41x Deleveraging: The Market Gives a Hybrid Three-Engine Entity an AI Average Score

COHR Optical Communications and Materials Platform Deep Dive Report

Analysis Date: 2026-04-14 · Data Cut-off: FY2026 Q2 (2025-12-31)

Chapter 1: Executive Summary

What the Market Sees in COHR. An AI optical communications growth stock—benefiting from the 800G/1.6T upgrade cycle, with a $2B endorsement from NVIDIA, priced at 41x Forward P/E, targeting FY2027E EPS of $7.47. On the financial statements, the D&C segment accounts for approximately 70%, and the Industrial segment for approximately 30%; among these, AI Datacom (estimated) accounts for approximately 55–60% of consolidated revenue, +34% YoY. The market is buying AI growth.

What This View Fails to Explain. COHR's 17.5% growth vs. LITE's 65.5%, but the Forward P/E difference is only 6x—the market's premium per unit of growth for COHR is 3 times that of LITE, indicating that a portion of the 41x P/E is not driven by growth. Our breakdown reveals that out of the $10.16 EPS increment from FY2025→FY2028E, approximately $3-4 stems from non-growth factors such as decreasing D&A, interest savings, and preferred share redemption. Approximately three tenths of revenue (Industrial segment, ~28% in FQ2'26; ~31% in Q1 FY26) is contracting by -10%, yet it is uniformly priced at an AI multiple P/E. The market is bundling three entirely different sources of value—AI growth, capital structure normalization, and shrinking businesses—and pricing them together with a growth stock valuation language.

So What Is It, Actually. COHR is not a pure AI growth stock but rather a "41x deleveraging"—a post-merger deleveraging hybrid entity: Its reported segments are D&C and Industrial, approximately 72% / 28% (FQ2'26); for valuation, it is still split into three lines—AI Datacom approximately 55–60%, Industrial segment approximately 28% (including SiC approximately 5–8% and industrial lasers, etc.), and accounting and capital structure release from D&A/deleveraging—sharing wafer infrastructure, using decreasing D&A + debt repayment + mix improvement to mechanically create an EPS trajectory. ROIC 4.2% < WACC 10%—every additional $1 of capital invested is consuming value, not creating it. The primary variable truly determining COHR's fate is not Networking growth rate, but when ROIC will exceed WACC. 30% growth but ROIC<WACC = burning cash to build capacity; 10% growth but ROIC>WACC = starting to create value for shareholders.

What This Redefinition Changes. The valuation method shifts from "a single P/E for the entire company" to a Sum of the Parts (SOTP): Bear $150.7 (30%) / Base $211.8 (45%) / Bull $344.4 (25%) → Weighted Average $226.6, vs. $307.50 = -26.3%. Even the most optimistic probability yields only $249. Rating: Cautious Watch. The most probable path is not a cliff dive, but a boiling frog scenario—gradual deterioration with 40-50% probability, a cumulative -36% over 3 years, and an annualized -14%. Kill Switch: If ROIC remains <WACC in FY2027 → Thesis systemically fails.

Chapter 2: Main Question — Is the Market Using One Label to Score Three Different Curves with an Average?

2.1 What the Market Sees in COHR

The market's default map: COHR is a vertically integrated leader in AI optical communications, possessing the industry's longest technology positioning (full InP stack from substrate to module), a $2B endorsement from NVIDIA securing demand until 2028, and management's "unprecedented visibility." Priced at 41x Forward P/E, implying an FY2027-2030 EPS CAGR of 25%+. Key sell-side analysts value it using Non-GAAP EPS × growth P/E, with core tracking variables being Networking/Datacom revenue growth and 800G/1.6T shipment volumes.

The cornerstone of this default map: If Hyperscaler CapEx continues at +30% or more annually for over 3 years (ensuring Networking demand), if COHR secures >20% module market share in the 1.6T era (ensuring revenue growth), and if deleveraging proceeds as planned (ensuring EPS floor), then a 41x P/E in FY2027 is reasonable—and 32x on FY2028E $9.64 also offers some safety margin.

2.2 Why This Default Map Is Starting to Loosen

We identified 4 anomalies. The loosening of the market's map is not a dramatic "mistake" but a series of minor yet cumulative inconsistencies:

Anomaly 1: P/E-to-Growth Mismatch. COHR's P/E/G of 1.64x is the highest among LITE (1.18x) and Lumentum (1.50x). The market assigns the highest valuation premium per unit of growth to COHR, yet COHR has the lowest growth rate, lowest AI purity (69% vs. LITE's 90%+), and lowest profit margin (GM 37-39% vs. LITE's 42.5%). This suggests that a portion of the 41x P/E is not driven by growth—it's buying deleveraging upside, SiC optionality, or simply a label premium.

Anomaly 2: Rapid Inventory Increase. FQ2'26 inventory stood at $1,848M, a +28.5% increase over 6 months, while revenue grew only +10.3% during the same period. Inventory growth was 2.8 times revenue growth. Analysis confirms COHR's inventory exhibits downward stickiness—during the FY23→FY24 cyclical downturn, inventory increased by only +1% while revenue declined by -9%. After InP wafer processing, it cannot be returned or resold (customized), making inventory impairment the only recourse. Probability-weighted impairment: ~$90M (-$0.44/share).

Anomaly 3: FCF Turns Negative, Yet Management Claims "Unprecedented." FQ2'26 FCF = -$96M (CapEx $154M, +48% QoQ). If management truly had unprecedented firm bookings, there would be no need to build $1.85B in inventory—a firm commitment implies customers will take delivery, allowing for production on demand. Extensive inventory build-up itself suggests that the "unprecedented" nature of bookings leans more towards volume guidance than firm commitment. The company has not disclosed the respective proportions of "hard constraints" (signed contracts, definitive commitments for executable delivery and payment) and "soft guidance" (intentions, production scheduling guidance, non-binding frameworks) within its orders; given reliance solely on public statements, we roughly estimate that the hard constraint portion accounts for approximately 30–45%, and the soft guidance portion for approximately 55–70%—this breakdown cannot be verified line-by-line from financial statements and is presented as a scenario assumption only.

Anomaly 4: EPS Growth Attribution Bias. FY2025 GAAP EPS -$0.52 → FY2028E $9.64, an increase of $10.16. Market narrative attributes this to AI revenue growth. However, upon disaggregation: D&A decreasing from $554M to $380M releases approximately $0.70-0.85 EPS; interest decreasing from $243M to $180-200M releases $0.22-0.32; preferred stock dividend savings of $130M releases $0.55; SBC adjustments and mix improvement ~$0.30. Non-growth factors collectively contribute $1.77-2.02, accounting for 17-20% of the total increase. This is not to say growth is unimportant, but rather that 15-20% of the "25% growth rate" implied in the 41x P/E is actually disguised by mechanical accounting effects.

2.3 Our Judgment: Mispricing from a Single Label for Three Engines

COHR's core contradiction is not "whether AI growth is sustainable" (which is LITE's core issue), but rather "whether the market is applying a single label to average three completely different curves".

Economic Characteristics Differences of the Three Engines (Scope Note: The first three rows in the table below are mutually exclusive breakdowns, summing to approximately 100%; the fourth row is a sub-breakdown of the Industrial segment (do not add to the third row again). Consistent with financial reports: "D&C ~72% / Industrial ~28%": D&C internal ≈ AI Datacom (~55–60%) + Telecom, etc. (~12–17%).):

Applying a uniform 41x P/E across these curves is equivalent to blending "high-growth AI" with "declining/investment phase" into the same multiple—which is illogical in any SOTP model. After separate valuation, the probability-weighted sum of the three engines is $226.6, 26% lower than the uniform P/E valuation.

This is the meaning of "41x deleveraging": The market, using an AI growth stock multiple of 41x, has bought a post-merger deleveraging machine. AI growth is one engine, but it is not the only engine, nor is it the sole force driving the EPS trajectory.

Chapter 3: The Nature of the Entity — What System is COHR, and Why Can't a Uniform P/E Be Used?

Main Question Advancement: If the three engines had identical growth rates/profit margins/cyclicality, a uniform P/E would be reasonable; it is precisely because they are completely different that a uniform P/E constitutes a mispricing. This chapter also answers "how much of the 41x is buying growth, and how much is buying deleveraging"—capital structure events (preferred conversion, NVIDIA investment, decreasing D&A) explain the contribution of non-growth factors to the EPS trajectory.
Implications for ROIC: Among the three engines, AI Networking is the only one expected to drive ROIC above WACC (high growth + high margin); Industrial is dragging down ROIC (low margin + contraction); SiC is consuming capital during its investment phase without generating returns. The premise for ROIC exceeding WACC is that the NOPAT growth rate of the AI engine surpasses the combined Invested Capital growth rate of the three engines.

3.1 Company Profile and One-Sentence Definition

3.1.1 One-Sentence Definition

COHR is a photonics + materials hybrid undergoing post-merger restructuring, its core economic nature is: using II-VI's materials technology base (InP/SiC/III-V compound semiconductors) to underpin three revenue engines, where the growth rates, profit margins, cyclicality, and capital intensity of these three engines are entirely different, but they share the same wafer fabrication infrastructure.

The investment implication of this definition is: the market is assigning a uniform 41x Forward P/E to COHR, but this multiple is effectively averaging three completely different curves. AI Networking, with +34% YoY growth, merits 50-60x; industrial lasers, with -10% YoY, merit 10-15x; SiC materials are in an investment phase and not yet profitable. Therefore, COHR's valuation problem is fundamentally a Sum-of-the-Parts (SOTP) issue, not a single P/E problem.

3.1.2 Why It Must First Be Disaggregated as a "Hybrid"

There are three reasons why the company must first be broken down into three business lines for separate discussion:

First, the growth logics of the three engines are unrelated. AI Networking demand stems from the Hyperscaler CapEx cycle, with the driving variable being the demand for optical interconnect bandwidth from GPU clusters. SiC materials demand arises from EV penetration rates and the power semiconductor replacement cycle, entirely unrelated to AI. Industrial laser demand comes from manufacturing CapEx and is currently in a cyclical downturn.

Second, the profit margin profiles of the three engines differ significantly. Datacom & Communications' overall Non-GAAP GM is approximately 39-42%, while the Industrial segment's GM is approximately 28-32% due to telecom decline + industrial cyclical trough, with a difference of approximately 10 percentage points in gross margin levels between the two segments. Under a simplified assumption—where 1 percentage point of total company revenue shifts towards the D&C side and correspondingly shifts out of the Industrial side (with other structures unchanged)—the consolidated gross margin would mechanically increase by approximately 0.1 percentage point, as ΔGM ≈ 10pp × 1% change in revenue share = 0.1pp. Note: The 0.1pp here corresponds to the mixing effect of D&C relative to Industrial; if the high-margin increment occurs only within D&C (e.g., AI Datacom replacing traditional telecom business within the same segment), the marginal uplift depends on the margin difference within the segment and requires separate estimation.

Third, the three engines are subject to entirely different valuation methodologies. AI Networking is subject to EV/Revenue or high-growth P/E; SiC materials are subject to option pricing (investment phase, zero profit); industrial lasers are subject to cyclical stock mid-cycle EV/EBITDA. A uniform Forward P/E confounds these three layers of pricing logic.

3.1.3 II-VI Merger: Strategic Logic and Actual Results

The strategic logic behind the merger of II-VI and legacy Coherent (total consideration approx. $6.56B), completed in July 2022, was vertical integration: II-VI possessed material-side capabilities from InP substrates to EML chips, while legacy Coherent had system integration capabilities for industrial lasers and precision optics. The theoretical advantage post-merger was "atom-to-module" full-stack control.

Positive evidence of actual results: (1) 6-inch InP wafer yield exceeding traditional 3-inch lines, a direct outcome of post-merger investment in the Sherman plant; (2) In-house R&D and manufacturing capabilities for 800G EML and 1.6T InP chips, giving COHR substrate autonomy in the AI optical module supply chain that LITE lacks.

Negative evidence of actual results: (1) The merger generated $4,463M in goodwill + $3,064M in intangible assets, totaling 49.9% of total assets; (2) Post-merger D&A was as high as $554M/yr, with a significant portion being intangible asset amortization, severely suppressing GAAP earnings; (3) Net Debt climbed from near zero leverage pre-merger to $3.67B, and remains at $2.68B today; (4) Merger integration took nearly 2 years (FY2023-2024), during which revenue fell from $5.16B to $4.71B.

Net Effect Judgment of Merger (Moderate Execution): The technical value of vertical integration is materializing (InP autonomy + 6-inch yield), but the financial costs (leverage + amortization + integration period decline) require sustained strong AI cycles to be fully absorbed. If the AI CapEx cycle slows significantly before FY2028, the $7.5B in goodwill + intangibles will face impairment risk.

3.1.4 Implications of Reorganization from 3 Segments to 2 Segments

Starting FY2026, COHR will reorganize its original three segments (Networking/Lasers/Materials) into two segments: Datacenter & Communications (D&C) and Industrial. At the same time, it divested its Aerospace & Defense business (approximately $400M) and its Munich materials processing business.

Interpretation of this reorganization:

Positive Signal: Management is shifting the narrative from a "diversified photonics company" to an "AI optical interconnect company." Carving out Datacom as a standalone primary segment indicates management's recognition of this as a core valuation driver.

Negative Signal: The two-segment reporting makes it harder for investors to disaggregate the specific contributions of AI Networking vs. Telecom vs. SiC. The D&C segment includes both AI Datacom (high growth) and traditional Telecom (in decline), blending the fastest-growing and slowest-growing areas, thereby reducing analytical transparency.

3.2 Three-Engine Hybrid: Segment Disaggregation

This chapter aims to clarify the distinct economic characteristics of each of the three engines, as the market currently applies a single P/E multiple across three vastly different curves.

3.2.1 Engine 1: AI Networking / Datacom

Revenue Scale and Growth: The Datacom & Communications segment reported revenue of approximately $1.2B in FQ2'26, up +33.6% YoY, accounting for 72% of total revenue. AI Datacom (800G/1.6T optical modules and components) is the growth engine, while the telecom portion (DWDM, access networks) is nearly flat or slightly declining. We estimate pure AI Datacom revenue to be approximately $900M-$1.0B per quarter, or $3.6-4.0B annualized, representing about 55-60% of total revenue.

Product Line Tiers:

• 800G EML: Currently the primary shipping product. COHR supplies EML (Electro-absorption Modulated Laser) chips and some complete modules. 800G modules use 4x200G lanes, each requiring one EML chip. COHR ranks #2-3 in this market share, trailing Innolight but with an advantage in in-house component R&D.

• 1.6T InP: The next-generation product, utilizing 8x200G lane InP chips. COHR's 6-inch InP wafers (Sherman factory) are a key differentiator: because a 6-inch wafer's area is four times that of a 3-inch wafer, more dies are produced per wafer, leading to lower unit costs. 1.6T is expected to enter qualification in 2026 and mass production in 2027.

• CPO (Co-Packaged Optics): Integrates optical modules directly alongside switch ASICs, reducing power consumption and latency. COHR demonstrated a 6.4T CPO solution at OFC 2026. CPO revenue timeline: scale-out interconnects starting in H2 2026, and scale-up interconnects starting in H2 2027. This revenue stream is not yet fully reflected in Wall Street consensus.

Customer Concentration: NVIDIA is one of the largest customers. In March 2026, NVIDIA invested $2B, accompanied by a "multi-billion dollar" multi-year purchasing commitment, effective from 2027-2030. Bookings extend to 2028, with the CEO citing "unprecedented business visibility." However, the non-exclusive agreement means NVIDIA is also investing in LITE ($2B) and other suppliers, so COHR does not hold an exclusive position.

Profitability Profile: The company does not separately disclose the Datacom segment's profit margin. Based on the overall Non-GAAP OPM for the D&C segment of approximately 18-22% (which includes the lower-margin telecom portion), we infer that the pure AI Datacom Non-GAAP OPM is about 22-28%. This is due to stronger pricing power for AI Datacom products (tight supply-demand, 25-30% supply gap) and the ongoing realization of economies of scale (capacity utilization climbing from 50% to 80%+).

Counter Considerations: (1) 800G module ASPs are declining; although 1.6T units command higher prices, a volume-price squeeze may emerge in FY2027-2028; (2) Hyperscaler CapEx growth (+82% in 2026E) is unsustainable, and growth is bound to decline in 2027-2028, which will directly impact the trajectory of optical module demand; (3) Innolight has a pricing advantage in 800G pluggable modules, and COHR's cost competitiveness hinges on the successful realization of 6-inch InP yield.

Cyclicality Assessment: AI Networking revenue is highly correlated with Hyperscaler CapEx. Current bookings extending to 2028 provide 2-3 years of visibility, but this is essentially visibility into a CapEx derivative, not recurring revenue. If Hyperscalers reduce AI CapEx growth in 2028 (from +80% to +10-20%), COHR's Datacom growth rate could sharply drop from over +30% to single digits.

3.2.2 Engine 2: SiC Materials / Power Electronics

Revenue Scale (Estimate): COHR does not report SiC revenue separately. The SiC business was previously included in the original Materials segment, which had revenue of approximately $950M in FY2025. SiC substrate/epitaxial wafer revenue is estimated to account for $300-400M of this, with the remainder being other III-V materials. After the reorganization, SiC is grouped into the Industrial segment, making it more challenging to track.

DENSO/Mitsubishi $1B Investment Structure: In December 2023, DENSO and Mitsubishi Electric each invested $500M, totaling $1B, acquiring a 12.5% non-controlling interest in the SiC business (6.25% each), while COHR retained 75% control. The significant implications of this structure are: (1) external strategic investors validated the standalone value of the SiC business (implied valuation of $8B, though this was a pre-money valuation in 2023 when SiC enthusiasm was higher); (2) the investment comes with long-term supply agreements, ensuring revenue predictability; (3) however, 12.5% of the profits also belong to minority shareholders and must be deducted in the consolidated financial statements.

150mm to 200mm Transition: COHR is transitioning from 150mm SiC wafers to 200mm at its Sherman, TX factory. A 200mm wafer has 1.78 times the area of a 150mm wafer, theoretically reducing unit die cost by approximately 40%. However, depreciation of the 150mm production line is not yet complete, and running both lines simultaneously during the transition period increases fixed costs.

Impact of Wolfspeed's Chapter 11: Wolfspeed filed for Chapter 11 bankruptcy protection at the end of 2025. The impact on COHR is twofold:

• Positive: Wolfspeed is the world's largest SiC substrate supplier; its constrained capacity/exit will reduce supply, enhancing COHR's relative position and pricing power in the SiC market. SiC substrate market concentration will increase.
• Negative: Wolfspeed's difficulties illustrate the extremely high capital intensity and very long return on investment cycle for SiC capacity expansion. COHR faces the same capital-intensive challenges in SiC. If EV penetration slows (signs already evident in 2025-2026), and the SiC TAM shrinks from $21B (2030) to $12-15B, COHR's SiC investment returns will be significantly delayed.

Growth Drivers and Profitability Timeline: SiC TAM is projected to grow from $3B (2022) to $21B (2030), a CAGR of ~28%. However, COHR's SiC business is currently in an investment phase: high CapEx, yield ramp-up in progress, and long customer qualification cycles (automotive-grade SiC certification requires 18-24 months). We estimate the SiC business will reach breakeven in FY2027-2028 and begin contributing positive profits in FY2029.

3.2.3 Engine 3: Industrial / Legacy (Lasers + Remaining Materials)

Revenue Scale and Growth: The Industrial segment reported revenue of $478M in FQ2'26, down -9.9% YoY. This segment includes: industrial lasers (for cutting/welding/processing), precision optics (for semiconductor lithography), and SiC materials. After divesting Aerospace & Defense ($400M sale) and the Munich materials processing business, the remaining pure industrial laser + optics business is approximately $300-350M per quarter, or $1.2-1.4B annualized.

Quantifying Drag on Overall Margin: The Industrial segment's Non-GAAP OPM is estimated to be approximately 8-12%, compared to the D&C segment's 18-22%. Taking FQ2'26 as an example, Industrial accounts for 28% of revenue but contributes approximately 15-18% of profit. If we assume a D&C segment Non-GAAP OPM of 20% and an Industrial segment Non-GAAP OPM of 10%, then for every 1% reduction in Industrial segment's revenue contribution, the overall Non-GAAP OPM improves by approximately 0.1 percentage points.

Should Further Divestitures Occur? Our assessment is "conditional" :

• For Divestiture: Industrial lasers are in a cyclical downturn (-10% YoY), dragging down overall growth and valuation multiples. If fully divested, COHR would become a pure AI optics + SiC company, giving the market reason to apply a higher P/E multiple.

• Against Divestiture: Industrial lasers contribute stable cash flow ($1.2-1.4B revenue, 8-12% OPM), and these cash flows are helping to deleverage. With Net Debt still at $2.68B, cutting off a source of cash flow is risky. Furthermore, some industrial laser technologies (e.g., precision optics for semiconductor lithography) have synergies with the AI supply chain.

3.2.4 Key Cross-Analysis

Definition Note: The diagram above separates SiC and Industrial Laser, etc. to illustrate their respective marginal contributions to the blended OPM; both belong to the Industrial segment (approx. 28%, FQ2'26) in financial reports, do not add 5–8% and ~20–23% to 28%. Note: In the D&C segment (approx. 72%), excluding AI Datacom (~55–60%), approx. 12–17% is Telecom, etc. (not shown as a separate block in this diagram).

Valuation Methodologies Applicable to the Three Engines:

Among the Three Engines: Synergy or Conflict? The relationship among the three engines is one of weak synergy + weak conflict, primarily due to shared infrastructure:

• Synergy: InP wafer manufacturing capability simultaneously serves AI Datacom (EML chips) and Telecom (DWDM lasers), allowing capacity utilization to buffer each other during demand fluctuations. Both SiC and InP require compound semiconductor epitaxy technology, leading to overlapping talent and equipment.
• Conflict: Capital allocation conflict is the biggest issue. FQ2'26 CapEx was $154M, and all three engines are competing for capital budget. AI Datacom has the highest ROI, but SiC also holds significant strategic importance, and maintenance CapEx for the industrial segment cannot be entirely eliminated. With FCF being negative (-$96M/Q), the opportunity cost of every dollar of CapEx is rising.
• Net Assessment: Unlike typical conglomerates (e.g., GE) with clear "drag sources to be cut," COHR's conglomerate issue is more about capital allocation efficiency and valuation methodology confusion, rather than business logic conflicts.

3.3 Deep Dive into Capital Structure Events

3.3.1 Deleveraging Trajectory

Net Debt decreased from $3.67B at the time of merger (FY2023) to the current $2.68B, totaling approximately $990M in deleveraging over 3 years. The deleveraging pace is approximately $330M/year, primarily from:

(1) EBITDA Growth: TTM EBITDA grew from approximately $800M in FY2023 to approximately $1,254M. Free cash flow's contribution to deleveraging was limited (FY2024 FCF $199M, FY2025 FCF $193M) because CapEx simultaneously accelerated.

(2) Asset Sales: A&D business $400M + Munich materials processing; these one-time proceeds were directly used for debt repayment.

(3) Current Leverage Level: Net Debt/EBITDA (TTM) is approximately 2.1x, below the investment-grade threshold of 3.0x. However, for a company requiring significant CapEx for expansion, leverage headroom is limited. Since FCF turned negative in FQ1'26 (-$58M) and FQ2'26 (-$96M), the path of debt repayment through cash flow has been paused.

Deleveraging Target: Management has not explicitly provided a target Net Debt/EBITDA multiple. However, considering the injection of NVIDIA's $2B investment (mostly for CapEx rather than debt repayment) and organic EBITDA growth, we anticipate Net Debt/EBITDA to decline below 1.5x by FY2027, provided that EBITDA reaches the consensus of $1.8-2.0B.

3.3.2 Preferred Stock from $2.5B to $0: Complete Breakdown

This is the most complex and easily overlooked event in COHR's capital structure.

Origin: At the time of the II-VI merger, Bain Capital, through its BCPE Watson entity, injected capital in the form of Series B convertible preferred stock, with a balance of approximately $2.5B. The preferred stock carried dividend rights and constituted a senior claim on common shareholders' equity before conversion to common stock.

November 2025: Bain Capital and COHR signed a waiver agreement, irrevocably relinquishing all future Series B preferred stock dividend rights. The implication of this action: Bain will no longer receive income from the preferred stock; its exit strategy relies entirely on the appreciation of the common stock price after conversion. Management stated this move "aligns Bain's interests with common shareholders."

December 10, 2025: BCPE Watson converted 36,162 shares of Series B-2 preferred stock into 5,000,000 shares of common stock, which were sold on the same day at $189.55 per share via a Rule 144 block trade, monetizing $948M.

December 15, 2025: The remaining Series B-2 preferred stock triggered a mandatory conversion, converting entirely into common stock. This explains why Preferred Stock on the FQ2'26 (December 2025 quarter) balance sheet plummeted from $2,505M to $0.

Impact on Common Shareholders: The conversion of preferred stock into common stock resulted in significant share dilution. After conversion, BCPE Watson held approximately 9.78M shares of common stock (representing 5.2%). The total diluted share count depends on the conversion ratio (how many common shares each preferred share converts into), but based on a conversion price of approximately $189/share for $2.5B of preferred stock, approximately 13.2M new common shares were added, diluting existing shareholders by approximately 8.5%.

Positives: Preferred stock extinguishment means (1) no more dividend outflows, with FY2025 preferred stock dividends at approximately $130M, which returns to common shareholders; (2) simplified capital structure, with no more preferred claims over common stock; (3) Bain's stake reduction will gradually alleviate overhang pressure (a block trade of 5M shares in December 2025 has already released some of this).

Negatives: Dilutive effects have already occurred. FQ2'26 diluted shares were approximately 155.5M, an increase of approximately 10M shares compared to approximately 142-145M before preferred stock conversion. Bain still holds approximately 9.78M shares (5.2%), and the subsequent sale of these shares constitutes ongoing overhang pressure.

