AI-rewritten: This is a summary of an article from Tom’s Hardware, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

AMD has revised its release date for the Helion reference design from the first half of 2027 to November 2026. This standardized rack-scale solution is based on the Open Compute Project’s ORv4 specification and supports 44 OpenU trays that are 21 inches wide. The design allows for both air-cooled and direct liquid-cooled deployments while remaining compliant with OCP’s DC-MHS standard for interoperability.

In a CPU-only configuration, Helion can house up to 32,768 Zen 6 cores using the EPYC ‘Venice’ 9996 processor. AMD describes this setup as resembling an agentic AI sandbox system, similar to models used in Meta’s Muse and OpenAI Dots where specific workloads run on dedicated CPU servers connected to separate inference systems. While Helion is primarily focused on CPUs, it also supports up to 16 MI350P GPUs and 102 Tbps of networking for hybrid enterprise use cases.

AMD clarified that it will not sell the racks directly; instead, this reference design is being contributed to OCP so that original equipment manufacturers can bring it to market. CEO Lisa Su plans to announce supporting partners during the upcoming OCP Global Summit. Although AMD did not disclose specific deployment partners in its pre-briefing, it suggested that existing Helios customers may adopt the new Helion architecture alongside their current AI training and inference infrastructure.