AMD unveils Helion rack blueprint for agentic AI

AMD introduced Helion, an open rack blueprint for agentic artificial intelligence, at the Open Compute Project Summit in San Jose on October 12, positioning the design as a way to configure CPU and GPU resources for systems that plan tasks, call tools and carry out multi-step workflows. The blueprint offers two broad configurations. AMD says a CPU-only rack can support up to 32,768 cores, while an enterprise configuration combines EPYC processors with as many as 16 Instinct MI350P GPUs. The company describes the CPU-focused option as suited to hyperscale operators running agent tasks alongside separate model-inference infrastructure. The mixed rack is intended to bring CPU, GPU and general-purpose computing together for enterprise use. Helion also includes an integrated token router, which AMD says can direct agent requests among different models, including cloud-hosted frontier models and locally run open-source systems. The design supports air or liquid cooling, according to the company. AMD says it plans to contribute the specification to the Open Compute Project for further development.