NVIDIA has announced a transformative end-to-end 800-volt high-voltage DC (HVDC) power infrastructure for next-generation data centers, targeting the immense demands of Artificial Intelligence workloads. Traditional 54 V DC in-rack power delivery systems, originally designed for only tens of kilowatts per rack, are unable to scale to the 1-megawatt range now required by cutting-edge AI facilities. NVIDIA introduced this new HVDC solution during GTC and Computex 2025, with its rollout planned for 2027, as the company’s GB200 and GB300 NVL72 systems already draw as much as 132 kW per rack—far more than the typical 50 to 80 kW capacity of most existing data centers.
The move to 800 V HVDC infrastructure addresses critical scalability and efficiency challenges. If conventional 54 V distribution were used for megawatt-class racks, racks would need nearly all available chassis space to accommodate massively thick copper busbars, leading to enormous material costs—potentially requiring nearly half a million metric tons of copper for a 1 GW installation. By converting 13.8 kV AC directly to 800 V DC at the grid edge, then distributing it through row-level busways and stepping down the voltage with in-rack DC-DC modules for GPUs, NVIDIA’s system streamlines the entire process. This consolidated approach removes redundant AC-to-DC and DC-to-DC conversions, reduces the number of fans and power supplies, and results in a much simpler electrical system.
Efficiency improvements are central to NVIDIA’s HVDC architecture, with up to a 5 percent increase in end-to-end efficiency and a 45 percent cut in copper consumption. The result is not only lower infrastructure and electricity costs but also a significantly reduced environmental and physical footprint. NVIDIA’s initiative is backed by an ecosystem of industry partners: Infineon, MPS, Navitas, ROHM, STMicroelectronics, and Texas Instruments supply the power electronics; Delta, Flex Power, Lead Wealth, LiteOn, and Megmeet handle power shelving; while Eaton, Schneider Electric, and Vertiv work on protective devices and infrastructure standardization. NVIDIA’s approach sets a new bar for the power delivery backbone demanded by the exponential growth of Artificial Intelligence data center deployments worldwide.
