Infineon and NVIDIA are expanding their collaboration on 800-volt direct current (VDC) electrical architectures for AI data centers, now focusing on gallium nitride (GaN) converters for one of the most demanding stages of the power chain. The German company announced it on October 21, 2026 in Munich, coinciding with NVIDIA’s GTC Berlin conference.
Infineon, NVIDIA, and GaN for 800 VDC in 30 seconds
- Infineon and NVIDIA are focusing on the intermediate bus converter (IBC), which steps down voltage before it reaches the accelerators.
- GaN allows switching at up to 1 MHz, with smaller components and higher power density.
- Infineon covers conversions from 800 V to 50 V, 12 V, and 6 V.
- It’s the third phase of a collaboration announced in May 2025, with no commercial dates or efficiency figures disclosed.
AI racks are consuming increasing amounts of power, and the industry is looking to raise distribution voltage to cut losses. In that transition, the intermediate bus converter (IBC) is the component that steps down the main bus voltage to a lower level, from which other stages generate the precise voltages processors need. Infineon describes it as one of the most critical stages in the entire power chain.
Why gallium nitride
According to Infineon, compared with silicon, GaN lets power transistors switch much faster and with lower losses, enabling frequencies of up to 1 MHz at the IBC stage. With higher frequencies, inductors and passive components can be much smaller, translating into more power in less space and lower-profile boards. In AI data centers, rack space is increasingly scarce.
The company adds that GaN’s low conduction and switching losses help limit the heat generated as power per rack approaches a megawatt. Its portfolio covers the conversion ratios this stage requires, from 800 V to 50 V, to 12 V, and to 6 V, rounded out with its PSoC Control C3 control family and EiceDRIVER gate driver family to offer the complete system. Infineon also makes IBC solutions using silicon and silicon carbide. The announcement does not disclose specific efficiency figures or availability dates.
Jochen Hanebeck, Infineon’s CEO, said AI is fundamentally changing what is demanded of electrical infrastructure, and that semiconductors are at the center of the solution. He added that GaN-based solutions are key to the next generation of data centers.
A three-phase collaboration
The relationship between the two companies began in May 2025, when they announced they were working together on 800 VDC architectures. In October 2025 they expanded it with power-path protection solutions, such as hot-swap controllers, so that 800 VDC racks can be serviced and have components replaced without shutting down. The current announcement adds high-voltage conversion using GaN.
Infineon stresses that the 800 VDC voltage is being introduced alongside the existing alternating-current and lower-voltage direct-current architectures, as one more option for high-density computing, not as an immediate replacement. CloudNews explained in September how AI is driving the shift to 800 VDC in data centers, and NVIDIA’s own plans to move to 800V racks as AI increasingly hits the power grid, not just GPUs. The German company positions itself as a supplier for the entire chain, from grid-connected power supplies down to the point of load next to the processor.
Frequently Asked Questions
What have Infineon and NVIDIA announced?
An expansion of their collaboration on 800 VDC architectures for AI data centers, focused on gallium nitride intermediate bus converters.
What is an intermediate bus converter?
It’s the stage that steps down the main bus voltage to a lower level, from which other stages generate the voltages AI processors need.
What advantages does GaN have over silicon?
According to Infineon, it switches faster and with lower losses, enables frequencies of up to 1 MHz, and reduces component size, which increases power density.
What conversions does Infineon cover?
From 800 V to 50 V, to 12 V, and to 6 V at the high-voltage stage of the intermediate bus converter.
Sources:
