800V DC transition strengthens the analog semiconductor recovery; TXN remains the top pick, and ON receives upside catalyst watch

Overview

This report discusses investment opportunities in the U.S. semiconductor industry amid upgrades in AI data center power architecture. The core view is that as rack power rises from about 120-130kW to 250kW in 2026, 600kW in 2027, and above 1MW in 2028, the traditional 48V architecture faces bottlenecks in current, busbars, and thermal losses, prompting data centers to gradually shift toward ±400V HVDC, +800V HVDC sidecar, and, over the longer term, native 800V SST architectures. This migration increases content for analog, power semiconductors, wide-bandgap materials, and connectors, while also coinciding with a cyclical recovery in the analog industry.

Core views

Citi is most constructive on TXN because it is expected to gain share in data center power starting in 2H26, while also benefiting from high-voltage GaN and internal manufacturing expansion; MPWR continues to benefit from increasing share in enterprise data and communications markets; ON has been added to the 90-day upside Catalyst Watch due to momentum related to SiC, vertical GaN, and 800V DC. The report also believes connector companies such as APH and TEL will benefit from higher voltage, higher density, more complex power interconnects, and liquid-cooling systems. At the industry level, analog unit shipments remain 30%-40% below prior peaks, mature-node utilization has recovered to about 80% and is approaching the optimal 85%-90% range, and together with foundry and analog supplier price increases, the analog upcycle still has room to continue.

Analysis framework

The report uses an analytical path from system architecture to semiconductor content: first assessing how rising AI rack power drives migration from 48V to HVDC, then breaking down the impact of ±400V sidecar, +800V sidecar, and SST architectures on GaN, SiC, VRM, Stage 1/Stage 2 power conversion, connectors, and industrial power infrastructure, and finally combining AI data center GW deployments, TAM, supplier qualification, analog-cycle utilization, and pricing changes to form views on TXN, ON, MPWR, APH, TEL, and other names.

Methodology notes

• Industry architecture analysisHVDC migration path

  Differentiate adoption cadence and semiconductor impact across three architecture types: ±400V HVDC, +800V HVDC sidecar, and native 800V SST.

  ±400V HVDC is more retrofit-compatible and is expected to be adopted earlier; +800V sidecar is expected to advance with Rubin Ultra after 2H27; SST is better suited for >1MW greenfield data centers and is expected to become more meaningful after 2029.

• Market size estimationData center analog and power semiconductor TAM model

  Derive TAM based on AI data center GW deployments, rack power, number of conversion stages, and higher semiconductor content.

  The report expects data center analog and power semiconductor TAM to reach about $35B by 2030, with a roughly 30% CAGR from 2025 to 2030, mainly driven by AI deployment growth and higher content from 800V architecture.

• Technology roadmap assessmentApplicability boundaries of GaN and SiC

  GaN is oriented toward high-frequency, high-density conversion below 650V, while SiC is oriented toward high-power infrastructure conversion above 1000V.

  Under the ±400V architecture, voltage stress on individual GaN devices does not exceed 650V, giving the mature 650V GaN ecosystem an advantage; in SST, central rectification, and facility-level conversion, SiC's high-voltage and efficiency advantages are more prominent.

• Cycle analysisAnalog semiconductor recovery framework

  Assess the stage of the analog cycle by combining unit shipments, mature-node utilization, inventory restocking, and pricing changes.

  Mature-node utilization is about 80%, approaching the optimal 85%-90% range; unit shipments are still 30%-40% below prior peaks, and price increases plus data center demand may extend the recovery.

Asset mapping & comparison

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

• Texas Instruments (TXN)

  Top beneficiary and top pick, with higher share in data center power and high-voltage GaN capability as the core thesis.

  Strengths

  Has high-voltage GaN supply capability, with internal GaN manufacturing capacity having expanded significantly since October 2024; expected to ramp in Rubin GPU Stage 1 and Stage 2 sockets starting in 2H26; also an important supplier for CPU VRM.

  Weaknesses

  Share gains in data centers still depend on qualification and customer ramp timing, while power semiconductor standardization and multi-sourcing strategies may limit pricing power.

  Comparison

  Relative to established suppliers such as Infineon, MPWR, and Renesas, the report views TXN as a relative share gainer.

