AMD introduces Helion open rack reference design for agentic AI

AMD has unveiled Helion, an open rack reference design tailored for agentic AI workloads that will be contributed to the Open Compute Project. The 21-inch, 44OU rack supports both direct liquid cooling and air cooling alongside up to 102Tbps of networking. Rather than selling the racks directly, AMD is offering the design as a blueprint for ecosystem partners, OEMs, and ODMs. [1][2][3]

Helion comes in two primary configurations: a hyperscale CPU-only setup and an all-in-one enterprise model. The CPU-only architecture can accommodate up to 128 sixth-generation EPYC Zen 6 Venice processors, delivering as many as 32,768 cores to handle agent execution and offloading inference to external accelerators. The enterprise configuration combines EPYC CPU blades with up to 16 Instinct MI350P GPUs for integrated inference and compute. [1][2][3]

Partners working on Helion designs include Cisco, Dell Technologies, Foxconn, HPE, Lenovo, Supermicro, and others. Expected availability varies across reports, ranging from November 2026 to the first half of 2027, while official pricing and full hyperscale power figures have not been disclosed. [2][3]