A radical change in the data center physical infrastructure is currently taking place globally. Traditional enterprise data centers had operated on average power densities between 5 and 15 kW per rack for many years. However, the development of novel AI models and their requirement for enormous GPU clusters and specific network fabric has shifted physical infrastructure requirements far beyond any known experience.
Modern AI factories need up to 100 kW to 250 kW per rack. Such a jump in the power density creates a significant operational bottleneck.
As soon as the architects try to implement next generation AI accelerators into the facilities of current configuration, they face major barriers related to thermal management and electric grid infrastructure. Existing cooling solutions cannot cope with the extreme heat load, PDUs get overheated, and facility owners are forced to go through complicated and time-consuming processes to prevent the system breakdowns.
To address the physical infrastructure challenges, Schneider Electric and semiconductor corporation Advanced Micro Devices, Inc., two leaders in the energy technology market, have introduced a new reference design for AMD Helios rackscale solution.
Engineering the Infrastructure for 246 kW Racks
The strategic collaboration marks the first major milestone in a broad technology alliance between Schneider Electric and AMD. Rather than treating silicon computing and physical facility power as separate engineering disciplines, the reference design integrates AMD’s hardware platform directly with Schneider Electric’s electrical, liquid cooling, and digital management suite.
Key technical and operational pillars of the Helios reference design include:
Extreme Rack Density: The design allows each individual AI rack to consume power at up to 246 kW densities while giving a scalable framework from single cluster up to 10.4 megawatts (MW) of IT.
Powered by AMD Helios Silicon: The underlying IT infrastructure utilizes AMD Instinct™ MI455X GPUs, 6th Gen AMD EPYC™ CPUs, AMD Pensando™ Vulcano network interface cards (NICs) along with the open source ROCm™ software stack.
Advanced Liquid Thermal Management: To manage extreme thermal load conditions, the design includes Motivair by Schneider Electric Coolant Distribution Units (CDUs) and hybrid air/water cooled designs capable of taking away heat of up to 84% at the rack level.
Digital Twin and Simulation Modeling: Prior to physical construction, the design goes through validation using ETAP electrical design simulation and EcoStruxure™ IT Design Computational Fluid Dynamics (CFD) simulation software in conjunction with real time operations management from the AVEVA™ Unified Operations Center.
Also Read: AMD and Anthropic Announce Strategic Partnership to Deploy 2 Gigawatts of Instinct MI450 Series GPUs
Benchmark Energy Efficiency: The co-designed solution allows achieving an outstanding Power Usage Effectiveness (PUE) of up to ~1.12.
Impact on the Power and Energy Infrastructure Industry
The alliance between Schneider Electric and AMD reflects fundamental structural shifts across the broader Power and utility management landscape:
1. Industrializing Liquid Cooling as an Essential Facility Standard
Historically, liquid cooling was viewed by facility engineers as an expensive, niche solution reserved for specialized supercomputing labs.
Designing a validated 246 kW rack reference design featuring Motivair CDUs formalizes liquid cooling as a primary, mainstream utility requirement. As chip power draw escalates, power and facility engineers must design electrical rooms and chilled water loops specifically around direct-to-chip and rear-door liquid cooling architectures.
2. Transitioning to Digital Twin Facility Simulation
Attempting to trial-and-error electrical and thermal configurations at 10+ megawatt scale introduces multi-million-dollar failure risks.
Integrating ETAP and EcoStruxure CFD modeling into the Helios platform establishes a new industry standard: Predictive Digital Twin Engineering. Facility operators can test electrical load distributions, thermal dissipation pathways, and predictive maintenance scenarios virtually before committing capital to physical hardware installation.
Impact on the Semiconductor and Accelerated Compute Sector
The joint blueprint delivers profound implications for the Semiconductors market, reshaping how chip designers evaluate system-level performance:
Shifting Silicon Design from Chip-Level to Rack-Scale Architecture: Semiconductor firms can no longer design GPUs and CPUs in isolation from facility power limits. AMD‘s co-development with Schneider Electric proves that future market leadership requires co-designing silicon, high-speed interconnects, and physical power delivery together.
Removing Thermal Throttling to Guarantee Peak Compute Yields: High-end AI accelerators underperform if thermal management systems cannot dissipate localized heat spikes. Providing pre-tested liquid cooling blueprints ensures that AMD Instinct accelerators operate continuously at maximum clock frequencies without thermal throttling.
Lowering Integration Friction for Open Silicon Ecosystems: Unvalidated hardware deployments often delay customer adoption. Offering a turnkey, ANSI-validated reference design gives enterprise buyers confidence that AMD Helios clusters can be installed into existing data centers swiftly, competing directly against proprietary bundled hardware stacks.
Conclusion
Moving from AI infrastructure to AI factories will be required, and for that purpose, the compute, networking, power and cooling needs to be integrated into one another from the start, according to Forrest Norrod, Executive Vice President and General Manager of the Data Center Solutions Business Group at AMD. “Through our collaboration with AMD, we’re offering an engineering validated reference design which fills the gap between advanced AI computing platforms, energy technology, and the practical deployment in the data centers,” said Manish Kumar, Executive Vice President, Secure Power & Data Centers at Schneider Electric.
The Helios platform reference design from Schneider Electric and AMD creates an operational framework for the age of AI. This partnership shows that the development of next-gen artificial intelligence requires solutions to its physical infrastructure problems first.