AMD has unveiled AMD Helion, an open blueprint for server racks built to run AI agents. Unlike most AI hardware news, which is all about GPUs, AMD Helion puts the CPU at the centre, with options ranging from a CPU-only rack with up to 32,768 cores to all-in-one racks that add GPUs. Here is what AMD Helion is, how it works and what is still unclear.
AMD announced Helion on October 12, 2026, in Santa Clara, California, around the Open Compute Project (OCP) Global Summit.
What is AMD Helion?
Helion is a rack-scale reference design, which means a ready-made blueprint that server makers can use to build complete racks of computers. It is based on an OCP specification and builds on AMD’s earlier Helios rack design.
AMD’s argument is that agentic AI, meaning AI that carries out tasks rather than just answering questions, is “not a single workload.” Dan McNamara, AMD senior vice president, said it spans personal agents, enterprise apps, code execution and complex workflows, each needing different computing power. When an agent runs code, calls tools or handles a workflow, much of that work runs on CPUs, while the AI model itself runs on GPUs.
AMD Helion configurations for agentic AI
AMD gives two example setups:
- A CPU-only rack with 6th Gen AMD EPYC processors and up to 32,768 CPU cores. It’s aimed at hyperscale companies that want dedicated capacity for running agents, connected to separate systems for AI models.
- An all-in-one rack that combines AMD EPYC CPU nodes, AMD Instinct MI350P GPU nodes and general-purpose computing. It’s aimed at enterprises.
Both liquid-cooled and air-cooled versions are possible. AMD also says Helion includes an integrated “token router” that directs agents’ requests to different AI models, both frontier and open-source.
Partners, open specification and availability
AMD plans to contribute the Helion specification to OCP so others can build on it. Several server and infrastructure companies backed it in the announcement, including Cisco, Dell, Hon Hai (Foxconn), HPE, Lenovo, QCT, Supermicro and Wiwynn.
According to AMD, Helion configurations are being developed with these partners and are planned to be available in November 2026.
Benefits and concerns
The open approach is the main selling point. Companies can mix CPUs and GPUs to fit their needs, use the infrastructure they already have and avoid being tied to one closed system. AMD says this helps customers make better use of the power available to them, which is a growing problem for data centres.
But there are things to keep in mind. AMD has not shared performance numbers, prices or independent benchmarks for Helion. The “planned for November” timeline is a forward-looking statement, and AMD’s own release lists risks such as supply problems and export rules. Being open at the rack level also doesn’t mean every part is interchangeable, since the example designs are built around AMD chips.
There is a security side too. AI agents that execute code are a new kind of risk: if an agent is misled by bad input, it could run harmful code or leak data. The token router sends agents’ requests between different models, including outside frontier models, so companies will need to control which data goes where. AMD’s release itself lists “potential security vulnerabilities” and “issues related to the responsible use of AI” among its risk factors.
The race to build hardware for agents is heating up, as we saw with NVIDIA’s Vera CPU for AI agents. Power remains the big limit, as covered in our look at 800-volt DC power and the $7 trillion data centre race.
As a writer, I feel AMD is right that agents need more than GPUs. A lot of the real work happens on ordinary processors. I like the open blueprint, but until there are real numbers and independent tests, I see Helion as a promising design, not a proven one.
FAQ: AMD Helion agentic AI rack
What is AMD Helion?
AMD Helion is an open, OCP-based rack blueprint for agentic AI infrastructure, announced on October 12, 2026.
How many CPU cores does AMD Helion support?
AMD’s CPU-only Helion rack, using 6th Gen AMD EPYC processors, can support up to 32,768 CPU cores.
When will AMD Helion be available?
AMD says Helion configurations are being developed with partners and are planned to be available in November 2026.