Data-center electrical equipment implications of NVIDIA 800VDC architecture Report Interpretation

Overview

This expert-call note examines how NVIDIA's proposed 800VDC architecture may reshape demand for data-center electrical equipment. Its central conclusion is that the transition is progressing more slowly than previously expected, leaving a substantial and potentially higher-margin market for conventional power architectures despite continued AI-driven data-center growth.

Core views

The report lowers its expectations for 800VDC adoption because AI inference is growing faster than expected and does not require the same power density as AI training. It now expects 800VDC to represent 21% of US new-build data centers in 2030, down from 25%, and 17% in Europe, down from 20%. Inference workloads require up to 150kW per rack at maximum, according to the expert, allowing traditional data-center power architecture to remain viable. For inference, the likely upgrade is liquid cooling rather than a full 800VDC redesign. The transition is also expected to be gradual. Full 800VDC architectures are not anticipated for at least five years; the more likely near-term configuration would use 800VDC for servers while networking, storage and lighting retain traditional architectures. NVIDIA's Kyber NVL144 rack systems, designed for Rubin Ultra GPUs, have reportedly been pushed to 2028, more than 12 months behind the original 2027 target. These timing shifts reinforce the report's view that existing electrical-equipment demand should remain important for longer. The expert expects non-AI data centers to remain dominant in 2032. Microsoft is cited as an illustration of the sector's scale, with a target of up to 38GW of installed power by 2032, implying roughly 20% CAGR over the next six years. Non-AI data centers are projected to account for two-thirds of that installed base by 2032, versus about 83% currently, still implying 17% CAGR. AI data centers would account for one-third, rising from about 17% currently and implying 35% CAGR. Thus, AI infrastructure grows faster, but traditional power architecture remains the larger addressable market and could retain higher margins. The report specifically notes that this supports suppliers such as Legrand even without announced SST development. Manufacturers are pursuing different paths to 800VDC. Delta Electronics is piloting SSTs in Asia and targets a 5MW design. Eaton plans to sample a 2MW SST with customers by end-2026, leveraging its 2025 acquisition of Resilient Power. Schneider Electric is promoting transformer rectifier units and DC rectifiers, while Vertiv and ABB are offering medium-voltage UPS products that remove some conversion steps but still rely on traditional technology. GE Vernova has announced a 175MW MV UPS, versus ABB's 1-2MW offering, and expects to introduce SSTs in 2027. Start-up Heron Power is developing a 5MW SST and expects its production facility to be ready by end-2027. The report sees high costs, qualification requirements and timing as constraints on SST deployment. Hyperscalers are issuing large SST RFPs, which the expert characterizes as price- and volume-sensitivity exercises while they pilot products from multiple vendors. SST pricing remains about $400-500k per MW because manufacturing scale is limited. After a preferred technology is selected, evaluation and ordering are expected to require a further 12-18 months, placing the earliest mass-deployment window in 2029. Solar entrants Enphase and SolarEdge may benefit from their DC-distribution knowledge, but face demanding data-center requirements for redundancy and field service. Attention is shifting toward other required components, particularly solid-state breakers. Traditional circuit breakers operate in milliseconds, whereas 800VDC requires microsecond switching. The expert estimates solid-state breakers at about $600k per MW, roughly five times the price of current breakers and above SST pricing. As higher-cost equipment enters the stack, total powertrain content for AI training is projected to reach about $4 million per MW by 2030, from $1.7-1.8 million per MW in 2025; for AI inference, it is projected to reach $2.6 million per MW, from $1.4 million per MW. The report also argues that enterprise inference workloads generally do not need massive high-density compute clusters outside specialized use cases such as pharmaceutical drug discovery. Near-term catalysts are the OCP Summit on 12-15 October and NVIDIA GTC on 20-22 October. The report expects OCP and NVIDIA to continue setting the technology agenda, while OEMs prioritize speed to market. It expects meaningful competitive differentiation to become a greater focus only in four to five years, when second-generation SSTs begin to be developed.

Analysis framework

The report synthesizes an investor webcast with data-center veteran Herve Tardy, comparing AI training and inference power requirements, revising regional 800VDC penetration assumptions, and tracing the implications through data-center capacity, equipment product roadmaps, component costs, vendor qualification and deployment timing.

