Tuesday 13 October 2026
Night Edition
ZOTPAPER

News without the noise

ZOTPAPER
Hardware & Devices

AMD unveils Helion rack design packing up to 32,768 Zen 6 cores for agentic AI servers

Reference design contributed to the Open Compute Project, with partners to be named at OCP Global Summit

By LineZotpaper
Published
Updated
Read Time2 min
Sources2 outlets
AMD has introduced Helion, a standardized 21-inch rack reference design for agentic AI servers that can house up to 32,768 Zen 6 cores in a CPU-only deployment, complementing the Helios rack it revealed at last year's OCP Global Summit.

AMD is expanding its rack-scale data center lineup with Helion, a reference design built around the Open Compute Project's ORv4 specification, which is slightly wider and taller than a standard IT rack. The design supports 44 OpenU trays and can be deployed with either air cooling or direct liquid cooling. It is also compliant with OCP's DC-MHS (Data Center Modular Hardware System) standard, which OCP says is intended to bring "interoperability between key elements of data center, edge, and enterprise infrastructure."

In a CPU-only configuration, Helion can hold up to 128 EPYC 9996 CPUs, the dense 256-core 'Venice' design, delivering the 32,768-core total. The reference design is more flexible than Helios, which is built specifically for AI training and inference, with Helion focused primarily on CPUs.

AMD will not sell racks itself. Helion is a reference design AMD is contributing to OCP, with OEMs and ODMs expected to bring it to market. The company did not name supporting partners during a press pre-briefing, but chief executive Lisa Su is scheduled to announce them during the lead keynote at the OCP Global Summit later today. AMD also did not disclose deployment partners, though it indicated that existing Helios customers are likely to also deploy Helion racks.

§

Analysis

Why This Matters

  • The design targets the growing market for CPU-based agentic AI inference, giving data center operators an open, standardized alternative to GPU-heavy training infrastructure.
  • By contributing the design to OCP, AMD is pushing an open alternative to rival Nvidia's proprietary rack-scale offerings.
  • The reference design could reshape how AI servers are built and serviced if OEMs and ODMs adopt it broadly.

Background

Helion follows AMD's Helios rack, unveiled at OCP Global Summit last year, which addresses AI training and inference workloads. Both designs are standardized through the Open Compute Project, an industry group that publishes open specifications for data center hardware. Helion centers on AMD's EPYC 'Venice' 9996, a high-density 256-core CPU, reflecting a push to serve AI workloads that do not require GPU accelerators.

Key Perspectives

AMD: Positions Helion as a flexible, CPU-focused sibling to Helios, offering rack-scale density while remaining compliant with open OCP standards. OEMs and ODMs: Stand to gain an open, interoperable reference design they can bring to market without developing rack infrastructure themselves. Critics and rivals: Competing vendors, particularly Nvidia, argue that GPU-based infrastructure remains superior for demanding AI workloads, and the practical uptake of new OCP designs will depend on real-world deployments.

What to Watch

  • Partner announcements from Lisa Su's OCP Global Summit keynote.
  • Whether existing Helios customers publicly commit to deploying Helion racks.
  • Adoption of the design by ODMs and the prices of resulting systems.

Sources

Zotpaper

Written by software from the reporting listed above, scored by an automated standards desk, and published without a person reading it first. If something here is wrong, tell the editor and it will be put right.

How we workSubscribe