AMD Introduces Helion Open Blueprint for Agentic AI Racks

Summary

AMD has introduced AMD Helion, an open rack-scale blueprint for infrastructure designed for agentic AI workloads. Based on an Open Compute Project specification design, it is intended to let customers match CPU performance, core density, power use and GPU acceleration to how their agents execute tasks and call models. The blueprint supports liquid- and air-cooled configurations and includes an integrated token router that directs agent model calls across frontier and open-source models. One proposed CPU-only rack uses sixth-generation AMD EPYC processors and can support up to 32,768 CPU cores for hyperscale agent execution connected to separate model-inference infrastructure. An all-in-one design combines AMD EPYC CPU nodes, AMD Instinct MI350P GPU nodes and general-purpose compute for enterprise deployments that combine agent execution and GPU inference. AMD says Helion can be adapted to existing IT infrastructure and workload, performance, power and cost requirements. The company plans to contribute the specification to the OCP community, while rack configurations are being developed with ecosystem partners and are planned for availability in November 2026. The announcement includes support from Cisco, Dell, Hon Hai, HPE, Lenovo, QCT, Supermicro and Wiwynn, although the stated availability remains planned rather than completed.