Global data center and AI power infrastructure: UBS stays constructive on AI data-center infrastructure, while tracking delays, power constraints and the shift toward 800VDC.

Overview

This fourth UBS Data Center Central consolidates data-center and AI-infrastructure research. Its central conclusion is that the structural expansion in AI power demand remains intact, but the investment case is increasingly shaped by project execution, grid access and the timing of the transition to higher-voltage power architectures.

Core views

UBS's project-delay and cancellation tracker suggests that disruption remains contained relative to the addressable buildout, although the trend merits monitoring. In the US, net currently affected capacity is 7,010MW and ultimate affected capacity is 16,910MW across nine reported cases: four confirmed delays account for 4,580MW currently and 14,480MW ultimately, while five cancellations or withdrawals account for 2,430MW. UBS Evidence Lab shows 57GW of operational and 130GW of pipeline capacity in North America, compared with 17GW and 30GW in Europe. On that basis, affected capacity represents 12% of operational and 5% of pipeline capacity in the US, versus 6% and 3% in Europe. The gap between current and ultimate US affected MW is largely driven by one 11GW project being delayed in a 1GW phase. UBS therefore sees delays and re-phasing as a more relevant risk than broad cancellation at present. The report remains positive on AI-infrastructure equipment but is increasingly selective on execution quality. UBS's European capital-goods preferences are Schneider, Halma, Prysmian and Siemens Energy, all Buy-rated; Siemens, Centiel and Munters are also Buy-rated, while ABB, Legrand, Belimo, Alfa Laval and Wartsila are Neutral-rated. In US electricals, UBS is Buy-rated on Vertiv, GE Vernova, nVent, Eaton and Modine, remains positive on Cummins in machinery, and expects Quanta Services, MasTec, Primoris and Dycom to benefit from tight labor supply in data-center fiber and electrical work. Delta Electronics is described as structurally well positioned in Asia. The common thesis is that AI investment shifts value beyond chips and servers into the equipment needed to distribute, convert, protect, cool and deliver increasingly dense power loads. The next near-term catalyst is Data Center World Singapore and Legrand's capital-markets day on 29-30 September. UBS expects announcements on sidecar power-distribution units and solid-state transformer offerings, plus commercialization updates from major electrical-equipment vendors. For Legrand, UBS believes the data-center segment's rise to about 30% of group sales from about 20% makes an organic-growth-target increase from the existing 3-5% range plausible. It sees a 4-6% range as mechanically supported by the changing mix and a potentially more bullish 5-7% range if management takes a less conservative view of data-center growth. UBS expects the company to retain 3-5% M&A CAGR and stable profitability. A sidecar offering for approaching 600kW racks and a clear roadmap toward 1MW racks are viewed as important for addressing concerns about Legrand's 800VDC position. The report notes that ABB launched its Infinitus solid-state-transformer range on 21 September and scheduled a 24 September investor webinar, while UBS expects Schneider to advance its sidecar and solid-state-product roadmap. UBS frames AI power as a potential constraint on compute growth and a large content opportunity for power-electronics suppliers. It estimates AI power-electronics TAM rises from US$6.6bn in 2025 to US$91.1bn in 2030, a 69% CAGR. Its model projects AI rack units to grow at a 12% CAGR from 2025 to 2030, but thermal design power per rack at a 39% CAGR, from 55.4kW to 282.8kW. AI IT capacity deployed rises from 11.5GW to 104.5GW, and power content per IT watt from US$0.57 to US$0.87. Thus, UBS argues the largest driver of TAM is rising watts per rack rather than equipment volumes. It believes 2027 growth does not require broad 800V adoption because higher power requirements for Rubin, AMD's move to rack-scale systems and custom-ASIC ramps at Google, Amazon and Meta can support demand. The longer-term 800VDC transition changes where value sits in the data-center power chain. UBS expects architectures to progress from server power-supply units and rack-level shelves toward dedicated 400V/800V power racks and, ultimately, solid-state-transformer systems. Higher voltage lowers current, copper use and transmission losses, while increasing the role of high-voltage distribution, protection, controls and conversion. UBS considers the initial transition especially content-additive because dedicated power racks and HVDC infrastructure are added before legacy low-voltage systems are fully removed. Schneider is regarded as the most directly exposed European capital-goods name because of its switchgear, distribution and controls positions and its 800V roadmap; ABB and Legrand should benefit from demand for medium-voltage equipment, busways, distribution infrastructure and power-management systems. The report also identifies constraints to this thesis. UBS's semiconductor assumptions imply 51GW of AI capacity additions in 2027 and 105GW by 2030, above market expectations of roughly 30-50GW annually, raising the risk of a future digestion period. A slower migration to 800VDC sidecars could reduce TAM by roughly US$0.5-1.0bn; estimated Rubin Ultra power-semiconductor content is about US$153m/GW with a sidecar against US$108m/GW with integrated power supply. Intermediate 50V sidecars may extend the transition even though increasing rack power makes higher-voltage delivery progressively necessary. On batteries, UBS rejects the view that NVIDIA's Rubin architecture eliminates rack-level demand. It argues that 800V designs primarily move backup power from the server tray to dedicated sidecars and power racks. While some hyperscalers could use more hall-level UPS, UBS assumes a 40-50% BBU attach rate for Rubin-class deployments and expects content per rack to increase with density. Its forecast has AI-server BBU TAM rising from US$0.7bn in 2025 to US$2.5bn in 2026 and US$12.8bn in 2030, with combined BBU and CBU opportunity reaching US$15.9bn by 2030. Panasonic is viewed as leveraged to cylindrical cells and sidecar backup modules, while Samsung SDI could benefit from both rack- and building-level UPS deployment. Newsflow reinforces both demand visibility and execution risk. Google announced at least €13bn of AI-infrastructure investment in Finland through 2028, including three northern data centers and an agreement for up to 50% of one Fortum nuclear plant's output for 22 years. UBS sees this as evidence that campus selection is becoming power-led and as support for Nordic generation and grid investment. Large compute contracts and regional expansion also support powered shells, electrical distribution, cooling and construction demand, but carry financing, customer-concentration and counterparty risks. At the same time, Texas reportedly froze new data-center grid connections while assessing project credibility after large-load requests exceeded 474GW, versus roughly 48GW in 2023; selected US-region requests exceed 700GW. UBS views stricter deposits and financial-security requirements as negative for speculative powered-land valuations but constructive for credible developers, established hyperscalers and suppliers attached to funded projects. UBS also addresses AI-governance developments. It notes proposals for slower frontier-model scaling, independent evaluator access, common safety standards and international coordination, alongside OpenAI's confirmation that it will not pursue an IPO in 2026. UBS does not see these voluntary developments as a material capex downgrade so far. The report says the critical question is whether labs adopt measurable capability thresholds, independent evaluation and enforceable commitments.

