AMD unveiled its open agentic AI infrastructure rack, AMD Helion, on Oct. 12 local time. Helion is an open agentic AI infrastructure blueprint designed to configure servers to match tasks handled by AI agents. AMD said it applies AMD computing technology to Open Compute Project (OCP) specification designs.

Helion follows the design direction of AMD Helios, a rack-scale solution that promotes an open platform approach. Enterprise customers can set CPU performance, core density and power based on the tasks handled by agents, and add GPU acceleration. It can be configured in various ways, including liquid and air cooling, and also supports open software, the company said. Partner companies can divide agent execution, model inference and supporting computing based on customer workloads.

The CPU-only rack uses 6th-generation AMD EPYC processors and supports up to 32,768 CPU cores. It targets hyperscale customers that operate large data centres. It provides dedicated computing capacity for agent execution and connects model inference to separate infrastructure for processing. The structure separates the CPU area where agents handle tasks from the model inference area.

The all-in-one rack combines EPYC CPU nodes, AMD Instinct MI350P GPU nodes and general-purpose computing on one platform. It is designed so partners and customers can build enterprise agentic AI systems at once. The company said enterprises introducing agentic AI into existing IT infrastructure can use equipment they have already invested in as the foundation for new AI functions.

Cisco, Dell Technologies, Hon Hai Technology Group, HPE, Lenovo, Supermicro and Wiwynn, a total of 7 companies, expressed their intention to support Helion. AMD plans to propose Helion specifications to the OCP so the community can further develop and adopt them. OCP is a collaborative organisation that opens and shares data centre equipment designs. AMD said it is developing agentic rack configurations with ecosystem partners, with a launch set for November.

Dan McNamara (댄 맥나마라), senior vice president and general manager of AMD's Compute and Enterprise AI group, said, "Agentic AI is not a single workload. It spans diverse areas such as personal agents, enterprise applications, code execution and complex workflows, and each has different computing requirements." He said, "To support this, you need computing and infrastructure that are optimised for each workload and can respond to diverse needs." He added, "AMD Helion combines the performance of AMD EPYC, open software and flexible rack design to help partners build systems tailored to customer needs and use available power efficiently."