  Risks

  Delayed Rubin/Rubin Ultra ramp, GaN costs or yields below expectations, rollback in analog price increases, and AI data center capex below expectations.

• ON Semiconductor (ON.O)

  A 90-day upside Catalyst Watch name, benefiting from SiC, vertical GaN, and the 800V DC migration.

  Strengths

  Differentiated in SiC and vertical GaN; infrastructure-layer applications such as SST, central rectification, high-density power shelves, and ESS may increase SiC adoption; the company noted that its ESS business delivered about 40% CAGR over the past five years with roughly 60% market share.

  Weaknesses

  Near-term incremental industrial opportunities have limited impact on analog companies, with more catalysts dependent on analyst day communication and the medium- to long-term rollout of 800V architecture.

  Comparison

  Compared with TXN's thesis centered on GaN and data center power share, ON is more leveraged to SiC, infrastructure-layer exposure, and vertical GaN differentiation.

  Risks

  ON's analyst day fails to meet market expectations, SiC demand is affected by EV/industrial cycles, or 800V/SST deployment is delayed.

• Monolithic Power Systems (MPWR)

  A continued favored beneficiary of power delivery and VRM.

  Strengths

  Previously held a near-exclusive position in the Hopper platform GPU VRM market, has shipped vertical power modules to Google, and is expected to benefit from share gains in enterprise data and communications markets.

  Weaknesses

  NVIDIA shifted to a multi-supplier strategy from late 2024 into 2025, adding Infineon, Renesas, and others, which may weaken MPWR's exclusivity advantage.

  Comparison

  Compared with TXN's new-share-gain thesis, MPWR already has a strong VRM base but faces multi-source competition.

  Risks

  Customer multi-sourcing strategies, price competition, and integrated premium compression from standardized specifications.

• Amphenol (APH)

  A beneficiary in connectors and power interconnects.

  Strengths

  Higher voltage, higher density, and more complex AI systems require more power interconnects, busbars, high-power cables, and thermal management connections, and APH has broad product coverage.

  Weaknesses

  The magnitude of benefit depends on the adoption speed of 800V architecture and the pace of data center projects.

  Comparison

  Unlike semiconductor suppliers, APH benefits from greater system complexity and higher connector value per system.

  Risks

  Slower AI data center construction, connector content increases below expectations, and customer pricing pressure.

• TE Connectivity (TEL)

  A beneficiary in connectors, data networking, and energy businesses.

  Strengths

  The report notes that TEL's DDN business is expected to deliver CAGR above 50% from FY25 to FY27, with power connectivity accounting for about 25% of segment revenue, and the 800V transition may bring more than 30% content uplift; its energy business also benefits from grid reinforcement and data center power access.

  Weaknesses

  Realization of benefits depends on execution in DDN and energy projects, and connector growth is tied to the data center construction cycle.

  Comparison

  Similar to APH, TEL benefits from greater power architecture complexity, but it also has energy business exposure.

  Risks

  Data center construction pace, delayed 800V adoption, and volatility in energy project orders.

• Infineon

  An incumbent supplier in GaN, SiC, and power delivery.

  Strengths

  Has scaled 300mm GaN and SiC capabilities, is an important participant in the Stage 1/Stage 2 power delivery market, and has industry sizing related to SST, ESS, and SSCB markets.

  Weaknesses

  There are many suppliers and overlapping supply relationships, and industry standardization may limit pricing power.

  Comparison

  Relative to TXN, Infineon is an incumbent strong supplier; relative to ON, it also has both SiC and GaN positioning.

  Risks

  Intensifying competition, slower-than-expected realization of the GaN cost curve, and customer multi-sourcing pressuring prices.

• Analog Devices (ADI)

  Enhancing integrated voltage regulator opportunities through the Empower Semiconductor acquisition.

  Strengths

  Empower is a pure-play integrated voltage regulator asset, and ADI noted that initial design wins are expected to emerge in 2H26.

  Weaknesses

  The deal still requires integration, IVR needs 1-2 years of engineering collaboration in advance, and commercial ramp is more medium- to long-term.

  Comparison

  Compared with current power delivery opportunities at MPWR and TXN, ADI is more leveraged to next-generation platforms and longer-term IVR positioning.

