Navitas And Nvidia Advance 800-Volt Power Design For AI Data Centers

Navitas Semiconductor began collaborating with Nvidia on power semiconductors for an 800-volt direct-current architecture that Nvidia plans to use in large AI data centers, while a separate Navitas-Microchip design targets direct rack-level power conversion.

The Nvidia architecture is designed for Kyber rack-scale systems and GPUs including Rubin Ultra. Nvidia plans to begin full-scale production alongside Kyber systems in 2027, with support for 1-megawatt IT racks and larger installations.

The architecture is designed to convert 13.8-kilovolt alternating-current grid power to 800 volts of direct current near the data center perimeter. Moving the conversion closer to the facility boundary is intended to reduce the number of power-conversion stages before electricity reaches the racks.

Nvidia projects that the approach could reduce copper requirements by 45%, improve end-to-end power efficiency by up to 5% and lower maintenance costs by up to 70%. Those figures are targets for the planned architecture rather than operating results from a live AI data center.

“800 VDC unlocks the compute performance and power density required for AI at scale,” Vladimir Troy, Nvidia’s vice president of data center infrastructure, said.

Navitas’ standalone 20-kilowatt board converts 800 volts to 6 volts and is designed to remove the conventional 48-volt intermediate stage inside data-center racks. The board uses 16 650-volt gallium nitride field-effect transistors (GaN FETs), switches at 1 MHz and targets 96.5% efficiency at full load.

The board is specified at 6.5 millimeters in height, approximately 20% thinner than a mobile phone. Navitas supplies gallium nitride and silicon carbide power semiconductors.

Separately, Navitas and Microchip developed an 800V-to-6V reference design for rack power applications. The platform uses two parallel 10-kilowatt modules and targets 96% efficiency at approximately 1 MHz. It combines Microchip’s dsPIC33AK digital signal controllers and TA100 security integrated circuit with Navitas’ 650-volt GaN devices.

The design was scheduled for exhibition at the OCP Global Summit in San Jose, California, Oct. 12-15, 2026. Joe Thomsen, Microchip’s corporate vice president, described the shift as a major change in data-center power architecture.

“AI infrastructure optimization is driving one of the most significant power architecture transitions the data center industry has experienced in decades,” Thomsen said.

Gene Sheridan, Navitas’ CEO and co-founder, said the company was selected to collaborate with Nvidia on the 800-volt initiative. The work follows growing attention to power constraints across AI infrastructure, including those outlined in an earlier report on AI power bottlenecks.