📚 My Bookmarks
No bookmarks yet
Use the chapter navigation to jump around this report.
AI Infrastructure Opportunities (1): NVIDIA Rubin and the 800V Power Chain
An investment map from NVIDIA's Rubin platform and 800V power architecture to the AI data center power chain
Chapter 1: Core View
The change in NVIDIA's next-generation platform is not just an improvement in GPU performance; it marks the beginning of a shift in how AI data centers are built. In the past, the market viewed AI servers as higher-value servers. After Rubin, high-power racks will push data centers toward a form closer to "high-density power factories." 800V is not an isolated new component, but a power-architecture signal within this round of change.
The previous round of AI investment mainly revolved around chips and servers. The next round must evaluate power, cooling, protection, energy storage, connectors, testing, and operations and maintenance at the same time. As rack power moves from tens of kilowatts to hundreds of kilowatts, and even close to 1 megawatt, the core question expands from "whether the chips are powerful enough" to "whether the entire system can supply power, dissipate heat, protect itself, be delivered, and be continuously maintained in a stable way."
Investment research needs to answer four questions:
- What exactly is changing in NVIDIA's next-generation platform?
- Why are 54V / 48V architectures beginning to approach their physical limits?
- Why will the industry move toward 800V?
- Within this new industry chain, which companies are only conceptually adjacent, and which are most likely to turn the opportunity into revenue first?
Chapter 2: Revenue Realization Sequence: Who Recognizes Revenue First
The 800V industry chain is better ranked by revenue path: who enters customer budgets first, who delivers first, and who recognizes revenue first.
The easiest mistake is to assume that because power semiconductors are important, they will necessarily recognize revenue first. System vendors and infrastructure providers are closer to customer procurement and may instead show the change earlier in orders, backlog, revenue, and cash flow.
| Revenue realization sequence | Representative companies | What they sell | Why they may recognize revenue earlier | Main signals to watch |
|---|---|---|---|---|
| First wave: already absorbing AI data center expansion budgets | Vertiv, Eaton, ABB, Schneider | Critical power, power distribution, protection, cooling, services | AI data center expansion already requires these systems, without waiting for full-scale 800V mass production | Orders, backlog, revenue growth, service revenue, operating cash flow |
| Second wave: companies that turn 800V into deliverable systems | Delta, LITEON, Flex, Vertiv | Power Rack, Power Shelf, Sidecar, BBU, CDU | Customers are not buying individual components, but systems that can be installed, tested, and brought online | Product specifications, customer validation, mass-production timing, monthly revenue, gross margin |
| Third wave: station-level heavy electrical equipment and campus power bottlenecks | GE Vernova, Hitachi Energy, ABB, Eaton, Schneider, Siemens, Mitsubishi Electric | Transformers, rectification, switchgear, in-station power distribution, grid connection, and services | Individual orders are large, but cycles are long, with the earliest signals appearing in production scheduling, lead times, and pricing | Delivery lead times, pricing, factory production schedules, project orders, backlog |
| Fourth wave: semiconductors with greater upside but longer validation cycles | TI, ST, Infineon, onsemi, ADI, MPS, Navitas, ROHM, Renesas, AOSL, POWI, Innoscience | Power devices, control chips, isolation, sensing, drivers, reference designs | They must first enter system solutions, BOMs, and customer certification before volume can ramp | Reference designs, BOM lock-in, customer certification, shipments, price volatility, gross margin |
Investment implications:
The first wave of financial-statement evidence may come from data center power infrastructure companies such as Vertiv, Eaton, ABB, and Schneider. The first wave of dedicated 800V upside may come from system companies such as Delta, LITEON, Flex, and Vertiv that can deliver Power Racks, BBUs, and CDUs. The semiconductor layer is technically indispensable and may also offer significant share-price upside, but for most companies, revenue and profit elasticity will be easier to assess only after system solutions, customer certification, and mass-production schedules become clearer.
Chapter 3: Expectation Gap: The First Wave of Revenue Is Not the Same as the First Wave of Investment Opportunities
The revenue sequence answers "which type of company recognizes revenue first." Investment opportunities also require a valuation constraint: whether the market has already priced in sufficiently high growth expectations in advance.
As of May 26, 2026, some 800V-related names have already been clearly priced by the market as AI infrastructure winners. Vertiv's share price is about USD 327, its market capitalization is about USD 128 billion, and its trailing P/E is above 80x; the midpoint of the company's 2026 adjusted EPS guidance is about USD 6.35, corresponding to about 52x forward adjusted earnings. Navitas already has a market capitalization of about USD 6.7 billion, but its revenue in the first quarter of 2026 was only about USD 8.6 million, so its valuation reflects more expectation for future design wins and successful mass production. High-quality analog/power companies such as Monolithic Power Systems and Analog Devices are also already trading at elevated valuations, showing that the market does not lack attention to AI power suppliers.
