AMD introduced the Helion Agentic AI Rack on October 12 as an open blueprint based on an Open Compute Project specification. It separates agent execution, model inference and general compute into configurations that can match workload and power needs. A CPU-only design can provide up to 32,768 sixth-generation EPYC cores, while an enterprise all-in-one option combines EPYC CPUs, Instinct MI350P GPUs and general-purpose nodes. An integrated token router directs model calls across frontier and open-source models. Air- and liquid-cooled versions are supported, AMD plans to contribute the specification to OCP, and partner systems are planned for November. It targets utilization and power problems in heterogeneous agent workloads, but pricing, field efficiency, delivery scale and software maturity remain unproven until deployment.