AMD Helios AI Factory Design Targets 246kW Racks and 10.4MW Clusters

Schneider Electric and AMD released a power-and-cooling reference design for Helios rack-scale systems, supporting up to 246kW per rack and modular AI clusters with 10.4MW of IT load.

Co-designing the chips and the facility

Schneider Electric and AMD released a joint AI-data-center reference design for the AMD Helios rack-scale platform. Helios combines AMD Instinct MI455X GPUs, 6th Gen EPYC CPUs, Pensando Vulcano NICs, and the ROCm software ecosystem. The blueprint covers facility power, cooling, IT space, and lifecycle software rather than treating compute separately from the building.

The published design supports modular clusters with up to 10.4MW of IT capacity and high-density loads of up to 246kW per rack. Motivair CDU-based and hybrid air/liquid cooling is designed to remove up to 84% of heat. Schneider Electric also says the design can reach a PUE of approximately 1.12 at full load.

Why it matters

At these rack densities, replacing servers inside a conventional room may not satisfy electrical distribution, piping, thermal, and safety requirements. A prevalidated reference design defines the interfaces among silicon, networking, power, and cooling before construction, potentially shortening planning for new sites and high-density retrofits.

Conditions behind the numbers

The 84% heat-removal figure and PUE of about 1.12 are vendor design capabilities under specified full-load conditions, not guaranteed results for every facility. The current framework is validated to U.S. ANSI standards, while IEC support is planned. Actual outcomes will vary with climate, water temperatures, power architecture, utilization, and operations.

Official source