Overview: The Shift to 800VDC and Solid State Transformers

Nvidia's August 2026 announcement that its 800VDC architecture is officially in production marks a pivotal moment, spearheading the industry's transition to 800V HVDC data center power infrastructure by 2027. Concurrently, solid-state transformers (SST) are emerging as the ultimate power supply solution, securing major orders for next-generation AI data centers. As AI rack power scales beyond 600kW, traditional 54V systems approach their engineering limits in copper loss and thermal management. This shift addresses the physical limits of traditional power delivery, ensuring sustainable scaling for AI computing.

Technical Details: Overcoming the Physics of AI Power Delivery

Traditional data center power chains rely on power-frequency transformers and multi-stage DC-DC conversions, yielding only 95% efficiency. The 800VDC architecture mitigates copper losses and thermal constraints by increasing voltage and decreasing current. The SST represents the final form of this evolution. By replacing traditional iron cores with silicon carbide (SiC) and gallium nitride (GaN) semiconductors, SSTs elevate frequencies from 50Hz to 10kHz. Utilizing a three-stage topology, SSTs enable a direct, one-step conversion from 10kV medium-voltage AC to 800V DC, achieving system efficiencies exceeding 98.5% and reducing equipment volume by over 50%. However, commercialization faces three core thresholds: high SiC device costs, complex DC fault isolation, and rigorous global certification cycles of 12 to 18 months.

Market Context: Projecting the 800V HVDC and SST Boom

The market for 800V DC distribution is expanding rapidly, though estimates vary by scope. Hengzhou Chengsi projects the global 800V DC distribution system revenue to grow from 521 million RMB in 2025 to 1.87 billion RMB by 2032, a 57.7% CAGR. QYResearch estimates the data center-specific 800V DC market at $316 million in 2025, reaching $663 million by 2032. Meanwhile, the SST market is poised for explosive growth. SemiAnalysis estimates that 800VDC-covered data center incremental capacity will reach 39GW by 2030, with the SST market hitting $13 billion. Huaan Securities forecasts the global SST market to surge from 280 million RMB in 2026 to 101 billion RMB in 2030, reflecting a 245% CAGR. Currently, SST unit costs remain 10% to 20% higher than traditional transformers, but economies of scale are expected to narrow this gap.

Industry Impact: Global and Domestic Competitive Landscape

The global landscape shows overseas leaders pioneering while domestic Chinese manufacturers rapidly catch up. Internationally, Eaton, ABB, and Infineon are deeply integrated with top AI clients. ABB's Infinitus 800VDC targets 1MW racks, while Infineon has secured multi-year capacity reservations with major AI cloud providers, involving potential revenues in the high billions of euros. In China, Sungrow leads industrialization with its EnerNeo series, achieving 98.5% conversion efficiency and boasting a 1.2GW capacity plan by late 2026 alongside 130MW framework agreements. China XD Electric secured the first overseas data center order for domestic SSTs, and Sifang Co. holds over 3 billion RMB in SST-related order reserves. Upstream, CR Micro is advancing high-voltage SiC devices to support this ecosystem.

Implications for AI Data Centers: The Transition Roadmap

The transition follows a clear four-stage roadmap. From 2026 to 2027, the focus is on white space retrofits using HVDC power cabinets. Between 2027 and 2028, native 800VDC computing emerges as rack-level BBUs and supercapacitors gradually replace traditional UPS systems. Facility-level 800VDC distribution will form between 2028 and 2029. Finally, post-2029, the industry will enter the SST end-state, completely restructuring data center power architectures. During this transition, 800V HVDC serves as a crucial bridge, with the sidecar power cabinet market projected to reach 103.5 billion RMB by 2028, offering immediate revenue realization for the supply chain.

Future Outlook: Commercialization Timelines and Investment Realities

While the consensus views SST as the definitive solution for AI data centers, commercialization timelines require careful calibration. Although 2026 is deemed the commercialization inception year, mass production is expected around 2027, with large-scale deployment delayed until 2028. Consequently, substantial financial contributions from SST to corporate earnings will likely not materialize until 2028 or later. The window period for transition solutions like HVDC should not be overlooked, as sidecar power cabinets offer immediate revenue realization. Investors and industry stakeholders must differentiate between technical milestones and revenue milestones. Ultimately, without SST, the power efficiency of megawatt-level racks cannot transcend physical limits, making it indispensable for the sustainable expansion of global AI computing power.