800V + Solid-State Transformers: The "Silicon-based Revolution" for AI Data Centers is Here!
With ChatGPT processing millions of requests per second and autonomous driving models performing terabyte-level data training, AI data centers are facing an unprecedented "power crisis." Over the past five years, GPU rack power has surged from under 10kW to a starting point of 100kW, with some high-end clusters even approaching the megawatt level.
At the 2025 OCP Global Summit, NVIDIA's white paper, 800VDC Architecture for Next-Generation AI Infrastructure, provided a clear answer: 800V DC power supply + SST (Solid-State Transformer) will become the ultimate power solution for next-generation AI data centers.
I. From AC to DC: A Disruptive Change in Power Architecture
Current data centers predominantly use AC distribution, where electricity undergoes multiple conversions before reaching the GPUs, resulting in low efficiency and significant losses. The new architecture will complete the centralized conversion from AC to DC at the facility level, establishing local 800V DC data centers:
- Medium-voltage AC is directly converted to 800V DC via high-capacity power conversion systems.
- High-voltage DC is distributed directly to each compute node.
- It is stepped down in a single stage to the 12V DC required by GPUs using high-ratio LLC converters.
This architecture not only simplifies the power delivery path but also reduces the footprint by 26%, freeing up valuable space around the processors.
II. From Concept to Deployment: AI Power Enters the Practical Stage
The 800V + SST solution is no longer just theoretical:
- Foxconn's Kaohsiung No. 1 Data Center has already adopted this solution for a 40-megawatt unit.
- The Sidecar architecture, jointly developed by Schneider Electric and NVIDIA, supports 1.2-megawatt racks.
- More than 20 companies, including CoreWeave, have initiated related designs.
Industry Predictions:
- 2025-2026: 800V HVDC enters the adoption phase.
- From 2027 onwards: SST will see large-scale application.
III. Beyond Power Supply: Green and Efficient
The significance of this transformation extends far beyond power supply itself:
- The inherent compatibility of SST with photovoltaics and energy storage helps data centers achieve "full green power coverage."
- Gradual phase-in of wide-bandgap semiconductor materials like Silicon Carbide (SiC) and Gallium Nitride (GaN) is driving upgrades across the power electronics industry.
As an NVIDIA executive stated: "A scalable power architecture is the core engine of the AI factory."
IV. Yiyue Supports Selection, Global Preferred Solutions Are Ready
As SST systems move from concept to deployment, the Hitachi Energy LinPak power module, represented by Yiyue Singapore, provides crucial support: its wide voltage range covers 1700V–3300V, with a maximum current of 1000A, offering a highly reliable, globally validated preferred power electronics solution for the efficient implementation of SST products.
From GPUs to power systems, the focus of value in the AI industry is gradually shifting towards infrastructure. When 800V current flows through SSTs into compute clusters, it carries not only electrical energy but also new opportunities in the global technological competition.
Our Official Alibaba Store：https://sg1568697071rwko.trustpass.alibaba.com/