Methodology notes

• Industry AnalysisSupply-demand framework

  Data-center demand and power-architecture adoption analysis

  The report compares the growth of AI and non-AI data-center capacity with the power-density needs of training and inference workloads to judge demand for traditional systems, 800VDC equipment and SSTs.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Powertrain component and vendor-transition analysis

  The report links hyperscaler procurement and workload requirements to equipment vendors' product choices, including SSTs, rectifiers, MV UPS and solid-state breakers.

Asset mapping & comparison

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

• NVIDIA

  Its proposed 800VDC architecture and delayed Kyber NVL144 timing are central to the sector transition discussed.

  Strengths

  Drives technology thought leadership in data-center power architecture.

  Weaknesses

  Kyber NVL144 systems were reported as delayed to 2028.

  Comparison

  Along with OCP, NVIDIA is described as a technology thought leader relative to OEMs focused on speed to market.

  Risks

  Slower 800VDC adoption limits the near-term pace of full-architecture deployment.

• Legrand

  Traditional electrical-equipment supplier positioned to serve the enduring conventional-architecture market.

  Strengths

  Can address a large traditional power-architecture market despite not announcing SST development.

  Weaknesses

  No announced SST development.

  Comparison

  Contrasts with manufacturers actively piloting SSTs or intermediate 800VDC products.

  Risks

  Longer-term architecture changes could increase the importance of SST-related capabilities.

• Eaton

  Electrical-equipment manufacturer developing an SST product.

  Strengths

  Plans customer sampling of a 2MW SST by end-2026 and can leverage the Resilient Power acquisition.

  Weaknesses

  Product remains in the sampling stage.

  Comparison

  Targets a smaller SST design than Delta Electronics' planned 5MW system.

  Risks

  Customer testing, technology selection and ordering cycles delay broad SST deployment.

• Schneider Electric

  Electrical-equipment manufacturer pursuing intermediate 800VDC products.

  Strengths

  Promotes transformer rectifier units and DC rectifiers for the transition.

  Weaknesses

  Its current approach is not a full SST solution.

  Comparison

  Similar transitional positioning to Vertiv and ABB, versus SST pilots at Delta and Eaton.

  Risks

  The pace and final design of the 800VDC transition remain uncertain.

• GE Vernova

  Electrical-equipment manufacturer offering MV UPS while pursuing SST development.

  Strengths

  Has announced a 175MW MV UPS and expects SST introduction in 2027.

  Weaknesses

  SST commercialization remains prospective.

  Comparison

  Its MV UPS scale is cited as larger than ABB's 1-2MW offering.

  Risks

  Mass SST deployment is not expected before 2029.

Key data

• US 800VDC share of new-build data centers in 203021%Revised down from 25%
• Europe 800VDC share of new-build data centers in 203017%Revised down from 20%
• Maximum AI-inference rack power requirementUp to 150kW per rackSupports continued use of traditional architecture
• Microsoft targeted installed power by 2032Up to 38GWCited as implying about 20% CAGR over the next six years
• Non-AI data-center share of the 2032 installed baseTwo-thirdsDown from about 83% currently but still implying 17% CAGR
• AI data-center share of the 2032 installed baseOne-thirdUp from about 17% currently and implying 35% CAGR
• SST priceAbout $400-500k per MWHigh because manufacturing scale remains limited
• Solid-state breaker priceAbout $600k per MWAbout five times current circuit-breaker pricing
• AI-training powertrain content in 2030About $4 million per MWUp from $1.7-1.8 million per MW in 2025
• AI-inference powertrain content in 2030$2.6 million per MWUp from $1.4 million per MW in 2025

Impact & implications

The report argues that lower and later 800VDC adoption does not weaken the broader electrical-equipment opportunity: conventional architectures should continue serving most data-center capacity through 2032, while the eventual transition raises equipment content per MW. Near-term benefits appear more favorable for established traditional-equipment suppliers, whereas broad SST deployment depends on cost reduction, qualification and customer selection.

Risks

• 800VDC adoption is progressing more slowly than previously anticipated because inference workloads can continue using traditional architectures.
• SST costs remain high and manufacturing scale is limited.
• SST suppliers must meet demanding data-center standards for redundancy and field service.
• Vendor testing, technology selection and ordering cycles could delay mass SST deployment until 2029.

What to watch

• The OCP Summit on 12-15 October.
• NVIDIA GTC on 20-22 October.
• Progress in SST pilots, customer sampling and preferred-vendor selection.
• Development of second-generation SSTs and competitive differentiation over the next four to five years.