SemiAnalysis In-Depth Analysis: Under the Wave of 800VDC, the Power Equipment Market Is Quietly Being Reshaped

SemiAnalysis In-Depth Analysis: Under the Wave of 800VDC, the Power Equipment Market Is Quietly Being Reshaped

The power architecture of data centers is at a historic turning point. As GPU rack power density leaps toward the 600kW range, the 800VDC direct current distribution revolution driven by physical laws has evolved from internal experiments by hyperscale cloud providers into an industry-wide systemic reconstruction that cannot be avoided. This transformation will profoundly reshape the multibillion-dollar power equipment market in the next few years.

According to a recent in-depth report by industry research firm SemiAnalysis, the core logic of 800VDC is based on physical constraints: At 600kW rack power, raising the distribution voltage from 54V to 800V reduces the current by approximately 15 times and conductor resistive heat losses by about 219 times, greatly cutting copper usage, lowering heat load, and reducing conversion losses. For a 1GW IT load, a full migration to an 800VDC architecture can result in about 69MW of continuous grid energy savings, amounting to tens of millions of dollars in annual electricity cost reductions, or an equivalent increase in computing capacity.

SemiAnalysis forecasts that by 2030, data centers covered by 800VDC will add about 39GW of capacity, with matching power equipment market forming two core categories: Power Rack/Sidecar market peaking at around $11 billion in 2028, and Solid State Transformer (SST) market at about $13 billion by 2030. This process will advance through four stages, each with different equipment changes and vendor reshuffling logic.

Why 800VDC is Inevitable: A Physical Constraint-Driven Architecture Revolution

Currently, most data centers use 415V or 480V three-phase AC distribution, with 48-54V DC power inside racks. This architecture works at today’s rack power levels, but with next-generation GPU clusters (such as Nvidia Kyber Ultra) pushing single rack power to 660kW, the existing low-voltage system faces three physical limits.

First is the runaway copper weight. At 48-54V, a 1MW rack needs about 200kg of busbar copper; at 1GW scale, total copper reaches hundreds of tons, with costs, weight, and installation complexity exceeding engineering feasibility. Second, rack space is swallowed by power devices: the current NVL72 rack already occupies up to 8 power shelves, and sticking to low voltage would see Kyber-class racks’ power hardware filling the entire rack, leaving no room for compute units. Third, current becomes a bottleneck: at 600kW, 54V distribution needs to carry about 12,500A; switching to 800V lowers current to about 750A, dramatically reducing conductor size and heat stress.

SemiAnalysis notes that larger-scale computing domains mean denser racks, denser racks mean the 600kW power envelope — and 800VDC is the physical enabling technology that makes this power envelope feasible.

Four-Stage Migration Path: From Rack Sidecar to Solid State Transformer

SemiAnalysis divides the 800VDC migration into four stages, stretching from 2026 to after 2029.

Stage 1 (2026/2027): White Space Retrofit. Migration is led by Google and Meta, who have been advancing 800VDC architecture via OCP working groups for over 18 months, jointly with Microsoft developing the Diablo 400 open standard. The core device is the row-level HVDC power sidecar: a 42U standalone cabinet receiving AC from the top busway, outputting 800VDC to neighboring IT racks, integrating rectification, BBU battery backup module, and optionally supercapacitor buffering. Existing transformers, UPS, and switchgear remain unchanged. SemiAnalysis estimates ASP for this stage to be about $400–500k/unit, or about $500k/MW, roughly 10 times higher than current AC power devices ($40k/unit).

Stage 2 (2027/2028): Native 800VDC Computing Arrives. As native 800VDC chip systems (like Kyber racks) enter mass production, 800VDC becomes a physical requirement instead of voluntary early adoption. Architecture is similar to Stage 1, but the voltage reduction point moves from the rack’s internal power shelf to onboard power modules on compute blades. At the same time, centralized low-voltage UPS systems gradually retire, replaced by rack-level BBU and supercapacitors, with Google and Meta already deploying this “distributed UPS” architecture.

Stage 3 (2028/2029): Full Electrical Architecture Rewrite. 800VDC distribution moves up from the rack row to facility level, with dedicated rectifiers in grey space directly converting 415V AC to 800VDC, distributed via DC busway throughout the data hall. AC panels and floor PDUs exit the main power loop. In white space, power sidecars are replaced by “battery racks” — no longer rectifying AC to DC, directly receiving 800VDC from grey space, retaining only DC distribution units, BBU shelves, and supercapacitors. SemiAnalysis estimates battery rack content at about $200k/MW.

Stage 4 (Post 2029): Solid State Transformer Final State. Solid State Transformers (SST) directly convert medium-voltage AC to 800VDC, replacing both low-voltage transformers and rectifiers in one device. In theory, this boosts system efficiency from the current ~82% to over 87%, achieves ~40x reduction in weight and 14x reduction in volume. SemiAnalysis forecasts SST market size to reach about $1.3B by 2030.

Power Sidecar and SST: Market Opportunities for Two Core Devices

At the device market level, SemiAnalysis’ industrial model breaks down 800VDC device content by MW for each stage.

