- Schneider Electric and AMD released their first co-engineered reference design for deploying Helios rack-scale AI clusters up to 10.4 MW
- The blueprint supports 246 kW racks with Motivair liquid cooling that removes up to 84% of heat and targets PUE as low as 1.12
- The move parallels rival Vertiv's Nvidia reference designs as infrastructure vendors race to support AMD's growing GPU share
AMD ADVANCING AI 2026, SAN FRANCISCO — Schneider Electric and AMD have released their first jointly engineered reference design for deploying Helios, AMD's rack-scale AI system, giving operators a prevalidated blueprint for high-density AI clusters. The companies unveiled the design Thursday, timed to AMD's Advancing AI event, where the Instinct MI455X-based Helios rack goes head to head with Nvidia's next-generation Vera Rubin platform.
The design supports AI clusters up to 10.4 MW of IT capacity and racks drawing as much as 246 kW — roughly 20 times the density of a conventional enterprise rack. It covers power distribution, liquid cooling, air cooling and electrical infrastructure, backed by engineering documents, floor layouts and thermal and electrical simulations.
Blueprints for a moving target
Reference designs matter because AI chips are advancing faster than data centers can be redesigned to power and cool them. Every new generation of GPU racks draws more power and throws off more heat, and operators struggle to keep their facilities' power and cooling designs current. A reference design hands them a validated starting point. The blueprint "takes a lot of the guesswork out" and could "take months off of the design cycle of a typical project," said Rob Bunger, global director of data center solution architecture at Schneider Electric.
The customers are neoclouds, engineering firms and enterprises building their own data centers, many of them new to the field. "If you're not a 20-year veteran in the data center industry and you're trying to deploy this stuff," reference designs improve the odds a deployment works as intended, Bunger said.
Chasing AMD's momentum
The move parallels work Vertiv has done with Nvidia, whose GB200 NVL72 reference designs support 132 kW racks. Nvidia dominates rack-scale AI, but AMD is gaining, landing 6-GW GPU commitments from both OpenAI and Meta. "It's not just going to be Nvidia," Bunger said, predicting customers will turn to AMD "in cases where Nvidia is not a fit." Schneider, which already partners with Nvidia, is positioning itself as a vendor-neutral supplier of the power and cooling underneath either platform.
Schneider's February 2025 Motivair acquisition underpins the cooling. The design specifies a Motivair coolant distribution unit and, for the first time in a Schneider reference design, rear-door heat exchangers as a retrofit option. Schneider claims the liquid cooling removes up to 84% of heat and reaches PUE — a measure of how much of a facility's power actually reaches its IT gear — as low as 1.12 at full load, against a data center industry average closer to 1.5.
Helios is expected to ship in volume next year, making the blueprint a bet that being ready early beats waiting for the racks to arrive.