Infineon has joined NVIDIA’s MGX AI Factory ecosystem, targeting power delivery for next-generation AI data centers that are moving toward 800 VDC architectures. The collaboration centers on Infineon power management solutions designed to support NVIDIA’s MGX architecture and NVIDIA’s 800 VDC MGX-compatible power racks.
The companies are aligning around an open, modular reference architecture intended for AI factory deployments, with 800 VDC power racks positioned as an upgrade path that can raise AI compute performance and power density in existing infrastructure while pointing toward future 800 VDC builds.
Power silicon for an 800 VDC stack
Infineon says its “grid-to-core” power conversion work spans silicon (Si), silicon carbide (SiC), and gallium nitride (GaN). In its description of the stack, Infineon points to GaN switching frequencies close to 1 MHz to enable compact bus converters. It also calls out a combination of its proprietary SiC JFET technology and dedicated control ICs for protection and hot-swap functionality on native 800 V server boards.
On voltage rails, Infineon’s power management solutions are described as converting from 800 V down to 50 V, 12 V, or as low as 6 V.
800 VDC distribution is fundamentally about moving more power with lower current, which can reduce conductor mass and distribution losses at very high rack densities. But the engineering work is in the details: where conversion stages sit, how protection and hot-swap are implemented, and how close DC power can be pushed toward the load without creating serviceability and safety headaches in the rack.
Within the MGX AI Factory ecosystem, Infineon says it supports the full 800 VDC power conversion flow down to intermediate bus and core voltages in NVIDIA MGX-based systems. The stated goal is to reduce conversion stages and deliver DC power closer to the rack to improve efficiency, simplify infrastructure, and support higher-density AI deployments.
“As AI models continue to grow in size and complexity, data centers must deliver dramatically more compute performance within the same physical, power, and cooling constraints,” said Adam White, Division President Power & Sensor Systems at Infineon.
Source: Infineon
