AMD has unveiled Helion, an open-standard rack reference design aimed at AI data centres, with a CPU-only configuration capable of holding up to 32,768 Zen 6 cores. The rack is planned for the first half of 2027, giving server makers a design to build from rather than a finished AMD rack to order today.
AMD Watch analysis
What happened
Tom’s Hardware reports that Helion uses a standardised 21-inch rack design based on the Open Compute Project’s ORv4 specification. It supports up to 44 OpenU trays and can be configured for air or direct liquid cooling. AMD is contributing the design to OCP; OEMs and ODMs, rather than AMD, are expected to bring racks based on it to market.
The CPU-only configuration can accommodate up to 128 EPYC “Venice” 9996 processors, adding up to 32,768 cores per rack. A different configuration can include up to 16 MI350P GPUs and 102 Tbps of networking, with less room for CPUs. AMD positions that hybrid option for enterprises seeking an all-in-one rack for data management, networking and AI.
Why it matters
Agentic AI is creating demand for more than accelerators: agents can also need CPU capacity for tasks such as building software in separate sandboxes, while inference runs elsewhere. Helion gives AMD a reference design for that kind of infrastructure as well as a more mixed CPU-and-GPU setup.
The important distinction is that this is a design contribution, not evidence that racks are already shipping or deployed. Its practical impact will depend on which manufacturers adopt it, what configurations they offer and when they become available.
Our read
Helion is AMD making a case that the infrastructure behind AI agents should not be all accelerators, all the time. The CPU-heavy option is the intriguing part: if agent workloads grow as expected, operators may need more flexible ways to add ordinary compute alongside inference hardware. For now, it is a plan for partners to turn into products, not a rack you can roll into the data centre next week.
What to watch
- Which OEMs and ODMs adopt the Helion reference design.
- Whether partners bring both CPU-only and hybrid configurations to market.
- Whether the planned first-half 2027 availability holds.
- What deployment details AMD shares about the design’s cooling and workload performance.
Discussion spark: Should agentic AI infrastructure be built around large pools of CPU capacity as well as accelerators, or will specialised AI chips still do most of the important work?
Sources and evidence
Independent WittyWires tracker for public updates about AMD. Not affiliated with or endorsed by AMD; this is not an official account.