Nvidia's 800 VDC Ecosystem Reshapes AI Data Center Power for the Megawatt Era

AI computing is pushing data center power infrastructure past its breaking point. The last major rack distribution standard, a 54-volt direct current system, was designed for an era of kilowatt-scale racks. With Nvidia's Blackwell GPUs and the forthcoming Vera Rubin platform driving single racks toward a megawatt or more, the old way is no longer viable—both physically and economically.

In response, Nvidia has been building out a new ecosystem around an 800-volt direct current (VDC) power architecture under its MGX AI Factory program. The initiative, designed to support 1 MW IT racks and beyond starting in 2027, is not just a simple voltage tweak. It is a fundamental re-architecture of how electricity moves from the grid all the way down to the silicon core of a GPU—and a coalition of the biggest names in power electronics has now signed on .

Infineon Joins the Grid-to-Core Flow

The most recent and comprehensive development came on May 28–29, 2026, when Infineon Technologies officially joined Nvidia's MGX AI Factory ecosystem . Infineon doesn't just supply a component; it aims to cover the complete 800 VDC power conversion flow. This means managing power from the high-voltage DC input at the rack, through an intermediate bus converter, and down to the final core voltages feeding the GPU .

The company is leveraging its entire materials toolbox—silicon (Si), silicon carbide (SiC), and gallium nitride (GaN)—to achieve this. On the cutting edge, Infineon is using GaN technology switching at frequencies near 1 MHz to build ultra-compact bus converters that maintain high efficiency . The company's hot-swap controller technology is also being adapted so that future server motherboards can operate natively at 800 VDC, a key step for serviceability and safety .

Why 54 V Had to Die

The efficiency and infrastructure gains from moving to 800 V are stark. The legacy 54 V bus was designed when a fully loaded rack might draw a few dozen kilowatts. Now, as GPU power density has increased by a factor of 3.4× from Hopper to Blackwell, those busbars are hitting hard physical limits .

Nvidia has detailed the core benefits of the 800 VDC shift: an end-to-end efficiency improvement of up to 5% across the power chain, and a drastic reduction in copper . A single 1 MW rack at 54 V requires up to 200 kg of internal copper busbars. At a gigawatt-scale data center, that need balloons to 200,000 kg. By drastically reducing current, the 800 V system collapses that copper requirement, while also allowing for a simpler three-wire (POS, RTN, PE) infrastructure instead of four conductors .

The architecture also saves space and improves reliability by eliminating a stage of AC-to-DC conversion. Power from the grid (13.8 kV AC) is converted to 800 VDC once at the facility's edge, removing the need for rectifiers at each rack . Nvidia estimates these changes can drive down total cost of ownership by up to 30% and maintenance costs by up to 70% . The system is also designed to handle the acute volatility of synchronous AI workloads, which can yank whole-facility power consumption from 30% to 100% in milliseconds .

The All-GaN Bet from Innoscience

While Infineon is pursuing a multi-material strategy, Innoscience has committed to a single-chain solution. The company announced an all-gallium nitride (GaN) power solution designed to convert power directly from the 800 VDC input rail down to GPU core voltages .

By using GaN for every conversion stage, Innoscience claims it can deliver higher power density and efficiency while lowering carbon emissions, enabled by GaN's ability to switch at high frequencies without the efficiency losses seen in silicon at elevated voltages. On the 800 V input side, Innoscience's GaN technology reportedly reduces driver losses compared to SiC-based designs .

The Broader Ecosystem Mobilizes

The 800 VDC push is too large for any single company. Nvidia has disclosed a broad list of ecosystem partners across three tiers:

• Silicon providers: Analog Devices, Infineon, Innoscience, Monolithic Power Systems (MPS), Navitas, OnSemi, Renesas, ROHM, STMicroelectronics, and Texas Instruments .
• Power system component makers: Delta, Flex Power, Lead Wealth, LiteOn, and Megmeet .
• Data center power system integrators: Eaton, Schneider Electric, and Vertiv .

Recent actions from these partners illustrate the ecosystem's momentum:

• Analog Devices (ADI) announced on May 28, 2026, that it is working alongside Nvidia to deliver the protection, conversion, and intelligence technologies required for AI factory deployments inside the MGX ecosystem .
• Eaton delivered a new 800 VDC reference architecture in support of Nvidia's design on May 28, 2026, a milestone in its "grid-to-chip" strategy for high-density AI facilities . This builds on an earlier collaboration with Nvidia on the Kyber rack-scale system for Rubin Ultra GPUs .
• LiteOn Technology showcased its next-generation 800 VDC power rack architecture, a 110 kW Power Shelf, and a 2.1 MW in-row coolant distribution unit (CDU)—all based on Nvidia's MGX modular design—at GTC 2026 .

This convergence of partners makes a clear statement about where the industry is headed. As AI factories scale from tens of kilowatts to megawatts per rack, the power delivery architecture must transform from a decentralized collection of low-voltage power supplies to a centralized, high-voltage backbone. The MGX ecosystem, with 800 VDC at its center, is now the industry's blueprint for that future.