NVIDIA & Navitas Announce Collaboration on 800V High-Voltage DC architecture

0
29

Navitas Semiconductor, known for its GaNFast GaN and GeneSiC SiC power semiconductors, has revealed a partnership with NVIDIA to develop next-generation 800V high-voltage DC (HVDC) architecture. This initiative aims to support NVIDIA’s ‘Kyber’ rack-scale systems, which power advanced GPUs like the Rubin Ultra.

The announcement comes shortly after Infineon also disclosed a collaboration with NVIDIA focused on HVDC solutions, leveraging silicon, SiC, and GaN technologies.

NVIDIA’s new 800V high-voltage DC (HVDC) architecture, designed for data center IT racks starting at 1 megawatt and beyond, is engineered to boost overall power efficiency by up to 5%, cut maintenance costs by 70% (largely due to fewer power supply failures), and significantly lower cooling expenses by delivering HVDC power directly to IT and compute racks.

In contrast, current data center designs rely on 54V in-rack power distribution, which is only capable of handling a few hundred kilowatts. This setup requires bulky copper busbars to carry low-voltage power from rack-mounted shelves to compute trays. As power demands exceed 200 kW, this architecture faces serious limitations in terms of power density, material requirements, and system efficiency.

To meet the massive energy needs of modern AI data centers—which require gigawatts of power—NVIDIA proposes converting 13.8 kV AC grid power directly to 800V HVDC at the data center’s edge using solid-state transformers (SSTs) and industrial rectifiers. This approach eliminates several traditional power conversion stages, improving both efficiency and reliability.

The higher voltage of the 800V HVDC system also allows for a significant reduction in copper usage—up to 45%—by minimizing I²R losses. Unlike traditional 54V systems, which would need over 200 kg of copper to power a 1MW rack, NVIDIA’s solution delivers the same power with thinner wiring, offering a more scalable and sustainable option for next-generation AI data centers.

LEAVE A REPLY

Please enter your comment!
Please enter your name here