AMD has announced an expansion of its long-standing partnership with Hewlett Packard Enterprise (HPE) to push the next wave of open rack-scale AI systems. As part of this move, HPE will become one of the OEMs to adopt “Helios”, AMD’s new open, full-stack AI architecture designed specifically for massive AI cluster deployments. It’s scheduled for delivery in late 2027, replacing HLRS’ current flagship Hunter.
Designed for cloud-scale AI without the cloud
The Helios platform combines AMD EPYC CPUs, AMD Instinct accelerators, AMD Pensando networking and the ROCM open software stack under a single architecture. All this technical jargon means the Helios platform is powerful enough to run huge AI workloads, the type usually done in big cloud data centres. However, this could mean that customers can run it entirely on premises. The result: easier deployment, higher efficiency, and better scalability for research labs, hyperscalers, and enterprise AI workloads.
HPE has also contributed purpose-built networking solutions built on HPE Juniper hardware, in collaboration with Broadcom, delivering high-bandwidth, low-latency Ethernet connectivity using the Ultra Accelerator Link over Ethernet (UALoE) standard. This helps keep the entire platform aligned with an open standard, rather than a proprietary, locked-down AI system.
AMD says a single Helios rack can deliver 2.9 exaFLOPS of FP4 compute, which is powered by AMD Instinct MI455X GPUs paired with AMD EPYC Venice CPUs. All of this is built on the OCP Open Rack wide design. This should make setup and scaling faster for large-cluster operators, while keeping flexibility for custom deployment needs.
As part of the collaboration, the High-Performance Computing Centre Stuttgart HLRS in Germany is building a new supercomputer called Herder. Powered by AMD Instinct MI430X GPUs paired with AMD EPYC “Venice” CPUs. The header is built on the HPE Cray GX5000 platform and will support both traditional HPC simulations and emerging generative AI workloads.
With Helios, AMD is positioning itself as a leading advocate for open AI infrastructure, while competitors largely build proprietary, closed systems. The HPE partnership brings:
- faster AI cluster deployment
- interoperability backed by open standards
- lower entry barrier for sovereign and enterprise AI workloads
The goal is to provide governments, research labs, and cloud providers with datacenter-class AI at scale without vendor lock-in.