  Risks

  Acquisition integration, design wins below expectations, and delays in customer platform timing.

Key data

• Report date2026-06-15Citi U.S. semiconductor industry research, 39 pages in full.
• Data-center-related analog salesAbout $17B, about +86% YoY in 2026EGrowing from about $9B to about $17B, reflecting expanding AI data center demand.
• Data center analog and power semiconductor TAMAbout $35B in 2030E, about 30% CAGR in 2025-2030Driven jointly by AI deployment growth and higher content from the 800V DC transition.
• AI data center deploymentsAbout 14GW in 2026E, about +58% YoY; about 19% CAGR in 2025-2030The report estimates AI deployments grow from about 9GW to about 21GW.
• Rack-level GPU power delivery TAMGrowing from about $2B to about $12B in 2028E, with about 70%-75% CAGRCovers conversion from 800V to sub-1V, with Stage 2 VRM accounting for about 60% of content.
• Rack power evolutionCurrent level of about 120-130kW, 250kW in 2026, 600kW in 2027, above 1MW in 2028Rising power drives high-voltage DC architectures to replace the traditional 48V architecture.
• Analog industry cycle indicatorsMature-node utilization about 80%; unit shipments 30%-40% below prior highsUtilization is close to the optimal 85%-90% range, and restocking remains insufficient.
• Incremental industrial opportunityAbout $2B, about 2% of the $100B industrial semiconductor TAMOpportunities exist in ESS, SST, and SSCB, but the impact on industrial exposure for analog companies is limited over the next 24 months.
• Target price revisionsTXN: raised to $345 from $280; ON: raised to $120 from $100Mainly driven by price increases, analog recovery, and demand for data center analog/power semiconductors.
• Key catalyst for ON2026-09-16 analyst dayCiti initiates a 90-day upside Catalyst Watch ahead of this event.

Impact & implications

The investment implication is that the 800V DC migration is not only a change in power architecture, but may also alter the growth profile of analog and power semiconductor companies. If data center or AI-related revenue rises to about 20% of total revenue, related companies may see valuation re-rating. In the short term, the more visible opportunities lie in the more retrofit-friendly ±400V HVDC sidecar, qualification progress and target price increases for TXN and ON, and APH and TEL benefiting from more complex connector content; in the long term, outcomes will depend on +800V sidecar, SST, integrated voltage regulators, and the cost curves of SiC/GaN.

Risks

• The rollout of 800V DC architecture may be delayed due to retrofit constraints, supply chain readiness, safety certification, and inconsistent customer standards.
• Native 800V SST depends more on greenfield data centers and facility-level redesign, and the pace of deployment after 2029 remains uncertain.
• AI data center deployments, GPU/ASIC shipments, or rack power growth below expectations would reduce TAM and content assumptions.
• A large number of power semiconductor suppliers, specification standardization, and multi-sourcing strategies may limit pricing power and margins for analog vendors.
• If analog restocking underwhelms, mature-node utilization declines, or price increases prove unsustainable, the recovery thesis would weaken.
• If GaN and SiC cost, yield, reliability, or qualification progress fall short of expectations, the technology roadmap may not materialize as expected.
• If ON's analyst day does not provide sufficiently clear catalysts on SiC, GaN, or 800V DC, the short-term upside thesis may be challenged.

What to watch

• Qualification and shipment progress in 2H26 for TXN and ON in Rubin/Rubin Ultra-related Stage 1 and Stage 2 power sockets.
• Updates at ON Semiconductor's analyst day on 2026-09-16 regarding SiC, vertical GaN, 800V DC, and long-term financial targets.
• The actual pace of adoption of ±400V HVDC in CSP and OCP Mount Diablo-related deployments.
• Changes in rack power and power architecture roadmaps for Rubin Ultra in 2H27 and Feynman around 2028.
• Project rollout of SST and >1MW greenfield data centers in 2029-2030.
• Whether mature-node utilization continues to rise from about 80% toward 85%-90%, and whether price increases for analog and lagging-edge nodes can continue.
• Whether AI data center deployment GW and the share of data-center-related sales in analog company revenue approach about 20%, and the resulting potential for valuation re-rating.
• Orders, content, and growth guidance for APH and TEL in datacom, DDN, and energy connectivity businesses.