3.3.3 Terms and Impact of NVIDIA's $2B Investment

On March 2, 2026, NVIDIA invested $2B in COHR at a price of $256.80/share. Key terms:

• Non-exclusive multi-year strategic agreement: With a "multi-billion dollar" procurement commitment, execution period 2027-2030
• Use of proceeds: Primarily for R&D + U.S. InP/SiPh capacity expansion (Sherman, TX factory expansion)
• New shares issued: $2B / $256.80 = approximately 7.79M shares, representing approximately 5.3% of shares pre-investment, approximately 5.0% dilution post-investment
• CPO Timeline: Scale-out CPO revenue starting from 2H 2026, Scale-up CPO revenue starting from 2H 2027

Conflicting signals from the investment: NVIDIA also invested $2B in LITE on the same day, totaling $4B in photonics commitments. This indicates NVIDIA is diversifying its optical supply chain, and neither COHR nor LITE are exclusive suppliers. The market's reaction that day was a 7% decline in COHR's stock price, indicating the market was more concerned about dilutive effects and the "non-exclusive" terms, rather than excited by the procurement commitment.

3.3.4 D&A Decreasing Trajectory

FY2025 D&A was $554M, a significant portion of which is amortization of intangible assets from the II-VI merger (customer relationships, developed technology, trademarks). Intangible assets of $3,064M are amortized using an accelerated method, typically fastest in the first 5-7 years, then decreasing.

The merger occurred in July 2022, and by FY2026, four years will have passed. We estimate the D&A decreasing trajectory as follows:

• FY2026E: ~$520M (D&A/Rev 7.5%)
• FY2027E: ~$450M (D&A/Rev 5.1%)
• FY2028E: ~$380M (D&A/Rev 3.6%)
• FY2029E: ~$300M (D&A/Rev 2.5%)

EPS Impact of Decreasing D&A: For every $100M reduction in D&A, after-tax EPS increases by approximately $0.50-0.55 (assuming a 20% tax rate and 155M diluted shares). Therefore, from FY2026 to FY2029, the decreasing D&A alone could contribute $0.70-0.85 to EPS growth, representing approximately 10-11% of FY2027E EPS of $7.47. This is 'accounting' EPS growth, not 'operational' EPS growth; investors need to differentiate.

3.3.5 Reduced Interest Expense

FY2025 interest expense was $243M. Primarily from:

• Term Loan B-2: SOFR+2.00%, floating rate
• Senior Notes 5.000% due 2029, fixed rate

Every $1B debt reduction from deleveraging saves approximately $50-70M in interest (depending on whether floating or fixed rate debt is repaid). If part of NVIDIA's $2B investment is used for debt repayment (though management stated it's primarily for CapEx), interest savings would further boost EPS. Our baseline assumption is that FY2027 interest expense decreases to $180-200M, a savings of $43-63M compared to FY2025's $243M, contributing approximately $0.22-0.32 to after-tax EPS.

3.4 Management and Execution

3.4.1 CEO Jim Anderson's Track Record

Jim Anderson was appointed CEO of COHR on June 3, 2024, prior to which he served as CEO of Lattice Semiconductor (September 2018 to 2024).

Lattice Performance: During Anderson's six-year tenure at Lattice, the company achieved a 10x increase in stock price. He transformed Lattice from a low-end FPGA provider to a mid-range FPGA supplier targeting communications and industrial markets, achieving record operating profit and gross margins. This demonstrates his ability to execute transformations in complex technology companies.

However, there is an important counterpoint: Lattice is a small company with $700M in revenue and a relatively focused product line. COHR is a post-merger hybrid with $6B+ in revenue, three business lines, and 40,000+ employees, with an order of magnitude higher complexity. Lattice's success cannot be linearly extrapolated to COHR.

Experience at AMD: Anderson previously served as SVP at AMD, responsible for the Computing & Graphics business group. This experience provided operational expertise in a large semiconductor company, but AMD at the time (2015-2018) was undergoing a massive transformation under Lisa Su's leadership, and Anderson's individual contribution and the overall company momentum are difficult to separate.

Appointment Terms: $48M in inducement equity awards, of which $36M was a sign-on bonus (partially compensating for equity lost upon leaving Lattice). This scale indicates COHR was serious about recruiting Anderson, but also means his compensation package is highly tied to the stock price.

3.4.2 Merger Integration Execution

The II-VI merger was completed in July 2022. Key milestones:

• FY2023 (1st year post-merger): Revenue $5.16B, declined (due to merger integration friction + industry cycle downturn)
• FY2024 (2nd year post-merger): Revenue $4.71B, continued to decline (-8.8% YoY), this was the trough of the integration period
• FY2025 (3rd year post-merger): Revenue $5.81B, recovered with growth (+23.4% YoY), exceeding pre-merger levels

Merger Synergies: The progress of achieving the synergy figures promised by management at the time of the merger (approximately $250M in annualized cost savings) is opaque. However, indirect evidence suggests execution is progressing: (1) Non-GAAP GM recovered from a FY2024 low of 30.9% to 39.0% in FQ2'26; (2) Organizational simplification from three segments to two; (3) Divestiture of A&D and Munich businesses inconsistent with core strategy.

Assessment: Anderson's execution trajectory since his appointment (June 2024) has been positive — clear business restructuring, decisive divestiture of non-core assets, and the successful landing of NVIDIA's strategic investment. However, these achievements largely benefited from the macroeconomic tailwind of the AI CapEx cycle, and the real test will come during a cycle downturn.

3.4.3 Capital Allocation Priorities

Current management faces a capital allocation trilemma:

1. CapEx acceleration: FQ2'26 CapEx $154M (annualized $616M), primarily for InP capacity expansion and SiC 200mm conversion. The arrival of NVIDIA's $2B investment will further accelerate CapEx.
2. Deleveraging: Net Debt $2.68B, but FCF has turned negative, so in the short term, deleveraging can only be achieved through EBITDA growth to reduce the leverage ratio, not through cash repayment.
3. Buybacks: Only $54M in buybacks in FY2025, and only $22M in FY2024. Buybacks at 41x Forward P/E are value destructive: $1 of buyback only acquires $0.024 of earnings ($1/41x), far below the $1 face value.

Our assessment: Management has made the correct choice — CapEx first, deleveraging second, buybacks last. Investing in capacity during a growth period is more rational than buying back shares at 41x P/E. However, if the AI cycle slows down, leading to CapEx ROI below expectations, this "correct" decision will turn into a burden of overinvestment.

3.4.4 Insider Trading Signals

Past 12 months: zero open market purchases, purely sell-on-vest mode.

Specific signals:

• 2026 Q1: 9 sells, 0 buys
• 2025 Q4: 16 sells, 0 buys
• Director Howard Xia: Exercise price $21.67, immediately sold at $236-258/share, these were low-cost options from the merger period
• Only 4 open market purchases in the past 8 quarters, all at 2024 lows (stock price $50-70)

Interpretation: This pattern is identical to LITE — management consistently sold off shares during the rally, with no meaningful additions. But it's important to distinguish: (1) most sales are sell-on-vest (immediate sale after option exercise), which is normal liquidity management and not necessarily a bearish signal; (2) zero open market purchases are a more significant negative signal, as it indicates no one is willing to use their own money to buy at the current price.

A/D Signal Strength Assessment: Zero buys + continuous sells = Neutral to slightly negative. Not as extremely bearish as LITE's A/D of 0.036, but certainly not a bullish signal. At a price of $307/share (41x P/E), management's actions express the view that "I'm not buying at this price."

3.5 What This Chapter Means for the Main Question

3.5.1 Key Findings

1. COHR's valuation issue is a Sum-of-the-Parts (SOTP) problem, not a P/E problem: A 41x Forward P/E is an average score assigned to three engines with distinctly different economic characteristics, and the explanatory power of this average is very weak.
2. AI Datacom is a valuation driver, but not the whole picture: Accounts for 55-60% of revenue (estimated), but the sustainability of growth depends on Hyperscaler CapEx, an exogenous variable.
3. Capital structure is simplifying: Preferred stock extinguishment + deleveraging + divestiture of non-core assets = structural improvement, but Bain overhang + NVIDIA dilution + negative FCF = short-term pressure.
4. Decreasing D&A is a "free" contribution to FY2027-2029 EPS growth: But this is accounting growth (Class B mechanical release), not business growth (Class A operational improvement), and needs to be differentiated in valuation. Decreasing D&A improves GAAP EPS but does not improve NOPAT, hence, it does not drive ROIC above WACC.
5. Management is doing the right things, but with significant tailwinds: Anderson's execution is verifiable (Lattice track record), but COHR's complexity far exceeds Lattice's, and current achievements largely benefit from the AI CapEx macroeconomic tailwind.

3.5.2 Implications for "When will ROIC exceed WACC?"

ROIC = NOPAT ÷ Invested Capital. For the current ROIC of 4.2% to exceed WACC of 10%, NOPAT needs to grow by approximately 2.4 times (from $1.0B to $2.4B) with no significant increase in Invested Capital. The contribution of the three engines to this goal is entirely different:

• AI Networking: The only engine capable of significantly increasing NOPAT. If Networking revenue grows from $4B to $6B and OPM expands from ~20% to 25%, NOPAT contribution will increase from $800M to $1,500M, an increment of $700M. This is the primary engine for ROIC to exceed.
• Industrial: NOPAT contribution is approximately $150M (OPM 10% × $1.5B), but it is shrinking, and its contribution is decreasing. It does not help ROIC exceed, but instead slows down the exceeding speed by dragging down the blended OPM. If Industrial were divested, Invested Capital would decrease by $3-4B, and ROIC would immediately improve—but management is unlikely to do so before deleveraging is complete.
• SiC: Current NOPAT contribution is negative (loss during investment phase). Invested Capital increases by $200-300M annually (CapEx). SiC is a pure drag on ROIC until it reaches breakeven—it increases the denominator (capital investment) and decreases the numerator (losses).

Conclusion: ROIC exceeding almost entirely depends on the profit growth rate of the AI Networking engine surpassing the combined capital consumption rate of the three engines. This means that the speed at which ROIC exceeds is entirely tied to whether Hyperscaler AI CapEx can sustainably avoid a cliff-edge decline for many years—this is **Belief B4** in the reverse DCF later (definition can be found at the beginning of §5, "Code Legend," and §5.2.5). It is not only the load-bearing wall for revenue but also a necessary condition for the ROIC exceeding narrative.

Chapter 4: Valuation Mechanism and Multiples Bridge — What Multiples Does This Moat Support?

Main Question Advancement: Moat assessment (six-dimension arithmetic mean 3.2/5) directly determines the multiples for each engine in SOTP—if 5/5 monopoly level, Networking gets 10x Rev (close to LITE); a 3.2/5 medium-to-strong moat implies 6x Rev is a reasonable upper limit. This chapter not only answers "how wide is the moat" but also "Why COHR does not deserve top-tier pure growth stock multiples, but deserves higher than Innolight".
Meaning for ROIC: Moat strength determines margin sustainability. 3.2/5 means: margin expansion is supported during periods of tight supply-demand (current), but margins will be eroded by price competition during periods of loose supply-demand (FY2028+). ROIC expansion requires sustained margin expansion, and the moat's sensitivity to supply-demand cycles is a key constraint on whether ROIC expansion can be sustained.

4.1 Six-Dimension Moat Analysis

4.1.1 Core Judgment Precedent

COHR's moat is a hybrid moat with moderate width but deepening, with an overall score of 3.2/5 (six-dimension arithmetic mean) and a trending improvement. The core characteristic of its moat is not a monopoly in a single dimension (the kind LITE has in 200G/lane EML), but rather systemic cost advantages created by deep vertical integration + customer lock-in. This means COHR's moat is highly valuable during periods of tight supply-demand (current) but will be eroded by price competition during periods of loose supply-demand (next 2-3 years).

4.1.2 Technology Barrier: 4/5 — InP Vertical Integration is the Deepest in the Industry, but Not Unique

Core Judgment: COHR possesses the most complete vertical integration chain in the optical communications industry—from InP substrates to epitaxy, to chips, to modules—which is increasing in value in the 1.6T/3.2T era, as the supply bottleneck for InP EML (industry projected 36% supply gap) grants a structural advantage to companies with in-house InP production capacity.

Quantifiable Advantages of 6-inch InP Wafers:

• Area: The area of a 6-inch wafer is 4 times that of a 3-inch wafer (π×3² vs π×1.5²), so the theoretical number of dies that can be cut from each wafer increases by 4 times
• Cost: COHR publicly claims that 6-inch InP has achieved a 60% reduction in die cost. This means that for the same EML chip, COHR's manufacturing cost is approximately 40% of competitors using 3-inch InP
• Yield Rate: COHR confirmed in its FQ2'26 earnings report that the yield rate of its 6-inch production line has exceeded that of traditional 3-inch lines. This is counter-intuitive—typically, larger wafer sizes have lower initial yield rates than smaller ones—indicating that the process maturity at the Sherman factory has crossed the inflection point of the learning curve
• Capacity Conversion Timeline: COHR plans to migrate most InP production from 3-inch to 6-inch over "the next few years," not yet fully switched, meaning the cost advantage is still gradually being realized

Comparison with LITE's EML Technology:

LITE in 200G/lane EML has a short-term technological monopoly, with a 50-60% global market share in high-end laser chips. COHR ranks #2-3 at the EML chip level. Key differences:

Therefore, LITE possessed a technological advantage at the EML chip level in the 800G era; however, in the 1.6T era (8×200G/lane), COHR's 6-inch InP cost advantage began to take effect; in the 3.2T+ era, both companies are transitioning from InP EML to hybrid SiPh+InP solutions, reshaping the competitive landscape.

SiC Technology Barrier: COHR is transitioning from 150mm to 200mm in the SiC domain. 200mm wafers have 1.78 times the area of 150mm wafers, theoretically leading to a roughly 40% reduction in unit die cost. Although Wolfspeed plans to file for Ch.11 by the end of 2025, it has announced the world's first 300mm SiC wafer (January 2026), technically remaining a leader. onsemi has successfully ramped 200mm SiC in Bucheon, South Korea, increasing the number of chips per wafer by approximately 80%. COHR ranks in the middle tier for SiC technology—not a leader, but with funding and demand assurance provided by DENSO/Mitsubishi investments.

CPO/SiPh Strategy: COHR demonstrated a 6.4T CPO solution at OFC 2026 and is collaborating with Tower Semiconductor to develop 400Gbps/lane silicon modulators for 3.2T modules on the SiPh roadmap. This means COHR simultaneously covers both InP (current) and SiPh (future) technology routes, reducing the risk of reliance on a single technology.

Counterpoints: (1) The 60% cost advantage of 6-inch InP is COHR's own claim, and actual competition also needs to consider module-level costs (Innolight has a labor cost advantage in module assembly); (2) If SiPh completely replaces InP EML in the 3.2T era, COHR's InP capacity investment in Sherman will face overcapacity risk; (3) LITE is also developing its own InP manufacturing capabilities, narrowing the technological gap.

Rating Basis: 4/5, because 6-inch InP offers a globally unique and proven cost advantage, but it still lags LITE in EML chip performance and faces long-term SiPh replacement risk.

4.1.3 Switching Costs: 3/5 — Customer Lock-in Exists but is Not Exclusive

Core Judgment: The qualification cycle for optical modules/components is 6-12 months, which constitutes a medium-strength switching barrier. NVIDIA's $2B investment + multi-year procurement commitment strengthens the lock-in, but non-exclusive terms limit the ceiling of stickiness.

Economic Implications of the Qualification Cycle:

• Time: Qualification for 800G modules typically takes 6-9 months; for 1.6T, as a new speed grade, the certification cycle extends to 9-12 months
• Cost: Includes sample testing, interoperability verification, high-temperature/high-humidity/vibration reliability tests, etc. For hyperscalers, the direct engineering cost for each qualification is approximately $0.5-2M, but indirect costs (delayed deployment plans) are significantly higher than this
• Outcome: Once qualified, customers tend to maintain the same supplier within the same speed generation, as switching means repeating the entire process, which is especially uneconomical in an environment of tight EML supply (36% gap)

True Stickiness of NVIDIA's $2B Lock-in:

• Positive: The procurement commitment provides 2-4 years of revenue visibility, and the CEO claims bookings extend to 2028. The investment aligns interests—NVIDIA, as a shareholder (approx. 5% stake), has an incentive to maintain the supply relationship
• Limitation: Non-exclusive—NVIDIA invested $2B in LITE on the same day, and a diversified procurement strategy means COHR does not enjoy an exclusive position. NVIDIA's procurement share allocation depends on technological progress and price competition, not a fixed proportion
• Causal Reasoning: NVIDIA's dual/triple sourcing is a rational behavior—optical modules are critical components for GPU clusters, and a single supplier poses too high a risk. Therefore, COHR's NVIDIA lock-in is more akin to a "baseline share guarantee" rather than an "exclusive supply lock-in"

SiC Customer Lock-in: DENSO and Mitsubishi Electric each invested $500M (totaling $1B) to acquire a 12.5% non-controlling interest. The certification cycle for automotive-grade SiC devices is 18-24 months and involves functional safety (ISO 26262) certification, making switching costs much higher than for optical modules. This makes the customer lock-in for the SiC business stronger than that for the AI Networking business.

Counterarguments: (1) Innolight and Eoptolink have secured 60% of NVIDIA's 800G SFP module orders, indicating that NVIDIA's procurement will not favor COHR/LITE simply due to investments; (2) The qualification window reopens in the 1.6T era, putting all suppliers back on an equal footing; (3) While SiC customer lock-in is strong, SiC revenue accounts for only 5-8%, contributing limitedly to overall switching costs.

Rating Basis: 3/5, qualification cycle provides a moderate barrier, NVIDIA's investment is supportive but not exclusive. SiC lock-in is strong but revenue contribution is small.

4.1.4 Economies of Scale: 3/5 — Large Scale but Not Fully Translated into Cost Leadership

Core Judgment: COHR's annual revenue of $6.7B is 2.5 times that of LITE, but the larger revenue scale comes from diversified businesses (Industrial/Materials), not from a leading single market share in optical communications. In AI Datacom, the most critical sub-market, COHR's scale advantage is limited.

Breakdown of Scale Components:

In AI Datacom, a valuation-determining segment, COHR's revenue scale is only 30-50% larger than LITE's, not 2.5 times. This suggests that the moats from economies of scale are weaker than what surface-level figures imply.

Scale Effects of Manufacturing Footprint:

• Sherman, TX: InP/SiC wafer manufacturing, location of 6-inch InP production line, and target for 200mm SiC expansion
• Ipoh, Malaysia: Optical module assembly, offering tariff immunity advantage (non-China manufacturing)
• Chambersburg, PA: Legacy II-VI compound semiconductor factory

Multi-factory operations provide supply chain resilience, and Malaysian manufacturing offers a differentiating advantage in the current tariff environment. However, multiple factories also mean higher fixed costs: COHR's D&A of $554M/yr (9.5% of Revenue) is significantly higher than the industry average, with a large portion attributable to manufacturing infrastructure depreciation.

The Paradox of Vertical Integration Scale: Vertical integration (from substrate to module) theoretically should reduce costs by eliminating external markups. In practice, however, vertical integration also means bearing the fixed costs and technological risks of the entire supply chain. Innolight chose to outsource InP chips and assemble modules itself, leveraging China's labor cost advantage to achieve lower module-level costs and capture the largest share in the 800G pluggable market (including Eoptolink, totaling approximately 60% of NVIDIA's orders). Therefore, COHR's vertical integration scale advantage is primarily reflected at the component (chip) level, not at the module (end-product) level.

Counterarguments: (1) Large scale equals high fixed costs; if the AI CapEx cycle slows down, declining capacity utilization would impact COHR more severely than asset-light Innolight; (2) While the Industrial segment's $1.2-1.4B revenue contributes to scale, its low-profit margins (OPM 8-12%) drag down overall return on investment.

Rating Basis: 3/5, overall large scale but limited scale advantage in the core AI Datacom market, and the scale effects of vertical integration are partially offset by high fixed costs.

4.1.5 Network Effects/Standard Participation: 2/5 — Industry Standard Participation but No Lock-in Effect

Core Judgment: Standards in the optical communications industry (MSA/QSFP/OSFP) are open standards, and participation in standard setting does not constitute a moat. COHR's competitive advantage comes from technology and manufacturing, not standard lock-in.

COHR participates in industry standard organizations such as OIF (Optical Internetworking Forum) and MSA (Multi-Source Agreement), holding influence in defining MSA pluggable module specifications. However, the original intent of MSA standards is to ensure multi-vendor interoperability, thus participation in standard setting reduces rather than increases vendor lock-in.

The only standard-related advantage: COHR's co-design relationship with NVIDIA in the CPO domain—CPO, unlike pluggable modules, does not have mature MSA standards, and early CPO deployments rely more on customized collaboration with ASIC vendors (Broadcom/NVIDIA). If COHR's CPO solution becomes the default option for NVIDIA's next-generation platform, it would create stronger lock-in than pluggable modules. However, CPO revenue is not expected to scale until 2027, so this moat has yet to materialize.

Rating Basis: 2/5, open standard industry, no network effects, CPO co-design relationship is a future option.

4.1.6 Brand/Reputation: 3/5 — Strong II-VI Materials Reputation, Brand Integration Ongoing Post-Merger

Core Judgment: II-VI has accumulated 30+ years of technical reputation in the InP/III-V compound semiconductor field, which is valuable among materials customers (DENSO/Mitsubishi for SiC) and optical communications customers. However, the merged "Coherent Corp" brand has only been in operation for 3 years (since July 2022), and brand integration is still ongoing.

Path for Reputation to Translate into Economic Value: In the fields of semiconductor materials and photonics, the core value of brand reputation is quality trust—when customers select suppliers, trust in long-term reliability and consistency constitutes implicit switching costs. II-VI's 30-year reputation in the InP materials supply chain gives COHR a trust advantage when pursuing new customer qualifications.

Counterarguments: (1) The "Coherent" brand name can easily be confused with the old Coherent company (lasers), and the new Coherent's AI optical communications identity has not yet been fully recognized by the market; (2) Brand reputation has limited value in the price-sensitive 800G pluggable market—Innolight won 60% of NVIDIA's orders with lower prices, indicating brand is not the decisive factor.

Rating Basis: 3/5, strong reputation in materials, but brand is not a primary competitive variable in the optical module market.

4.1.7 Barriers to Entry: 4/5 — InP Manufacturing Requires a Decade of Accumulation, New Entrants are Virtually Non-Existent

Core Judgment: Establishing a competitive InP EML production line requires 5-10 years and $1B+ in investment, making this COHR's most certain moat dimension. The barriers to entry for 200mm SiC production lines are similarly high; Wolfspeed's Chapter 11 filing demonstrated the financial pressure from capital intensity.

Quantifying InP Barriers to Entry:

• Time: Building InP EML manufacturing capability from scratch, even with a technical team, requires 5-7 years to achieve mass production yields. COHR (including II-VI) has accumulated 20+ years of experience in InP.
• Capital: Investment for a 6-inch InP production line is estimated at $500M-$1B (including equipment, cleanrooms, process development); ongoing expansion investments for the Sherman factory are partly funded by NVIDIA's $2B.
• Talent: InP epitaxial growth and chip fabrication require highly specialized engineering teams, and the global talent pool with this experience is extremely small (concentrated in a few companies in the US, Japan, and Europe).

Pace of Catch-up by Chinese Manufacturers:

Accelink and Innolight are already highly competitive at the optical module assembly level, but there is still a gap in InP chip self-development. Innolight's strategy is to procure InP chips (from COHR/LITE/Mitsubishi/Sumitomo, etc.) and perform module packaging itself, competing at the module level with scale and cost advantages.

This means the threat from Chinese manufacturers is primarily at the module level (competing with COHR's module business), rather than at the chip level (where COHR's core technological barrier lies). However, if the Chinese government promotes InP chip self-sufficiency (similar to the SiC path), this barrier could face erosion risk in 5-10 years.

SiC 200mm Barriers to Entry: 2026 marks a watershed year for the SiC industry, transitioning from capacity expansion to cost efficiency. STMicro is building a vertically integrated 200mm SiC factory in Catania, Infineon's Module 3 in Kulim, Malaysia, has begun to ramp, and onsemi has successfully ramped 200mm in Bucheon, South Korea. While there are many entrants, each requires $2-5B in investment, and Wolfspeed's bankruptcy demonstrates that even industry pioneers cannot withstand the capital pressure of expansion. For COHR, the SiC barrier to entry is not "competitors won't come" but rather "competitors will come but also face immense capital pressure."

Counterarguments: (1) Broadcom bypasses InP barriers via the SiPh route, capable of manufacturing optical engines without InP manufacturing capabilities; (2) If SiPh becomes mainstream in the CPO era, the InP barrier to entry would become "a barrier to entry into a shrinking market."

Rating Basis: 4/5, InP and SiC barriers to entry are both high, but the rise of the SiPh route means new entrants may not necessarily need to overcome the InP barrier.

4.1.8 Overall Moat Assessment

Comparison with LITE:

Moat Quality Conclusion: COHR's moat is "system-level" rather than "single-point"—no single dimension has the same monopoly as LITE in EML, but the superposition of moderate advantages across multiple dimensions forms a broader, comprehensive barrier. This moat's effect is amplified during periods of tight supply and demand (current EML shortage of 36%) and its effect diminishes during periods of relaxed supply and demand (FY2028+).

4.1.9 Multiple Bridge: What Valuation Multiple Does a 3.2/5 Moat Support?

The ultimate goal of moat evaluation is not to assign a score, but to answer: What multiple does this moat strength support in a SOTP (Sum-of-the-Parts) valuation?

COHR is assigned a 6x EV/Rev multiple for AI Networking in SOTP (Base case), and this multiple precisely corresponds to a 3.2/5 moat assessment: higher than Innolight's 3-5x (due to chip-layer cost advantage + InP autonomy + geographical diversification), and lower than LITE's implied 8-15x (due to no 200G/lane EML performance exclusivity, 21pp lower AI pure-play exposure, and 43pp lower growth rate).

If the moat quality declines from 3.2 to 3.0 in FY2028+ due to easing supply and demand, and the multiple drops from 6x to 4-5x, the Networking valuation alone would decrease by $5-7B, with a per share impact of -$29 to -$40. This is the direct transmission of the moat's cyclical sensitivity to SOTP—moats are not static, and neither should multiples be.

4.2 Competitive Landscape Depth

4.2.1 800G/1.6T Optical Module Competition: Three Layers, Three Battlegrounds

Optical module competition is not one-dimensional; it needs to be broken down into three layers:

Layer 1: InP Chip Competition — COHR Ranks #2-3, Catching Up to LITE

The global EML chip market is dominated by five suppliers: LITE, COHR, Broadcom, Mitsubishi, and Sumitomo. LITE holds a first-mover advantage in 200G/lane EML and approximately 50-60% market share, and is considered the "gold standard" for the 1.6T era.

