Open architecture combines flexible CPU and GPU configurations to support hyperscale agent execution and enterprise AI deployments.
AMD has introduced AMD Helion, an open rack-scale infrastructure blueprint designed to address the diverse computing requirements of agentic AI, giving enterprises and hyperscalers greater flexibility to balance performance, power consumption and deployment costs.
Built around an Open Compute Project (OCP) specification-based design, Helion extends AMD’s rack-scale infrastructure strategy with configurations tailored to different AI workloads. These range from CPU-only racks for hyperscale agent execution to integrated CPU and GPU systems designed for enterprise deployments.
The CPU-only configuration, powered by sixth-generation AMD EPYC processors, can support up to 32,768 CPU cores per rack. It is designed to handle agent execution while connecting to separate infrastructure for model inference. An alternative configuration combines AMD EPYC CPU nodes, AMD Instinct MI350P GPU nodes and general-purpose compute within a single platform.
“Agentic AI is not a single workload. It spans personal agents, enterprise applications, code execution and complex workflows, each with different compute requirements.” — Dan McNamara, Senior Vice President and General Manager, Compute and Enterprise AI, AMD
The architecture reflects an emerging infrastructure challenge: agentic AI requires more than accelerated model inference. AI agents also need compute resources to execute tasks, run code and coordinate complex workflows. AMD is positioning Helion as a way to match these requirements to different hardware configurations while supporting air- and liquid-cooled designs.
Helion also combines open hardware and software with an integrated token router designed to direct agents’ model calls across frontier and open-source models. AMD plans to contribute the specification to the OCP community for further development and adoption.
Industry partners, including Cisco, Dell Technologies, HPE, Lenovo, QCT, Supermicro, Wiwynn and Hon Hai Technology Group, have expressed support for the architecture, highlighting opportunities for customised rack integration, networking, cooling and deployment.
AMD said Helion configurations are under development with ecosystem partners and are planned to become available in November 2026. The open blueprint could give organisations another option for scaling agentic AI while adapting infrastructure to existing data-centre investments and workload demands.