Details
- NVIDIA said its 800 VDC power architecture is designed to help AI factories scale beyond the limits of traditional AC distribution, with fewer conversion stages between the grid and GPUs.
- The effort involves NVIDIA, Google, and Microsoft working through the Open Compute Project, with more than 80 equipment and infrastructure companies already building to the specification.
- NVIDIA’s MGX-compatible 800 VDC power rack is slated for the second half of 2026, while a row power center is expected in 2027 and is described as supporting up to 2 megawatts per row.
- The company says the hybrid rack approach can be deployed in existing AC facilities without changing building electrical systems, preserving prior infrastructure investments while increasing compute density.
- Web material around the architecture says 800 VDC reduces conversion losses, copper use, and cable bulk versus 54 VDC rack systems and 480 VAC distribution, and NVIDIA frames it as part of a broader shift toward native DC data centers.
Impact
The announcement reinforces a broader industry move toward high-voltage DC power for AI infrastructure, especially as rack densities approach the 1 MW range. If the ecosystem coalesces around shared specifications, it could accelerate vendor interoperability, reduce deployment friction, and shape data-center capex plans over the next 12–24 months.
