Nvidia and Infineon Technologies have joined forces to modernize the electrical backbone of AI data centers, aiming to replace today’s complex network of low-voltage power supplies and cables with a single, high-voltage direct current (DC) architecture.
As the computational demands of AI systems accelerate, so too does their power consumption. Current-generation GPUs now draw over one kilowatt each, pushing traditional rack power requirements from 120 kW to as high as 500 kW—with projections surpassing one megawatt before the end of the decade. This growth has placed immense stress on existing 54-volt infrastructures, increasing both power losses and failure rates.
Infineon’s Power & Sensor Systems division, working alongside Nvidia, proposes a centralized 800-volt DC system to deliver electricity more efficiently to the GPUs directly on the server boards. By converting power at the point of use, the design minimizes transmission losses, reduces the need for bulky copper wiring, and significantly improves thermal management and reliability.
This approach also consolidates multiple power supplies into a single high-voltage backbone, cutting down the number of components that can fail and freeing up valuable rack space. However, Infineon notes that adopting such an architecture will require new safety protocols and conversion technologies to ensure stable operation and protect against electrical hazards in high-density environments.
The 800V DC design was first presented by Nvidia at Computex 2025 as part of its initiative to support next-generation AI clusters. The new standard aligns with Nvidia’s Kyber platform, which connects up to 576 Rubin Ultra GPUs and targets large-scale inferencing workloads.
Industry analysts see this as a logical evolution for AI data centers. Higher voltage reduces energy losses across long power paths and simplifies maintenance while enabling more sustainable, serviceable power delivery at scale.
The transition is already gaining momentum. More than 50 Nvidia MGX partners are reportedly preparing infrastructure and hardware optimized for 800V systems, while over 20 companies showcased compatible silicon and power components at the OCP Global Summit in Germany.
Together, Nvidia and Infineon are positioning this overhaul as a foundation for the coming “gigawatt era” of AI computing—where smarter, cleaner, and more reliable power delivery will be as critical as GPU