COHR's competitive strategy at the EML chip level is to leverage its 6-inch InP cost advantage to gain market share: A 60% reduction in die cost means that even if performance metrics (bandwidth, temperature stability) are close to but do not surpass LITE's, COHR can win over price-sensitive customers with its pricing.

Key Judgment: In the 800G era, LITE's EML technological leadership is clear. In the 1.6T era (2026-2027), the key competitive variable shifts from "who has better EML performance" to "who can mass-produce at a lower cost," because 1.6T requires eight 200G/lane EMLs (twice that of 800G), increasing the proportion of chip cost in the module's Bill of Materials (BOM). This favors COHR's cost advantage.

Layer 2: Module Assembly Competition — Chinese Manufacturers Dominate, COHR Ranks #3

In NVIDIA's 800G SFP module supply chain, Innolight + Eoptolink together capture approximately 60% of the market share, with the remaining 40% shared by US-based manufacturers such as COHR, LITE, and Broadcom. Innolight's competitive advantages are:

• Cost: Labor and operational cost advantages of Chinese manufacturing; even with externally sourced InP chips, module-level costs remain lower than those of US-based manufacturers
• Speed: Short cycle time from sample to mass production; has established a first-mover advantage in 800G LPO (Linear Pluggable Optics)
• Scale: Shipped over 500,000 400G modules in H1 2024; 800G capacity continues to expand

COHR's differentiation at the module level: (1) Its Ipoh, Malaysia factory provides non-Chinese manufacturing supply chain security, appealing to Western hyperscalers; (2) Vertical integration allows COHR to use self-produced InP chips in its modules, enhancing supply chain autonomy; (3) However, its cost competitiveness remains weaker than Innolight's.

Market Share Evolution Forecast (Medium Confidence in Execution): Innolight's leading market share position in the 800G era may not persist into the 1.6T era. This is because the supply bottleneck for EML chips in 1.6T modules (a 36% shortfall) will limit Innolight's module output—Innolight relies on externally sourced EML chips, and if LITE/COHR prioritize supplying their own modules, Innolight's 1.6T module shipments will be constrained.

800G ASP Trend:

800G module ASPs are declining, which is a typical pattern in speed upgrade cycles. The industry expects 800G ASPs to fall by 20-30% in 2026 compared to 2025, further decreasing to near 400G levels by 2027. Initial 1.6T ASPs are approximately 1.8-2.2 times that of 800G, but will also decline rapidly as mass production scales up.

Investment Implications of Volume-Price Dynamics: The volume growth masking price declines pattern in the optical module market (unit shipments +60%, ASP -30%, revenue growth +12%) means that merely looking at revenue growth would overestimate market health. When shipment growth decelerates (2028+), the negative impact of declining ASPs will become evident.

4.2.2 CPO Competition (2027+): Broadcom is the Biggest Threat

Basic Economics of CPO:

The core advantage of CPO (Co-Packaged Optics—packaging optical engines directly next to switch ASICs) is power consumption. Broadcom claims CPO achieves approximately 5.5W per 800Gb/s port, whereas equivalent pluggable modules consume about 15W, representing a power reduction of approximately 3 times. On a 64-port (800G per port) switch, this translates to saving hundreds of watts, which is immensely valuable for power-constrained AI data centers.

Broadcom's CPO Strategy:

Broadcom is the biggest proponent of CPO, with its Bailly CPO platform adopting an open ecosystem approach. The Tomahawk 6 "Davisson" 102.4 Tb/s switch co-packages 16 6.4 Tb/s optical engines, utilizing TSMC's COUPE (Compact Universal Photonic Engine) photonic engine.

The threat of Broadcom's CPO strategy to COHR lies in Broadcom's use of SiPh (Silicon Photonics) instead of InP as the optical engine foundation. If SiPh CPO becomes the mainstream solution for data center interconnects, the importance of InP will decline—InP will still be needed as a light source (because silicon cannot emit light efficiently), but InP's value share in the CPO architecture will be lower than in pluggable modules.

COHR's Competitive Position in CPO:

COHR showcased its 6.4T CPO solution at OFC 2026, covering both InP and SiPh routes. COHR's CPO strategy is a "two-pronged approach": (1) providing InP-based CPO optical engines to customers like NVIDIA (leveraging existing InP manufacturing advantages); (2) developing SiPh solutions through collaboration with Tower Semiconductor (hedging against technology route risk).

Large-scale commercial deployment of CPO is projected for 2028-2030 (Yole Group estimate), rather than 2026-2027. COHR's scale-out CPO revenue begins in H2 2026, and scale-up in H2 2027, but initial revenue will be small and not yet fully reflected in Wall Street consensus.

Key Judgment: CPO will not "kill" pluggable; both forms will coexist long-term—CPO for high-density interconnects within switches, pluggable for long-distance transmission between data centers. COHR's simultaneous pursuit of both forms is the correct strategy. However, if Broadcom's SiPh CPO becomes the dominant standard, COHR needs to ensure its SiPh capabilities keep pace, otherwise its market share in the CPO era will be constrained.

4.2.3 SiC Competitive Landscape: Wolfspeed Falls, But Many Alternatives

2022 SiC Power Semiconductor Market Share:

COHR's position in the SiC market is as a substrate and epitaxial wafer supplier, not a SiC power device manufacturer. Therefore, COHR is not in direct competition with STMicro/Infineon/onsemi, but rather is an upstream supplier. Wolfspeed is COHR's direct competitor in the SiC substrate market.

Quantifying Wolfspeed's Impact:

Wolfspeed filed for bankruptcy protection at the end of 2025, but its Mohawk Valley 200mm factory remains operational, and in January 2026, it announced the world's first 300mm SiC wafer. Wolfspeed's bankruptcy was not due to technical failure, but because a $9B+ debt burden overwhelmed its balance sheet—expansion capital requirements far exceeded cash flow.

Impact on COHR:

• Short-term positive: Wolfspeed's capacity constraints/declining customer confidence, with some SiC substrate demand shifting to COHR, enhancing COHR's relative position in the SiC market
• Medium-term uncertainty: If Wolfspeed successfully streamlines through restructuring (reducing $5B+ debt), its technological advantage (300mm SiC) will still lead COHR, and competitive pressure will not disappear
• Long-term lesson: Wolfspeed's failure model is a warning to COHR—SiC expansion is extremely capital-intensive, and COHR's investment in SiC (200mm conversion) also requires significant CapEx; if EV penetration slows, it faces a similar risk of delayed return on investment.

200mm SiC Competitive Progress:

Key Judgment: COHR is a competitive substrate supplier in the SiC market, but not a market leader. The DENSO/Mitsubishi investment provides $1B in funding and long-term demand lock-in, which differentiates COHR from purely commercial competitors. However, the competitive landscape of the SiC market is rapidly changing—2026 is the year when multiple manufacturers simultaneously ramp 200mm production, and cost efficiency and yield will become decisive factors.

4.2.4 Key Competitive Judgment: Where is the True Advantage, Where is it Just a Participant?

Competitive Landscape Summary:

COHR has a true competitive advantage at the component layer (InP chips), is a follower at the module layer (pluggable), and is an early but unproven participant at the system layer (CPO). The implication of this stratification for valuation is: if the value in the optical module industry shifts upstream to the component layer (e.g., EML supply shortage in the 1.6T era), COHR benefits; if the value remains at the module layer (where Innolight locks in customers with low costs), COHR's component advantages cannot be fully realized.

4.3 Technology Roadmap and Risks

4.3.1 InP vs SiPh: Not Substitution, but Convergence

Core Judgment: InP and SiPh are not an "A replaces B" relationship, but rather an "A and B converge at different levels" relationship. Because silicon cannot emit light efficiently (indirect bandgap semiconductor), even the most advanced SiPh solutions require InP as the light source. The question is not whether InP will be replaced, but rather whether InP's value share in the optical module BOM will shrink.

Technical Physics Constraints:

• Irreplaceability of InP: InP (Indium Phosphide) and GaAs (Gallium Arsenide) are direct bandgap semiconductors, capable of efficient light emission and detection. Silicon is an indirect bandgap semiconductor and cannot be used for lasers or efficient photodetectors. Therefore, all SiPh solutions require "hybrid integration" to combine InP/GaAs light sources with silicon photonics circuitry.
• Advantageous Scenarios for SiPh: SiPh offers cost advantages in modulation and routing functions (utilizing mature CMOS processes), suitable for high-density, low-power CPO applications. COHR has partnered with Tower Semiconductor to achieve 400Gbps/lane silicon modulators.
• Advantageous Scenarios for InP: For long-distance (>2km), high-power, and high-temperature environments, InP EML remains the optimal choice. The mainstream technical solution for 1.6T pluggable is 8×200G/lane InP EML.

Technical Choices for Different Rate Generations:

4.3.2 1.6T Competitive Time Window

1.6T is a critical window for COHR to catch up with LITE. Reasons:

First, the doubling of EMLs amplifies cost advantages: 800G requires 4 EMLs, while 1.6T requires 8. The cost share of EMLs in the module BOM increases from approximately 30% for 800G to over 40% for 1.6T. COHR's 60% reduction in 6-inch InP die cost translates to absolute savings in the 1.6T era that are double those of 800G.

Second, supply bottlenecks reshuffle the deck: The industry projects an EML supply gap of 36%, limiting Innolight's model of externally sourcing EMLs. COHR's self-sufficiency in EML supply becomes a greater competitive advantage in the 1.6T era.

Third, the qualification window reopens: 1.6T is a new rate generation, requiring all suppliers to re-enter the hyperscaler qualification process. LITE's first-mover advantage for 800G cannot be directly transferred to 1.6T.

Counterpoint: (1) LITE still leads in performance metrics (bandwidth, signal-to-noise ratio, temperature range) for 200G/lane EMLs; if hyperscalers prioritize performance over price, COHR's cost advantage may not necessarily translate into market share; (2) Goldman Sachs projects the "main ramp-up phase" for 1.6T to be in 2026, and COHR needs to pass qualification by FY2027 to capture this window.

4.3.3 CPO vs Pluggable: Coexistence, Not Replacement

Industry Consensus (Yole Group/IDTechEx): Large-scale commercial deployment of CPO is expected in 2028-2030, not 2026-2027. In the current phase (2026-2027), CPO's revenue contribution is small – COHR's scale-out CPO will launch starting in 2026H2, and scale-up starting in 2027H2, but initial scale will be limited.

Logic for CPO and Pluggable Coexistence: CPO is suitable for within switches (short-distance <100m, high-density, power-sensitive), while pluggable is suitable for inter-data center applications (long-distance >100m, maintainability requirements). Therefore, CPO will not replace pluggable but rather expand the total optical interconnect market.

Pros and Cons of COHR's Two-pronged Approach:

• Pros: Possessing both InP (pluggable) and SiPh (CPO) technology routes allows COHR to participate regardless of which route becomes mainstream. This technological hedge is a unique capability of vertically integrated companies.
• Cons: Both routes require significant R&D and CapEx investments, diluting resources. Broadcom's investments in SiPh/CPO are more focused, potentially leading to deeper technical advantages in the CPO era.

4.3.4 Maximum Technical Risk: Broadcom SiPh + CPO Disruption

Risk Description: If Broadcom's SiPh CPO platform (Bailly + TSMC COUPE) becomes the default interconnect standard for AI data centers, the following consequences would be unfavorable for COHR:

1. Shrinking InP Value Share: In a CPO architecture, InP only provides the light source (laser), while modulation/routing/detection are all performed by SiPh. InP's value share in the module BOM decreases from 30-40% in pluggable modules to 10-15% in CPO.
2. Weakened Vertical Integration Advantage: COHR's vertical integration is built around the InP value chain (substrate → epitaxy → chip → module). If InP's value share shrinks, the economic return of this vertical integration chain decreases.
3. CapEx Becomes Sunk Cost: The investment in expanding 6-inch InP production at the Sherman factory (partially funded by NVIDIA's $2B) is based on the assumption of InP's sustained high value. If SiPh CPO dominates the market in 2028-2030, these InP capacities will face the risk of underutilization.

Risk Probability Assessment (Medium Execution Difficulty): The probability of Broadcom SiPh CPO completely disrupting InP before 2030 is low (15-20%) because:

• Historical Baseline: Technology replacement in the optical communication industry typically requires 2-3 rate generations (10-15 years). Moving from 800G (InP-dominant) to CPO-dominant will require at least two generations, 1.6T and 3.2T.
• Current Evidence: Even Broadcom's SiPh CPO requires an InP light source. Solutions completely bypassing InP (e.g., silicon light sources) are not commercially viable before 2030.
• Natural Experiment: In 2026, NVIDIA simultaneously invested $2B in COHR (InP route) and LITE (InP route). If NVIDIA believed SiPh was about to replace InP, it would not have made such investments.

However: Even if InP is not completely replaced, the shift of InP's value share towards SiPh is a definite trend. COHR's countermeasures (collaboration with Tower on SiPh, in-house CPO development) are correct, but it needs to elevate its SiPh capabilities from "demonstration-level" to "production-level" before 2027-2028.

4.3.5 Technical Roadmap Risk Summary

The greatest technical uncertainty is not whether "InP will be replaced" (it won't, at least not before 2030), but rather "whether InP's value share will shrink from 40% to 15%" (a strong trend). If the latter occurs, COHR's vertical integration will shift from "full-stack value capture" to "light source value capture only," necessitating a re-evaluation of its valuation logic.

Chapter 5: Key Constraints — What Is Preventing ROIC from Exceeding?

Advancing the Main Problem: Three major constraints—①ROIC<WACC (value destroyed daily) ②Bookings quality black box + inventory hard constraint (bear case speed unknown) ③Networking is essentially a Hyperscaler CapEx derivative (exogenous variable uncontrollable). Other risks (Industrial drag, SiC delays, dilution, technological substitution) are secondary constraints, impacting through these three primary constraints.
Meaning for ROIC: R1 (AI CapEx Downturn) directly reduces the NOPAT numerator, **R3 (Deleveraging Stall)** hinders the Invested Capital denominator from shrinking; when both act simultaneously, ROIC may decline instead of increasing. **B4 (the market's implied belief that "CapEx won't collapse for years")** is the external condition most often relied upon in the ROIC crossover narrative, and historical experience shows a base rate of approximately 75% for "significant downward revisions within three years after high growth" (see §5.2.5).

5.1 Risk Topology

5.1.1 Core Judgments Preceded

The risk structure for COHR is not an independent checklist, but a **mutually amplifying system**. The most dangerous aspect is not any single risk, but the multiplier effect generated when **R1 (AI CapEx Downturn)** and **R2 (Valuation Label Collapse)** are triggered simultaneously: when Networking growth slows by -20%, the P/E will not drop proportionally from 41x to 33x. Instead, it will fall to 20-25x because the market label collapses from "AI growth stock" to "post-merger cyclical stock." With both layers combined, the stock price downside could reach -40% to -55%.

5.1.2 AI CapEx Cycle Downturn (Risk R1, Maximum Risk)

Core Judgment: This is COHR's number one risk because Networking/Datacom accounts for 72% of revenue, and its +34% YoY growth is the core pillar of valuation. Once Hyperscaler CapEx growth decelerates from the current +82% YoY to +10-20%, COHR's Networking revenue growth will plummet from over +30% to single digits, and the P/E multiple will non-linearly contract due to **R2 (Label Collapse)**.

Unsustainability of Current Hyperscaler CapEx:

The combined CapEx of the four major Hyperscalers (MSFT/GOOGL/AMZN/META) in 2026 is approximately $690B, a YoY increase of +82%. This growth rate is mathematically unsustainable—if it were to maintain +80% for another year, CapEx in 2027 would reach $1.24 trillion, and its proportion to the combined revenue of the four companies would rise from the current 35% to over 50%. Alphabet has already hinted in Q3 2025 that CapEx growth will "gradually return to normalization," and FCF once declined by 90% YoY.

Therefore, the question is not "whether" the CapEx cycle will slow down, but "when" and "how quickly."

Historical Base Rate (One of Three Anchors):

The optical communications industry has experienced three significant CapEx-driven downturn cycles:

In the three cycles, the average revenue decline for optical communications companies was approximately -15 to -30%, and the stock price decline was about -35 to -55%. Based on 3 out of 3 occurrences, the base rate is 100% (CapEx cycles will eventually slow down). The question is the timing and intensity.

Counter-Example Conditions (One of Three Anchors):

Historically, conditions where CapEx cycle downturns had less-than-average impact on optical communications include: (1) Simultaneous ramp of new rate generations (e.g., the 400G ramp in 2020 partially buffered demand decline); (2) User expansion from telecom to data centers (diversified demand sources). Currently: 1.6T is ramping in 2026-2027, and CPO will start contributing incremental growth from 2027. These two conditions are partially met, meaning the intensity of the next adjustment might be lower than 2001 but higher than 2023.

Natural Experiment (One of Three Anchors):

The April 2025 tariff impact provided a small stress test. Optical communications stocks fell 13-40% within a week (COHR -40%, LITE -13%, Lumentum -35%). This demonstrates: (1) COHR's Beta is significantly higher than LITE's (because the Industrial segment in its hybrid structure is tariff-sensitive); (2) Under pressure, the market tends to sell hybrid stocks first, then pure AI plays.

Quantification of Transmission Mechanism:

• Networking Revenue Impact: Drops from +30% YoY to +5-10% YoY, as bookings provide some buffer until 2028
• OPM Impact: Capacity utilization drops from 80%+ to 60-70%, GM contracts by 2-4pp, OPM contracts by 3-5pp
• EPS Impact: FY2028 EPS drops from consensus $9.64 to $6.50-7.50 (vs current buy-in at $9.64)
• Valuation Impact: P/E contracts from 41x to 25-30x (if AI label is maintained) or 20-25x (if label collapses to cyclical stock)
• Combined Stock Price Impact: 25x × $7 = $175 (-43% from $307.50) to 20x × $6.5 = $130 (-58%)

Probability Assignment: 25-35% for a significant adjustment to occur within the next 18 months (based on a 100% base rate that it will eventually happen, but timing is uncertain; 1.6T ramp provides a 1-2 year buffer).

Downside: CEO claims bookings extend to 2028. If bookings are contractually obligated (take-or-pay), revenue visibility indeed spans 2-3 years. However, we don't know how much of these bookings are firm commitments vs. soft indications (black box).

5.1.3 Valuation Gravity / Label Collapse (Risk R2)

Core Assessment: A 41x forward P/E implies an assumption of 25%+ EPS CAGR sustained for over 3 years. This multiple is valid only if the market categorizes COHR in the "AI growth stock" valuation bucket (average P/E within the bucket is 35-60x). Once networking growth falls below 15%, COHR will be reclassified as a "post-merger hybrid," with an applicable P/E decreasing from 40x to 18-25x. Label collapse is a swifter stock price killer than earnings decline.

Implied Assumption Breakdown:

Current share price of $307.50 = 41.2x × FY2027E EPS $7.47. PEG Anchoring:

• COHR: 41x / 25% growth (consensus FY2026-2028 EPS CAGR) = PEG 1.64x
• LITE: 47x / 40%+ growth = PEG 1.18x
• Lumentum: ~30x / 20% growth = PEG 1.50x

COHR's PEG is the highest among the three, implying the market assigns the highest premium per unit of growth to COHR. This includes: (a) accelerated EPS expectations from deleveraging; (b) SiC option value; (c) passive fund inflow premium after S&P 500 inclusion.

Quality Premium and Margin of Safety:

Even if COHR meets consensus expectations (FY2028 EPS $9.64), investors buying at the current share price would still face a P/E of 32x two years from now. For an investment return of >10%/yr, the P/E would need to be maintained at ≥29x two years later, which is difficult when EPS growth slows to below 15%—historically, industrial technology stocks with 15% EPS growth have a median P/E of approximately 22-28x. Therefore, even if performance meets targets, the current valuation has already consumed most of the margin of safety.

If Networking Growth Slows to 15%:

• P/E shifts from the AI bucket (35-60x) to the hybrid industrial bucket (18-28x)
• Neutral scenario: 25x × $8 EPS = $200 (-35%)
• Bear case: 20x × $7 EPS = $140 (-54%)

Probability Assignment: 30-40% probability of P/E compression to below 30x within 18 months (Based on historical benchmarks: P/E compression after growth slowdown in the optical communications sector typically takes 2-3 quarters to shift from the high-growth bucket to the mid-growth bucket; Counter-example condition: If SiC accelerates profit contribution during the same period, it could partially offset the growth slowdown).

5.1.4 Deleveraging Stall (Risk R3, CapEx competing for cash flow)

Core Assessment: COHR's deleveraging narrative (Net Debt from $3.67B → $2.68B) is one of the important pillars of its valuation, but accelerating CapEx (from ~$95M/Q to $154M/Q) has already caused FCF to turn negative (-$96M in FQ2'26), threatening the pace of deleveraging.

Structural Analysis of Negative FCF:

OCF in FQ1'26 and FQ2'26 was unusually low ($46M/$58M vs. previous ~$180-190M), requiring verification for working capital changes (inventory +$215M QoQ) or one-off factors. If OCF returns to $200M/Q, but CapEx remains at $150M/Q, FCF would only be +$50M/Q, or $200M annualized—it would take 13 years to repay the $2.68B Net Debt.

Three Possible Deleveraging Paths:

1. NVIDIA $2B Funding: If a portion of NVIDIA's investment flows in as an upfront payment/capital injection, Net Debt could be reduced by ~$1-2B in a single instance. However, the $2B is for purchasing equity at $256.80/share, not debt repayment—meaning the $2B enters equity, not debt reduction. Deleveraging would be via equity dilution, not cash.
2. Asset Sales: The Munich business and A&D ($400M) have already been sold. If residual assets in the Industrial segment (valued at $2-3B) continue to be sold, significant deleveraging is possible. However, this would reduce the revenue and profit base.
3. EBITDA Growth: Net Debt/EBITDA decreased from 3.2x in FY2024 to 2.1x currently, primarily driven by EBITDA growth rather than debt repayment. If EBITDA continues to grow by 20%+/yr, Net Debt/EBITDA could fall to ~1.2x by FY2028, achieving "deleveraging" without debt repayment.

Interest Rate Risk Quantification:

COHR's Term Loan B will see its interest rate decrease to SOFR+2.00% (from SOFR+2.50%) in January 2025. Current SOFR is approximately 4.3%, so the effective interest rate is around 6.3%. On $3.5B of total debt, annualized interest is approximately $220-244M. If the Federal Reserve cuts rates by 100bp in 2026-2027, interest savings would be approximately $35M/yr (per $1B floating rate × 1%). Conversely, if interest rates remain high, interest will continue to suppress GAAP EPS by approximately $1.40/share/yr.

Probability Assignment: 20-30% probability of significant deleveraging stall before FY2027 (Based on: FCF turning negative is a fact; Counter-example: If OCF returns to $200M+/Q and CapEx stabilizes, there would still be $200-400M annually available for debt repayment; Natural experiment: Net Debt decreased by $990M from FY2023 to FQ2'26, indicating the deleveraging mechanism is in operation).

5.1.5 Technology Substitution and InP Value Dilution (Risk R4)

Core Assessment: COHR's core competency is built on InP vertical integration, but in the CPO (Co-Packaged Optics) era, the proportion of InP chips in the overall optical module BOM (Bill of Materials) will decrease from 30-40% in the pluggable era to 10-15%. If Broadcom's SiPh (Silicon Photonics) solution succeeds in 2028-2030, InP will be downgraded from a core component to an auxiliary component, diluting COHR's vertical integration value.

Transmission Timeline:

• 2026-2027: Pluggable remains dominant (800G/1.6T), with InP EML having the highest value
• 2027-2028: CPO begins mass production, InP chips and SiPh chips run in parallel, but InP still holds a performance advantage on high-speed lanes
• 2028-2030: In the 3.2T+ era, SiPh + external laser solutions mature, and the irreplaceability of InP declines

COHR's Hedging Strategy: COHR is simultaneously pursuing both InP and SiPh routes (collaborating with Tower Semiconductor on SiPh), but if SiPh becomes mainstream, COHR's $2B+ InP capacity investment in its Sherman factory will face overcapacity, increasing its fixed cost burden.

Probability Assignment: 15-25% probability of significant InP value dilution in 2028-2030 (Based on: Technology substitution cycles typically last 5-7 years, and current SiPh has already been laboratory-validated for 400G/lane; Counter-example: InP's physical property advantages in high-speed transmission may persist up to 6.4T+; Natural experiment: Broadcom's SiPh CPO prototype was showcased at OFC in 2025 but has not yet reached mass production).

5.1.6 Execution and Integration Risk (Risk R5)

Core Assessment: CEO Jim Anderson simultaneously manages three businesses with vastly different growth curves (high-growth AI + SiC investment phase + Industrial divestiture), making attention dilution a reasonable concern. However, the FQ2'26 earnings beat (Revenue $1.69B vs consensus $1.64B, EPS $1.29 vs $1.21) and 8 consecutive quarters of OPM improvement (from 2.8% to 11.8%) indicate that execution currently has no issues.

Key Monitoring Points: (1) Whether Industrial segment OPM continues to deteriorate (if management neglects this business); (2) After segment reorganization (3 segments → 2 segments), whether SiC's financial performance is obscured by the D&C segment or Industrial segment, reducing transparency; (3) Whether NVIDIA's committed CPO deliveries (starting 2027) proceed on schedule.

Probability Assignment: 20-30% probability of significant execution issues arising (Based on: Historical failure rate for post-merger integration is approximately 30-40%; Counter-example: COHR has been integrating for 2 years and OPM has continuously improved; Jim Anderson is from Lattice Semiconductor and has successful restructuring experience).

5.1.7 Equity Dilution (Risk R6)

Core Assessment: Dilution has already occurred, it is not a future risk.

Preferred Stock Conversion: The FQ2'26 Balance Sheet shows Preferred Stock decreased from $2,505M (FQ1'26) to $0. This is the conversion of Series B-2 convertible preferred stock, and the S-3ASR registration statement indicates that approximately 9,775,846 common shares were issued from the Series B-2 conversion, available for sale in the secondary market (as of December 16, 2028). Including other parts of Series B, total dilution of approximately 29.9M shares has been incorporated into the FY2026 diluted EPS calculation.

