{"slug": "amd-and-schneider-electric-release-246kw-helios-ai-rack-reference-design-for", "title": "AMD and Schneider Electric Release 246kW Helios AI Rack Reference Design for Data Centers", "summary": "AMD and Schneider Electric have jointly released the first reference design for AMD's Helios rackscale AI platform, providing a validated blueprint for data center developers covering facility power, cooling, IT space, and lifecycle software for high-density AI deployments. The design supports individual rack power densities up to 246kW and modular AI clusters scaling to 10.4MW IT load, with liquid cooling removing up to 84% of heat and targeting a PUE of approximately 1.12 at full load. Volume deployments are expected in the second half of 2026.", "body_md": "[AI](/news/ai/)\n\n[AI](/news/tag/ai/)\n\n[AI INFRASTRUCTURE](/news/tag/ai-infrastructure/)\n\n[AMD](/news/tag/amd/)\n\n[DATA CENTER](/news/tag/data-center/)\n\n[DATA CENTER DESIGN](/news/tag/data-center-design/)\n\n[LIQUID COOLING](/news/tag/liquid-cooling/)\n\n[SCHNEIDER ELECTRIC](/news/tag/schneider-electric/)\n\n[SUPPLY CHAIN](/news/tag/supply-chain/)\n\n# AMD and Schneider Electric Release 246kW Helios AI Rack Reference Design for Data Centers\n\n- Reference design supports individual rack densities of up to 246kW and modular AI clusters scaling to 10.4MW IT load\n[[1]](https://www.datacenterdynamics.com/en/news/amd-and-schneider-electric-launch-reference-standards-for-helios-ai-rack/) - Liquid cooling via Motivair by Schneider Electric CDUs removes up to 84% of heat, targeting a PUE of approximately 1.12 at full load\n[[2]](https://ca.finance.yahoo.com/news/schneider-electric-amd-release-first-190000719.html) - The Helios rack packs 72 AMD Instinct MI455X GPUs delivering 2.9 exaFLOPS of FP4 compute and 31TB of HBM4 memory\n[[3]](https://www.amd.com/en/products/rackscale-solutions/helios.html) - Design is validated to ANSI standards for U.S. deployments, with IEC standards planned for global expansion\n[[2]](https://ca.finance.yahoo.com/news/schneider-electric-amd-release-first-190000719.html) - Volume deployments of the Helios platform are expected in the second half of 2026\n[[1]](https://www.datacenterdynamics.com/en/news/amd-and-schneider-electric-launch-reference-standards-for-helios-ai-rack/)\n\nAMD and Schneider Electric have jointly released the first reference design for AMD's Helios rackscale AI platform, providing data center developers with a validated blueprint covering facility power, cooling, IT space, and lifecycle software for high-density AI deployments. The announcement was made at AMD's Advancing AI 2026 event in San Francisco on July 24 [1].\n\nThe reference design supports individual rack power densities of up to 246kW and modular AI cluster configurations scaling to 10.4MW of IT load, addressing the growing challenge of integrating next-generation AI accelerator platforms into physical data center infrastructure [2]. The design incorporates advanced liquid cooling from Motivair by Schneider Electric and targets a power usage effectiveness (PUE) of approximately 1.12 at full load\n\n.\n\n[[2]](https://ca.finance.yahoo.com/news/schneider-electric-amd-release-first-190000719.html)The Helios platform itself is built around 72 AMD Instinct MI455X GPUs based on the CDNA 5 architecture, paired with 6th Gen AMD EPYC CPUs and AMD Pensando Vulcano NICs in a double-wide OCP Open Rack Wide (ORW) form factor. At rack scale, the system delivers up to 2.9 exaFLOPS of FP4 compute and 31TB of HBM4 memory [3].\n\n## What the Reference Design Covers\n\nThe reference design spans four technical domains: facility power, facility cooling, IT space configuration, and lifecycle software integration. Electrical and thermal designs have been validated using ETAP power analysis and Schneider Electric's EcoStruxure IT Design CFD simulation tools, with integrated Electrical Digital Twin capabilities for ongoing monitoring [2].