Infineon Technologies and Nvidia are teaming up to build power conversion products for data centers. The goal is to keep AI servers fed as their electricity appetite heads somewhere alarming.
The partnership centers on an 800 V high-voltage direct current (HVDC) architecture. It is designed to replace the patchwork of power systems most data centers run on today.
What the partnership actually does
Infineon first announced the collaboration on May 20, 2025. The aim was next-generation power delivery for AI data centers.
Technically, the plan moves away from decentralized alternating current (AC) distribution and lower-voltage DC setups. It shifts toward a centralized 800 VDC framework. Power conversion would happen closer to the GPUs, the processors doing the heavy lifting in AI workloads.
That is where Infineon comes in. The company brings expertise across three semiconductor materials: silicon, silicon carbide (SiC), and gallium nitride (GaN). SiC and GaN are newer materials that can handle higher voltages and switch faster than traditional silicon. They also lose less energy along the way.
Efficiency projections reach up to approximately 98% per conversion stage using Infineon’s technologies.
Both companies framed the work in terms of reliability. Nvidia’s Gabriele Gorla highlighted the 800V HVDC system’s ability to deliver high reliability and energy efficiency across data centers. Infineon’s Adam White emphasized intelligent systems that can meet rising AI power demands while minimizing downtime.
The power problem behind the deal
The numbers explain the urgency. Rack power demands are currently around 120 kW. They are expected to rise to over 500 kW, and could reach up to 1 MW by around 2030.
Nvidia unveiled the 800 VDC architecture at Computex, and Infineon publicly backed it.
The two companies have also collaborated on advanced components. These include hot-swap controllers using SiC JFET technology. A hot-swap controller lets operators replace hardware while the system stays live.
Infineon has also developed 27 kW power supplies aligned with the 800 VDC architecture.
A growing ecosystem around 800 VDC
Infineon officially joined Nvidia’s MGX AI Factory ecosystem on May 29, 2026. MGX is Nvidia’s modular reference design program. Joining it means Infineon’s 800 VDC work gets embedded into designs for what the industry calls AI factories.
In September 2026, a memorandum of understanding with Skeleton Technologies was signed. It covers exploring solid-state transformers and GaN-based supercapacitor sidecars to enhance 800 VDC systems. A memorandum of understanding is an agreement to explore, not a finished product, so this remains early-stage work.
The broader pitch is that these systems could significantly reduce power distribution losses and infrastructure complexity. The target is reliable, energy-efficient power for the coming era of AI “gigafactories.”
What this means for the AI infrastructure race
Nvidia’s interest is straightforward. Its customers want to pack more chips into each rack. If power delivery becomes the bottleneck, GPU sales eventually feel it. Backing a new architecture, and recruiting suppliers into its MGX program, helps Nvidia shape how future data centers get built around its hardware.
For Infineon, the partnership offers a seat at the design table early in a potential industry-wide transition. If 800 VDC becomes a standard, suppliers with SiC and GaN expertise may be well placed to benefit.
There are real hurdles. Moving an entire facility to a new voltage architecture is not a weekend project. Many existing data centers were designed around AC distribution, and retrofitting them is a different challenge from building new ones.
Things to watch include how many hardware makers adopt MGX designs built around 800 VDC, the outcome of the Skeleton Technologies exploration, and whether rival chip and power companies line up behind the same architecture or push alternatives.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.