AMD launches Helion open rack for massive agentic AI workloads
AMD unveiled the Helion platform, an open rack system for agentic AI that can scale to 32,768 EPYC CPU cores and combine CPUs with Instinct GPUs, with availability slated for November 2026.
AMD announced the Helion platform, an open-rack solution built to support agentic artificial intelligence, based on an Open Compute Project spec. The system can be deployed as a CPU-only rack using sixth-generation EPYC silicon, scaling up to 32,768 cores for hyperscale customers that need dedicated agent execution separate from inference workloads. An alternative all-in-one rack merges EPYC CPU nodes with Instinct MI350P GPU nodes and general-purpose compute, targeting enterprise AI deployments.
Both configurations are offered in air-cooled and liquid-cooled variants and feature an integrated token router that directs AI agents' model calls across frontier and open-source models. AMD intends to submit the Helion design to the OCP so other vendors can adopt it, and launch partners such as Cisco, Dell Technologies, Hon Hai Technology Group, Hewlett Packard Enterprise, Lenovo, QCT, Supermicro and Wiwynn are developing their own versions for a November 2026 rollout. Executives Dan McNamara, Arun Narayanan and Trish Damkroger highlighted the platform's flexibility, performance and power efficiency.
Why it matters
Helion provides a scalable, open-standard rack for large-scale AI agents, potentially lowering costs and speeding adoption across data centers.
In this story