Analysis framework

UBS combines a proprietary delay/cancellation tracker with UBS Evidence Lab regional-capacity data, company guidance and consensus forecasts, sector research, industry feedback and comparative coverage maps. It then links project activity, rack-power density and technology architecture to demand for electrical, thermal, semiconductor and backup-power equipment, while testing the thesis against grid, financing, adoption and regulatory constraints.

Methodology notes

• Industry AnalysisSupply-demand framework

  Data-center capacity, project-delay and grid-demand tracking

  UBS compares delayed or cancelled capacity with operational and pipeline capacity, then considers power availability and credible project funding as constraints on future supply.

• Industry AnalysisVolume-price decomposition

  AI power TAM build-up driven by rack units, watts per rack and power content per IT watt

  The report separates equipment-volume growth from rising power density and content intensity, concluding that higher watts per rack drive most of the TAM expansion.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  AI power-architecture transition from server power supplies to HVDC power racks and solid-state transformers

  UBS traces how a higher-voltage architecture shifts demand and value toward distribution, protection, controls, conversion, cooling and energy-storage suppliers.

Asset mapping & comparison

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

• Vertiv Holdings (VRT)

  Buy-rated US electricals supplier exposed to AI-infrastructure investment and the convergence of power and cooling.

  Strengths

  Strong positioning in rack conversion, 800V/HVDC, cooling and grey-space/medium-voltage systems in UBS's competitive map.

  Comparison

  UBS identifies power-and-cooling convergence as its distinguishing position.

  Risks

  Project delays, power constraints and a slower 800VDC transition could affect demand.

• Schneider Electric

  Buy-rated European capital-goods preference and the report's most directly exposed European name to the 800VDC trend.

