AMD And HPE Expand Collaboration To Advance Open Rack-Scale Ai Infrastructure

AMD has expanded its work with HPE to push forward a new generation of open, rack-scale Ai systems, positioning both vendors to meet fast-growing demand for large training and inference clusters across cloud, research, and enterprise environments.

HPE will be among the first OEMs to bring AMD’s Helios architecture to market. The platform integrates EPYC CPUs, Instinct GPUs, Pensando networking, and the ROCm software stack into a single open design aimed at simplifying deployment and improving efficiency at scale.

Moving To Rack-Scale Ai

AMD described Helios as a full-stack architecture designed for large Ai workloads running across massive clusters. The platform builds on the OCP Open Rack Wide standard and includes next-generation EPYC Venice CPUs, Instinct MI455X GPUs, and Pensando Vulcano NICs.

The company also worked with Broadcom on a purpose-built HPE Juniper Networking switch for Helios. The switch uses the Ultra Accelerator Link over Ethernet standard to support high-bandwidth connectivity across training clusters.

Dr. Lisa Su, Chair and CEO at AMD, said the long-running relationship with HPE is moving into a new phase as demand for open infrastructure climbs.

“With Helios, we’re taking that collaboration further, bringing together the full stack of AMD compute technologies and HPE’s system innovation to deliver an open, rack-scale Ai platform that drives new levels of efficiency, scalability, and breakthrough performance for our customers in the Ai era,” Su said.

Antonio Neri, President and CEO at HPE, said the two firms have spent more than a decade working together on systems that support high performance computing.

“With the introduction of the new AMD Helios and our purpose-built HPE scale-up networking solution, we are providing our cloud service provider customers with faster deployments, greater flexibility, and reduced risk in how they scale Ai computing in their businesses,” Neri said.

HPE plans to offer Helios worldwide in 2026.

Architecture Aimed At Large Models

AMD said Helios can deliver up to 2.9 exaFLOPS of FP4 performance per rack. The platform targets organisations looking for open standards rather than proprietary interconnects, with the ROCm software stack positioned as a key component for flexibility across both Ai and HPC workloads.

The design is intended to support faster deployment timelines for providers rolling out large clusters. The Open Rack Wide format also gives partners room to integrate custom components, such as the HPE and Broadcom switch developed for the platform.

New Hardware For European Research

The collaboration extends into high performance computing. Herder, a new system for the High-Performance Computing Center Stuttgart in Germany, will be built on HPE’s Cray Supercomputing GX5000 platform and powered by AMD Instinct MI430X GPUs alongside next-generation EPYC Venice CPUs.

HLRS expects the system to support workloads ranging from numerical simulation to machine learning and hybrid Ai workflows.

Prof. Dr. Michael Resch, Director of High-Performance Computing Center Stuttgart (HLRS), said the combined design offers flexibility for researchers dealing with a mix of traditional and emerging workloads.

“The pairing of AMD Instinct MI430X GPUs and EPYC processors within HPE’s GX5000 platform is a perfect solution for us at HLRS,” Resch said. “Our scientific user community requires that we continue to support traditional applications of HPC for numerical simulation. At the same time, we are seeing growing interest in machine learning and artificial intelligence.”

He said Herder’s architecture will allow users to run larger simulations and develop more efficient computational methods that benefit from next-generation hardware.

Delivery of Herder is scheduled for the second half of 2027, with the system expected to go into service by the end of that year. It will replace the current HLRS flagship, Hunter.

Further details on AMD’s Ai and HPC technologies are available at https://www.amd.com.

Last Updated on December 3, 2025 by Nick Ross