Based on the current outstanding share count of 155.5M, total dilution from 29.9M shares is approximately 19.2%. However, these shares will convert and be sold in tranches (until 2028), therefore: (1) FY2026 consensus EPS has already partially reflected dilution (diluted shares ~165M); (2) By FY2028, if fully converted, diluted shares could reach 175-185M.

SBC: Based on MCP data, quarterly SBC fluctuates significantly: Q1'25=$35M, Q2'25=$41M, Q3'25=$41M, Q4'25=$160M (abnormal, includes one-time item), Q1'26=$44M, Q2'26=$87M (includes NVIDIA-related). Normalized SBC is approximately $40-45M/Q, annualized $170M, accounting for about 2.5% of Revenue. This ratio is relatively low in the semiconductor/optical communications industry (LITE is about 3-4%), so SBC is not COHR's primary source of dilution.

NVIDIA $2B Investment Dilution: NVIDIA invested $2B at $256.80/share, acquiring approximately 7.8M shares, representing about 5% ownership. This dilution is already reflected in the increase in outstanding shares.

Overall Dilution Impact: From 153M diluted shares in FY2025 to a possible 185M diluted shares in FY2028E, total dilution is approximately 21%. If EPS grows from $5.35 to $9.64 (+80%) during the same period, the impact of dilution is offset by growth. However, if growth falls short of expectations, dilution will amplify the downside pressure on EPS.

5.1.8 Risk Synergy/Anti-Synergy Matrix

Synergy ①: R1+R2 (CapEx Downturn + Label Collapse, Most Dangerous, Probability 20-25%): When networking growth slows from +30% to +10%, P/E will not adjust linearly—the market will re-evaluate "is this an AI growth stock or a cyclical stock?" Once the label collapses, P/E dropping from 40x to 20x is non-linear. Historical reference: II-VI's stock price went from $47 to $27 (-43%) in 2018, while revenue only declined by -12%, but its P/E compressed from 35x to 18x.

Synergy ②: R1+R3 (CapEx Downturn + Deleverage Stalling, Second Most Dangerous, Probability 15-20%): A CapEx downturn leads to reduced Operating Cash Flow (OCF), while the company continues to maintain high CapEx (NVIDIA's expansion commitment might be a contractual obligation), further deteriorating Free Cash Flow (FCF). Net Debt increases instead of decreasing, and interest expenses continue to suppress EPS.

Hedge: R1 partially hedged by rate upgrades (related to R4): Even if the AI CapEx cycle slows, if the 1.6T ramp proceeds simultaneously, COHR's InP chip demand may not decline in parallel—ASP increases driven by rate upgrades can partially offset the decline in volume.

Synergy ③: R3+R5 (Deleverage Stalling + Execution Below Expectation, Moderate, Probability 10-15%): If integration efficiency falls short of expectations (R5) leading to slower Operating Profit Margin (OPM) expansion, while capacity expansion consumes cash and deleveraging slows (R3), both factors jointly suppress FCF, extending the deleveraging timeline by 2-3 years. This is not a fatal combination, but it will undermine the credibility of the "deleveraging unlocks EPS" narrative, indirectly impacting P/E support.

Synergy ④: R4+R1 (Long-term, Probability 10-15%): If Silicon Photonics (SiPh) replaces Indium Phosphide (InP) between 2028-2030 (R4), and simultaneously the AI CapEx cycle has passed its peak (R1), COHR will face both declining demand and core technology devaluation. The InP capacity investment ($2B+) at the Sherman factory would become a sunk cost in this scenario, and the risk of goodwill impairment ($4.5B goodwill) would significantly increase. However, this combination occurs after 2028, giving management a 2-3 year adjustment window.

5.1.9 Quantitative Risk Summary (Preliminary Estimates, Subject to Further Validation)

Note: R1 (CapEx downturn) and R2 (label collapse) are highly correlated (estimated correlation coefficient >0.7) and cannot be simply added. Their joint probability is approximately 20-25%, with a joint impact of -40% to -55%, thus the combined expected downside is about -8% to -14%. This figure will be refined in the second round (financial modeling) using Python Monte Carlo simulations.

5.1.10 "Boiling the Frog" Scenario (Gradual Deterioration, No Single Kill Switch Triggered)

The most insidious risk is not a cliff-edge drop, but rather the following gradual combination:

• FY2027: Networking growth slows from +30% to +20%, "still growing" so P/E maintains 35x
• FY2027H2: CapEx decreases from $150M/Q to $120M/Q, but still higher than FY2025's $100M; FCF recovers to slightly positive but insufficient for rapid deleveraging
• FY2028: Growth further declines to +12%, P/E compresses to 28x, the market begins to question "is this a cyclical stock?"
• FY2028H2: SiC is still in its investment phase (200mm ramp delayed by 6 months), and the Industrial segment is still declining
• Result: Stock price gradually declines from $307.50 to $200-220 (-28% to -35%); no single quarter triggers a "Kill Switch", but holders are trapped in a situation that is "not bad enough to sell, but not good enough to buy"

This scenario is dangerous because it does not trigger stop-loss discipline—each quarter is "okay," but cumulatively, it results in persistent valuation erosion.

5.2 Reverse DCF

5.2.1 Six Beliefs Implied by Current $307.50

We use the Reverse DCF method to infer market assumptions from the current stock price. The following is the set of implied assumptions required for $307.50 (EV $50.5B) to hold true:

5.2.2 Revenue CAGR ~21.7% (Belief B1, FY2025→FY2028)

Implicit Assumption: Revenue needs a 21.7% CAGR to grow from FY2025 $5.81B to FY2028E $10.46B.

Segment Breakdown:

• Networking/Datacom: Needs to grow from $4B (annualized) to $7.5B, CAGR ~23% — Requires sustained 800G growth + 1.6T ramp + CPO contribution.
• Industrial: Stable or slight growth from $1.8B to $2.0B — Not overly difficult, but relies on a recovery in the manufacturing cycle.
• SiC: Growth from $0.5B to $1.0B — Requires successful 200mm ramp and recovery of EV demand.

Our Assessment: A Networking CAGR of 23% is achievable with the support of the 1.6T ramp, but predicated on Hyperscaler CapEx not experiencing an annual decline of >20%. If CapEx growth decelerates from +80% to +20%, Networking CAGR would drop to 15%, and total revenue CAGR would fall to 14-16%, corresponding to FY2028 Revenue of ~$8.5-9.0B versus the consensus of $10.5B.

Fragility: ⭐⭐⭐⭐/5 — Depends on external variables (Hyperscaler CapEx), which COHR cannot control.

5.2.3 Non-GAAP OPM Expansion to 18–20% (Belief B2)

Implicit Assumption: Non-GAAP OPM expands from the current 12% (GAAP 11.8%) to 18-20%. FY2028 EPS of $9.64 on $10.5B revenue requires net profit of approximately $1.8B, corresponding to an OPM of about 17-19% (after deducting interest and taxes).

Drivers: (1) Improved product mix (higher proportion of AI Datacom, higher GM); (2) Operating leverage (R&D and SG&A growth slower than revenue); (3) D&A step-down; (4) Decreased interest expense (deleveraging).

Counterarguments: SiC in its investment phase (high CapEx, low utilization) will drag down overall OPM; 1.6T competition may compress ASP and GM; if the Industrial segment is not divested, its low GM will continue to be a drag.

Fragility: ⭐⭐⭐/5 — Directionally correct (improved product mix + D&A step-down), but magnitude depends on B1 (revenue growth) and SiC progress.

5.2.4 Natural D&A Step-Down (Belief B3)

Implicit Assumption: Of the FY2025 D&A of $554M/yr, most is intangible asset amortization from the II-VI merger (customer relationships/technology/trademarks, typically amortized over 5-15 years). The merger was completed in July 2022, so by FY2028, it will be the 6th year, and some short-term intangible assets (5-year term) will have completed amortization, leading to a D&A decrease of $100-200M/yr.

Mechanical Effect on GAAP EPS: For every $100M reduction in D&A, after-tax EPS increases by approximately $0.45-0.50 (assuming a 25% tax rate and 185M diluted shares). This is the most certain EPS driver — it does not require growth or margin expansion; it is purely a function of time.

Fragility: ⭐⭐/5 — High certainty, unless goodwill impairment occurs (which would increase expenses in the opposite direction).

5.2.5 'AI CapEx Sustained for Years' (Belief B4, Most Fragile)

Implicit Assumption: The $307.50 valuation requires Networking to maintain 20%+ growth until FY2028. This demands Hyperscaler AI CapEx to grow by at least +15% annually (assuming stable COHR market share) and not experience any single year with a >30% decline.

Historical Baseline Rate: Over the past 20 years, major tech CapEx cycles (2000-2001/2007-2009/2018-2019/2022-2023) have shown an approximately 3/4 = 75% probability of experiencing an annual decline of >30% within 3 years after the high-growth phase ends. The only counterexample was the steady growth period from 2010-2015, which was due to mobile internet providing continuous demand increments.

Is current AI CapEx a "mobile internet-like" structural shift? If so, CapEx growth might gradually decline from +80% to +15-20% without a cliff. However, if AI ROI is not proven by 2027-2028, cuts could be severe. We do not know the answer; this is the biggest black box in this report.

Fragility: ⭐⭐⭐⭐⭐/5 — The most fragile belief, as it relies entirely on external variables, and the historical baseline rate (75% probability of significant adjustment) is unfavorable.

5.2.6 SiC Transitions from Drag to Positive Contribution (Belief B5)

Implicit Assumption: The SiC business moves from a loss to breakeven in FY2027-2028, no longer dragging down overall EPS. If SiC continues to incur annual losses of $50-100M, it would be equivalent to an EPS drag of $0.25-0.50 per year.

Fragility: ⭐⭐⭐/5 — Wolfspeed's exit improved the competitive landscape, but the 200mm ramp and EV demand still present uncertainties.

5.2.7 Dilution Covered by Earnings Growth (Belief B6)

Implicit Assumption: Diluted shares grow from 155M to 185M (+19%), but EPS grows from $5.35 to $9.64 (+80%), meaning growth > dilution.

Fragility: ⭐⭐/5 — If B1 (revenue growth) holds, B6 automatically holds.

5.2.8 Implicit Assumptions vs. Consensus

Our largest disagreement with the market/consensus lies in B1 (revenue growth) and B4 (CapEx sustainability) — these two beliefs are interdependent and both point to the same external variable: Hyperscaler AI CapEx. This makes COHR's investment thesis highly concentrated on a variable that we cannot control, cannot predict, and for which the historical baseline rate is unfavorable. This in itself is a risk signal.

5.2.9 Valuation Impact of Single Belief Failure

Conclusion: Belief B4 ("Hyperscaler AI CapEx sustains for many years without a sharp decline") is the most vulnerable and the single variable with the greatest impact on valuation. If tracking only one variable, prioritize watching for a sharp downward revision in cloud providers' AI CapEx (same exogenous source as Risk R1).

5.3 Core Questions (CQ) & Confidence Levels

CQ refers to the Core Questions listed in Chapter 1. This section provides the confidence levels (our subjective probability, 0–100%, that the proposition holds true or is verifiable) derived from the preceding analysis and the main rationale.

5.3.1 Eight Core Questions

CQ1: Can Networking growth accelerate from approximately 17% to the sell-side consensus of 25%+ (around FY2027)?

• Confidence: 35%
• Main Rationale: Hyperscaler CapEx +82% YoY is difficult to sustain long-term; historically, optical communication revenue often slows down approximately 2–3 quarters after a CapEx peak; 1.6T ramp-up and qualification limit FY2027 contribution. Growth acceleration not only requires product ramp-up but also a CapEx cycle that does not decelerate – we judge the probability of the latter to be low.

CQ2: How much EPS can be released by deleveraging and declining D&A combined? Is the path reliable?

• Confidence: 50%
• Main Rationale: FCF has been negative for two consecutive quarters; deleveraging relies more on EBITDA growth than on cash debt repayment; NVIDIA's $2B is an equity investment, not debt repayment capital; quarter-over-quarter CapEx increase and deleveraging present tension. The direction of D&A decline is certain, but a precise amortization schedule is needed in subsequent chapters; the overall EPS "release pace" is slower than optimistic narratives suggest.

CQ3: Does the Sum-of-the-Parts (SOTP) valuation for the three engines support the value implied by the current market capitalization?

• Confidence: 40%
• Main Rationale: Based on the preceding business segmentation and conservative multiples, AI Datacom, SiC, Industrial, and net debt roughly sum to an enterprise value range of approximately $26–32B, with the midpoint significantly lower than implied by the current price; if a higher EV/Revenue is used for Networking, the range can shift upward – multiple selection is sensitive. Conclusion: The current price implies a significant "unified label" premium, which requires verification through a subsequent financial model.

CQ4: In the 1.6T era, can COHR significantly narrow the performance gap with LITE in EML?

• Confidence: 40%
• Main Rationale: LITE still has approximately a 12–18 month lead in 200G/lane EML; COHR's 6-inch InP cost reduction is valuable for 1.6T, but "catching up" requires simultaneously meeting performance and yield targets; at the module level, it faces cost competition from companies like Innolight, and its chip-level advantage may be partially diluted.

CQ5: Can SiC become a sustainable growth engine? When will 200mm contribute to profits?

• Confidence: 35%
• Main Rationale: Wolfspeed's exit improves the competitive landscape; DENSO/Mitsubishi investments and long-term agreements provide demand anchors; onsemi's 200mm precedent supports the cost curve. Counter-factors are EV penetration rates and industry price wars.

CQ6: Has the market fully understood the impact of Preferred shares disappearing/converting on per-share value?

• Confidence: 55%
• Main Rationale: The mechanism is clear: dilution from conversion of convertible preferred shares (tens of millions of shares); consensus diluted shares already partially reflect this; remaining uncertainty primarily lies in the pace of secondary market sales.

CQ7: Is the trade-off between expansion CapEx and deleveraging beneficial to shareholders?

• Confidence: 40%
• Main Rationale: FCF is negative and CapEx is high; NVIDIA's capital injection cannot replace cash for debt repayment; under the combination of OCF and CapEx, the path to net leverage reduction is relatively long. Management is "betting" on sustained AI demand – if it slows down, both cash efficiency and the deleveraging narrative will be pressured.

CQ8: Do non-China capacity and supply chain layouts bring a structural advantage relative to tariffs/geopolitics?

• Confidence: 60%
• Main Rationale: Malaysian assembly indeed offers a relative advantage for Networking, but COHR's decline in the 2025 tariff impact was greater than some peers', reflecting continued exposure to China-related supply chains in the Industrial/materials segment, with advantages and drags largely offsetting each other.

5.3.2 Summary

COHR's core profile from the initial analysis: a post-merger hybrid operating on three parallel tracks: "AI growth + deleveraging + SiC option", whose $48B valuation is highly dependent on the sustainability of AI CapEx (B4 belief). This belief is the most vulnerable; historical base rates show that 75% of CapEx cycles experience an annual adjustment of >30% within 3 years after the high-growth period ends. Current valuation has consumed most of the margin of safety (PEG of 1.64x is the highest among peers), and the preliminary SOTP ($26-32B) is also below market capitalization, indicating that the market is paying a $16-22B premium for a unified AI label.

The core question for subsequent validation is: How much of this $16-22B label premium can be justified by the 1.6T ramp + CPO + SiC option + declining D&A?

Chapter 6: Supplementary Analysis A — Factors Supporting ROIC Uplift

Meaning for ROIC: SiC option realization and BOM cost reduction are two accelerators for ROIC uplift – SiC transitioning from investment phase to profitability can stop consuming ROIC's numerator (NOPAT), and BOM reduction directly improves margin, boosting NOPAT. However, current evidence does not support realization before FY2027; these are Bull case conditions (25% probability), not base case assumptions.

6.1 SiC Standalone Financial Modeling

6.1.1 SiC Revenue Breakdown: Substrate vs. Epitaxy vs. Other Materials

COHR's Role in the SiC Value Chain: Substrate + epitaxy wafer supplier, not a device manufacturer. The device side (MOSFET/Diode) belongs to the domain of STMicro/Infineon/onsemi. This means COHR's SiC revenue comes from "selling shovels," not "digging for gold" – substrates and epitaxy wafers are upstream materials for EV power devices.

Global SiC Substrate Market Size: In 2024, global SiC substrate revenue was $1.04B, a YoY decline of -9%, due to slowing EV demand and customer inventory adjustments.

COHR Market Share: In 2024, COHR's market share in the SiC substrate market was 13.9%, ranking fourth. First is Wolfspeed at 33.7%, second is TanKeBlue (Tianke Heda) at 17.3%, and third is SICC (Shandong Tianyue) at 17.1%. This means COHR's SiC substrate revenue in 2024 was approximately $1.04B × 13.9% ≈ $145M (medium confidence inference).

SiC Revenue Structure Inference:

Key Calibrations: COHR's Materials segment FY2025 full-year revenue is approximately $970M (Q3 $237M × 4Q estimate, considering Q4 decline). The proportion of SiC within the Materials segment is not disclosed, but Materials includes SiC + rare earth magnetic materials + industrial precision optical materials. We estimate SiC accounts for 22-30% of the Materials segment revenue, or $210-290M, which is consistent with the bottom-up market share estimate above (medium confidence).

Why This Number Matters: If SiC annual revenue is ~$250M, it only accounts for 3.7% of COHR's $6.7B TTM total revenue. Even if SiC has a high standalone valuation, its weight in COHR's overall valuation is limited—unless the market prices SiC as an option (see Section 1.4).

6.1.2 200mm Yield Curve and Cost Advantage

The Economics of 150mm → 200mm: The area of a 200mm wafer is 1.78 times that of a 150mm wafer (π×100² / π×75² = 10000/5625), so the number of dice that can be cut per 200mm wafer is approximately 1.78 times that of a 150mm wafer. However, the cost of a 200mm substrate is $1,300-1,800/wafer, while a 150mm wafer is approximately $800/wafer, making the cost 1.6-2.25 times higher.

The Math of the Cost Crossover Point:

When 200mm yield catches up to 150mm (i.e., when die yield is consistent), the cost per die for 200mm will be approximately 9% lower than 150mm. If 200mm substrate costs decrease from $1,550 to $1,200 with mass production (expected during maturity in 5-7 years), the cost per die will be approximately 15-20% lower than 150mm (medium confidence inference).

Current Yield Status: By early 2026, AI-assisted process control has improved 200mm SiC yield by 10-15 percentage points. COHR's 6-inch InP wafer yield at its Sherman factory has surpassed traditional 3-inch lines, indicating that the company has transferable experience in managing large-diameter wafer yield, but InP and SiC are entirely different material systems, so yield experience cannot be directly transferred.

Competitive Comparison:

Key Causal Chain: onsemi's vertical integration (substrate to device) means a large proportion of its SiC substrates are consumed internally and do not participate in open market competition. Therefore, COHR's competitors in the external substrate market are primarily Wolfspeed (capacity constrained after bankruptcy) and Chinese suppliers (whose quality still lags but prices are 30-40% lower). Wolfspeed went bankrupt in mid-2025, and Mohawk Valley's utilization rate was extremely low. This created market share opportunities for COHR and Chinese manufacturers—but this window is temporary, as Wolfspeed emerged from restructuring in September 2025 and re-secured $2.1B in financing.

6.1.3 Break-Even Analysis

Wolfspeed as a Cautionary Tale: Wolfspeed's financial data before bankruptcy provides a reference for SiC business break-even. For the first 9 months of FY2025 (ending March 2025), revenue was $560.6M, net loss was $939.9M, and operating loss margin exceeded -60%. Wolfspeed's failure pattern: $6.7B debt + $400M annual interest + negative EBITDA + Mohawk Valley capacity utilization <30%. The core issue was not SiC technology failure, but a collapse in capital structure—before 200mm capacity reached full utilization, fixed costs (depreciation + interest) consumed all profits.

COHR SiC Break-Even Estimate:

COHR's SiC business structure has a key difference compared to Wolfspeed: COHR does not produce devices; it only manufactures substrates and epitaxy. Therefore, its CapEx intensity and fixed cost scale are significantly smaller than Wolfspeed's (Wolfspeed built a complete 200mm device fab, while COHR only requires material growth and epitaxy equipment).

(Low confidence: The above model contains multiple assumptions that cannot be independently verified.)

Break-Even Timeline Estimate:

Honest Disclosure: The biggest black box in the above break-even model is SiC-specific costs. COHR does not separately disclose SiC's gross margin, CapEx, and OpEx—all figures are derived from our inferences based on industry benchmarks. This places the entire break-even analysis at a speculative level and it should not be included in the primary valuation conclusions. Its value lies in providing a sense of magnitude: the SiC business's current loss is not substantial (at ~$15M level), it is not a significant drag on COHR's overall profitability, but it is also not a profit contributor.

6.1.4 SiC Standalone Valuation

Method 1: EV/Revenue Comparable Approach

COHR SiC Revenue ~$250M × 2.0-3.0x = $500M-$750M

Method 2: DENSO/Mitsubishi Investment Implied Valuation

In December 2023, DENSO and Mitsubishi Electric each invested $500M (totaling $1B), acquiring a 25% non-controlling interest in COHR's SiC business (12.5% each), with COHR retaining 75%.

Adjusted Fair Valuation: $4.0B × 0.5-0.7 (30-50% discount reflecting market changes) = $2.0-2.8B (COHR's 75% Stake = $1.5-2.1B)

Method 3: DCF (Highly Simplified)

Highly sensitive to growth rate and steady-state margin assumptions

SiC Standalone Valuation Summary:

What does the significant divergence (4.5x) among the three methods indicate: EV/Revenue and DCF provide the value of "SiC as a loss-making substrate business" ($500-750M), while the DENSO investment implies the value of "SiC as a strategic materials platform + EV supercycle option" ($2-3B). This difference itself is an investment judgment: If you believe SiC substrates will become the "silicon wafers" of the EV era (similar to SUMCO's role in semiconductors), the DENSO valuation is reasonable; if you think Chinese substrates will drive prices down to rock bottom (similar to the fate of solar-grade silicon wafers), EV/Revenue is more realistic.

Our Judgment: Within COHR's $48B market capitalization, the standalone value of SiC (COHR's 75% stake) is approximately $0.5-2.0B, accounting for 1-4% of the market cap. This implies that SiC is not the primary driver of COHR's current stock price—AI optical communications is. However, SiC is a meaningful option: If EV demand explodes in 2027-2028 + 200mm cost advantages are realized + Chinese competition is constrained by tariffs, SiC could transform from a "drag" (industrial segment drag) into a "new growth engine." The current market's implied weighting for SiC within the $48B valuation does not exceed 5%, which is consistent with our standalone valuation—it is neither undervalued nor overvalued; it represents fair option pricing.

6.2 Segment OPM Breakdown: Hard Data

6.2.1 Segment Revenue: Known Data

Starting from FY2026, COHR reorganized from three segments (Networking/Lasers/Materials) into two segments (Datacenter & Communications, or D&C / Industrial).

FQ2'26 (Dec 2025) Two-Segment Revenue:

FQ1'26 (Sep 2025) Two-Segment Revenue:

FY2025 Old Three-Segment Structure (Last Year, Q3 FY2025 Data):

6.2.2 Segment OPM Reconstruction: Back-calculation from Known Constraints

Known Hard Data:

• Company-wide Non-GAAP OPM FQ2'26: 19.9% ($336M / $1,686M)
• Company-wide Non-GAAP GM FQ2'26: 39.0%
• Company-wide GAAP OPM FQ2'26: 11.8% ($199M / $1,686M)
• FY2025 Full-year Non-GAAP operating income: $1.04B, OPM 17.8%
• D&C Revenue $1,208M (72%), Industrial Revenue $478M (28%)

Missing Segment Margin Disclosures: COHR does not separately disclose the operating income or margin for D&C and Industrial in its earnings releases. While segment disclosures required by ASC 280 in 10-K theoretically include segment profit/loss, the company discloses "segment earnings," whose definition may differ from standard operating income (typically excluding corporate costs and amortization of intangible assets). We are unable to obtain these precise figures from public searches.

Therefore, we use a constraint-based back-calculation method:

Constraint 1 — Gross Margin Differential Inference:
The D&C segment primarily includes AI optical communications (high-speed transceivers, 800G/1.6T), data center optical components, and telecom network products. LITE, as a nearly pure-play AI optical communications company, has a Non-GAAP GM of 42.5%. COHR's D&C segment, in addition to AI optical communications, also includes traditional telecom (DWDM, etc., which have lower gross margins); thus, the D&C segment's GM should be lower than LITE's but higher than the company's overall GM.

The Industrial segment includes industrial lasers (material processing, highly competitive, lower GM), SiC materials (currently loss-making/marginally profitable), scientific instrument optics, and more. The common characteristics of these businesses are strong cyclicality and intense competition, with GMs typically ranging from 25-35%.

Constraint 2 — OpEx Allocation Inference:
Non-GAAP OpEx = Revenue × (GM% - OPM%) = $1,686M × (39.0% - 19.9%) = $1,686M × 19.1% = $322M. This $322M in OpEx includes R&D and SGA. The D&C segment has high R&D intensity in AI optical communications (6-inch InP wafers, CPO, 1.6T), and the Industrial segment also has R&D needs for SiC and lasers. Assuming OpEx is allocated proportionally to revenue (no better information available):

However, allocating OpEx proportionally to revenue is the simplest assumption; in reality, the R&D intensity of the D&C segment should be higher than Industrial (due to the extremely rapid technological iteration in AI optical communications). Therefore, the actual OPM for D&C is likely below 21.8%, and Industrial's is above 14.2%. Adjusting R&D weighting (D&C R&D accounts for 60-65% of revenue instead of 72%):

Best Estimate for Segment OPM (FQ2'26):

6.2.3 Quantifying Industrial's Drag on Overall Margins

Core Question: How much does the Industrial segment's low margin drag down the company's overall OPM?

OPM Increment for every 1% of revenue shifted from Industrial to D&C: (23.5% - 10%) × 1% = +0.135pp.

This means that if D&C increases from 72% to 80% over the next 2 years (driven by AI demand), the company's overall Non-GAAP OPM will rise from 20% to 21%, with mix shift alone contributing +1pp. Therefore, COHR's OPM expansion story is not about Industrial improvement, but rather the increasing proportion of D&C.

