AMD Debuts Helios Rack-Scale Architecture to Challenge Nvidia's AI Dominance
The new unified infrastructure integrates Zen 6 CPUs and MI400 GPUs into a high-density reference design with Schneider Electric.
AMD has introduced Helios, a unified rack-scale AI architecture that integrates compute, networking, power, and cooling into a single system. The move signals a strategic shift for the company as it attempts to compete with Nvidia not just on individual chips, but through comprehensive, turnkey infrastructure solutions.
The Helios system utilizes a double-wide rack configuration featuring AMD Zen 6 "Venice" EPYC processors, MI400 series GPUs—specifically the MI455X—and Pensando "Vulcano" network interface cards. At rack scale, the architecture delivers peak performance of 2.9 exaFLOPS in 4-bit (FP4) and 1.4 exaFLOPS in 8-bit (FP8) precision, supported by 31 TB of HBM4 memory.
To streamline deployment in hyperscale data centers, AMD has partnered with Schneider Electric to provide a joint reference design. This collaboration addresses the extreme physical requirements of modern AI workloads, supporting rack power densities of up to 246 kilowatts and AI clusters that can consume up to 10.4 megawatts. Thermal management is handled through advanced liquid cooling technologies provided by Schneider Electric's Motivair division, specifically utilizing the MCDU-70.
The Shift to AI Factories
This transition reflects a broader industry trend where standalone hardware is being replaced by "AI factories." In this model, compute and networking are designed in tandem with facility engineering to meet the massive power and cooling demands of frontier model training and large-scale inference.
"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," said Forrest Norrod, EVP and GM in the data center solutions business group at AMD.
Breaking the Infrastructure Monopoly
By offering a validated reference design, AMD aims to reduce the engineering risk and time-to-deployment for hyperscalers. This strategy is a direct attempt to break Nvidia's dominance by providing a viable, integrated alternative for massive AI clusters.
Manish Kumar, EVP for secure power and data centers at Schneider Electric, noted that the collaboration "bridges the gap between advanced AI compute platforms, energy tech, and real-world data center implementation."
What to Watch
As hyperscalers evaluate the Helios architecture, the industry will be watching for real-world deployment benchmarks and the speed at which these 10-megawatt clusters can be scaled. While the technical specifications are confirmed, the primary challenge remains whether AMD can convince the largest cloud providers to pivot their facility designs away from the established Nvidia ecosystem toward this new rack-scale standard.