AMD unveiled Helion on October 12 at the OCP Summit in San Jose, an open rack blueprint for agentic AI derived from its Helios design and built on an Open Compute Project specification. AMD plans to contribute the specification to OCP.
Helion comes in two main configurations. One is a CPU-only rack of 6th-generation EPYC processors, up to 32,768 cores, aimed at hyperscalers running agents on dedicated capacity. The other is an all-in-one rack that combines EPYC CPU nodes with Instinct MI350P GPU nodes for enterprises. Liquid- and air-cooled versions are offered, and an integrated token router sends agents' model calls to both frontier and open-source models.
Supermicro and QCT have backed the design, and partner configurations are expected in November 2026. The figures come from AMD's own announcement and have not been independently tested.