From an investment perspective, companies can be divided into three states:
| Expectation-gap state | Representative companies | Investment implications | Metrics that need validation |
|---|---|---|---|
| Strong fundamentals, but price already reflects them fairly fully | Vertiv, Delta, MPWR, TXN, ADI, Navitas | They can still grow, but incremental returns need stronger orders, margins, and cash flow to support valuation | 2027 order visibility, gross-margin expansion, FCF/share, customer concentration |
| Evidence is strengthening, while market expectations remain debated | Eaton, ABB, Schneider, Flex, LITEON, GE Vernova | There is already order or product evidence, but the market is still judging whether this is cyclical ordering, manufacturing revenue, or sustainable high-quality revenue | Backlog conversion to revenue, segment margin, service attach, system gross margin, cash flow |
| The expectation gap may be larger, but evidence is insufficient | Power Integrations, AOSL, BizLink, onsemi, Infineon, ROHM, Renesas, Innoscience, Megmeet | Share prices may not fully reflect specific 800V opportunities, but companies must first prove concrete SKUs, customer certification, and a mass-production path | Design wins, BOM lock-in, customer validation, mass-production shipments, value per unit |
This dimension changes the earlier conclusion: the companies with the clearest first wave of financial-statement realization are not necessarily the companies with the largest first wave of expectation gaps. Vertiv remains one of the clearest AI infrastructure beneficiaries, but its valuation already requires it to keep delivering high growth, high margins, and strong cash flow. Companies such as Flex, LITEON, Power Integrations, and BizLink could instead see larger expectation revisions if they later prove that they have secured high-value system roles or certified positions.
Market expectations show a similar structure. Names such as VRT, NVTS, and MPWR are already frequently placed into AI power, 800V, and data center power narratives, with the pricing debate focused more on "whether valuation is too full" and "whether realization can keep pace." Flex, POWI, AOSL, BizLink, and the protection/telemetry links receive less attention, and the main question is still "whether they have truly entered the mass-production BOM." This expectation structure cannot replace fundamental validation, but it can help judge trade crowding and expectation gaps.
Chapter 4: Layered Table of 29 Companies
NVIDIA's official ecosystem list is highly valuable, but it is only the starting point for "who has entered the collaboration ecosystem," not the endpoint for "who has already secured mass-production orders." This information is used to locate each company within the industry chain.
| Layer | Company | Stock mapping | Its position in the industry | Investment implications | Expectation-gap state |
|---|---|---|---|---|---|
| Power semiconductors / control chips | Texas Instruments | TXN |
800V reference architecture, control, isolation, power chips | Very strong technical benchmark, but not necessarily the greatest upside | Expectations are already high; more a direction to validate |
| Power semiconductors / control chips | STMicroelectronics | STM |
SiC, power devices, 800V power-board solutions | Board-level solution evidence is relatively clear, while revenue still depends on system adoption | Expectations are moderate; the key is customer adoption |
| Power semiconductors / control chips | Infineon | IFX.DE / IFNNY |
SiC, GaN, drivers, power modules | A long-established leader; certainty is stronger than upside | AI power upside may be obscured by automotive/industrial cycles |
| Power semiconductors / control chips | onsemi | ON |
SiC, power devices, sensing | There are already clues of NVIDIA collaboration, but AI purity is not the highest | If the AI data center mix rises, expectations may be revised |
| Power semiconductors / control chips | Analog Devices | ADI |
Isolation, sensing, monitoring, hot swap | Low-profile but high quality; may capture protection and telemetry value | Company quality is already priced in, while the specific value of 800V is still not fully explicit |
| Power semiconductors / control chips | Monolithic Power Systems | MPWR |
Power management, secondary power supply, digital power | More oriented toward in-system power management, not simply main power devices | AI power expectations are high and require continued earnings follow-through |
| Power semiconductors / control chips | Navitas | NVTS |
GaN / SiC, high-voltage conversion, reference boards | High upside, high validation risk | The theme is heavily priced in, creating the greatest pressure for mass-production realization |
| Power semiconductors / control chips | Alpha and Omega Semiconductor | AOSL |
Power-device candidate | Needs to prove concrete SKUs and customer lock-in | The expectation gap may be large, but evidence is insufficient |
| Power semiconductors / control chips | Power Integrations | POWI |
High-voltage power IC candidate | Current evidence remains relatively early-stage | The expectation gap depends on whether the NVIDIA design socket converts to mass production |
| Power semiconductors / control chips | Renesas | 6723.T |
MCU, control, power management | May participate in the control chain, but purity is relatively low | A low-purity watch item unless a clear 800V role is disclosed |