Impact of Munich Business Sale: Management has announced the sale of the Munich materials processing business (part of the Industrial segment). This will directly reduce Industrial revenue while divesting a low-margin business, leading to a positive impact on overall OPM. The specific impact depends on the scale of the Munich business (undisclosed), but the direction is clear: OPM accretion.

6.2.4 Margin Comparison with LITE

Key Insight: The COHR D&C segment's margin is already close to LITE's, but its growth rate is only half that of LITE (33.6% vs 65.5%). This indicates two things:

1. COHR's D&C segment is not a "low-cost version of LITE": Its margin capability is similar to LITE's; therefore, the margin gap stems from the drag of the Industrial segment, not an efficiency issue within the D&C segment itself.
2. The growth rate difference is a product mix issue: LITE is almost 100% exposed to the fastest-growing sub-segment of AI optical modules (200G/lane EML, 800G→1.6T), whereas COHR's D&C segment also includes slower-growing telecom networks and some traditional data communication products. If COHR's AI sub-business growth rate were similar to LITE's, it would be dragged down by 10-15pp by the telecom portion.

Valuation Implications: The market assigns COHR a Forward P/E of 41x, and LITE a Forward P/E of 47x. If COHR successfully divests its low-margin Industrial segment businesses (or allows them to shrink to <20% of revenue), the company's overall OPM would increase from 20% to 22-24%, more closely resembling LITE's profitability structure. In this scenario, COHR's P/E should converge towards LITE's—implying that the current 41x includes a 10% "conglomerate discount" (conglomerate discount). Whether this discount is justified depends on management's willingness and ability to execute a focused strategy.

6.3 Upside Scenario Analysis — What the Market Underestimates?

The initial analysis primarily leaned bearish (consistent with the confidence levels of the core issues listed in §5.3, generally cautious), as the evidence chain focused on three suppressive factors: CapEx cyclicality, difficulty in valuing a conglomerate, and negative FCF. However, a bearish analysis that does not systematically examine upside dimensions constitutes a risk of one-sided argumentation. The following identifies 3 upside dimensions that are not yet fully priced in by market consensus, as well as one catalyst that is priced in but whose supportive strength is underestimated.

6.3.1 CPO Revenue: 2027-2028 Incremental Revenue Not Yet in Consensus

Core Judgment: COHR's CPO (Co-Packaged Optics – integrating optical modules directly next to the switch ASIC package to reduce power consumption and latency) revenue is expected to scale from FY2027H2 onwards, but the current sell-side consensus (FY2028E revenue $10.46B) lacks explicit modeling of CPO's contribution, as TAM forecasts for the CPO market itself remain highly dispersed (2028 forecasts ranging from $49M to $5.5B, a difference of over 100x).

Timeline and Product Progress:

COHR demonstrated a 6.4T CPO solution at OFC 2026. Management's CPO revenue timeline is divided into two phases:

• Scale-out CPO (for inter-GPU cluster interconnect): Commencing shipments in 2026 H2, meaning initial revenue contribution in H2 of FY2027 (ending June 2027)
• Scale-up CPO (for intra-GPU cluster interconnect, with higher bandwidth demands): Starting 2027 H2, covering the full FY2028

NVIDIA's $2B investment (March 2026) came with a "multi-billion dollar" multi-year purchase commitment, with an execution period from 2027-2030. A significant portion of this purchase commitment is for CPO products, as NVIDIA's next-generation platform (Rubin architecture, expected 2027-2028) will substantially increase the CPO deployment ratio.

CPO TAM Uncertainty:

Market research firms' forecasts for the CPO market size vary widely, which itself is evidence of it being "under-modeled":

The forecast discrepancy exceeds 100 times, indicating a lack of industry consensus on CPO definition and adoption speed. We take a mid-case scenario: the CPO module + component market in 2028 is approximately $1.5-3.0B, with COHR market share of 15-25%, corresponding to CPO revenue of $225M-$750M/yr.

Incremental Impact on EPS:

If CPO contributes $500M in incremental revenue in FY2028 (mid-to-optimistic scenario), calculated with a Non-GAAP OPM of 20-25%, the incremental Operating Income would be approximately $100-125M. After-tax, this would contribute approximately $0.45-0.55 to EPS. This implies that FY2028E EPS could be revised up from the consensus of $9.64 to $10.1-10.2, approximately +5%.

Downsides: (1) The CPO supply chain (SiPh engine + CW laser source + advanced packaging) is not yet fully mature, and yield ramp-up delays are the biggest risk in 2026-2027; (2) In the CPO architecture, ASIC vendors (Broadcom) develop their SiPh optical engines in-house, and COHR only provides the CW laser source and some optical components, resulting in a lower value share per port than pluggable modules; (3) CPO and pluggable are expanding in parallel rather than substituting, but if CPO cannibalizes pluggable's market share, COHR's pluggable revenue would be offset.

Is this dimension adequately priced in? Not entirely. In the consensus FY2028E Revenue of $10.46B, most of the growth is derived from projected volume increases of 800G/1.6T pluggables. The incremental contribution of CPO as a "new category" is assigned conservative assumptions (~$200-300M) or not separately broken out by most sell-side models due to TAM uncertainty and revenue timing risks. If CPO progress aligns with NVIDIA's Rubin timeline, there is an upside potential of approximately $200-450M, corresponding to an EPS revision of $0.15-0.40.

6.3.2 Underestimated EPS Contribution from Declining D&A

Core Judgment: COHR's FY2025 D&A is $554M, with a large portion being intangible asset amortization arising from the II-VI merger (July 2022). Starting in the fourth year post-merger (FY2026), some short-lived intangible assets (customer relationships/existing technology, typically amortized over 5-10 years) will begin to decline at an accelerated pace, which will "freely" unlock GAAP EPS, while sell-side models have limited accuracy in modeling this gradual improvement.

Logical Derivation of Amortization Decline:

Intangible assets generated from the II-VI merger were approximately $5.3B (recognized in FQ1'23), including:

• Existing Technology/Patents: Typically amortized over 5-10 years; batches begin to expire in the 4th-5th year post-recognition (FY2026-2027)
• Customer Relationships: Typically amortized over 7-15 years; decline is slower
• Trademarks/Brands: Typically 10-20 years or indefinite; slowest amortization

COHR's FQ2'26 (December 2025 quarter) GAAP vs Non-GAAP EPS gap is $0.53/Q ($0.76 GAAP vs $1.29 Non-GAAP). Of this $0.53/Q gap, intangible asset amortization accounts for approximately $0.55-0.65 (annualized approximately $340-400M), and SBC accounts for approximately $0.25-0.35.

Assuming amortization declines from $554M in FY2025 to:

• FY2027E: ~$450M (-19%), as some 5-year existing technology assets expire in FY2027
• FY2028E: ~$350M (-37%), as more batches expire
• FY2029E: ~$270M (-51%), leaving only long-lived customer relationships and goodwill impairment provisions

Impact on GAAP EPS: Every $100M reduction in D&A contributes approximately $0.51 to GAAP EPS after-tax (assuming a 21% tax rate) ($100M × 0.79 / 155M shares). From FY2025 to FY2028, if D&A decreases by $204M (from $554M to $350M), the incremental GAAP EPS would be approximately $1.04. This means FY2028 GAAP EPS will improve by an additional $1.0+ compared to the current trajectory, purely from the expiration of accounting amortization, requiring no business-level changes.

Is this adequately modeled? Partially modeled. The Non-GAAP EPS consensus ($9.64 FY2028E) already excludes the impact of amortization, so Non-GAAP EPS is unaffected by D&A changes. However, the improvement in GAAP EPS will narrow the gap between GAAP and Non-GAAP, which is meaningful for two types of investors: (1) investors who screen using GAAP P/E, as the GAAP P/E will accelerate its compression from the current "meaningless" (304x trailing); (2) institutional investors focused on GAAP earnings quality.

L1 Principle Test: D&A amortization decline is an accounting variable, not a business variable (L1 Principle #2). Therefore, this should not be the core support for investment judgment. However, its investment implication is indirect: accelerating GAAP EPS improvement changes COHR's visibility in quantitative screening, attracting more institutional attention.

6.3.3 S&P 500 Inclusion Effect

Core Judgment: COHR was added to the S&P 500 on March 23, 2026. Passive fund buying has occurred but has not been fully digested, and the ongoing rebalancing effect will provide structural liquidity support over the next 6-12 months.

Event Timeline:

• March 6, 2026: S&P Global announced COHR's inclusion in the S&P 500, +8% on that day
• March 23, 2026: Official inclusion effective date
• March 24, 2026: First trading day post-inclusion closed at $272.33, +6.78% on that day
• April 10, 2026: Current price $307.50, up approximately 22% from pre-announcement

Estimate of Passive Fund Size:

Assets tracking the S&P 500 index are approximately $7.8 trillion (ETFs + index funds). COHR's market cap of $48B accounts for approximately 0.10% of the S&P 500's total market cap. Therefore, passive funds need to allocate approximately $7.8B to COHR stock. Based on COHR's average daily trading volume of approximately $600-800M, full absorption would require about 10-13 trading days.

Is the initial buying complete? The rebalance buying by core passive funds (SPY, IVV, VOO, etc.) was concentrated and executed around 3-5 days before and after the inclusion date; this part is largely complete. However, portfolio adjustments by enhanced index funds, Smart Beta products, and active funds benchmarked to the S&P 500 are gradual and typically continue for 2-6 months post-inclusion.

Ongoing Effects: The long-term effects of S&P 500 inclusion include: (1) improved liquidity (narrower bid-ask spread, easier execution of block trades); (2) increased analyst coverage (from 15 to 20+ firms); (3) increased options market depth (beneficial for institutional hedging); (4) a "must-own" effect (benchmark-aware funds cannot have completely zero allocation). These effects are not one-off but provide structural valuation support.

Downsides: (1) Historical research shows that excess returns after S&P 500 inclusion tend to disappear after 6-12 months (mean reversion); (2) The inclusion effect only impacts fund flows, it does not change fundamentals; (3) If the AI CapEx cycle weakens, passive fund support cannot prevent valuation contraction.

6.3.4 Quantification of Bull Case

Superimposing the three undervalued dimensions:

Probability Anchoring (Triple):

Basis for CPO 30% probability:

• Historical Benchmark Rate: New optical packaging technologies typically require 3-5 years from demonstration to scaled revenue (400G pluggable: Demo 2018 → Scale 2021). CPO Demo 2026 → Scale 2028 = 2 years, faster than historical, but NVIDIA's capital injection accelerates this timeline
• Counter-Case Conditions: If SiPh engine yield falls below 80%, CPO scaling will be delayed by 12-18 months, pushing revenue contribution to FY2029. Current public information on SiPh yield is insufficient, making it a black box
• Natural Experiment: The degree of CPO integration in NVIDIA's Rubin architecture is the best validation. Uncertainty is expected to significantly narrow when Rubin technical details are disclosed in 2027

D&A has a higher probability of 50%, as this is an accounting certainty (amortization plans are disclosed in 10-K footnotes); the only uncertainty is whether management will add new intangible assets due to acquisitions.

Bull Case Synthesis:

If the above three items are simultaneously optimistic, FY2028E EPS is raised to $10.5-11.0 (vs consensus $9.64, +9-14%). Calculating with 35x P/E (reasonable given FY2028 25%+ growth):

• Bull Case Fair Value: 35x × $10.75 = $376 (vs current $307.50, +22%)
• Base Case (Consensus): 35x × $9.64 = $337 (+10%)
• Bear Case (CapEx Deceleration + CPO Delay): 25x × $8.50 = $213 (-31%)

6.4 Quantity-Price Scissors Gap Quantitative Analysis

6.4.1 Scissors Gap 1: Optical Module Quantity vs. Price — Where is the threshold where "volume growth cannot offset price decline"?

Core Judgment (Medium Execution Probability): 800G optical modules are in the early stages of a typical "volume up, price down" scissors gap. Shipment volume YoY growth is +85-100%, but ASP has declined by 30-40% from its 2024 peak. Currently, volume growth far outpaces price decline, resulting in positive net revenue growth (+34% YoY). However, when 800G shipment volume growth slows to below +20% (estimated FY2028), if ASP continues to decline at -15-20%/yr, net revenue growth will compress to single digits.

Quantified Time Series:

Drivers for 800G ASP Decline:

(1) Increased Supply: Innolight and Eoptolink are expanding production capacity for 800G pluggable modules. The labor cost advantage of Chinese manufacturers puts continuous downward pressure on module ASPs. 800G QSFP-DD module ASP is expected to decrease from a 2024 peak of approximately $700-800 to around $450-550 in 2026, and further to $350-450 in 2027.

(2) Technological Maturity: As EML chip yields improve and packaging processes standardize, manufacturing cost reductions are passed on to ASPs.

(3) Generational Replacement: After 1.6T modules begin shipping, some high-end demand shifts, and 800G faces "previous generation price reduction" pressure.

1.6T ASP Buffering Effect:

1.6T modules currently have an ASP of approximately $1,200-2,000, about 2.5-3 times that of 800G. This is because 1.6T uses 8×200G/lane (vs 800G's 4×200G/lane), doubling the number of InP chips per module, and requiring more complex packaging.

1.6T is expected to begin mass shipments in FY2027, contributing $2.0-3.0B in revenue by FY2028 (COHR's estimated figures). The high ASP of 1.6T offsets the 800G ASP decline, allowing overall AI Datacom revenue growth to remain at +15-25% in FY2027-2028, rather than the +0-10% calculated for 800G alone.

Historical Reference: The transition period from 400G to 800G (2022-2024) showed similar quantity-price dynamics. 400G ASP decreased from approximately $300 (2022) to $150-200 (2024, -40-50%), but the simultaneous ramp-up of 800G led to continued growth in total industry revenue. This means the quantity-price scissors gap for COHR on 800G is real, but the generational relay of 1.6T acts as a natural hedge.

Investment Implications: Looking solely at the 800G quantity-price scissors gap would lead to the conclusion of "significant growth slowdown in FY2028", but with the 1.6T relay, overall AI Datacom growth has more resilience. The key risk is not the quantity-price scissors gap itself, but the timing of the 1.6T ramp – if 1.6T mass production is delayed by 6 months in FY2027, the 800G ASP decline would create a "growth vacuum" during the transition period.

Downside Risks: (1) Initial 1.6T yields may be low, leading to actual shipments falling below plan; (2) The high ASP of 1.6T may decline rapidly as mass production matures, limiting the duration of the buffering effect; (3) If Hyperscalers delay 1.6T deployment (because 800G is sufficient), the generational relay would be delayed.

6.4.2 Scissors Gap 2: Hyperscaler CapEx Growth vs. COHR Revenue Growth — Is market share rising or falling?

Core Judgment (Medium Execution Probability): The combined CapEx of the top four Hyperscalers is estimated at approximately $690B in 2026 (+82% YoY), while COHR Networking revenue growth is only +34% YoY. This 48pp growth differential suggests that COHR's "wallet share" within Hyperscaler CapEx is declining. However, the reason is not weakened competitiveness, but rather that most of the incremental CapEx is directed towards GPUs themselves (approximately 60-70%), while the proportion of optical modules remains relatively stable.

Hyperscaler CapEx Allocation Breakdown (2026E Total Approximately $690B):

COHR's market share in optical modules + components is approximately $4.0-4.8B (TTM AI Datacom revenue) / $15-25B (Optical Module TAM) = approximately 16-32%. This share range is broad because we do not know how much of COHR's AI Datacom revenue comes from hyperscalers (vs. enterprise/telecom).

Transmission Coefficient Estimate:

If Hyperscaler CapEx growth decelerates from +82% to +20% (2027-2028 scenario), optical module demand growth will not decline proportionally, because:

(1) Installed base requires upgrades: Demand driven by the 800G to 1.6T upgrade cycle is independent of new CapEx
(2) Proportion of optical modules in CapEx is rising: The optical interconnect density of AI clusters (number of optical modules required per GPU) increases with cluster scale
(3) However, growth slowdown is certain: The transmission coefficient is approximately 0.5-0.7, meaning if CapEx growth drops from +82% to +20% (a 62pp decline), COHR Datacom growth would decline by approximately 31-43pp, from +34% to around -9% to +3%

This Implies: If Hyperscaler CapEx growth sharply drops to +20% in 2027, COHR AI Datacom revenue growth would compress from +34% to low single digits or even negative growth. This aligns with the cyclical risk assessment from the initial fundamental analysis but quantifies the magnitude of the transmission.

Downside Risks: (1) The transmission coefficient of 0.5-0.7 is a historical inference; if the growth in optical interconnect density for AI clusters exceeds expectations (CPO+1.6T), the transmission coefficient could be lower (better); (2) Hyperscalers may not reduce CapEx growth to +20%; current CEO statements generally imply >+30% for 2027; (3) COHR's diversification (SiC+Industrial) provides some hedge when AI CapEx slows down (though limited in magnitude).

6.4.3 Scissors Gap 3: GAAP vs. Non-GAAP EPS Difference — Narrowing or Expanding?

Core Judgment (Medium Execution Probability): The difference between GAAP and Non-GAAP EPS is slowly narrowing. This is because the declining D&A (a negative force) is slightly greater than the increasing SBC (a positive force). By FY2028, the difference will narrow from $4.70 in FY2025 to approximately $2.5-3.0, making GAAP P/E shift from "meaningless" to having reference value.

Gap Breakdown:

Note: FQ1'26 GAAP EPS ($1.19) being higher than Non-GAAP ($1.16) is an outlier, as a one-time gain (gain from A&D business sale) in that quarter inflated GAAP. Excluding one-time items, the underlying gap remains approximately $0.50-0.60/Q.

Breakdown of Narrowing Drivers:

(1) Decreasing D&A (Narrowing Force): As analyzed in Section 3.2, amortization decreases from approximately $554M/yr to approximately $350M/yr by FY2028, an annualized narrowing of $204M. Post-tax, this narrows EPS by approximately $1.04.

(2) SBC Growth (Widening Force): FY2025 SBC was approximately $161M, but Q4'25 ($160M) and FQ2'26 ($87M) data were unusually high because a NVIDIA investment triggered accelerated recognition of equity compensation. Normalized SBC is approximately $180-200M/yr, projected to increase to $250-300M by FY2028 (compensation cost driven by revenue growth). SBC growth of approximately $90-140M/yr, post-tax, this widens the EPS gap by approximately $0.46-0.71.

(3) Net Effect: D&A narrowing of $1.04 > SBC widening of $0.46-0.71, a net narrowing of approximately $0.33-0.58/yr.

Investment Implications: GAAP EPS approaches $7.0 by FY2028 (vs Non-GAAP $9.64), corresponding to a GAAP P/E of approximately 44x (vs Non-GAAP P/E 32x). A 44x GAAP P/E is not unreasonable for an optical communications company with 25%+ growth, but it remains higher than LITE's GAAP P/E (due to LITE having no acquisition-related amortization burden). Investment signal from the narrowing gap: For quantitative strategies that screen based on GAAP, COHR is moving from a "GAAP loss/uninvestable" category into a "GAAP profit/investable" category, which expands the potential buyer base.

Counterpoints: (1) If COHR makes new large acquisitions in FY2027-2028, it would add new intangible assets and amortization, reversing the narrowing trend; (2) SBC growth could exceed expectations, especially if NVIDIA investment terms include additional equity compensation arrangements; (3) The improvement in GAAP P/E is accounting-driven and does not change the company's true cash flow generation capability.

6.4.4 Divergence 4: R&D vs. Revenue Growth Rate — OpEx Leverage Release or Insufficient Technology Investment?

Key Assessment (Execution Level: Medium): COHR's R&D/Revenue ratio decreased from approximately 12% in FY2024 to approximately 11% in FY2025, while revenue growth was +23%. This means R&D absolute spending growth (approx. +10% YoY) was lower than revenue growth, and OpEx leverage is being released. However, if the R&D/Revenue ratio continues to decline to <10%, in the three tracks (1.6T/CPO/SiC) that simultaneously require R&D investment, it may face the risk of eroding its technological competitiveness.

R&D Spending Trend Projection:

Positive Aspect of OpEx Leverage: R&D growth (+10-13%) is lower than revenue growth (+20-26%), releasing approximately 0.5-1.0 percentage points (pp) of OPM improvement annually. This aligns with the trend of COHR's Non-GAAP OPM improving from 12% in FY2024 to 18% in FQ2'26. If R&D growth maintains at +13% while revenue growth maintains at +20%, R&D/Revenue would decrease to approximately 8.5% by FY2028, cumulatively releasing approximately 2pp of OPM.

Negative Considerations for Technological Competitiveness: COHR simultaneously faces three R&D tracks: 1.6T InP mass production, CPO SiPh development, and SiC 200mm transition. If R&D/Revenue falls below 9%, the R&D budget for each track would be approximately $260M (split equally), whereas LITE, focused on a single track (AI optics), has an R&D budget of over $200M. This implies that COHR's R&D intensity per track is lower than that of its more focused competitors.

Investment Implications: In the short term (FY2026-2027), the release of OpEx leverage is a reliable source of OPM expansion. In the long term (FY2028+), if R&D/Revenue falls below 8%, it will be necessary to monitor whether the progress of the three technology tracks shows signs of falling behind (1.6T yield, CPO customer qualification, SiC 200mm yield).

6.5 InP BOM Value Share Derivation

6.5.1 Current Pluggable Module BOM Breakdown

The total BOM (Bill of Materials) cost for an 800G QSFP-DD optical module is approximately $200-300 (depending on supplier and volume) and is roughly broken down as follows:

Key Data Source: Industry analysis indicates that "laser chips account for approximately 60% of optical module costs" (Deep Fundamental Substack). However, this figure includes all optical components (EML+PD+optical elements), not just InP EML. Pure InP EML accounts for approximately 20-25% of the 800G module BOM, which is the core value share COHR can capture as an InP chip supplier.

COHR's Value Capture in 800G Pluggables:

As an InP EML+PD chip supplier, COHR's value capture per 800G module is approximately $70-110 (total optical chip value). If COHR also provides complete modules (rather than just chips), its ASP would increase to $200-300/module, but its gross margin (GM) would decrease from chip-level 40-50% to module-level 25-35%.

6.5.2 BOM Restructuring under CPO Architecture

CPO fundamentally reconfigures the optical module value chain. In pluggable modules, the optical module is an independent, pluggable component containing full transmit/receive/signal processing functions. In CPO, these functions are separated:

Changes in InP's Role in CPO:

In pluggable modules, InP chips perform three functions: transmission (EML) + partial modulation (electro-absorption modulation) + partial detection (PD). InP accounts for 30-40% of the BOM.

In CPO, the SiPh (silicon photonics) engine takes over modulation and detection functions, and InP retains only a single function: providing a CW (continuous wave) laser source. The value of InP lasers diminishes from "transmission + modulation" to "providing a light source only."

Derivation of InP's Proportion in CPO BOM (estimate, approximately 6.4T per CPO tile):

InP BOM Share: Pluggable 30-40% → CPO 10-15%, a decrease of approximately 60-65%.

6.5.3 Investment Implications of Value Share Changes

Will the decrease in value share per port be offset by total market growth?

Assumptions for per-port equivalence (COHR InP/CW value, model scope):

InP Value Share per Port: Pluggable $80 → CPO $7.5–11, a decrease of approximately 86–91%

This magnitude of decrease is more severe than the "30-40% down to 10-15%" cited in the initial analysis, because the initial analysis compared BOM share (module level), whereas this analysis compares the absolute dollar value per port (direct impact on COHR's revenue).

However, the total CPO market size may compensate (2030 market size and illustrative COHR revenue):

Net Effect: CPO will add $0.24-0.45B in revenue for COHR by 2030, but if CPO simultaneously cannibalizes pluggable market share, the net increase may be smaller. Key variables are whether pluggable and CPO experience "parallel expansion" (both grow) or "substitution" (CPO cannibalizes pluggable). Current industry consensus points to parallel expansion in 2027-2028, with CPO beginning to substitute some pluggable modules in 2029-2030.

Investment Judgment (Medium Execution Risk): The decline in COHR's InP value share in the CPO era is structural, with per-port contribution decreasing from $80 to $7.5-11. However, COHR is hedging through two strategies: (1) providing CW laser sources in CPO (leveraging InP manufacturing capabilities), which, despite lower unit value, benefits from higher supplier concentration (only 3-4 global players with this capability); (2) simultaneously investing in SiPh (in collaboration with Tower) to capture a larger share of the CPO BOM (SiPh engines account for 35-45%). The success of strategy (2) depends on the progress of the Tower collaboration, which is a hypothesis requiring validation in FY2027-2028.

Chapter 7: Value Creation vs. Mechanical Release — How Much of EPS Growth Is Real?

Advancing the Main Question: This is a quantitative validation of the "41x deleveraging." Revenue attribution reveals that 80%+ of incremental growth comes from a single AI engine. An EPS waterfall analysis demonstrates that approximately 48% of GAAP improvement stems from non-growth factors. An SOTP of $226.6 proves that $307.50 cannot be justified.
Operational Improvements vs. Mechanical Release: Category A (Operational) = AI growth + mix improvement + economies of scale, contributing 52% of EPS growth. Category B (Mechanical) = declining D&A + interest savings + disappearance of preferred shares + tax rate normalization, contributing 48%. The current 41x P/E prices the stock as a pure growth stock—the market has valued Category B mechanical release at the same growth multiple as Category A growth. This is the precise financial meaning of "41x deleveraging."
Implications for ROIC: Declining D&A improves GAAP profit but does not improve NOPAT (D&A is an accounting allocation of invested capital and does not affect operating cash flow). Interest savings improve EPS but do not improve ROIC (interest is a financing cost, not an operating cost). Therefore, Category B mechanical release does not drive ROIC upward—only Category A operational improvements can drive ROIC above WACC.

7.1 Financial Attribution Analysis

7.1.1 Revenue Attribution Waterfall: FY24→FY27E

COHR's revenue growth narrative appears simple: "AI-driven surge in optical module demand." However, a closer look reveals significant differences in the sources, quality, and sustainability of this incremental growth.

FY24→FY25 Revenue Bridge ($5,810M - $4,708M = +$1,102M, +23.4%)

Key Judgment (Medium Execution Certainty): Of the $1.1B increment in FY25, 80%+ comes from AI Networking (800G shipments + portfolio improvement). This means COHR's revenue growth is almost entirely solely dependent on the AI CapEx cycle. Neither the industrial segment nor the telecom segment contributed positive increments. If the AI CapEx cycle slows down before FY27, COHR has no "backup growth engine" to compensate.

