If the v10 royalty rate can be determined in advance from ARM Tech Day or patent analysis, it will directly change the confidence level of B3/CQ-4, potentially triggering a rating change. We recommend allocating the most research resources.

35.4 Intersection of OQ and Kill Switch

OQ	If Answer is Optimistic	If Answer is Pessimistic	Related KS
OQ-1 (SBC)	SBC<10%→NM Path Confirmed	SBC>15%→H1 Deepened	KS-3
OQ-2 (Related Contracts)	Proportion of related-party transactions decreases→Governance Improvement	Proportion of related-party transactions increases→Governance Deteriorates	—
OQ-3 (v10)	Royalty >1.5x→Pricing Power Confirmed	Royalty <1.2x→B3 Weakened	KS-1/KS-4
OQ-4 (CSS Royalty)	>$80/chip→DC Model Validated	<$30/chip→DC Revenue Below Expectations	—
OQ-7 (SB Margin)	Trigger Line $60→Large Safety Buffer	Trigger Line $100→Dangerous	KS-2

© 2026 Investment Research Agent. All rights reserved.