| Power semiconductors / control chips | Richtek | Parent company 2454.TW |
Power management ICs | Already integrated into MediaTek, with limited upside for the parent company | Indirect mapping makes it difficult for the expectation gap to be reflected in the parent company |
| Power semiconductors / control chips | ROHM | 6963.T |
SiC power devices | The technical logic is strong, but AI 800V evidence still needs to be strengthened | The technical direction is valid; awaiting AI customer evidence |
| Power semiconductors / control chips | Innoscience | 02577.HK |
GaN power devices | A domestic GaN upside clue, with certification and mass production still to be verified | Upside comes from certification breakthroughs; the risk is insufficient evidence of global supply |
| Power semiconductors / control chips | Efficient Power Conversion | Private | GaN FET | A non-listed target, mainly useful for validating the GaN path | Can only serve as path validation |
| Power system components | Delta Electronics | 2308.TW |
660kW Power Rack, BBU, CDU, microgrids | Among the strongest system-evidence names; focus on orders and gross margin | Market capitalization already reflects considerable AI expectations; new-product revenue breakdowns need to be watched |
| Power system components | LITEON | 2301.TW |
110kW Power Shelf, 800V Power Rack, BBU, CDU | Mass-production cadence should be tracked as a priority | If validation and mass production advance on schedule, dedicated revenue expectation gaps may remain |
| Power system components | Flex | FLEX |
800V Power Rack, manufacturing, and system integration | The key question is whether margins are system margins or manufacturing margins | The expectation gap depends on SpinCo and the quality of system gross margins |
| Power system components | BizLink | 3665.TW |
High-voltage harnesses, connectors, cable assemblies | May enter the certification scope, but product evidence still needs improvement | The connector/harness link receives less attention, so evidence breakthroughs could bring greater upside |
| Power system components | Megmeet | 002851.SZ |
Industrial power supplies, data center power supplies | A domestic upside clue; local beneficiaries must be distinguished from NVIDIA's global supply system | Domestic AI power expectations and NVIDIA's global supply system need to be validated separately |
| Power system components | Lead Wealth | No confirmed independent listing | Power-component partner | No clear listed-company mapping; observe it as a supply-chain node | No direct public-market mapping |
| Data center power systems | Vertiv | VRT |
Critical power, thermal management, services, 800V portfolio | One of the most direct financial-statement beneficiaries, but valuation is the most likely to become crowded | Highly priced in; later performance must be assessed through FCF/share and service gross margin |
| Data center power systems | Eaton | ETN |
Power distribution, protection, busway, backup power, thermal management | An early revenue platform for AI power expansion; focus on margins and cash flow | Expectations already exist, but valuation pressure is lower than VRT; the cash bridge is key |
| Data center power systems | ABB | ABB / ABBN.SW |
Power distribution, DC protection, MV UPS, system integration | Order evidence is strong, while cash-flow transmission still needs validation | Data center strength is already partly priced in, while high-voltage protection may still offer an expectation gap |
| Data center power systems | Schneider Electric | SU.PA |
800V sidecar, energy management, software, services | More of a recurring-service platform; valuation depends on whether expectations are too full | The platform attributes are recognized by the market, while the 800V sidecar still needs revenue evidence |
| Data center power systems | GE Vernova | GEV |
Transformers, rectification, station-level power, grid interface | Large orders with long-cycle realization; prioritize lead times and backlog | Broad electrification has already been priced in significantly, and dedicated 800V exposure is not the main driver |
| Data center power systems | Hitachi Energy | Parent company 6501.T |
Transformers, substations, grid connection | A representative heavy-electrical bottleneck, with a slow revenue cadence | The group-level mapping is not pure; more of a long-cycle validation case |
| Data center power systems | Mitsubishi Electric | 6503.T |
Critical power, UPS, industrial electrical equipment | A watch item among traditional large manufacturers; direct 800V evidence is still insufficient | Direct expectations are not high, but clear catalysts are also lacking |
| Data center power systems | Siemens | SIE.DE |
In-station power distribution, automation, digital operations and maintenance | Strong system capabilities, but dedicated 800V exposure still needs to be separated out | The theme exposure is not highly pure; more tied to the overall industrial electrical cycle |
| Data center power systems | Heron Power | Private | High-voltage power conversion | A non-listed target, showing that part of the profit pool remains in private markets | Can only serve as path validation |
Assessment method:
When assessing a company, first confirm its position in the customer budget. Is it selling core devices, deliverable systems, or station-level infrastructure? Different positions imply completely different revenue timing, gross margins, cash flow, and valuation logic.