FY25→FY26E Revenue Bridge ($6,959M - $5,810M = +$1,149M, +19.8%)

Key Judgment (Medium Execution Certainty): FY26E growth rate declines from +23% to +20%, as 800G ASP erosion (-$200M) begins to partially offset volume growth. This is the first signal of a volume-price squeeze. The mass production timeline for 1.6T must be on schedule, otherwise FY26 growth will fall below 15%.

FY26E→FY27E Revenue Bridge ($8,763M - $6,959M = +$1,804M, +25.9%)

Consensus accelerates to +26% in FY27, with the core assumption being full mass production of 1.6T + CPO beginning scaled shipments. This acceleration requires three things to materialize simultaneously:

1. 1.6T yield reaching mass production standards (Sherman fab 6-inch InP)
2. CPO transitioning from proof-of-concept to volume production
3. Hyperscaler CapEx growth rate remaining >20%

Probability of all three conditions materializing: We assign 55-60%, because 1.6T yield has verified progress as mentioned earlier, but both CPO and CapEx have uncertainties.

7.1.2 Gross Margin Bridge: From Trough to Recovery

FY24 GM 30.9% → FY25 GM 35.4% (Improvement +4.5pp) Driver Breakdown:

Quarterly Progress Validation (High Execution Certainty, Hard Data):

Interpretation: Within 6 quarters, GM improved from 34.1% to 37.0%, an average of +0.5pp/quarter. If this pace is maintained, full-year FY26 GM ≈37.5%, and FY27E ≈39.0%.

Counterpoints: GM improvement cannot be extrapolated beyond 40%, because: (1) 800G commoditization will compress module GM; (2) Early 1.6T yields are typically lower than mature 800G, which may cause a short-term dip in GM; (3) Industrial segment recovery could lower blended GM (if industrial recovery growth rate > AI growth rate). Our FY27 GM upper bound estimate is 39.5%, rather than the 40%+ implied by the sell-side.

7.1.3 EPS Waterfall: The Bridge from Loss to Profit

COHR's GAAP EPS dramatically shifts from -$0.52 in FY25 to a consensus Non-GAAP EPS of $5.35 in FY26E. But how much of this $5.87/share "improvement" is due to genuine operational improvement, and how much is accounting adjustments?

FY25 GAAP Profit Dissection (High Execution Certainty, 10-K Data):

FY25 $554M D&A Dissection (High Confidence):

• Intangible Asset Amortization (Merger-related): ~$336M (60.7%)
• Tangible Asset Depreciation (PP&E): ~$218M (39.3%)
• Intangible asset amortization is a one-time cost: $3,064M of intangible assets generated by the merger are amortized over 10-15 years, decreasing to ~$200M by FY29
• Thus, D&A will decrease from $554M (FY25) → $480M (FY26E) → $420M (FY27E) → ~$300M (FY29E), and this $254M reduction will directly increase GAAP EPS by approximately $1.3/share

FY25→FY26E EPS Bridge (Medium Confidence, Model Inference):

Key Judgment: Of the $4.77 improvement in GAAP EPS from -$0.52 to +$4.25, $2.50 (52%) comes from revenue growth + GM improvement (true operational improvement), $0.82 (17%) comes from D&A + interest expense decrease (dissipation of merger legacy costs), $0.80 (17%) comes from tax rate normalization (FY25 anomaly), and $0.40+(-$0.50)=-$0.10 comes from preferred conversion (slight negative net effect). Approximately half is true improvement, and half is accounting/capital structure normalization.

7.1.4 Three P/E Presentation

SBC/Revenue = $160M/$5,810M = 2.8% > 5% threshold? It does not actually trigger, but the D&A difference is huge, so we still present the three P/E ratios:

Key Finding: Non-GAAP P/E of 57.5x is the figure quoted by sell-side, but Owner P/E of 93.8x reveals an overlooked fact: after deducting SBC, the true return to shareholders per dollar of market capitalization is extremely low. The reason is not excessive SBC (only 2.3%), but rather GAAP earnings being suppressed by D&A. An Owner FCF perspective is more useful: FY25 Owner FCF = OCF $634M - CapEx $441M - SBC $160M = $33M, Owner FCF Yield is only 0.06%.

GAAP/Non-GAAP Gap Trend (High Confidence):

The gap is narrowing ($3.20→$2.36, -26% over 4 years) due to decreasing intangible asset amortization. By FY29, the gap between Non-GAAP and GAAP will mainly be SBC (~$200M/yr), approaching "normal" tech company levels.

7.2 Scissors Gap Analysis

7.2.1 Scissors Gap #1: CapEx Intensity vs. FCF Generation

COHR's CapEx intensity in FY25 was $441M, representing 70% of OCF and 7.6% of revenue. This is significantly higher than typical optical components companies (Lumentum 4-5%, II-VI pre-merger 5-6%), because COHR is simultaneously making three investments: (1) Sherman 6-inch InP capacity expansion ~$150M; (2) SiC material production capacity construction ~$100M; (3) Regular maintenance + industrial facility upgrades ~$190M.

Why This is a Scissors Gap Problem: CapEx rebounded from $347M (7.4% rev) in FY24 to $441M (7.6% rev) in FY25, but FCF conversely decreased from $199M to $193M. Revenue grew +23% while FCF remained flat—because $880M of incremental revenue was absorbed by $94M of incremental CapEx and working capital build.

Assessment (Medium Execution Risk): If the AI CapEx cycle peaks before FY28, COHR's Sherman InP expansion ($200M+) and SiC build-out (cumulative $300M+) will face capacity utilization risks. Historical analogy: After optical communications CapEx peaked in 2019, Lumentum's CapEx decreased from $188M (FY19) to $100M (FY20), but utilization also fell from 85% to 60%, leading to an 8pp decline in GM due to fixed cost deleveraging. COHR has a larger fixed cost base ($1.9B PP&E), meaning a similar decline in utilization would have a greater impact.

7.2.2 Scissors Gap #2: GAAP vs Non-GAAP EPS (Narrowing)

This is a positive scissors gap.

FY25 GAAP EPS -$0.52 vs Non-GAAP $3.50, a gap of $4.02/share. By FY26E, GAAP ~$4.25 vs Non-GAAP ~$5.35, narrowing the gap to $1.10. By FY29E, the gap will further narrow to $1.20 (primarily driven by SBC).

Mechanism: The core driver for the narrowing gap is the roll-off of intangible asset amortization. The $3,064M in intangible assets generated by the II-VI merger are amortized over 10-15 years, peaking at approximately $400M in FY23 and decreasing to ~$190M by FY29. This $210M reduction is "automatic"—it requires no operational improvements, purely a function of time.

Investment Implication: GAAP P/E declines from 72.4x (FY26E) to 40x (FY28E), even without operational improvements, purely decreasing D&A contributes 15% of P/E compression. This is a favorable re-rating catalyst for investors focusing on GAAP earnings (e.g., index funds). However, it also means: sell-side analysts valuing COHR using Non-GAAP P/E are effectively "double-counting" some good news—Non-GAAP already adds back amortization, but the decreasing amortization is then presented again as a "growth story."

7.2.3 Scissors Gap #3: Hyperscaler CapEx Growth vs COHR Revenue Growth

Hyperscaler CapEx is approximately $380B in 2025, and approximately $690B in 2026E, +82%. COHR FY25 revenue is $5.81B, FY26E $6.96B, +20%. Hyperscaler growth is 4 times that of COHR.

Why this gap? Because optical components only account for 3-5% of Hyperscaler CapEx. Of the $690B CapEx, ~$250B is for construction/power infrastructure, ~$300B is for GPU/servers, ~$100B is for networking (including optics), of which optical transceivers + components are approximately $20-35B. COHR holds a 15-20% share in this $20-35B market.

Risk of this scissors gap: When Hyperscaler CapEx decelerates from +82% to +10% (2028E consensus), the deceleration of optical components will be amplified, because: (1) Inventory cycle—Hyperscalers over-procure optical transceivers (buffer stock) during CapEx peaks, then digest inventory before placing new orders when CapEx slows; (2) Price elasticity—CapEx slowdown implies supply-demand balance, eliminating pricing power.

Historical analogy: In 2019, Hyperscaler CapEx was -3%, while Lumentum Datacom revenue was -22% and Inphi optical chip revenue was -18% in the same year. The amplification factor was approximately 6-7x. If Hyperscaler CapEx growth decreases to +5% in 2028, COHR Datacom revenue growth could fall from +25% to between -5% and +5%.

7.2.4 Scissors Gap #4: Inventory Turnover vs Revenue Growth

FY25 inventory is $1,438M, +12% YoY, while revenue is +23%. Superficially healthy (inventory growth < revenue growth). However, DSI (Days Sales of Inventory) is still 140 days, significantly higher than Lumentum (85 days) and Innolight (60 days).

Why is DSI so high? Because COHR is vertically integrated—from InP substrates to chips to modules, every layer has work-in-progress (WIP). Lumentum primarily makes chips, and Innolight primarily does module assembly, resulting in shorter inventory cycles. Of COHR's 140 days DSI, ~50 days are raw materials (InP, SiC substrates), ~60 days are WIP (wafer processing), and ~30 days are finished goods.

Risk: If AI demand slows, approximately $400-500M of WIP and finished goods within the $1.44B inventory face markdown risk. If 800G modules become commoditized, inventory write-downs could reach $100-200M. This risk is not evident in FY25 because demand remains strong (25-30% supply deficit), but it is a key performance indicator (KPI) to monitor in FY27-28.

7.2.5 Scissors Gap #5: R&D Investment vs Revenue Growth (Positive Signal)

FY25 R&D is $582M, accounting for 10.0% of revenue. FY24 R&D is $479M, accounting for 10.2% of revenue. The absolute value of R&D grew +22%, but its proportion slightly decreased by -0.2pp. This indicates that COHR is gaining OpEx leverage—R&D output efficiency is improving.

However, there is a hidden concern: Of the $582M in R&D, how much is allocated to maintaining existing products (800G EML) and how much to next-generation products (1.6T, CPO, SiC devices)? The company does not disclose these breakdowns. If >50% of R&D is for maintenance, then "efficiency improvement" is actually a decline in innovation investment. Considering LITE's R&D/Revenue was 14.2% during the same period, whether COHR's 10% is sufficient to maintain technological leadership is an open question (Medium execution risk, requires more evidence).

7.3 SOTP Valuation — Three Engines Independently Priced

7.3.1 Engine 1: AI Networking (FY27E basis)

Revenue Estimate: FY27E AI Networking revenue of $5.0-6.5B, depending on 1.6T ramp speed and CPO contribution.

Multiple Selection:

• Comparable Companies: LITE trades at ~8-10x EV/Rev (FY27E), but LITE has 200G/lane EML performance leadership → COHR should be discounted by 10-20%
• AI Optical Communications Pure-play: Market awards 6-9x EV/Rev
• COHR Specific Factors: (1) 6-inch InP cost advantage (positive); (2) Intense competition at the module layer (negative); (3) NVIDIA investment lock-in (positive but non-exclusive)
• We use a 5.0-8.0x range, corresponding to bear/base/bull scenarios

7.3.2 Engine 2: SiC Materials (Option Pricing)

SiC is the most challenging part of COHR to value, as it is in an investment phase (break-even or slight loss) but has a massive potential market ($20B+ by 2030).

Comparable Anchors:

• Wolfspeed (WOLF): Focuses on SiC substrates + devices, market cap ~$2B, but faces balance sheet challenges (high leverage, potential restructuring)
• ON Semi SiC business: Implied valuation ~$5-8B (SOTP breakdown), but ON has an established silicon business foundation
• COHR SiC characteristics: In-house substrate manufacturing (II-VI legacy), but scale is far smaller than ON Semi, devices are still in early stage

Option Valuation Logic: SiC revenue FY26E ~$450M. If successfully expands to $1B+, it would be worth $5-8B; if SiC oversupply leads to a price war (Wolfspeed is already cutting prices), it might only be worth $1.5B (approx. 3x revenue, typical for low-margin materials companies).

7.3.3 Engine 3: Industrial Segment (Cyclical Stock Mid-Cycle)

The Industrial segment includes industrial lasers, precision optics, and materials processing, which are the core businesses of traditional II-VI/Coherent. FY26E revenue ~$1.9B, mid-cycle OPM 8-12%.

Multiple Selection: Comparable industrial lasers (IPG Photonics, Trumpf implied): 8-12x mid-cycle EV/EBITDA.

Note: EBITDA = Revenue × (OPM + 5% D&A/Rev), where 5% is an estimate of D&A as a percentage of revenue for the Industrial segment.

7.3.4 SOTP Assembly + Probability Weighting

Sum-of-the-Parts EV (Bear / Base / Bull), diluted share count at 165M (consistent with the matrix below):

Probability Assignment (Triple Anchoring):

Probability-Weighted Fair Value: $249.7/share
Current Price: $307.50
Downside Potential: -18.8%

7.3.5 Reverse DCF: What is $307.50 Buying?

Using a WACC of 10%, terminal growth of 3%, and terminal EBITDA margin of 20%, a reverse DCF implies FY30 revenue of $27.1B for $307.50, corresponding to a 5-year CAGR of 36.1%.

Is this implied growth rate reasonable? The consensus 3-year CAGR (FY25→FY28) is 21.7%. To achieve a 5-year CAGR of 36.1%, FY29-FY30 would need to maintain a growth rate of 61%—this is almost impossible, unless terminal value assumptions are completely changed.

Even if assumptions are relaxed (WACC 9%, terminal EBITDA margin 25%, terminal multiple 18x), the implied 5-year CAGR still needs to be ~22%, slightly higher than the 3-year consensus. Conclusion: The current price at least prices in the perfect execution of the consensus trajectory and implies a premium valuation in the terminal phase.

7.4 Capital Efficiency and Owner Economics

7.4.1 Owner FCF: The Reality Obscured by CapEx and SBC

FY25 Owner FCF Breakdown (OCF − CapEx − SBC):

Owner FCF is essentially zero, meaning that at a market cap of ~$50.7B, the true cash return for shareholders is <0.1%. This is not because the business is unprofitable (EBITDA $1.1B), but because: (1) $441M CapEx is invested for future growth (Sherman InP, SiC); (2) $160M SBC is diluting existing shareholders; (3) $243M in interest is servicing merger leverage.

What is needed for Owner FCF to improve? (The following is an illustrative projection, not company guidance)

7.4.2 ROIC: Still Below the Hurdle

FY25 ROIC and Invested Capital (Illustrative Basis):

ROIC of 4.2% < WACC of 10% means that COHR is currently destroying value on its invested capital base. This is primarily driven by the substantial post-merger invested capital ($11.1B, including $7.7B goodwill + intangibles).

When will ROIC exceed WACC? NOPAT needs to reach $1.1B, corresponding to an operating profit of $1.3B+ (assuming a 15% tax rate). At an 18% OPM, revenue of $7.2B is needed; at a 20% OPM, $6.5B is needed. FY27E consensus of $8.76B + anticipated OPM improvement could allow ROIC to exceed WACC for the first time.

Counterpoint: Of the $11.1B in invested capital, $7.7B is goodwill + intangibles, which one school of thought believes should be excluded from invested capital (as they represent M&A premiums, not operating assets). If excluded, invested capital = $3.4B, ROIC = 13.7% → already exceeds WACC. We believe they should not be excluded because: (1) management chose this acquisition, and investors' true capital includes this decision; (2) if goodwill is impaired, the loss is real.

7.4.3 Debt Deleveraging Progress

Deleveraging progress: FY23 ND/EBITDA 3.8x → FY25 2.7x → FY27E ~0.7x. In FY25, $435M of debt was repaid, and interest expense decreased from $289M (FY24) to $243M (FY25), estimated to be $180M in FY26E and $120M in FY27E.

This is the most certain positive factor in the COHR story: The reduction in interest of $120M (FY25→FY27E) directly translates into an EPS improvement of $0.62/share and does not depend on any operating assumptions.

7.5 The 5 Most Important Findings (Ranked by Decision Value)

1. SOTP Weighted $250, Current -18.8% Discount — The market is assigning a uniform P/E to three segments, essentially valuing the Industrial and SiC segments as if they were AI. Breaking it down, only the bull case ($358) supports the current price

2. Owner FCF ≈ Zero, ROIC 4.2% < WACC — At a ~$50B market cap, shareholders' true cash return is <0.1%. Investors are not buying current cash-generating capability, but rather future earnings 3-5 years out

3. Over 80% of Revenue Growth Relies on a Single AI Engine — There is no backup growth engine. AI CapEx cycle peaking risk = COHR growth engine stalling risk, with a historical amplification factor of 6-7x

4. Narrowing GAAP/Non-GAAP Gap is the Most Certain Catalyst — D&A decrease from $554M → $300M happens automatically and does not require operational improvement. This is favorable for GAAP P/E compression and for attracting passive funds

5. Price-Volume Scissors Gap to Appear in FY27-28 — 800G ASP is already declining; 1.6T must take over in time, otherwise revenue growth will be significantly below consensus

Chapter 8: Addressing Financial Gaps

This chapter further validates three key valuation judgments: (1) Owner FCF for FY25 is only $33M (yield 0.06%), confirming that the "deleveraging" component of "41x deleveraging" has not yet translated into true cash returns; (2) Although SBC of $170M/year accounts for only 2.3% of revenue, its relative dilutive impact is enormous given the near-zero Owner FCF base (Owner P/E 93.8x vs. Non-GAAP P/E 57.5x); (3) Inventory of $1,848M with a DSI of 159 days far exceeds peers; vertical integration explains the structurally high DSI, but downward rigidity (inventory +1% while revenue -9% from FY23→FY24) means impairment is unavoidable during a cyclical downturn. These three findings collectively point to the same conclusion: Non-GAAP P/E of 57.5x conceals the true state of Owner Economics.

8.1 Owner FCF Forward Projection FY25→FY28E

8.1.1 Owner FCF Projection Model

Key Assumption Notes:

• OCF/EBITDA Conversion Rate: FY25 actual 57% (OCF $634M / EBITDA $1,106M), as working capital build and interest payments consumed 43% of EBITDA. With slower WC growth and declining interest, the FY28E conversion rate is expected to rise to ~57%
• CapEx: FY26E increases to $500M (peak for Sherman+SiC expansion), gradually falling back to $450M in FY27-28 (maintenance+selective growth), CapEx/Rev decreasing from 7.6% to 4.3%
• SBC: Maintained at 2.5%/Rev, absolute value slowly increasing

Assessment (Medium Execution): Owner FCF only reaches $480M in FY28E, with a yield of merely 0.95%. Even after 3 years, COHR's Owner FCF yield is less than 1% — In contrast: LITE's estimated Owner FCF yield is ~2-3% for the same period, and optical communications ETFs average 3%. Investors in COHR are buying forward earnings for the 4th-5th year, not cash generation for the next 3 years.

8.1.2 Owner FCF vs. Traditional FCF Differences

Traditional sell-side uses FCF (=OCF-CapEx) as a metric; FY25 FCF=$193M, FY26E $250M, FY27E $465M. However, these figures do not deduct SBC, which is misleading for true shareholders: SBC dilutes shares by 1-1.5% annually, accumulating to 5% dilution over 4 years, directly eroding per-share value.

The difference entirely stems from SBC, and it is widening (SBC absolute value growth). When the sell-side uses traditional FCF for valuation, they overstate it by $160-200M annually, accumulating to an overstatement of $725M over 4 years — using a 10x FCF multiple, this overvalues equity by $7B (approximately $42/share).

8.2 SOTP Sensitivity Analysis

8.2.1 Single Variable Sensitivity: AI Networking EV/Revenue Multiple

AI Networking accounts for 85-88% of SOTP, making its multiple the most sensitive variable:

Note: The table shows fair value per share (deducting net debt of $2.2B, SiC $3.0B, Industrial $2.9B constant). Shaded areas represent prices above the current $307.50.

Key Insight: The current $307.50 is only justifiable under a combination of AI Networking multiple ≥9x + Base revenue or multiple ≥8x + Bull revenue. A 9x EV/Rev is a level currently only justifiable by LITE and Broadcom's optical communications businesses in the market. COHR lacks LITE's EML performance leadership at the 800G component level and Broadcom's DSP integration at the system level, thus lacking support for a 9x multiple.

8.2.2 Two-Variable Sensitivity: AI CapEx Growth vs. 1.6T Share

Conclusion: $307.50 requires a combination of CapEx +30% and 1.6T share ≥25% — this necessitates optimal results on both the demand side (sustained strong CapEx) and the supply side (COHR winning more 1.6T share). If either side falls short of expectations, the valuation will not support the current price.

8.3 Debt Structure and Deleveraging Path

8.3.1 Debt Maturity and Refinancing

Based on FY25 10-K and Q2 FY26 earnings report:

Deleveraging Trajectory (Q2 FY26 Confirmed):

Quantification of Interest Expense Savings:

This is one of the most certain EPS drivers in the COHR story: FY25→FY28E interest savings of $143M, directly translating to +$0.74 EPS, independent of any revenue or margin assumptions.

8.3.2 Refinancing Risk Assessment

Short-term Risk (FY27 Revolver Maturity): The $500M revolving credit facility matures in 2027, but with COHR's current credit profile (ND/EBITDA ~1.5x) and $864M in cash, an extension is almost certain.

Mid-term Risk (FY29 Term Loan B): The $2.8B Term Loan B matures in 2029. At the current repayment rate, the balance at maturity will be approximately $1.5-2.0B. Considering FY29E EBITDA of ~$2.0B, ND/EBITDA <1.0x, the refinancing risk is low. However, if the AI cycle peaks in FY28 and EBITDA declines to $1.2B, ND/EBITDA could rise to 1.5x — still manageable, but interest rates might increase by 50-100bp.

8.4 In-depth Working Capital Analysis — Inventory Red Flags

8.4.1 Abnormal Inventory Growth in Q2 FY26

FY25 data (inventory $1,438M, +12% YoY, "healthy"), but the Q2 FY26 earnings report (December 2025) shows inventory has surged to $1,848M, an increase of +$410M (+28.5%) in six months.

Concurrent revenue growth: Q4 FY25 $1,529M → Q2 FY26 $1,686M, +10.3% in six months.

Inventory growth (+28.5%) is 2.8 times revenue growth (+10.3%) — this is the 6th, and most alarming, divergence in addition to the five divergences summarized previously.

8.4.2 Two Interpretations of Inventory Growth

Positive Interpretation (Management Narrative): 1.6T is nearing mass production (FY26H2), requiring advance material preparation (InP substrate processing cycle is long, ~3-4 months). NVIDIA's $2B investment includes multi-year purchase commitments, necessitating safety stock. The Sherman factory production line ramp requires WIP (Work In Progress) fill.

Negative Interpretation (Our Concerns): Of the $410M increase, if >$150M consists of 800G finished goods/WIP → 800G will face impairment after commoditization. FY25 DSI was already 140 days (key-metrics calculation), and Q2 FY26 is even higher. Historical analogy: In 2018, Lumentum's inventory rose from $120M to $280M after the acquisition, followed by a $40M impairment in 2019 when demand weakened.

8.4.3 Comprehensive Working Capital Analysis

Implications of significant AP increase: AP surged from $632M in FY24 to $1,119M in Q2 FY26 (+77%), far exceeding revenue growth. This indicates COHR is extending its payment terms with suppliers (DPO from 70 days → ~90 days), a strategy to manage cash flow by leveraging its negotiation power within the supply chain. While beneficial for OCF in the short term, this is unsustainable—if suppliers start demanding shorter payment terms (e.g., during a demand slowdown), OCF will face an adverse impact.

8.4.4 Inventory Impairment Sensitivity

Probability-weighted impairment: $100M×30% + $300M×15% + $500M×5% = $30M + $45M + $25M = $100M (or -$0.52/share). This risk is not reflected in current consensus EPS.

Chapter 9: Market Expectation Validation — What Is COHR Buying, and What Is It Missing?

Advancing the Core Questions: The market is buying AI growth and also quietly buying InP scarcity premium (70% supply-demand gap), but it is not separately discounting Industrial or transparently valuing the SiC option. Competitive analysis confirms two judgments—COHR's advantage lies in chip-level cost, not module-level share (thus margin expansion comes from cost reduction, not pricing power), and CPO does not kill InP but changes InP's role (BOM proportion decreases but unit value increases).
Implications for ROIC: The 70% InP supply-demand gap is the only structurally supported part of current ROIC—it allows COHR to maintain margins during periods of tight supply. However, the gap will narrow in 2028+ with SiPh substitution and capacity expansion, at which point ROIC's margin support will weaken. ROIC must exceed its cost of capital before the gap narrows, otherwise the window closes.

9.1 1.6T Competitive Landscape — Who Is Winning, and Who Will Win?

9.1.1 800G→1.6T Share Migration: From "Volume Game" to "Layer Differentiation"

The competitive structure of the 800G market is migrating towards 1.6T, but the migration is not a simple share transfer—the market is splitting into two competitive layers along technology lines:

Layer 1: Assembled Module Layer — Low barriers, share follows capacity and price

• Innolight + Eoptolink collectively account for ~60% of NVIDIA's 800G orders
• Because: Chinese manufacturing costs are 20-25% lower, capacity expansion is rapid, and module assembly does not require core IP
• 1.6T continuation: Innolight has completed NVIDIA 1.6T qualification tests and is expected to account for 50-60% of module share
• Conversely: Geopolitical risk is a structural discount for Chinese suppliers—NVIDIA's $4B investment in Western suppliers is a hedging signal

Layer 2: Core Chip/Laser Layer — High barriers, share follows technology and capacity

• Global EML capacity gap: Demand ~40 million units vs insufficient capacity
• COHR 6-inch InP wafers: 4x output per unit area (vs 3-inch), manufacturing cost down 60%+
• LITE 200G/lane EML: Currently the only mass-production supplier, 12-18 months performance lead
• Therefore: The laser layer is the real bottleneck, and COHR and LITE have pricing power in this layer—Innolight does not.

This stratification explains a seemingly contradictory fact: Innolight has the highest market share but its valuation multiple is lower than COHR/LITE. This is because the market is pricing "Layer 2 barriers" rather than "Layer 1 share." If only module share were considered, COHR should be much cheaper; if pricing power in the laser layer is considered, the current valuation has its logical basis.

