AI Data Center Power Upgrades Open Long-Term Incremental Upside for Analog and Power Semiconductors

Overview

This report examines the trend of AI data centers evolving from low-power cloud racks to high-power, MW-class racks. It analyzes how power is becoming the core bottleneck for AI expansion and argues that 800V DC, hybrid microgrids, SSTs, SSCBs, ESS/UPS, and related new architectures will reshape the supply chain from the grid to the GPU/XPU core. The report contends that rising power density will make data center power systems far more semiconductor-intensive, creating meaningful incremental markets for analog ICs, power devices, SiC, GaN, sensing, protection, and control chips.

Core views

The core view is that AI compute expansion is no longer constrained only by GPU supply. Power access, transformers, gas turbines, power conversion efficiency, rack space, and copper usage have all become key constraints. Traditional 48V/54V architectures are difficult to sustain at 600kW to 1MW rack levels, while 800V DC is expected to support higher compute density by reducing conversion stages, lowering current, cutting copper losses, and freeing up rack space. Value is expected to shift from traditional heavy power equipment and low-voltage distribution toward high-voltage power conversion, protection, control, sensing, and wide-bandgap semiconductor devices.

Analysis framework

The report uses a bottom-up AI analog semiconductor model that combines GPU/XPU and proprietary ASIC platform demand, rack power roadmaps, component content at different rack power levels, power-chain components, and supplier share assumptions to estimate TAM changes from rack-to-core to grid-to-data hall, while breaking out SI, SiC, GaN, VRM, IBC, PSU, SST, SSCB, and ESS/UPS segments.

Methodology notes

• Market size estimationBottom-up AI analog semiconductor model

  Map accelerator and rack demand to components, device types, and supplier revenue shares

  The report translates GPU/XPU and proprietary ASIC demand into low-power and high-power rack shipments and power requirements, then breaks this down into content pools such as VRM, IBC, PSU, SST, protection, sensing, optical communications, and wide-bandgap devices.

• Technology roadmap analysisGrid-to-core power chain analysis

  Evaluate conversion stages, efficiency losses, and physical constraints from the grid to the GPU/XPU core

  The report compares the traditional 415VAC/48V or 54V architecture with the 800V DC architecture, emphasizing that the latter can reduce conversion steps, lower current, cut copper usage, and improve reliability.

• Platform roadmapAI rack power roadmap

  Project rising rack power density from Hopper, Blackwell, Rubin, to Feynman

  The report argues that both Nvidia and alternative platforms will see higher per-rack power as GPU counts, TDP, CPUs, switches, and network topologies expand.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• TXN

  Primary beneficiary

  Strengths

  It has a leading power semiconductor business and a broad analog product portfolio, and the report believes it has the highest market share.

  Weaknesses

  The report does not identify specific weaknesses, but its growth leverage may be limited by its already high share base.

  Comparison

  Compared with other suppliers, TXN has a stronger base share in power semiconductors.

  Risks

  AI data center power upgrades may proceed more slowly than expected, or 800V DC adoption may lag.

• Infineon

  Primary beneficiary

  Strengths

  It has a broad AI power product portfolio across Si, SiC, and GaN, covering multiple links in the grid-to-core chain.

  Weaknesses

  The high-voltage data center market is still early, and actual share will depend on product qualification, reliability, and customer adoption.

  Comparison

  The report believes it has the broadest AI power portfolio and may achieve the largest share gains in CY25-CY30.

  Risks

  SiC/GaN cost, yield, reliability, or capacity expansion may fall short of expectations.

• ADI

  Beneficiary

  Strengths

  It has strong capabilities in analog, signal chain, and power management, and the Empower acquisition expands its AI power opportunity.

  Weaknesses

  The report does not quantify its coverage gap versus TXN and Infineon across the full power chain.

  Comparison

  The report says ADI has the third-largest revenue opportunity and may gain share.

  Risks

  Adoption of server board-level power and high-voltage conversion designs may be slower than expected.

• ON

  Beneficiary

  Strengths

  It has high leverage to SiC and GaN, the new wide-bandgap technologies, and may increase wallet share with customers.

  Weaknesses

  Relative to diversified analog suppliers, its portfolio breadth and system-level coverage may require further validation.

  Comparison

  It is more exposed to wide-bandgap materials and new technology leverage than to broad analog coverage.