\n\nSchneider Electric's AVEVA Unified Operations Center provides real-time monitoring and AI-driven predictive maintenance support, linking the physical infrastructure to software-layer management [2]. The design is currently validated to ANSI standards for U.S. deployments, with plans to extend the framework to IEC standards for international implementations\n\n.\n\n[[2]](https://ca.finance.yahoo.com/news/schneider-electric-amd-release-first-190000719.html)## Power and Cooling Architecture\n\nThe cooling architecture relies on Motivair by Schneider Electric coolant distribution unit (CDU)-based liquid cooling with hybrid air/liquid approaches, capable of removing up to 84% of rack-generated heat via direct liquid cooling [2]. This hybrid approach allows operators to handle the extreme thermal loads of 246kW racks without requiring fully sealed rear-door or immersion systems.\n\nAt approximately 140kW per Helios rack in its base GPU configuration, the platform draws meaningfully less power than competing rack-scale systems — NVIDIA's Vera Rubin NVL72, for instance, is estimated at 190 to 230kW per rack [3]. The 246kW ceiling in the reference design accounts for additional networking, storage, and management infrastructure surrounding the core GPU compute.\n\n## The Helios Platform\n\nAMD's Helios is the company's answer to NVIDIA's rack-scale DGX platforms, built on open standards including OCP Open Rack Wide, Ultra Accelerator Link (UALink), and Ultra Ethernet Consortium (UEC) specifications [3]. Each rack integrates 72 Instinct MI455X accelerators with 31TB of HBM4 memory providing 23.3 TB/s of aggregate memory bandwidth\n\n.\n\n[[3]](https://www.amd.com/en/products/rackscale-solutions/helios.html)Microsoft has already been confirmed as a Helios customer, and volume deployments are expected to begin in the second half of 2026 [3]. The open-standards approach is designed to give operators flexibility in component sourcing — a competitive differentiator against NVIDIA's more vertically integrated NVLink-based ecosystem.\n\n## Why It Matters for Operators\n\nReference designs of this type serve a specific function in the data center supply chain: they reduce the engineering burden on operators and colocation providers who need to integrate new high-density platforms into existing or greenfield facilities. By pre-validating the electrical, thermal, and spatial requirements, AMD and Schneider Electric are effectively compressing the timeline between silicon availability and production deployment.\n\nThe 10.4MW cluster specification is notable because it maps closely to common modular data center building blocks — many hyperscale and neocloud operators deploy in 10-15MW increments. A validated reference design at that scale allows developers to plan facility builds around confirmed power and cooling envelopes rather than working from preliminary GPU specifications [2].\n\n## Companies mentioned\n\n## Further sources\n\nThe stories that matter, in one email. Free — unsubscribe anytime.", "url": "https://wpnews.pro/news/amd-and-schneider-electric-release-246kw-helios-ai-rack-reference-design-for", "canonical_source": "https://mlq.ai/news/amd-and-schneider-electric-release-246kw-helios-ai-rack-reference-design-for-data-centers/", "published_at": "2026-07-24 15:22:49.077505+00:00", "updated_at": "2026-07-24 15:22:51.404050+00:00", "lang": "en", "topics": ["ai-infrastructure", "ai-chips", "ai-products", "ai-research", "artificial-intelligence"], "entities": ["AMD", "Schneider Electric", "Helios", "Motivair", "AMD Instinct MI455X", "AMD EPYC", "AMD Pensando Vulcano", "OCP Open Rack Wide"], "alternates": {"html": "https://wpnews.pro/news/amd-and-schneider-electric-release-246kw-helios-ai-rack-reference-design-for", "markdown": "https://wpnews.pro/news/amd-and-schneider-electric-release-246kw-helios-ai-rack-reference-design-for.md", "text": "https://wpnews.pro/news/amd-and-schneider-electric-release-246kw-helios-ai-rack-reference-design-for.txt", "jsonld": "https://wpnews.pro/news/amd-and-schneider-electric-release-246kw-helios-ai-rack-reference-design-for.jsonld"}}