IN Brief:
- Infineon has joined NVIDIA’s MGX AI Factory ecosystem for next-generation AI server racks.
- The work supports 800VDC power architectures using silicon, silicon carbide, and gallium nitride technologies.
- AI data-centre power design is moving toward higher-voltage distribution, fewer conversion stages, and denser rack-level delivery.
Infineon Technologies has joined NVIDIA’s MGX AI Factory ecosystem to support power delivery architectures for next-generation AI data centres and 800VDC server-rack systems.
Infineon’s power management solutions will support NVIDIA’s MGX architecture and 800VDC power architecture, an open modular reference architecture developed for AI factory infrastructure. The 800VDC MGX-compatible rack approach is designed to help existing AI infrastructure scale compute performance and rack power density while creating an upgrade path for future deployments.
The company is supporting power conversion from grid to core using silicon, silicon carbide, and gallium nitride. Infineon says its GaN technology operating at switching frequencies close to 1MHz enables compact bus converters, while its SiC JFET technology and dedicated control ICs support protection and hot-swap functions on native 800V server boards.
The power-management chain is designed to convert 800V to intermediate bus voltages including 50V and 12V, and in some cases down to 6V. By supporting more of the conversion flow within MGX-based systems, Infineon is working to reduce conversion stages, deliver DC power closer to the rack, improve efficiency, simplify infrastructure, and enable higher-density AI deployments.
Adam White, Division President Power & Sensor Systems at Infineon, said: “As a member of NVIDIA’s ecosystem, Infineon is working with NVIDIA to redefine power delivery systems from the grid to the processor core, which is required for this next phase of AI innovation.”
The move follows Infineon’s wider power-infrastructure focus for PCIM Europe, where AI data centres, robotics, electromobility, solid-state switching, GaN, SiC, and software-defined power systems formed part of a broader technology programme.
AI data centres are forcing power architecture to change at rack level. Accelerated compute platforms require much higher power density than traditional server deployments, while physical space, cooling capacity, copper volume, and conversion losses remain hard engineering constraints. Delivering very high current at low voltage across a rack becomes increasingly difficult as power levels rise, which is why higher-voltage distribution is moving into the centre of AI infrastructure design.
Moving to 800VDC reduces distribution losses and copper burden, but it also changes the protection and conversion problem. Isolation, hot-swap behaviour, fault response, monitoring, service procedures, and safety handling become more demanding. Semiconductor devices are only one part of that architecture; control ICs, gate drivers, sensing, firmware, packaging, thermal paths, and validation methods all have to operate as a complete system.
Wide-bandgap devices are central to the transition because they support higher switching frequencies, lower losses, and improved power density when the surrounding circuit is engineered correctly. GaN is suited to compact high-frequency conversion stages, while SiC is increasingly used where higher voltage, efficiency, and temperature performance are required. Silicon remains important across control, sensing, and cost-sensitive stages of the same system.
AI rack power is becoming an ecosystem-level design problem rather than a set of isolated power-supply choices. Processor performance, rack density, energy efficiency, cooling, and power delivery are now tightly linked, making the path from grid input to processor core a strategic part of AI infrastructure design.