Power Sidecar (Sidecar/Power Rack) is the core incremental device for Stages 1 and 2. SemiAnalysis expects its market to peak at about $11 billion in 2028, then decline as facility-level 800VDC distribution becomes widespread in Stage 3. Diablo 400 specification, jointly developed by Google, Meta, and Microsoft, establishes a multi-vendor interoperability standard, with Delta, Advanced Energy, TE Connectivity, etc. participating. Notably, Nvidia did not adopt Diablo 400; instead, it independently developed a 660kW single-pole 800V reference design, with air-cooled version mass produced mid-2026, and liquid-cooled samples expected by the end of 2026.

Solid State Transformer (SST) is the Stage 4 endpoint device, and the hottest area for fundraising. SemiAnalysis reports that SST startups raised over $320 million in the 12 months ending March 2026. Key players include: DG Matrix (backed by ABB, SiC supply deal with Infineon, the only SST product listed in Nvidia MGX reference architecture, aiming for UL certification in late Q2 2026); Amperesand (targeting 30MW commercial deployments in 2026); Heron Power (building a US factory with 40GW annual capacity, focused on 4.2MW direct medium voltage input products); Novos Power (focused on direct medium-voltage to 800VDC conversion, claims 50% space reduction). Among incumbents, Eaton acquired Resilient Power Systems for SST tech in August 2025.

On efficiency benchmarks, ETH Zurich’s best public test (INTELEC 2025) showed a prototype achieving 98% efficiency converting 13.2kVAC to 800VDC at 400kW. DG Matrix, Amperesand, Heron Power all claim 98.5% efficiency, but actual data center deployments require 3–6MW units sustaining 99% efficiency under continuous load — a target not yet met by any vendor.

Four Major Challenges: Key Variables Determining 800VDC Penetration Speed

SemiAnalysis identifies four core barriers which will directly affect how quickly 800VDC moves from hyperscale cloud trials to broader markets.

Regulation and Safety. National Electrical Code (NEC) in the US aims to fully support 800VDC in NEC 2029 version; deployments before 2029 require site-by-site local authority approval. SemiAnalysis expects some 800VDC code coverage in NEC 2029, with complete maturity possibly not until NEC 2032/2035. On safety, IEEE 1584 does not cover DC systems, NFPA 70E lacks PPE tables for 600–1000VDC, and UL Solutions has begun DC safety research consortiums to fill these gaps.

Cooling and Auxiliary AC Loads. Cooling is the largest AC load in an 800VDC data center, and no vendor yet offers a full DC-native cooling ecosystem. Delta released a 2.4MW in-row cooling distribution unit supporting 800VDC at GTC 2026 — the first major DC-native cooling component — but the full cooling stack (chiller, compressor, pumps, building controls) still depends on AC supply. Nvidia explicitly stated at OCP Global Summit 2025 that their 800VDC reference design will retain an auxiliary AC bus.

Supply Chain and Standards Lag. Tech innovation in DC distribution runs ahead of standard-setting. For example, UL 857’s 14th edition (2025) raises the covered voltage from 600V to 1000VDC, with the 15th edition still being developed. In the SST field, as of May 2026, no vendor has completed UL certification for data center deployment. OCP working group is coordinating with regulators to push early standards by the end of 2026.

Grid Interconnection and Regulatory Pressure. NERC issued a level-three “basic action alert” in May 2026 covering large computing loads, with mandatory response deadline of August 3, 2026, and proposed “computing load entity” registration for data centers consuming over 1MW. ERCOT’s NOGRR282 adds voltage and frequency ride-through requirements and mandates large loads submit PSS/E and PSCAD electromagnetic transient models. SemiAnalysis notes that grid response behaviors in 800VDC sites depend on SST control algorithms, storage charge state, and GPU load curve, making them much more complex than traditional AC data centers and spawning new AI-native EPC providers like Aran Industries.

Total Electrical Cost Remains Stable, Content Structure Deeply Reshaped

SemiAnalysis’ model shows that across all four stages, total electrical equipment cost per MW stays within $3.6–4.8 million range, with little overall change. What does change is the structural migration of equipment content: grey space equipment value shrinks as centralized UPS ($1.2M/MW) is phased out, white space value peaks in Stage 1 with HVDC power sidecars, and rises again in Stage 4 as SST ($1–1.5M/MW) replaces low-voltage transformers and rectifiers.

Paths for improved efficiency are also clear: baseline AC architecture end-to-end efficiency is about 82%; Stage 2 (eliminating double UPS conversion) jumps to 86.5%; Stage 3 reaches 86.9%; and Stage 4 goes to 87.4%. For a 1GW IT load, Stage 4 efficiency boost versus baseline is about 5 percentage points, matching publicly cited Nvidia data.

For investors, the core logic of this transformation is: total market size does not expand dramatically, but redistribution of equipment content will profoundly alter income trajectories of existing vendors — some legacy equipment categories will shrink or disappear, while new categories’ market space is rapidly opening.

Risk Disclosure and DisclaimerThe market entails risks, and investments should be made cautiously. This article does not constitute personal investment advice, nor does it take into account individual users’ specific investment objectives, financial situations, or needs. Users should consider whether any opinions, views, or conclusions in this article fit their particular circumstances. If you invest based on this, you do so at your own risk.