AMD has introduced Helion, an open rack design that lets infrastructure partners tailor CPU and GPU resources to customers’ agentic AI workloads.
Announced Oct. 12, AMD Helion is a rack-scale reference design based on an Open Compute Project (OCP) specification. It is intended to help hyperscalers and enterprise customers build infrastructure for AI agents that execute tasks, run code and manage complex workflows.
For system integrators and infrastructure resellers, the opportunity is to help customers determine which workloads need CPU capacity, GPU acceleration or both, then match those requirements to their budgets and data center facilities.
AMD outlined two example configurations. The first uses sixth-generation AMD EPYC processors in a CPU-only rack capable of supporting up to 32,768 CPU cores. This setup targets hyperscalers that need dedicated computing capacity for agent execution, with AI model inference handled by separate infrastructure.
The second combines AMD EPYC CPU nodes, AMD Instinct MI350P GPU nodes and general-purpose computing resources in one rack. This configuration targets enterprises that want to run agentic AI workloads on an integrated platform. AMD says the blueprint supports air-cooled and liquid-cooled versions, allowing system builders to tailor deployments to their data center environments and power requirements.
AMD is also incorporating a token router that directs agents’ model requests across frontier and open-source models. This could help organizations coordinate access to different models within an agentic AI system.
AMD plans to contribute the Helion specification to OCP for further development and adoption. AMD is providing a reference design that hardware manufacturers and system integrators can use to build their own implementations.
The company says Helion configurations are in development with ecosystem partners and are planned for availability in November 2026. The design builds on AMD’s Helios rack-scale platform, extending its open infrastructure approach to workloads that require substantial CPU capacity alongside, or instead of, GPU acceleration.
Why CPU capacity matters for agentic AI
Agentic AI systems do more than generate responses. They can break tasks into steps, execute code, call tools and coordinate actions, creating demand for computing resources beyond the GPUs used to run large AI models.
Helion’s CPU-only configuration addresses that need by separating agent execution from model inference. For hyperscalers building services around large numbers of agents, this could make it easier to allocate computing resources to each part of the workload rather than relying on a single infrastructure design.
However, the 32,768-core figure describes the rack’s maximum capacity, not a guarantee of performance for every agentic workload. Actual results will depend on software, memory, networking and how efficiently applications use the available processors. Customers will also need to evaluate power consumption, cooling costs and the expense of integrating the system into existing environments.
What Helion means for enterprise channels and partners
For enterprise channel partners, Helion creates an opportunity to build AI infrastructure offerings around customers’ actual workloads rather than sell accelerator hardware alone. System integrators can help businesses determine how much CPU capacity they need for agent execution, where GPU acceleration makes sense and how to fit the equipment into existing data centers.
AMD has support from Cisco, Dell Technologies, Hon Hai Technology Group, Hewlett Packard Enterprise, Lenovo, QCT, Supermicro and Wiwynn. Their roles across networking, servers, storage, rack integration, cooling and manufacturing could help turn AMD’s reference design into complete enterprise deployments.
Dell, for example, says deployments can be extended with PowerRack networking and storage, while Supermicro intends to bring Helion to market through its Data Center Building Block Solutions.
For resellers, managed service providers and IT consultants, the commercial opportunity could extend to infrastructure assessment, system integration, deployment and ongoing support. Before committing to customer delivery dates, partners should confirm vendor-specific configurations, pricing, lead times and support responsibilities. Workload testing should also establish whether the proposed CPU/GPU mix meets the customer’s performance needs within available power and cooling capacity.
Read more: Beyond compute capacity, enterprise AI deployments also need access and spending controls, as illustrated by Nutanix’s latest tools for governing AI agents and private inference.