  Strengths

  Positions in switchgear, distribution and controls, supported by a stated 800VDC roadmap.

  Comparison

  UBS views Schneider as the most directly exposed European capital-goods company to the architecture transition.

  Risks

  Commercialization timing for sidecar and solid-state offerings.

• ABB

  Neutral-rated supplier regarded as strongly positioned for the 800VDC transition.

  Strengths

  Infinitus solid-state-transformer launch and exposure to medium-voltage equipment, distribution infrastructure and power management.

  Comparison

  UBS sees ABB as well positioned but maintains a Neutral rating.

  Risks

  Dependence on the pace of higher-voltage adoption and data-center project execution.

• Legrand

  Neutral-rated supplier for which the CMD could clarify growth and 800VDC positioning.

  Strengths

  Data-center sales are about 30% of group sales, compared with about 20% previously.

  Weaknesses

  Investor concerns over rack-level distribution and 800VDC positioning.

  Comparison

  UBS sees a sidecar launch and 1MW-rack roadmap as important relative to competitors.

  Risks

  Failure to provide credible product and growth-roadmap updates could leave positioning concerns unresolved.

• Panasonic

  Beneficiary of continued demand for rack-level cylindrical cells and sidecar backup modules.

  Strengths

  Leverage to rack-level batteries and sidecar backup modules.

  Risks

  Backup-power placement may shift increasingly toward facility-level systems.

• Samsung SDI

  Potential beneficiary of both rack and building-level UPS battery deployment.

  Strengths

  Exposure to alternative battery-placement outcomes.

  Risks

  Demand depends on the extent to which reserve energy migrates to facility-level systems.

Key data

• US affected data-center capacity7,010MW current; 16,910MW ultimateNine confirmed delay/cancellation cases; four delays and five cancellations/withdrawals.
• Affected capacity versus US base12% of operational capacity; 5% of pipeline capacityUBS Evidence Lab shows 57GW operational and 130GW pipeline capacity in North America.
• Affected capacity versus Europe base6% of operational capacity; 3% of pipeline capacityUBS Evidence Lab shows 17GW operational and 30GW pipeline capacity in Europe.
• AI power-electronics TAMUS$6.6bn in 2025 to US$91.1bn in 203069% CAGR according to UBS estimates.
• Rack power density55.4kW in 2025 to 282.8kW in 203039% CAGR in the UBS AI power TAM model.
• AI IT capacity deployed11.5GW in 2025 to 104.5GW in 203055% CAGR in the UBS model.
• AI-server BBU TAMUS$0.7bn in 2025; US$2.5bn in 2026; US$12.8bn in 2030Combined BBU+CBU opportunity is estimated at US$15.9bn in 2030.
• Legrand consensus organic growth9.6%, 8.7%, 6.5%, 5.6% and 4.7% for FY2026-FY2030Visible Alpha consensus; UBS sees scope for a mid-term organic-growth-target upgrade.

Impact & implications

UBS argues that AI buildout continues to favor electrical distribution, power conversion, controls, cooling and backup-power suppliers. However, the benefit will vary with the ability to execute against increasingly constrained grid capacity, secure credible funded projects and participate in the evolving 800VDC architecture.

Risks

• Grid availability and hyperscaler cash flows could moderate AI-infrastructure growth later in the decade.
• US semiconductor assumptions imply AI-capacity additions above prevailing annual market expectations, creating a potential future digestion period.
• A slower transition to 800VDC sidecars could reduce power-semiconductor TAM by roughly US$0.5-1.0bn.
• Large compute agreements increase financing, customer-concentration and counterparty risks.
• Project regulation, grid-connection screening, re-phasing and relocation can increase timing and execution risk.
• AI-governance commitments could matter more if they become measurable and enforceable restrictions on capability scaling.

What to watch

• Changes in the project-delay/cancellation tracker and whether delays remain contained relative to operational and pipeline capacity.
• Data Center World Singapore and Legrand's 29-30 September product, growth-guidance and 800VDC-roadmap announcements.
• ABB's 24 September investor webinar and further solid-state-transformer commercialization updates.
• The pace of sidecar and 800VDC adoption versus intermediate 50V solutions.
• Grid-connection policies, deposits and financial-security requirements, especially in Texas and other high-demand US regions.
• Evidence of measurable AI safety thresholds, independent evaluators and enforceable commitments from frontier labs.