Representative End User Clients
Representative Automation Clients
Representative Software Clients
The modular blueprint supports AI racks of up to 246 kW and multi-cluster deployments with up to 10.4 MW of IT capacity.
Schneider Electric and AMD announced that they have jointly developed and validated a reference design for the AMD Helios rack-scale platform, providing data center architects and operators with a blueprint for deploying high-density artificial intelligence (AI) infrastructure.
This new jointly developed reference design is the first milestone in the companies’ collaboration and is intended to help reduce the complexity and integration risks associated with deploying AI factories. It combines AMD’s compute, networking, and software technologies with Schneider Electric’s power, cooling, simulation, and digital infrastructure capabilities.
The reference design supports greenfield AI factories as well as retrofit projects where operators need to introduce higher-density computing into existing data center environments. It helps to address facility power, facility cooling, IT space, and lifecycle software requirements within a single infrastructure framework.
Supporting High-Density AMD Helios Deployments
The reference design supports the AMD Helios rack-scale platform, which incorporates AMD Instinct MI455X GPUs, 6th Gen AMD EPYC CPUs, AMD Pensando Vulcano network interface cards, and the open ROCm software ecosystem.
AMD Helios supports large and complex AI workloads through advances in compute performance, interconnect bandwidth, memory capacity, and system-level integration. However, deploying rack-scale AI systems also creates significant requirements for power distribution, heat removal, physical space, controls, and infrastructure management.
Pre-validated reference designs can help operators address these requirements earlier in the planning process by defining how IT equipment, electrical systems, cooling infrastructure, and operational software should be configured. Modeling the physical infrastructure before deployment can also help identify capacity constraints, thermal risks, and potential integration issues.
Modular Power and Cooling Architecture
The Schneider Electric and AMD reference design supports modular AI clusters with up to 10.4 MW of IT capacity for greenfield deployments and rack densities of up to 246 kW. This modular approach allows organizations to add capacity in stages rather than requiring each expansion to be a separate infrastructure project.
Key elements of the reference design include:
Modular AI Clusters: The architecture supports multiple AI clusters with up to 10.4 MW of IT capacity for large-scale greenfield deployments.
High-Density Racks: The architecture accommodates AI workloads requiring up to 246 kW of power per rack.
Liquid Cooling: Motivair by Schneider Electric coolant distribution units and hybrid air-and-liquid cooling configurations are capable of removing up to 84 percent of the heat generated by the IT equipment.
Power and Cooling Alignment: The supporting electrical and thermal infrastructure is configured around AMD Helios platform requirements to help reduce integration complexity.
Energy Efficiency: It is specified to achieve a power usage effectiveness (PUE) rating as low as approximately 1.12 when operating at full load.
The cooling architecture reflects the increasing use of direct liquid cooling in AI and high-performance computing environments. As rack densities rise, conventional air cooling alone may not provide sufficient heat-removal capacity, requiring operators to coordinate liquid cooling systems closely with facility power, controls, and IT equipment.
Digital Design, Simulation, and Operations
The reference design uses ETAP and EcoStruxure IT Design computational fluid dynamics simulation tools to model electrical and thermal performance. These capabilities help to enable architects and operators to evaluate power flows, equipment configurations, cooling capacity, and airflow before construction or installation begins.
Integrated electrical digital twin capabilities can be used to model, analyze, and manage infrastructure performance throughout the data center lifecycle. AVEVA Unified Operations Center provides an additional layer of operational visibility by consolidating real-time information from power, cooling, and IT systems.
The digital infrastructure can support real-time monitoring, analytics, predictive maintenance, and system-level optimization. Connecting design models with operational data can also help data center teams to compare expected performance with actual conditions and respond to changing workload, power, and cooling requirements.
By treating compute, networking, power, cooling, simulation, and operations as interdependent elements, the reference design provides a more integrated approach to AI infrastructure deployment. This is increasingly important as AI factories move toward larger clusters, higher rack densities, and greater demands on electrical and thermal systems.
The reference design has been validated against ANSI standards for deployments in the US. Schneider Electric and AMD also plan to extend the framework to support IEC standards for deployments in other regions.
Related ARC Insights
The Schneider Electric and AMD reference design reflects a broader shift toward integrated AI factory architectures in which compute platforms, power distribution, liquid cooling, digital twins, and operational software are designed together. Related ARC Advisory Group coverage includes:
Schneider Electric Advances AI Factory Infrastructure: Design, Power, and Operations Converge
Schneider Electric and Foxconn Collaborate on AI Data Center Infrastructure
Follow ARC Advisory Group for the latest trends in sustainability, automation, and next-generation technology.