Schneider Electric, AMD unveil blueprint for AI data centers

Schneider Electric SE and Advanced Micro Devices Inc. rolled out a jointly built reference design Thursday aimed at helping data centre operators deploy AI infrastructure faster, as demand for computing power tied to artificial intelligence continues to strain electrical grids and cooling systems worldwide.

The design, the first product of a broader partnership between the French energy management company and the Santa Clara, California-based chipmaker, provides a tested template for building out AMD’s Helios rack-scale system. It marks an attempt by both companies to shorten the runway between planning an AI data centre and actually switching it on, a process that has grown more complicated as power and thermal demands per rack have climbed sharply in the past two years.

Helios is powered by AMD’s Instinct MI455X graphics processors, its sixth-generation Epyc server chips, Pensando Vulcano networking cards and the open-source ROCm software stack. AMD has positioned the system as a rival to Nvidia Corp.’s dominant AI hardware platforms, betting that data center operators will pay a premium for systems that squeeze more performance out of every watt of power and square foot of floor space.

The reference design covers four areas that have become chokepoints in AI buildouts: facility power, cooling, physical IT layout and the software used to manage systems over their lifecycle. Schneider Electric said the design supports AI clusters of up to 10.4 megawatts for new-build data centres and can handle individual racks drawing up to 246 kilowatts — roughly 10 to 20 times the load of a conventional server rack from a decade ago.

To manage the heat generated by that density, the design relies on liquid cooling equipment from Motivair, a company Schneider Electric acquired, paired with hybrid air-and-liquid approaches capable of removing up to 84% of heat generated inside a rack. Schneider Electric said facilities built to the specification could achieve a power usage effectiveness ratio — a standard measure of data center energy efficiency — as low as 1.12 at full load, a figure that would place such a facility among the more efficient large-scale data centers currently in operation.

“Today organizations require comprehensive, AI-ready reference designs that can take them from planning to deployment faster and with less risk,” said Manish Kumar, executive vice president of Secure Power & Data Centers at Schneider Electric.

Forrest Norrod, who leads AMD’s data center solutions business, said the push reflects a shift in how AI infrastructure gets built. “AI infrastructure is rapidly moving to full-scale AI factories, and that requires compute, networking, power and cooling to be designed together from the start,” Norrod said.

The design also folds in digital modeling tools, including Schneider Electric’s EcoStruxure IT software and ETAP simulation platform, along with an “electrical digital twin” meant to let operators model infrastructure performance before construction begins. AVEVA, the industrial software company majority-owned by Schneider Electric, will provide monitoring through its Unified Operations Center platform.

The companies said the design has been validated against ANSI standards for U.S. deployments and plans to extend it to meet IEC standards for international markets.

The partnership underscores how thoroughly the AI buildout has reshaped the data center supply chain, pulling power-equipment makers like Schneider Electric and its rivals Vertiv Holdings Co. and Eaton Corp. into closer coordination with chipmakers earlier in the design process than has traditionally been the case. Data center capacity tied to AI has become one of the most closely watched drivers of electricity demand growth in the U.S. and Europe, prompting utilities and infrastructure providers alike to race to keep pace.