9.1.2 NVIDIA's $4B Photonics Investment: Supply Chain Lock-in or Label Gilding?

On March 2, 2026, NVIDIA simultaneously announced a $2B investment in COHR + $2B investment in LITE:

COHR deal structure:

• Purchased ~7.8 million shares @ $256.80/share (cash)
• Non-exclusive multi-year supply agreement + multi-billion dollar procurement commitment
• Execution period: Early 2027 to 2030
• Funds directed towards: Sherman, Texas InP capacity expansion

LITE deal structure:

• Purchased ~2.876 million shares of Series A convertible preferred stock @ $695.31/share
• Similar structure: Non-exclusive + multi-year procurement commitment + capacity access rights
• Funds directed towards: New US laser factory

What NVIDIA did not obtain: Exclusivity, board seats, price control. The agreement explicitly states "nonexclusive."

The true meaning of the investment — not "labeling" COHR/LITE, but rather:

1. CPO Capacity Reservation: The $4B secures high-power InP CW laser capacity required for CPO from 2027-2030, not current pluggable modules.
2. Geopolitical Hedging: Shifting the 60% reliance on Chinese suppliers in the optical module supply chain towards Western vendors.
3. Supply Chain Insurance: EML lasers are the true bottleneck for AI infrastructure; NVIDIA using capital to secure capacity is more effective than price negotiation.

Implications for COHR valuation: The $2B investment, calculated at $256.80/share, corresponds to a market capitalization of ~$40B. NVIDIA deemed COHR worthy of a $2B strategic investment at a $40B market capitalization. This is not proof that "COHR is worth $40B" (NVIDIA is buying supply chain security, not investment returns), but it confirms COHR's irreplaceability in the AI optical communications supply chain. If COHR were easily replaceable, NVIDIA would not use $2B to secure it.

9.1.3 Competitor Benchmarking

Key Insight: The competition between COHR vs LITE is not zero-sum—NVIDIA investing in both indicates: (1) Their technologies are complementary, not substitutable (2) The market is large enough to accommodate both (3) The real competition is between "NVIDIA-locked Western supply chains" vs "non-NVIDIA-locked Chinese supply chains."

9.1.4 1.6T Technology Path Debate: SiPh vs EML

Conclusion: SiPh holding 60% of the 1.6T market means EML will not disappear, but it will not grow. Because SiPh still requires an InP CW laser as an external light source, COHR's InP manufacturing capability is valuable in both paths—as an EML supplier (Path 1) or a CW laser supplier (Path 2). LITE is more reliant on the EML path, concentrating its path-specific risks.

9.2 Supply Chain Cross-Validation

9.2.1 InP Substrate: The True Bottleneck for AI Optical Communications

Severe imbalance in InP substrate supply and demand:

Three Cross-Validation Signals:

Signal 1: LITE Backlog Extends to 2028 — Lumentum reported its backlog extending beyond 32 months as of April 2026. This means capacity through the end of 2028 has been locked in. If LITE (which does not self-produce InP substrates) can fill its backlog, it indicates that InP supply is not a constraint for LITE—or that LITE has already secured supply from AXT/Sumitomo.

Signal 2: AXT Geopolitical Exposure — China implemented export licensing controls on indium in February 2025. AXT's 60-70% market share + Beijing production = the largest single point of failure in the global InP supply chain. If China restricts AXT exports, COHR's 6-inch line immediately becomes "irreplaceable" rather than just a "cost advantage."

Signal 3: COHR Capacity Expansion Plan — COHR plans to double its InP capacity in CY2026, and then double it again. OFC 2026 announced that a third 6-inch line is under construction in Zurich, Switzerland. Three 6-inch lines (Sherman TX + Järfälla Sweden + Zurich Switzerland) are distributed across three continents, providing geopolitical diversification.

Supply Chain Validation Conclusion: COHR's InP vertical integration has upgraded from a "cost advantage" to a "strategic imperative." Reasons: (1) 70% supply/demand gap → capacity is a moat (2) AXT geopolitical risk → scarcity of non-China capacity (3) NVIDIA $2B investment → customers are willing to use capital to secure capacity. This explains why COHR's 41x P/E looks expensive but the market is still buying—the market is pricing in InP capacity scarcity, not just optical module growth.

9.2.2 Hyperscaler CapEx: Demand Engine or Ticking Time Bomb?

2026 Big Five CapEx Guidance:

Goldman Sachs predicts: Cumulative Hyperscaler CapEx for 2025-2027 will reach $1.15 trillion, 2.4 times the $477B from 2022-2024.

Scissors Gap Alert (R-2 Extension): Hyperscaler revenue growth rate of 16.5% (2025) vs CapEx growth rate of 60% (2025). 2026 is even more extreme: 15.5% revenue growth vs 80% CapEx growth. The CapEx growth rate is 5 times the revenue growth rate—this ratio has historically never sustained for more than 6-8 quarters.

• If the CapEx growth rate in 2027 drops to the same magnitude as the revenue growth rate (approx. 15%), it means that the absolute value of CapEx could still grow, but the growth rate would decrease by about 65 percentage points compared to the previous year
• Optical module demand lags GPU installation by 6-9 months—even if CapEx peaks in mid-2026, optical module demand could sustain through 2027 H1
• However, COHR's revenue is 80%+ dependent on a single AI engine (confirmed by analysis in Chapter 8), and a CapEx slowdown would have a 6-7x amplified impact on COHR (confirmed earlier in Chapter 8)

Microsoft Pause Signal: TD Cowen reported that Microsoft canceled multi-gigawatt data center leases in the US and Europe and paused construction. This is not a full retreat (more about power availability management), but it is the first Hyperscaler deceleration signal.

9.2.3 Customer Concentration Cross-Validation

Innolight's NVIDIA dependency of >50% is extremely concentrated. Due to COHR's diversification in Industrial/SiC, single customer risk is relatively lower. However, COHR's AI Networking segment (55-60% of total revenue) is also highly dependent on NVIDIA.

9.3 CPO (Co-Packaged Optics) — COHR's Second Growth Engine or Moat Erosion?

9.3.1 CPO Timeline: Further Than Market Expectations

Key Judgment: CPO will not kill the pluggable market in 2026-2027. COHR management's CPO timeline (Scale-out H2'26, Scale-up H2'27) represents the start of revenue generation, not a pluggable replacement. The primary demand for pluggable up to 1.6T is still accelerating in 2026-2028, and CPO represents incremental growth from 2028+.

9.3.2 What CPO Means for COHR: InP Lasers Are Not Dead, Perhaps Even More Important

Core Argument: Broadcom's CPO architecture (External Laser Source, ELS) still requires InP CW lasers — and the requirements are higher:

• Pluggable: Each module requires 1-8 low-power EMLs
• CPO ELS: Each light source requires fewer but higher power (400mW) CW lasers
• COHR is developing 400mW CW InP lasers specifically designed for CPO

Precise Analysis of InP BOM Changes:

Conclusion: CPO is not as simple as "InP BOM decreasing from 30-40% to 10-15%." This is because:

1. Lasers do not disappear — Silicon cannot efficiently generate photons, InP remains the only choice
2. CPO unit price is higher — High-power CW lasers are more expensive than low-power EMLs, partially offsetting the decrease in BOM percentage
3. Overall volume growth — The number of optical connections in the CPO era will increase by 10-100x; even if the unit InP content decreases, total InP demand will still grow
4. COHR's 6-inch advantage is amplified in the CPO era — because CPO requires more consistent large-scale laser production, the yield and cost advantages of 6-inch lines are magnified

9.3.3 Ayar Labs: The True Disruptor Risk

Ayar Labs completed a $500M Series E in March 2026 (valuation $3.75 billion), with investors including NVIDIA, AMD, and Intel Capital.

Key Differences Between Ayar and Broadcom CPO:

• Broadcom CPO: SiPh PIC placed next to switch chip + external InP laser → InP still required
• Ayar CPO: Optical I/O directly integrated into chip package → InP demand minimized

If the Ayar model becomes mainstream, COHR's InP laser business faces a greater impact. However:

• Ayar is currently pre-revenue, with mass production earliest in 2027-2028
• NVIDIA has invested in both Ayar and COHR/LITE — indicating NVIDIA is hedging, not betting on a single path
• Even if Ayar succeeds, the transition period (2028-2032) will still require a significant amount of pluggable + Broadcom-style CPO

Probability Assessment: The probability of the Ayar model accounting for >30% of CPO market share before 2030 is: ~15-20%. Based on: (1) Historically, the average time from PoC to mass production is 5-7 years (2) NVIDIA investing in three paths simultaneously = high uncertainty (3) TSMC/advanced packaging capacity constraints.

9.4 Game Theory Lens — NVIDIA vs. Optical Module Supplier Power Balance

9.4.1 Identification of Game Structure

Players: NVIDIA (Buyer) vs COHR+LITE (Western Suppliers) vs Innolight+Eoptolink (Chinese Suppliers)
Game Type: Incomplete Information Multi-Party Bargaining Game + Exogenous Geopolitical Risk Shock

NVIDIA's Strategy Space:

1. Maintain Status Quo (60% China/40% West) — Lowest cost, highest geopolitical risk
2. Shift to Western Suppliers (Increase COHR/LITE share) — Cost increases by 20-25%, geopolitical risk decreases
3. Full Lock-in with Western Suppliers (Actual choice with $4B investment) — Highest cost, but ensures supply security

NVIDIA choosing Strategy 3 tells us:

• NVIDIA's internal probability estimate of geopolitical risk is higher than market consensus (otherwise the cost increase is unreasonable)
• NVIDIA's estimated loss from optical module supply disruption is extremely high (daily GPU production halt loss >> $4B annualized interest)
• The $4B investment is an "insurance premium", ROI comes from avoiding supply disruption, not from COHR/LITE stock price appreciation

9.4.2 Pricing Power Game: Who Controls Whom?

Key Game Insight: NVIDIA's $4B investment changed the game structure—from "suppliers having pricing power due to capacity scarcity" to "suppliers having capacity assurance due to NVIDIA's lock-in, but NVIDIA gaining price terms in the procurement commitment". This is a classic case of Nash Bargaining: neither party is entirely satisfied, but both are better off than without an agreement.

COHR Gains: Capacity expansion funding + Demand certainty (multi-year purchase commitment) + Valuation support
NVIDIA Gains: Supply security + Capacity priority + Geopolitical hedging
COHR Forfeits: Partial pricing flexibility (purchase commitments typically include price caps)
NVIDIA Forfeits: $4B capital + Non-exclusivity (competitors can also buy COHR products)

9.4.3 Chinese Suppliers' Prisoner's Dilemma

Innolight and Eoptolink face a classic Prisoner's Dilemma:

• Cooperate (Maintain Share + Price): Continue low-price mass production for NVIDIA, retain 60% share
• Defect (Raise Prices): Take advantage of current supply-demand imbalance to raise prices, but NVIDIA accelerates shift to COHR/LITE
• Betrayed (Geopolitical Shock): Chinese policy restricts exports, forcing share transfer

Current Equilibrium: Chinese suppliers choose "cooperation" (low prices to maintain share) because: (1) The benefit of raising prices < the loss from being replaced (2) NVIDIA's $4B investment is a "credible threat"—if you raise prices, I will use the already locked-in COHR/LITE capacity to replace you (3) Geopolitical risks are beyond Innolight's control

Implication for COHR: Chinese suppliers are locked into a low-price strategy, meaning module-layer ASPs will continue to decline. However, COHR does not primarily compete at the module layer—COHR competes at the laser layer, where the supply-demand imbalance is more severe and pricing power is stronger. Layer mismatch is COHR's structural advantage relative to Innolight.

9.5 Moat Evolution — From "Cost Advantage" to "Strategic Necessity"

9.5.1 Moat Timeline: Varying Strengths Across Different Time Scales

9.5.2 Moat Upgrade: Revisions to Initial Assessment

Using a baseline of Chapter 4 Six-Dimensional Arithmetic Mean 3.2/5 (improving). Third-stage new evidence revisions:

Upward Revision Factors:

• NVIDIA $2B investment confirming irreplaceability: +0.3
• InP substrate 70% supply-demand gap: +0.2 (Upgraded from cost advantage to capacity barrier)
• AXT geopolitical risk: +0.2 (COHR non-China capacity becomes strategic asset)

Downward Revision Factors:

• SiPh share in 1.6T rises to 60%: -0.1 (EML roadmap pressured)
• Ayar Labs funding acceleration: -0.1 (Long-term disruptive risk)

Revised Composite Moat: 3.2 + 0.7 - 0.2 = 3.7/5 (Upgraded from "Medium-High, improving" to "Medium-Strong, rapidly strengthening short-term")

9.6 Chapter Summary: What Changed in Round 3 (Competition and Scenarios)

What Changed:

1. Moat upgraded from "cost advantage" to "strategic necessity" — NVIDIA $2B + InP scarcity + geopolitical hedging
2. CPO is not a threat but an opportunity (2026-2028) — ELS architecture still requires InP CW lasers
3. Competitive layering: COHR has pricing power at the laser layer, does not directly compete with Innolight at the module layer

What Didn't Change:

1. SOTP -18.8% valuation pressure
2. AI single-engine 80%+ revenue dependency
3. CapEx scissors gap (5x growth vs revenue) cyclical risk
4. CQ still <60%

What Round 4 (Adversarial Review) Needs to Answer:

1. Red Team 7 Questions — Especially "AI CapEx Collapse Scenario" and "Tag Collapse"
2. Two-way calibration: Upside (AXT regulations + COHR benefit) vs Downside (CapEx collapse + 6-7x amplification)
3. Probability weighting needs revision (moat upgraded but valuation pressure unchanged)

Chapter 10: Revisions to Competition and Valuation Assumptions

This chapter consolidates a re-examination of six key assumptions from the preceding analysis: competitive layering, gross profit sources, NVIDIA investment, Industrial business, valuation difference relative to LITE, and several tail probabilities.

10.1 Module Layer Competition and the "Pricing Power" Statement

Reasoning: After COHR produces InP, it assembles 800G/1.6T pluggable modules, competing with Innolight, LITE, and others for the same customers and product forms; chips are not primarily sold separately at a distinct price. If "laser layer pricing power" is understood as the ability to raise chip prices for downstream module manufacturers, it would misinterpret the competitive boundaries.

Conclusion: The advantage lies in cost and supply certainty from vertical integration (6-inch, internal transfer pricing), reflected in module gross margins rather than module list prices. Innolight gains market share with module prices approximately 20–25% lower; COHR finds it difficult to defend by raising prices, but can maintain higher gross margins at comparable selling prices. NVIDIA's strategic investment purchases capacity and delivery certainty, not a premium for high-priced modules.

10.2 Gross Profit and Revenue: Mix, Cost, and Volume-Price Environment

Reasoning: The price range for 800G multimode/single-mode coexists with discounts from Asian suppliers, leading to a generally flat or downward trend in ASP. Gross margin improvement primarily stems from the product mix shift from 800G to 1.6T, 6-inch cost reduction, decreasing depreciation and amortization, and improved capacity utilization, rather than "pricing power."

Conclusion: The "scissor gap" between shipment volume and ASP is consistent with competitive pricing pressure—growth relies on volume and structural upgrades, making it difficult to offset slowdowns in volume with price increases. During a CapEx downturn phase, the moat lies in cost efficiency and product structure, not in pricing.

10.3 NVIDIA Investment and Per-Share Valuation

Reasoning: Approximately 7.78 million shares, representing about 5% dilution, corresponds to an approximate dilution of ~$0.25/share on an annualized EPS basis. If enterprise value remains constant and the share count increases from approximately 165 million to 173 million, the SOTP per share would decline to around $239, representing a discount of approximately −22% relative to the current price (previously approximately −19%). If the purchase commitment translates into incremental Networking revenue, it could partially offset this.

Conclusion: The overvaluation trend remains unchanged; the reasonable range should be expressed as a discount band of **−18% to −22%**, reflecting the uncertainty of dilution and order fulfillment.

10.4 Industrial: Approximately One-Third of Revenue and Structural Discount

Reasoning: Industrial accounts for approximately one-third of revenue, with growth close to zero and gross margin roughly 10 percentage points lower than Data Center. It overlaps with TRUMPF, IPG, Chinese manufacturers, etc., in various sub-segments. While excimer and other sub-segments offer differentiation, it is insufficient to drive the overall segment.

Conclusion: This segment is not a short-term "fixable item," but rather a structural factor that drags down blended growth and profit margins, thereby depressing comparable P/E multiples. The valuation gap with LITE highlights that business purity is one of the main explanations.

10.5 Relative to LITE: Why the P/E Difference is Understandable

Reasoning: LITE has higher revenue growth and AI exposure, and its Non-GAAP gross margin is already around 42.5%. The market is paying a premium for non-linear growth and "realized margins." Although COHR is larger in scale, its growth rate and mixed business operations depress its valuation multiple.

Conclusion: The P/E difference primarily stems from (1) growth rate, (2) business purity, and (3) position on the margin curve. The 6-inch cost advantage has not yet been fully reflected in reported gross margin. If gross margin cannot approach 42%+ by around FY28, the discount may become long-term rather than automatically disappearing with the 6-inch ramp-up.

10.6 Tail Probabilities and Recent Data Adjustments to Our Judgment

Reasoning: Three tail risks require individual ranges rather than point estimates: Ayar CPO market share >30% before 2030: approximately **15–20%** (historical tech penetration pace has been slow, ecosystem not standardized); AXT export controls to China escalate: approximately **20–30%** (frequency of material controls rising, insufficient geopolitical de-escalation conditions); AI CapEx significantly declines YoY (>30%): approximately **15–20%** (historical crisis samples are low, but current CapEx intensity amplifies tail risk). Q2 FY26 revenue and EPS slightly exceeded expectations, Data Center was +44% YoY, and Industrial turned from negative to flat. Management guided FY27 growth higher than FY26 and gross margin approaching 42%—favorable for the bull case, but does not change that the probability-weighted SOTP remains below the current price: the beat came from earnings volume, not multiple re-rating.

Conclusion: Under the above revisions, the confidence level for core issues allows for a slight upward adjustment to growth visibility (e.g., approximately +2pp in CQ1) and a downward adjustment to defensiveness under the "moat = pricing power" narrative (e.g., approximately −2pp in CQ4). The SOTP discount band remains **−18% to −22%**; overall confidence level is still <60%, and the investment stance remains cautiously monitored.

Chapter 11: Counter-Scrutiny — Where Are Our Judgments Most Vulnerable?

11.1 RT-1 Load-Bearing Wall Test — Is "AI CapEx continuing for 3+ years" a true load-bearing wall or wishful thinking?

11.1.1 Load-Bearing Wall Identification

The entire argument structure of Chapters 5–10 relies on a core pillar: Belief B4 — namely, that Hyperscaler AI CapEx can maintain an annual growth rate of approximately +15% or more in FY2027-2029 (consistent with the definition of B4 in §5 "Code Quick Reference"). If this pillar breaks:

• Networking revenue falls from +30% YoY to +5-10%
• P/E drops from 41x to 20-25x (label collapse)
• Overall stock price impact: -43% to -58% (RT-1 worst-case scenario)

Test Method: Not testing "if CapEx will collapse," but rather testing "how much evidence already suggests this pillar is cracking."

11.1.2 List of Cracking Evidence (As of April 2026)

Counter-evidence (evidence that the pillar is still holding):

11.1.3 RT-1 Verdict

Load-bearing wall status: Cracked but not broken.

Among the 6 cracks, F3 (mathematically unsustainable) and F5 (100% historical benchmark) are structural—not about "if it will happen" but "when it will happen." F6 (abnormal inventory) is a leading indicator at the COHR level and does not rely on macroeconomic judgment.

Among the 5 supporting facts, S1 (no cuts after tariffs) and S5 (Datacenter acceleration) are the strongest current counter-evidence. However, they indicate "it hasn't broken yet," not "it won't break in the future." S3 (1.6T ramp) is the most powerful structural counter-argument—because 1.6T represents a technological generational upgrade, not purely cyclical expansion.

Key Judgment: The probability of CapEx collapse is not high (17%±3%, triple anchored and calibrated), but the probability of CapEx slowing from +80% to +10-20% is very high (>70%). The difference between these two is that the former leads to COHR revenue -30%, while the latter leads to COHR Networking growth slowing from +30% to +10-15%. The arguments in Chapters 5–10 implicitly conflated "slowdown" and "collapse," overestimating the destructive power of a slowdown scenario and underestimating its probability.

Revision: Treat "CapEx slowdown (not collapse)" as part of the base case rather than a risk case. The base case should assume FY2028 Networking growth moderates to +12-18%, not the +22% implied by the Chapter 8 model. This would reduce the SOTP base case by approximately 10-15%.

11.2 RT-2 One-Sided Argument Detection — Did Chapters 5–10 Focus Solely on Bull or Bear Cases?

11.2.1 Bearish Argument Detection (Were Chapters 5–10 Overly Bearish?)

The rating direction in Chapters 5–10 is "cautiously monitored" (SOTP -18% to -22%). Check if any bullish arguments were overlooked:

Underestimated Bullish Arguments:

Conclusion: Chapters 5–10 are indeed bearish, but the reasons for the bearishness are rational—SOTP weighting does indicate overvaluation. The omissions above are not enough to reverse the rating direction but warrant a correction of the SOTP bear case probability (too high) and bull case probability (too low).

11.2.2 Bullish Argument Check (Did Chapters 5–10 give COHR too much credit?)

Conclusion: Chapters 5–10's interpretation of NVIDIA's investment is overly optimistic (should not be considered a unique advantage), and its assessment of the moat is overly optimistic (cost advantage does not equal a moat), but its interpretation of Industrial is overly pessimistic (ignoring the counter-cyclical buffer). The three partially offset each other.

11.2.3 RT-2 Integrated Bias Correction

Net Bias: Chapters 5–10 are slightly bearish (approximately 5-8% SOTP impact). Primarily due to:

1. The 2-3 year structural advantage of InP supply monopoly is underestimated (largest omission)
2. The certainty of D&A decline is underestimated (GAAP→Non-GAAP convergence is a buying catalyst)
3. However, NVIDIA's investment is overestimated, and the moat is overestimated, partially offsetting.

Action: SOTP base case is slightly adjusted to $245 (from $239), reflecting the structural premium from tight InP supply. The bear case remains unchanged at $158 (InP premium would disappear during a CapEx collapse). The bull case is slightly adjusted to $365 (from $358, SiC option value increased).

11.3 RT-3 Implied Assumption Test — What is $307.50 betting on?

11.3.1 Reverse DCF Belief Fragility Ranking (previously listed, re-ranked in forensic round)

11.3.2 Quantifying the Transmission of Belief B4 Breakdown (RT-3; corresponds to downside risk R1)

If Belief B4 breaks down (i.e., Hyperscaler AI CapEx experiences a cliff-edge YoY decline of >30%, consistent with the definition in §5.2.5):

Layer 1: Revenue Impact

• Networking annualized from $4.5B to $3.2-3.5B (-22~-29%)
• Industrial flat (~$1.8B)
• SiC slightly down (~$0.4B)
• Total: From $6.7B down to $5.4-5.7B (-15~-19%)

Layer 2: Margin Impact

• Capacity utilization drops from 80%+ to 55-65%
• Fixed cost proportion rises, GM compressed from 37-39% to 30-33%
• OPM compressed from 12-14% to 6-8%
• EPS Impact: From FY2027E $7.47 down to $3.50-4.50

Layer 3: Valuation Impact (Most Critical)

• Label collapse: "AI Growth Stock" (40x) → "Post-Merger Cycle Hybrid" (18-22x)
• Sherman factory $2B+ InP capacity → excess → impairment risk
• Goodwill $4.5B → impairment test triggered
• P/E × EPS: 20x × $4 = $80 (-74%), worst case 18x × $3.50 = $63 (-80%)

Is this too extreme? It is a fact that optical communications companies fell -80~-90% during the 2001 telecom bubble. However, that was a shift from extreme oversupply to demand disappearance. Current AI demand is supported by actual inference/training use cases (unlike the "build it and they will come" mentality of 2001 for fiber optics). Therefore, -74% to -80% is a tail-end scenario; a more reasonable impact of B4 breakdown is -43% to -58% (initial estimate for Risk R1·CapEx downturn), corresponding to $130-175.

Belief B4 Breakdown Probability (re-assigned in forensic round):

The aforementioned baseline range for AI CapEx cliff-edge probability is 17%±3%. Forensic calibration:

• Historical baseline rate: 12% (3/25 years) → Unchanged
• Environmental adjustment: CapEx/Revenue ratio 5x (unprecedented) → +5% → Unchanged
• New evidence (April 2026): CapEx guidance unchanged after tariff impact → -2%; but Microsoft pausing signal → +1%
• After calibration: 16%±3%, largely consistent with the baseline range, slightly reduced (-1pp)

11.4 RT-4 Probability Assignment Audit — Does every probability have three anchors?

11.4.1 Three-Anchor Audit of All Probability Assignments

11.4.2 Probability-Weighted SOTP Recalculation

Chapter 8 Original: Bear $158(25%) + Base $251(50%) + Bull $358(25%) = $249.7

After Forensic Calibration:

• Bear $158(30%) + Base $245(45%) + Bull $365(25%)
• = $47.4 + $110.3 + $91.3 = $249.0

The direction remains unchanged (still indicates overvaluation), but the base case decreased from $251 to $245 (RT-1 growth rate adjustment), and bear probability increased from 25%→30% (overly high growth rate adjustment), with the overall result largely unchanged (-0.3%). This indicates that the valuation direction judgment in Chapters 5–10 is robust and insensitive to probability assumptions.

Sensitivity Analysis: What conditions are required for SOTP to be ≥$307.50?

• If Bull probability=45%, Base=40%, Bear=15%: $158×0.15 + $245×0.40 + $365×0.45 = $285.9, still <$307.50
• If Bull=$400, other conditions unchanged: $158×0.30 + $245×0.45 + $400×0.25 = $257.7, still <$307.50
• Requires Bull>$450 and Bull probability>30%, which means COHR needs to achieve $12+ EPS by FY2028 and maintain a P/E of 37x+

Conclusion: The current $307.50 is almost impossible to justify within a probability-weighted framework unless extremely optimistic assumptions are made for the bull case. The valuation direction (overvalued) is further confirmed after the forensic red team review.

