Summary

• AMD and Schneider Electric have developed reference designs for deploying AMD’s Helios AI rack architecture.
• The designs cover power, cooling, IT space, and lifecycle software for high density AI clusters.
• AI infrastructure competition is shifting from chip roadmaps into deployable building systems.

AMD and Schneider Electric have developed reference designs for deploying AMD’s Helios AI rack architecture, connecting accelerated compute with the power, cooling, space, and software systems needed to make dense AI infrastructure buildable.

The designs cover facility power, facility cooling, IT space, and lifecycle software, with support for AI clusters of up to 10.4MW and individual rack densities of up to 246kW. They are intended to help developers and operators integrate AMD’s rack scale AI systems into datacentres with less design uncertainty and lower integration risk.

The announcement reflects a wider change in AI infrastructure. Much of the market has been framed around GPUs, model demand, and hyperscaler capital spending, but deployment is increasingly constrained by the building. A rack that looks compelling in a product roadmap still has to fit into an electrical design, reject heat, operate safely, and remain maintainable under production workloads.

Helios is AMD’s open rack scale architecture for AI training and inference, combining CPUs, GPUs, networking, and system design into a larger infrastructure proposition. Schneider Electric’s involvement pulls that proposition into the physical environment, where liquid cooling, power distribution, monitoring, and lifecycle management shape whether high density compute can move from order book to live capacity.

Reference designs are valuable because datacentre projects are increasingly exposed to coordination risk. Operators have to align chip supply, rack layouts, cooling systems, power equipment, contractors, grid constraints, and customer demand. A validated design gives them a more repeatable path, especially when AI customers expect fast deployment but sites still depend on engineering, permitting, procurement, and utilities.

For AMD, the collaboration supports a wider attempt to compete in AI infrastructure where Nvidia has dominated both hardware mindshare and ecosystem depth. Silicon performance is only one part of that contest. Buyers also need confidence that hardware can be deployed, cooled, managed, and upgraded without requiring bespoke engineering for every site.

Schneider Electric’s role gives the work a strong European relevance. The company has been expanding its AI factory and datacentre design activity at a time when European projects are running into power availability, planning constraints, grid queues, water scrutiny, and local consent issues. A clearer facility blueprint will not solve those constraints on its own, but it can reduce uncertainty inside projects already struggling with physical complexity.

The move also points towards a more integrated supplier market. AI infrastructure buyers are not simply purchasing chips, racks, cooling units, or software in isolation. They are buying a system that has to run at high utilisation, meet uptime expectations, control energy costs, support service contracts, and remain adaptable as accelerator generations change.

The next phase of AI infrastructure competition will therefore be shaped by buildability as much as by raw compute. Helios gives AMD a rack architecture, while Schneider Electric helps define the building envelope around it. If the reference designs reduce risk for operators, they could make AMD based clusters easier to deploy in a market where engineering friction has become one of the industry’s most expensive bottlenecks.