The Pulse
AMD Puts CPU-Only and Mixed Compute in Helion Rack Blueprint
AMD introduced Helion, an open rack blueprint for agentic AI, with both CPU-only and combined CPU-GPU configurations. The company says partner-developed systems are planned for November 2026 and that it intends to submit the specification t
AI.info Team · Written with AI ·
AMD gives agentic AI two rack designs
AMD’s new Helion blueprint puts a hardware choice at the center of agentic AI infrastructure: dedicate a rack to CPU-based agent execution, or combine CPUs and GPUs in a single system. The company introduced the design on October 12, describing it as an open rack blueprint based on an Open Compute Project specification. Its pitch reflects a distinction between coordinating agents and running the models they call—tasks that may need different mixes of compute.
“Agentic AI is not a single workload. It spans personal agents, enterprise applications, code execution and complex workflows, each with different compute requirements,” said Dan McNamara, AMD’s senior vice president and general manager of Compute and Enterprise AI. AMD says customers can tailor CPU performance, core density and power alongside GPU acceleration to suit those workloads.
A 32,768-core option for agent execution
One proposed Helion configuration uses 6th Gen AMD EPYC processors in a CPU-only rack supporting up to 32,768 CPU cores. AMD positions that design for hyperscale customers who want dedicated capacity to execute agents, with connections to separate infrastructure for model inference. The distinction matters: the rack is not described as an all-purpose substitute for GPU systems, but as a way to allocate compute to a different part of an agent workflow.
The second configuration combines AMD EPYC CPU nodes, Instinct MI350P GPU nodes and general-purpose compute in one rack. AMD presents it as an all-in-one concept for enterprise agentic AI, where customers may want agent execution and model inference on the same platform. The company says Helion designs may use either liquid or air cooling, giving partners options for different deployment requirements.
Partners want more than processor speed
AMD also describes Helion as a way to fit agentic workloads into existing IT infrastructure. Partners can adjust CPU and GPU configurations around each customer’s performance, power and cost requirements, the company says. That leaves system design and integration to the ecosystem rather than prescribing a single fixed rack configuration.
Cisco’s support statement makes the deployment question broader than the processor mix. Jeremy Foster, senior vice president and general manager of Cisco Compute, said: “It’s not just about compute performance, but how the entire system is connected, managed and secured.” AMD’s release also names Dell Technologies, HPE, Lenovo, QCT, Supermicro, Hon Hai Technology Group and Wiwynn among companies offering statements of support or collaboration.
OCP contribution and November target remain ahead
Helion builds on AMD’s Helios rackscale approach, but the two names refer to separate designs in the announcement: Helios is AMD’s existing rackscale solution, while Helion targets agentic AI configurations. AMD says it plans to contribute the Helion specification to the Open Compute Project for further development and adoption. That contribution is a plan, not a completed standards approval.
The rack configurations are still under development with ecosystem partners, and AMD plans to make them available in November 2026. The announcement does not give pricing, specific customer deployments or performance benchmarks for Helion. For now, the concrete offer is a design blueprint with two paths: a CPU-led rack for hyperscale agent execution and a mixed CPU-GPU rack aimed at enterprise deployments.