AMD introduced Helion, an open blueprint for agentic AI infrastructure. It lets customers match CPU core density, performance and power to agentic workloads. Configurations include CPU-only racks for hyperscale and integrated CPU/GPU racks for enterprises. A CPU-only rack can support up to 32,768 cores using 6th Gen AMD EPYC processors. An all-in-one rack combines EPYC CPUs, Instinct MI350P GPUs and general-purpose compute. AMD plans to contribute the Helion specification to the Open Compute Project. Helion racks are under development with partners and planned for November 2026. The design includes an integrated token router for model calls across frontier and open-source models. Partners can tailor configurations to workload performance, power and cost needs. Existing IT infrastructure can be repurposed for agentic AI with Helion.
What changed
Agentic AI infrastructure required custom, closed designs for each workload.
What it unlocks
Tailoring CPU core density, performance, and power to agentic workloads in open rack-scale designs.
- 32,768 CPU cores per rack
- planned availability November 2026
What you need to act on it
- OCP spec-based rack blueprint
- AMD EPYC or Instinct hardware
Sources