11.5 RT-5 Deep Dive into Label Collapse — The Most Underestimated Risk

11.5.1 Why "Label Collapse" (R2) is Sometimes More Dangerous Than "CapEx Cliff" (R1)

Previously, R1·Hyperscaler CapEx Cliff-like Decline was listed as the number one risk, and R2·Valuation Label Collapse as the second. However, after a forensic re-examination, it is believed that: Label collapse is an independent risk that can occur even if cloud CapEx has not yet declined sharply.

COHR currently enjoys the "AI growth stock" label, which gives it a 40x P/E. The conditions for this label to hold are not only "sustained AI CapEx" but also require COHR's AI growth rate to remain >20%.

Label collapse can occur without a CapEx collapse:

Scenario: Hyperscaler CapEx maintains +20%, but COHR's Networking growth rate drops from +34% to +12%:

• Reason 1: Market share loss to Innolight — Innolight module share 50-60% in the 1.6T era with 20-25% lower ASP
• Reason 2: Sustained ASP decline — 800G ASP flat to declining, 1.6T initial ASP high but quickly pushed down by competition
• Reason 3: Industrial drag not improving — 31% revenue with 0-4% growth continuously pulls down blended growth rate
• Result: Networking +12% × 69% proportion + Industrial +3% × 31% proportion = blended +9.2%

A company with 9% growth should not trade at 40x P/E. Historical reference:

If the growth rate drops below 10%, COHR's "reasonable" P/E is 15-22x, corresponding to a share price of $112-$164. This does not require a CapEx collapse; it only requires a combination of market share loss + ASP decline + Industrial drag.

11.5.2 Independent Probability Assignment for Label Collapse

11.5.3 Distinction Between Label Collapse (R2) and CapEx Cliff (R1)

Key Insight: Label collapse is COHR's most underestimated risk, as it doesn't require changes in external conditions—an **endogenous combination** of Innolight's market share erosion + natural ASP decline + Industrial drag is sufficient to trigger it. Chapters 5–10 link label collapse to CapEx downturns, underestimating the endogenous trigger path.

11.6 RT-6 Bidirectional Calibration — Upside Scenarios Also Need Challenging

11.6.1 Upside Scenario 1: AXT Export Controls → COHR InP Monopoly

If China imposes comprehensive export controls on InP substrates (probability 25%±5%):

• AXT (Beijing, 60-70% of global InP substrates) export disruption
• COHR 6-inch InP line becomes the largest non-Chinese capacity globally
• InP substrate prices could double (supply-demand gap widens from 70% to 85%+)
• COHR's short-term revenue might not increase (due to chip → module, not selling substrates), but its **cost advantage doubles** (competitors' chip costs rise significantly)
• Innolight and other Chinese module manufacturers might have InP supply cut off → market share shifts to COHR/LITE

Quantification: If Innolight's market share drops from 50% to 30% (supply disruption), COHR's module share rises from 15-20% to 25-30%, Networking revenue increases by $500M-1B/yr, EPS +$1.50-3.00. The corresponding SOTP bull case should be $380-420 (vs. $365 from forensic analysis).

But—Red Team Challenges Upside:

1. Chinese export controls are usually not outright bans, but rather license delays/quantity restrictions—AXT was still operating after indium controls in Feb 2025.
2. II-VI/Sumitomo/Sumitomo Electric and other Japanese suppliers also have InP capacity; COHR is not the sole beneficiary.
3. Control → Counter-control cycle: China might demand "quota for Chinese customers if buying COHR chips" as a condition.

After Calibration: AXT controls' actual benefit to COHR is lower than intuitive judgment. Raise bull case to $375 (+$10, from $365), not $420.

11.6.2 Upside Scenario 2: CPO Acceleration → COHR CW Laser Demand Exceeds Expectations

If Broadcom CPO begins large-scale replacement of pluggable from 2027 (probability 20-25%):

• InP CW laser demand surges (each CPO module requires a high-power 400mW CW laser)
• COHR might transition from a "module manufacturer" to a "CW laser supplier" (higher profit margins, less competition)
• Although BOM share drops from 30-40% to 10-15%, the **absolute volume (number of dies) could increase 3-5 fold** (CPO density is much higher than pluggable)

Red Team Challenges:

1. In the CPO era, COHR's role shifts from "selling modules" to "selling components," and the revenue multiplier is not necessarily larger—price declines might offset volume increases.
2. Broadcom might self-develop CW lasers (already has SiPh capability), not necessarily sourcing from COHR.
3. Ayar Labs' fully integrated CPO solution might not require external InP.

After Calibration: CPO is neutral to slightly positive for COHR (not a major boon), because CPO changes COHR's position in the value chain, rather than simply increasing demand. Maintain bull case $375 unchanged.

11.6.3 Downside Scenario: "Boiling Frog" (Described in Chapter 8, Forensically Quantified)

The gradually worsening scenario described previously is the most realistic threat:

• FY2027: Networking +20%, P/E 35x → $262 (-15%)
• FY2028: Networking +12%, P/E 28x → $224 (-27%)
• FY2029: Networking +8%, P/E 22x → $198 (-36%)

3-year cumulative -36%, annualized -14%, no single quarter triggers a Kill Switch, but annualized returns for holders are deeply negative.

Probability: 40-50% (highest single scenario), as this is the "growth naturally reverts to mean" base case narrative.

Chapter 12: Supplementary Analysis D — Factors Hindering ROIC Uplift

12.1 S1: Inventory Analysis — Is $1.85B a Proactive Build-up or a Cycle Top Trap?

12.1.1 Inventory Composition Breakdown

COHR does not disclose inventory details (raw materials/WIP/finished goods), but structural inferences can be made based on business nature and DSI:

12.1.2 Historical Comparison of Inventory Growth vs. Revenue Growth

Key Insight: During the FY23→FY24 cyclical downturn, inventory barely decreased (+1%) while revenue fell -9%—this indicates COHR's inventory is **downward rigid**. Because InP wafers cannot be returned or resold after processing (customization), inventory impairment is the only recourse. Of the current $1.85B inventory, if 10-15% consists of customized WIP/finished goods that cannot be absorbed due to slowing demand, the impairment amount would be $185-277M (-$0.91~$1.36/share).

12.1.3 II-VI Historical Impairment Reference

Historical benchmark: COHR/II-VI inventory impairment during cyclical downturns is approximately 3-6%. Based on the current $1.85B, probability-weighted impairment is approximately $56-111M (taking the midpoint of $80M, probability 40%).

Comparison with Probability-Weighted Impairment in Chapter 8: Chapter 8 estimated impairment at ~$100M (-$0.52/share). After forensic inventory analysis, revised to:

• Base scenario (CapEx slowdown but no collapse, probability 50%): impairment $50-80M, EPS impact -$0.25~$0.39
• Cyclical downturn (significant CapEx correction, probability 30%): impairment $150-250M, EPS impact -$0.74~$1.23
• Probability-weighted: ~$90M (-$0.44/share), largely consistent with Chapter 8's probability-weighted estimate, **estimate is robust**.

12.2 S2: Bookings Quality Investigation — Available Evidence for Firm vs. Soft

12.2.1 Public Information Review

Management claimed "unprecedented visibility with bookings extending to 2028" on the Q2 FY26 Earnings Call. What can we infer from public information?

Evidence 1: Differences in bookings statements between COHR vs LITE

Key Observation: LITE's statement is more specific than COHR's (provided months), while Lumentum used almost identical wording to COHR ("strong visibility") in 2022, but the soft commitment eventually evaporated. COHR's vague wording itself is a negative signal – if bookings were firm take-or-pay, management would be motivated to disclose amounts to support the stock price. Non-disclosure = likely contains a large number of soft LOIs.

Evidence 2: Clues from NVIDIA's Investment Terms

NVIDIA's $2B investment announcement (March 2026) stated "multi-year, multi-billion dollar commitments." This suggests that at least a portion of NVIDIA's purchase commitments are contractual obligations, not soft LOIs. However, "multi-billion" spread over 3-5 years = $400-700M annually, accounting for ~10-15% of COHR's Networking revenue. Even if NVIDIA's portion is firm, it only covers a small fraction of COHR's total bookings.

Evidence 3: Inverse Inference from Inventory Behavior

If management truly had "unprecedented" firm bookings, they would not need to build $1.85B in inventory – firm commitment means customers will pick up the goods, allowing for production on demand. Significant inventory build itself suggests that the "unprecedented" nature of bookings is more about volume guidance than firm commitment. This is because management might worry about not being able to meet demand if they don't build inventory in advance (behavior typical of supply-constrained eras), but this contradicts the narrative of "firm commitment through 2028" – if it were truly firm, there would be no need to worry about "not being able to meet demand"; delivery could be made as required.

12.2.2 Bookings Quality Rating

Based on the three pieces of evidence above:

Impact on Valuation: Bookings quality does not alter the base case (as the base case already assumes growth slowing from 22% to 17-19%), but it impacts the speed of the bear case – if soft bookings are canceled, the bear case would not be a gradual deterioration ("boiling frog"), but rather a sudden deceleration within a quarter, similar to Lumentum 2022-2023. This implies that the P/E compression speed in the bear case might be faster than initially estimated (label collapse could be completed within 1-2 quarters, rather than 3-4 quarters).

12.3 S3: SOTP Valuation Recalibration — Validating Identified Gaps

12.3.1 Comparison with SOTP Valuation

The Red Team estimated the probability-weighted SOTP at $249.0 (-19%), but after precise Python calculation, it is $226.6 (-26.3%). Sources of difference:

12.3.2 Python Precise Results

vs $307.50 = -26.3%

12.3.3 Why a $22.4 Difference?

In the forensic review text, I used simple extrapolation (fine-tuning from Chapter 8's $251) without fully re-running the model. After precise Python calculation, three changes combined:

1. Dilution Effect: 173M vs 165M → Per share decrease of $12-15
2. Networking Reduction: Rev × Mult from $37.7B→$33.0B → Per share decrease of $27
3. Probability Shift: Bear+5pp → Further weighted decrease of $3-4

Golden Rule K Requirement: The final draft must use the Python-adjusted figures ($226.6), not the approximate value from the forensic review text ($249).

12.3.4 Sensitivity Validation

Even with the most optimistic probability combination (Bear 15%/Base 50%/Bull 35%), the weighted SOTP is only $249.1 – still 19% below $307.50. A Bull case >$500 (i.e., Networking 8x × $6500M + SiC $8B+) would be needed to approach $307.50, which would simultaneously require:

• 1.6T share >30% (currently 15-20%)
• CPO to generate an additional $1B+ in revenue
• SiC to become industry's second largest (only after ON Semi)
• All catalysts to materialize simultaneously, probability <10%

Conclusion: $307.50 cannot be justified under any reasonable probability assumption; the overvaluation is greater than estimated in Chapter 8 (-26% vs -19%).

12.3.5 Inventory Writedown Sensitivity

12.4 Revised Rating Judgment

Rating Direction: Cautious Attention unchanged, but with stronger evidence—$307.50 is overvalued by 26%, not 19%.

Three-dimensional Status: [Expensive × Improving × Potential Catalysts] → Cautious Attention, but the degree of "Expensive" is revised from "Slightly Overvalued" to "Significantly Overvalued".

Chapter 13: Investment Masters' Roundtable — 5/5 Unanimous Cautious Attention (R-3)

This chapter's "Roundtable Discussion" is based on the published works and known investment philosophies of investment masters, simulating their potential evaluations of the subject analyzed in this report. These views are not actual statements from the aforementioned individuals, but rather represent reasonable deductions based on their investment frameworks. Readers should consider them as diverse perspectives for reference, not as investment advice.

13.1 Round 1: Independent Methodology Application

13.1.1 Warren Buffett (Weight 16%)

Circle of Competence Test: COHR is a hybrid (reportedly D&C ~72% / Industrial ~28%; of which AI Datacom accounts for approximately 55–60% of consolidated revenue, AI-related revenue purity approximately 69% compared to LITE), with multiple curves having different economic logics. When a company needs to be broken down into multiple engines to be clearly explained, the hybrid nature itself is a red flag.

Owner Earnings: GAAP NI ~$350M + D&A $554M - Maintenance CapEx ~$250M - SBC $170M = ~$484M. Market cap $48B, Owner Earnings yield approximately 1.0%. ROIC 4.2% < WACC 10% = Value destruction, not creation. Owner Earnings yield of 1.0% + ROIC of 4.2% does not align with "buying excellent companies at a fair price".

Pricing Power: Innolight is priced 20-25% lower, COHR's ASP is flat to declining. 6-inch cost advantage ≠ pricing power. Moat width should expand over time, but technological iteration (8-inch/SiPh) will narrow it. A cost advantage without pricing power is a fragile moat, which does not support a 41x P/E.

Capital Allocation: Negative FCF (-$96M/Q) + Inventory +28.5% + Net Debt $2.68B; all indicators point in the same direction. NVIDIA's $2B investment means COHR relies on the goodwill of a single major client—a moat that requires client investment to sustain is itself evidence of moat weakening.

Stance: Do not buy, too hard + value destruction

13.1.2 Li Lu (Weight 14%)

Variable Purification: After ±20% sensitivity analysis on 10 candidate variables, only 1-2 truly drive valuation: 800G/1.6T ASP trend + duration of InP substrate capacity shortage. The other 8 are noise.

Cognitive Discount Decomposition: Of the $58.50 premium ($307.50 vs SOTP $249), fundamental premium (InP shortage scarcity pricing) is approximately $15-20, while the remaining $38-43 is an AI label premium + cognitive bias from the D&A illusion. Declining D&A is an accounting improvement, not a cash flow increase; Owner Earnings remain largely unchanged.

Second-Level Thinking: Three incorrect directions of the consensus "high growth in AI optical communication"—CPO accelerating past pluggable; consensus is correct but 41x P/E is already priced in; Hyperscaler CapEx is mathematically unsustainable. With CQ at 53.4%, a 41x P/E offers no margin of safety. The consensus is not wrong enough, but the price does not allow the consensus to be correct, much less for partial errors.

Stance: Cautious, cognitive premium needs to be digested for an opportunity to arise

13.1.3 Stanley Druckenmiller (Weight 18%)

Expectation Gap: Consensus FY2028 EPS $9.64, my estimate $8.20-8.60 (deviation -11% to -15%). Sources of deviation: 1.6T ASP decline (25-35% ASP drop within 6-9 months after generational leap), CPO timeline pushed back to 2028H1, GAAP gross margin continuously suppressed by D&A.

Catalyst Calendar: Upside catalysts (1.6T order visibility/FCF turning positive) are already priced in (COHR rebounded 40%+ from its low). Downside tail risks (Hyperscaler CapEx downgrade/tariffs) are not priced in. +82% YoY CapEx growth is an outlier; optical communication demand will feel pressure within 3-6 months as it reverts to the mean of +25-30%.

Convexity: Bull case $375 upside +53%, Bear case $158 downside -35%, N/M=1.5:1 barely passing. PEG 1.64x (highest among the three), COHR -40% vs LITE -13% during the April shock, Beta is structurally too high. Insufficient convexity, too high Beta, crowded positioning; not a time for a heavy position.

Stance: Do not participate, insufficient odds

13.1.4 Ray Dalio (Weight 16%)

Debt Cycle: $3.5B total debt, effective interest rate 6.3%, annualized interest $210M, EBITDA coverage 2.8x. Interest rate +200bp → FCF deteriorates to -$146M, NAV -12%. -200bp → FCF nears breakeven, NAV +16%. COHR is a directional interest rate bet.

AI CapEx Cycle: CapEx/Revenue ratio of 35% exceeds the "overbuilding" threshold of 30%. COHR's inventory +28.5% is an early warning—if demand were truly visible, there would be no need to build up so much inventory in advance. Transmission lag is 3-5 quarters; H1 2027 is the most sensitive window.

Macro Linkage: COHR is simultaneously a currency bet (40% non-U.S. revenue, strong dollar -4% revenue) + debt cycle bet + geopolitical bet (25% tariffs, -40% in April already priced in but impact on gross margin will be felt 6-9 months later). Any one of the three gears reversing → 41x P/E repricing.

Stance: Do not participate, excessive macro risk exposure

13.1.5 Bear Prosecutor (Weight 20%)

D&A Placebo: The market has already priced in a Non-GAAP P/E of 57.5x (D&A already stripped out), so declining D&A provides almost no impetus to the Non-GAAP narrative. GAAP/Non-GAAP convergence does not bring a single dollar of new cash flow, does not improve ROIC, and does not reduce dilution. It catalyzes the narrative, not the fundamentals.

InP Shortage Questionable: SiPh substitution (Intel/Broadcom/Ayar Labs) and price competition (Innolight 20-25% lower) narrow COHR's addressable market without easing capacity constraints. Demand-side contraction is more dangerous than supply-side expansion.

NVIDIA: Hedge, Not Endorsement: Investing $2B in both COHR and LITE, and also in Ayar Labs (CPO route, potential disruptor). Investing in suppliers with one hand, and disruptors with the other = hedge portfolio management; NVIDIA's loyalty is zero.

Owner P/E 93.8x: 63% higher than Non-GAAP P/E (57.5x), SBC erosion far exceeds what the narrative presents. Total dilution 21% (FY25→FY28), ROIC 4.2%<WACC 10%, every new investment destroys value. Historical analogy: II-VI (COHR's predecessor) fell from $120 to $40 (-65%) after a $7B acquisition in 2021, at which time the thesis was also "D&A + optical communication demand," highly isomorphic.

Most Hollow Supporting Pillar: The true quality of bookings through 2028. The ratio of firm commitments vs. soft LOIs has never been disclosed (black box 30%+ largest contributor). Lumentum's "strong visibility" in 2022-2023 → soft commitments evaporated instantly during a demand pullback → fell from $84 to $43 (-49%).

Stance: Sell, multiple supporting pillars are hollow

13.2 Round 1 Summary: Core Fissures

The five masters converge on one deepest fissure: COHR's entire bull thesis is built on the "time for space" assumption—D&A reduction needs time, InP shortage protection needs time, deleveraging needs time, 6-inch yield ramp-up needs time. However, "time" itself is precisely COHR's most scarce resource, because ROIC<WACC means the longer the time, the more value is destroyed.

13.3 Round 2: Interrogation and Challenge

Li Lu → Bear: Is D&A a placebo? Li Lu points out that GAAP/Non-GAAP convergence would change index fund screening criteria, triggering passive fund buying as GAAP P/E drops from 72x to 50x—this is not cash flow improvement, but a flow catalyst. Bear counters: S&P 500 inclusion has already occurred, with limited incremental impact; quantitative strategies use trailing P/E, and convergence won't be significant until FY28-29; during the most sensitive window of H1 2027, the flow catalyst has not yet arrived. Conclusion: D&A as a flow catalyst is a story for FY28+, but CapEx cycle risk is closer in H1 2027.

Buffett → Dalio: The overlooked interest rate channel. Dalio only considered the debt side, but COHR also has valuation-side interest rate sensitivity—a 41x P/E implies a perpetual growth rate of 4.5%; a +100bp increase in the risk-free rate → P/E mechanically drops from 41x to 35x (-15%). Debt side + valuation side superimposed: interest rate +100bp → stock price approximately -20%. COHR is a dual-channel interest rate Beta, more dangerous than just looking at the debt side.

Druckenmiller → Buffett: Would you buy if the price were low enough? If COHR falls to $180-200 (between SOTP bear-base), PEG drops from 1.64 to 0.8-1.0, convexity N/M=2.5:1. Buffett responds: At $180, I would seriously consider it, but would more likely choose LITE—higher growth rate, higher GM, purer AI exposure, lower PEG. Within the same industry chain, choose the simpler, easier-to-understand one. Even if COHR falls to $180, LITE is the simpler choice.

13.4 Round 3: Colliding Insights

Insight 1: Systemic Time Fragility. The waiting period is not free. ROIC 4.2% < WACC 10% implies a daily consumption of 5.8% annualized implicit cost. If CapEx 2027 slows down + interest rates persist, all catalyst windows will close simultaneously.

Insight 2: D&A Pricing Vacuum. Decreasing D&A provides near-zero impetus to the Non-GAAP narrative (already added back). GAAP-convergent cash flow catalysts are FY28+. D&A should not be counted as a "definite catalyst"; it is downgraded to "narrative improvement".

Insight 3: NVIDIA's Hedging Attributes. NVIDIA simultaneously invests in COHR + LITE + Ayar = classic hedge, with zero loyalty. The investment provides a revenue floor ($500M-1B/yr), not a source of P/E premium.

Insight 4: Interest Rate Dual-Channel Beta. Interest rate +100bp → stock price approx. -20% (FCF -$25M + P/E -15%). COHR is essentially a leveraged directional bet on interest rates, disguised as a tech growth stock.

13.5 Roundtable Verdict

5/5 Unanimous: Cautious Watch, zero dissent. 0/5 dissent → No need to mark "(Critical)" according to R-3. The roundtable's unanimity itself is a signal—all 5 gurus, despite their varied styles, are bearish, indicating that the current valuation issue is structural, not perspective-dependent.

Chapter 14: Quantifying Cognitive Boundaries

14.1 Three-Dimensional Quantification

Deducibility 65%: COHR discloses segment revenue but not segment profit margins (requires reverse engineering); does not disclose inventory composition (raw materials/WIP/finished goods); does not disclose CapEx breakdowns (AI vs SiC vs maintenance); the ratio of firm/soft bookings is a black box. AI Datacom revenue is mixed with telecom in the D&C segment.

Business Complexity 4/5: Three engines (AI+SiC+Industrial) × Post-merger integration × Complex capital structure (preferred conversion + NVIDIA dilution + Net Debt) × Technological roadmap uncertainty (InP vs SiPh vs CPO) × Geopolitical risks (AXT/InP supply chain/China export controls).

Black Box Ratio 27%: Among the key variables impacting valuation, the proportion unverified by public data is—(1) firm/soft bookings ratio (critical for bear case speed); (2) specific amount and execution conditions of NVIDIA's purchase commitment; (3) precise timing of SiC business breakeven; (4) segment-level CapEx allocation; (5) precise yield figures for 6-inch InP.

Black Box 27%: A single-point price target is not prohibited but requires a "low confidence (black box 27%)" label. We opt for a three-point valuation (Bear/Base/Bull) instead of a single point because the combined valuation uncertainty of the three engines naturally leads to greater dispersion.

14.2 Black Box Variable Itemized Analysis

14.3 Impact of Cognitive Boundaries on Investment Judgment

Deducibility of 65% means our understanding of COHR has a 35% structural blind spot. Of this 35%, the most dangerous part is booking quality—because it directly determines whether the bear case is a "boiling frog" scenario (soft bookings slowly shrinking, annualized -14%) or a "Lumentum-style cliff" (soft bookings evaporate within a single quarter, revenue -20%+). The investment strategies for these two scenarios are completely different: the former allows holding and waiting for a stop-loss trigger, while the latter requires exiting before signals appear.

We do not know the answer; this is an honest cognitive boundary. Giving a high-confidence judgment with a 27% black box ratio is irresponsible. Therefore:

• Rating "Cautious Observation" instead of "Sell"—because we lack conviction on the downside speed.
• Using three-point valuation instead of single-point—because SOTP dispersion ($151-$344) reflects real uncertainty.
• Kill Switch uses observable metrics rather than forecasts—because we can monitor signals, but cannot predict timing.

This approach sacrifices "decisiveness" but protects investors from making decisions based on false certainty. COHR's black box area precisely covers the most critical transmission speed of the bear case—in this situation, "I don't know how quickly the bear case will arrive" is more valuable than "I predict the bear case will arrive in 2027Q2."

Chapter 15: Action Layer — How the preceding analysis translates into action judgment

Because the underlying asset is not a pure growth stock, a uniform P/E is not applicable → SOTP must be used. Because ROIC has not yet exceeded WACC, a "Time for Value" strategy is still eroding value → waiting has a cost. Because approximately 48% of EPS improvement is mechanically released, a 41x P/E is not purely buying growth → P/E requires a discount. Because Industrial and SiC are not correctly valued segmentally, SOTP is more reasonable than a single P/E → $226.6 vs $307.50 = -26.3%. Because key constraints remain unresolved (ROIC < WACC + bookings black box + CapEx derivatives), the most reasonable current action is cautious observation, not chasing the price.

Below is the specific signal monitoring system.

15.1 Kill Switch Red Lights (Any trigger → thesis broken)

15.2 Yellow Light Signals (Monitor but no action)

15.3 Green Light Signals (Upgrade Signals)

15.4 Interest Rate Risk Quantification

+100bp → stock price approx. -20%: FCF -$25M (increase in interest expense on floating-rate debt) + P/E -15% (high-growth stocks' sensitivity to interest rates). Term Loan B-2 SOFR + 2.00%, current effective interest rate ~6.3%.

Chapter 16: If You Can Only Remember One Thing

This report analyzes COHR's three-engine structure, moats, competitive landscape, financial attribution, valuation, risks, red team, and roundtable – but all these analyses ultimately converge on the same conclusion:

As a post-merger hybrid, COHR's true valuation re-rating does not occur when revenue continues to grow rapidly, but when its return on invested capital truly surpasses its cost of capital.

In the quarter when ROIC climbs from 4.2% to above 10%, COHR will transform from "41x deleveraging" into "an AI photonics company that is creating value" – the P/E multiple anchor will shift from "hybrid discount" to "value creation premium." Before that quarter arrives, a 41x P/E is a bet that it will happen, not a payment for what it already has.

Three Questions to Ask Next Time You Encounter a Similar Company

Question 1: How much of the EPS growth is real? Break down EPS increments into Category A (operational: revenue growth + margin improvement + scale effect) and Category B (mechanical: decreasing D&A + interest savings + capital structure normalization). If Category B > 30%, you are buying accounting normalization with growth stock multiples – the P/E needs a discount. COHR's answer: Category B is approximately 48%.

Question 2: Which is higher, ROIC or WACC? If ROIC < WACC, the cost of "waiting for catalysts" is not zero. Every year spent waiting consumes the implicit cost of WACC-ROIC. COHR's answer: Negative spread -5.8pp, consuming 5.8% annually.

Question 3: Unified P/E or SOTP? If a company has two or more segments with completely different growth trajectories, a unified P/E is a mathematical average, not an economic valuation. COHR's answer: Three engines (+34%/-10%/investment phase) using a unified 41x P/E = pricing shrinking businesses at AI multiples.

© 2026 Investment Research Agent. All rights reserved.