AMD and HPE Announce Plans for a New AI Platform and a Major Supercomputer

AMD and Hewlett Packard Enterprise are deepening their long-running collaboration, this time focusing on large-scale AI systems. The two companies announced a plan to build out “Helios,” an open rack-scale architecture designed for organisations that want to deploy and scale AI clusters without relying on closed or proprietary setups.

The move ties together AMD’s compute hardware with HPE’s system design and networking expertise, offering a more modular approach to running high-demand AI and HPC workloads.

Building An Open Stack For Large AI Clusters

Helios brings multiple AMD technologies under one roof, including EPYC CPUs, Instinct GPUs, Pensando networking components, and the ROCm software stack. Instead of treating each part as a separate layer, the platform aims to give cloud providers and enterprises a unified path for designing and rolling out multi-rack AI systems.

HPE will be among the first system builders to adopt Helios. The company is integrating a scale-up Ethernet switch developed with Broadcom, using the Ultra Accelerator Link over Ethernet (UALoE) standard. The goal is to provide high-bandwidth, low-latency connectivity inside large GPU clusters while sticking to open standards rather than one-vendor solutions.

Executives from both companies framed this as a continuation of their work in high-performance computing, with Helios being a more AI-driven evolution of earlier projects.

Performance Targets And Global Availability

Helios is built around the OCP Open Rack Wide format, a design that allows companies to deploy systems with less custom engineering. According to the provided specifications, each rack can deliver up to 2.9 exaFLOPS of FP4 compute, powered by AMD’s upcoming Instinct MI455X GPUs and EPYC “Venice” CPUs.

Pensando Vulcano NICs handle scale-out networking, while the ROCm software environment serves as the common layer for AI and HPC workloads.

HPE plans to make Helios-based systems available globally in 2026.

New “Herder” Supercomputer To Support European HPC And AI

Alongside the Helios update, AMD and HPE revealed a new supercomputer called Herder for the High-Performance Computing Center Stuttgart (HLRS) in Germany. The system uses Instinct MI430X GPUs and next-generation EPYC “Venice” CPUs inside the HPE Cray GX5000 platform.

Herder is designed for hybrid workloads that combine traditional simulation with machine learning, a direction many research institutions are now moving toward. HLRS expects the system to help scientists handle larger simulations while also experimenting with AI-driven techniques.

The system is scheduled for delivery in the second half of 2027 and is expected to enter service by the end of the year.