AMD, HPE team up on open rack-scale AI Platform built on Helios

AMD (Advanced Micro Devices, Inc.) and HPE (Hewlett Packard Enterprise) have expanded their partnership. Now, the companies are slated to deliver an open, rack-scale AI platform built on AMD’s upcoming Helios architecture.

The new platform will target cloud, research, and enterprise-level AI workloads. With this, HPE is set to be among the first OEMs to adopt Helios.

HPE is set to combine AMD’s EPYC ‘Venice’ CPUs, Instinct accelerators, Pensando networking, and ROCm software. This will be utilized with HPE’s Juniper Networking switches using Broadcom technology and Ultra Accelerator Link over Ethernet for high-bandwidth, low-latency connectivity.

For reference, the Helios-based racks are designed around the OCP Open Rack Wide standard. It can deliver up to 2.9 exaFLOPS of FP4 performance per rack, aimed at hyperscalers and service providers building large-scale AI clusters.

HPE aims to make systems based on the AMD Helios AI Rack-Scale Architecture available globally in 2026. The company will also pitch them as a faster path to deploying standards-based AI infrastructure.

Both companies have also announced ‘Herder’, a new supercomputer for Germany’s HLRS that will use HPE Cray GX5000 systems with AMD Instinct MI430X accelerators and the aforementioned EPYC ‘Venice’ processors.

This is scheduled for delivery within the second half of 2027, expected to go online by year-end. Herder will succeed the current ‘Hunter’ system, supporting both traditional simulation and emerging AI and machine learning workloads.