800 VDC Power for AI Data Centers: What to Design Now
Quick answer: 800 VDC power distribution converts utility power to 800-volt direct current once, at the edge of the data hall, instead of stepping through multiple AC conversions. It cuts copper, equipment and losses for 600 kW to 1 MW racks. Facilities designed today should reserve space for sidecars, DC busways and energy storage.
When a single rack draws as much power as a small neighborhood, the way electricity reaches it has to change. The industry's answer is 800-volt direct current, and the facilities going into design this year need to be ready for it.

Why doesn't today's AC power chain work for megawatt racks?
A conventional data center steps grid power down to 415 V AC, runs it through a UPS and PDUs, then converts it again to low-voltage DC inside every server (NVIDIA). Each conversion wastes energy and adds equipment. At 600 kW to 1 MW per rack, the copper needed to carry that current at 54 V becomes physically impractical (Converge Digest).
NVIDIA's 800 VDC design converts 13.8 kV AC to 800 VDC at the perimeter of the data hall and removes most intermediate steps. NVIDIA is targeting 1 MW racks and beyond starting in 2027, with Eaton, Schneider Electric and Vertiv among its power partners (NVIDIA).

When will 800 VDC become standard?
In August 2026, NVIDIA, Google and Microsoft shared version 0.3 of a Solid-State Transformer specification through the Open Compute Project (Data Center Richness). The shift is expected to take hold with NVIDIA's Vera Rubin Ultra platform in 2027, at up to 600 kW per rack. Solid-state transformers promise better efficiency and less floor space, but their supply chain is still maturing.
What is a power sidecar?
Few owners will replace an AC plant on day one. The bridge is the power sidecar, a dedicated 800 VDC cabinet beside the IT rack. Vertiv's PowerDirect 5000 sidecar delivers 400–900 kW into the GPU rack over a busbar (Vertiv). Owners keep their upstream AC gear and bring DC in only where the compute needs it. For site-wide 800 VDC, Vertiv describes two paths running in parallel: a medium-voltage DC UPS and the solid-state transformer.
What should you design in today?
- Space for sidecars. Lay out white space so every AI rack can gain a neighbor cabinet without re-planning the row.
- Busway routes, not just cable. Overhead DC busways and rack busbars change tray, clearance and structural loads.
- Room for storage. GPU clusters swing power in under a second, and NVIDIA's architecture uses energy storage to absorb those swings (NVIDIA).
- DC-qualified people. 800 VDC arc-flash, isolation and lock-out procedures differ from AC. Build training into the commissioning and turnover plan.
How do you align the facility with the rack roadmap?
Power architecture decisions should follow the hardware you'll actually deploy. Rack-scale vendors like MiTAC now ship complete liquid-cooled racks with integrated monitoring across racks, power and cooling (MiTAC Computing), so the electrical design can be matched to a known rack rather than a guess. Upstream, our sister division GRID7 designs microgrids that pair on-site generation with battery storage, the same storage 800 VDC architectures rely on to absorb GPU load swings. And our affiliate Gray Wolf Data Centers builds high-density power distribution into its standard 2–10 MW AI/HPC facility design.
A hall built today can run on AC and still be ready for DC. The decisions that make that possible are cheap on paper and expensive in concrete.
Frequently asked questions
Why are AI data centers moving to 800 VDC?
At 600 kW to 1 MW per rack, low-voltage power needs impractical amounts of copper. 800 VDC carries the same power with less current, fewer conversion steps and lower losses.
Can an existing AC data center support 800 VDC racks?
Yes, through power sidecars: an 800 VDC cabinet beside each AI rack that is fed from the existing AC system. It lets owners add DC only where the compute needs it.
What is a solid-state transformer?
A power-electronics transformer that can convert medium-voltage AC directly to 800 VDC in less space than conventional gear. An OCP specification is in development, but supply is still limited.
How INTUVA helps: our design and engineering team plans power architectures that serve today's racks and adapt to tomorrow's. Schedule a call to review your electrical roadmap.
Disclosure: Pete Sacco is the Founder and CEO of PTS Data Center Solutions and Gray Wolf Data Centers. INTUVA and GRID7 are divisions of PTS. Gray Wolf Data Centers is a partnership between PTS, ReNew Developers, Tonomia and MiTAC.