AMD introduces Helion, an open rack-scale blueprint for agentic AI infrastructure

AMD

AMD introduced Helion, a rack-scale design for agentic AI built on an Open Compute Project specification and positioned as a follow-on to its Helios rack. It comes in a CPU-only version for agent execution and an all-in-one version that adds Instinct GPUs for enterprise workloads, and it includes a token router that sends agents' model calls across frontier and open-source models.

• CPU-only racks use 6th Gen EPYC processors and scale to 32,768 CPU cores for hyperscale agent execution
• The all-in-one rack combines EPYC CPU nodes with Instinct MI350P GPU nodes, in liquid-cooled and air-cooled versions
• Cisco, Dell, HPE, Lenovo, Supermicro and others are partners tailoring configurations to customers
• Agentic rack configurations are planned for November 2026; the announcement gave no pricing, power draw or benchmarks