AMD unveils Helion, a 32,768-core CPU rack built for AI agents
Representative image: kewl / Pixabay
AMD's Lisa Su unveiled Helion, a CPU-centred rack with up to 32,768 EPYC cores for AI agents, at the OCP Global Summit on 12 October 2026.
What Happened
AMD chief executive Lisa Su unveiled Helion, a new rack-scale server design built mainly around processors rather than graphics chips, during her keynote at the Open Compute Project (OCP) Global Summit in San Jose, California, on Monday, 12 October 2026. AMD presented Helion as a system for "agentic" artificial intelligence, the kind of AI that plans and carries out multi-step tasks on its own. The company said the rack supports up to 32,768 processor cores using its sixth-generation EPYC "Venice" server chips, and that it will contribute the Helion design to the OCP so other companies can build on it.
Key Facts
- Helion was announced on 12 October 2026 at the OCP Global Summit in San Jose.
- It is built on sixth-generation AMD EPYC "Venice" server processors and supports up to 32,768 CPU cores in a rack.
- Reports describe both CPU-only and combined CPU-GPU configurations.
- AMD plans to contribute the Helion specification to the Open Compute Project.
- Server makers' versions are expected from November 2026; reports name partners including Dell, HPE, Lenovo, Supermicro and Cisco.
- Su also revealed "Verano", a later generation of EPYC chips on AMD's roadmap.
- Helion sits alongside Helios, AMD's GPU-focused rack-scale system launched earlier in 2026.
Why It Matters
For the past three years, the story of AI hardware has been about graphics processors, the chips that train and run large language models. Helion points to a shift. AI agents do not just generate a single answer; they call tools, search databases, run code and pass results between programs. Much of that work is ordinary computing that runs on central processors. As companies deploy more agents, chipmakers see demand rising for racks packed with general-purpose CPUs that sit next to, or in front of, the GPU clusters. Su told the summit audience that traffic from AI agents is rising sharply, according to reports of her keynote.
The move also matters in the contest between chipmakers. By designing a full rack around its EPYC chips and opening the design through the OCP, AMD is trying to make its processors the default choice for the agent layer of AI data centres, and to give big cloud companies a standard design they can customise rather than buy as a closed system.
Some details reported after the keynote vary between sources, including the exact rack standard used and the specific accelerators offered in mixed configurations, and should be confirmed when AMD publishes full specifications.
| Item | Detail |
|---|---|
| Product | AMD Helion rack-scale design |
| Announced | 12 October 2026, OCP Global Summit, San Jose |
| Processors | 6th Gen AMD EPYC "Venice" |
| Maximum cores | 32,768 per rack |
| Partner systems | Expected from November 2026 |
Impact
Short-term: Cloud companies and server makers can begin planning Helion-based systems, with partner products expected from November 2026.
Long-term: If agent workloads grow as chipmakers expect, data centres may need far more CPU capacity alongside GPUs, which could reshape spending and strengthen AMD's position in AI infrastructure.
Who is affected: Hyperscale cloud providers, enterprise data centre operators, server makers such as Dell, HPE, Lenovo and Supermicro, rival chipmakers Nvidia and Intel, and companies building AI agent services.
Key Takeaway
AMD's Helion, unveiled on 12 October 2026, is a CPU-centred rack with up to 32,768 EPYC cores aimed at AI agents, and its design will be shared through the Open Compute Project.
Questions and Answers
What is AMD Helion?
It is a rack-scale server design announced by AMD on 12 October 2026, built around sixth-generation EPYC "Venice" processors with up to 32,768 CPU cores, aimed at running AI agent workloads.
Why would AI agents need CPUs rather than GPUs?
Agents spend much of their time on general computing tasks such as calling tools, querying databases and running code, which processors handle. GPUs remain central for training and running the underlying AI models.
When will Helion systems be available?
Server makers' versions are expected from November 2026, according to reports of the announcement.
How is Helion different from AMD Helios?
Helios is AMD's GPU-focused rack-scale system launched earlier in 2026, while Helion is centred on CPUs, with CPU-only and CPU-GPU configurations reported.
Disclaimer: Prepared by the Peepals newsroom from publicly available sources with AI assistance. Information is accurate to the best of our knowledge at the time of publication and may change. Images may be representative. Not professional advice. Report an error via our contact page.
Sourced and fact-checked by the Peepals Global Editorial Team
Zoho skips annual pay hikes as AI costs soar; Vembu calls software job outlook grim
Miko maker Inner Sky Labs unveils India-built physical AI models and Kavach safety layer
Reported, fact-checked and published by the Peepals Global Editorial Team.