  Risks

  Actual adoption rates or commercialization speed of wide-bandgap devices in high-voltage data center architectures may be slower than expected.

• STMicro

  Potential beneficiary

  Strengths

  The report mentions STMicro in optical infrastructure and related analog content.

  Weaknesses

  The report provides less company-specific detail than for TXN, Infineon, ADI, and ON.

  Comparison

  Its opportunity is more concentrated in specific components and optical infrastructure rather than being one of the report's core share winners.

  Risks

  The value contribution or share gains of related components in AI power upgrades may be limited.

• Renesas

  Potential beneficiary

  Strengths

  It is mentioned in multi-phase VRM and server board-level power-related segments.

  Weaknesses

  The report does not provide a detailed conclusion on its overall share or growth magnitude.

  Comparison

  It participates alongside Infineon, TXN, and ADI in the power-management value pool closer to the accelerator.

  Risks

  Competition in VRM and board-level power may intensify, or changes in customer platform architectures may affect content value.

Key data

• Traditional cloud server rack powerabout 10-15kWServes as the baseline for the step-up in AI rack power.
• Blackwell-generation rack powerabout 100-120kWGPU power budgets and liquid-cooled rack demand rise significantly.
• Rubin Ultra rack powermore than 600kWThe report expects 800V DC could begin to be deployed at this stage.
• Feynman-era rack powerabout 1.5MWThe report says rack power could be about 100x higher than traditional servers.
• Cumulative incremental AI power demand in CY25-CY30about 233GWBottom-up estimate based on accelerator and rack demand.
• AI analog semiconductor market sizefrom about $7.9 billion to about $27.0 billion by CY30Implied CAGR of about 28%.
• rack-to-core TAMfrom about $7.6 billion currently to about $25.0 billion by CY30Implied CAGR of about 27%.
• grid-to-data hall TAMfrom about $245 million currently to about $1.8 billion by CY30Implied CAGR of about 49%.
• 800V DC copper savingsabout 45%Based on Nvidia's estimate of lower copper usage.
• TCO improvementup to about 30%Driven by a simpler power architecture, lower maintenance costs, and efficiency gains.

Impact & implications

The investment implication is that AI data center power architecture upgrades may shift analog semiconductor demand away from automotive and industrial cycle exposure toward longer-duration, higher-growth AI infrastructure demand. Companies with complete power-tree product offerings, high-voltage reliability, SiC/GaN material capabilities, and system-level design strength are more likely to win share. TXN is seen as leading in power semiconductor share, Infineon may capture the largest share gains in CY25-CY30 thanks to the most complete Si, SiC, and GaN combination, ADI benefits from the Empower acquisition and a large revenue opportunity, and ON has high leverage to SiC and GaN new technologies.

Risks

• 800V DC architecture deployment may proceed in stages, and ecosystem maturity, equipment availability, and customer qualification could all slow adoption.
• SSTs still face reliability, thermal, and electrical-stress challenges at multi-MW scale, and MV rectifiers may handle more of the role in the near term.
• Greenfield data center, microgrid, and high-voltage DC buildouts have long lead times and are constrained by grid interconnection, transformers, gas turbines, and construction cycles.
• AI training and inference workload volatility can create rapid load swings, requiring extra storage and overdesign, which raises cost and complexity.
• If AI accelerator demand, GPU/XPU roadmaps, or proprietary ASIC deployments come in below expectations, related power semiconductor TAM could be revised down.
• The cost, reliability, yield, and supply-chain capability of wide-bandgap materials such as SiC and GaN may affect the pace of commercialization.

What to watch

• Changes in per-rack power and network topology for Nvidia Rubin Ultra, Feynman, and AMD, Intel, and proprietary ASIC platforms.
• The actual rollout timetable for 800V DC in new data centers and white-space retrofits.
• Qualification and deployment progress for MV rectifiers, SSTs, SSCBs, and ESS/UPS in GW-scale AI data centers.
• Design wins for TXN, Infineon, ADI, ON, Renesas, STMicro, and others in AI power management, SiC, GaN, and high-voltage protection products.
• Whether data center power interconnection, transformer, and gas turbine lead times continue to constrain AI infrastructure expansion.
• Whether grid-to-rack power efficiency, copper usage, maintenance costs, and TCO improvements can meet the report's assumptions.