AMD Unveils Helion Open Rack Design Supporting 32,768 EPYC Cores
AMD has unveiled Helion, an open data center rack design capable of supporting up to 32,768 EPYC processor cores, with backing from major server manufacturers Cisco, Dell, HPE, Lenovo, and Supermicro, and availability slated for November 2026.
What Helion Offers
The Helion rack is built around AMD’s EPYC server processor line and is designed to pack an unusually high core count into a single rack, aimed at data center operators running large-scale computing workloads, including AI training and inference, high-performance computing, and large cloud services. Supporting tens of thousands of cores in a single rack reflects the broader industry push toward higher density infrastructure that can deliver more compute per square foot of data center space.
Broad Server Manufacturer Support
Unlike a proprietary design limited to a single hardware vendor, Helion has been adopted as an open design by five major server manufacturers, meaning data center operators will be able to source Helion-compatible equipment from multiple suppliers rather than being locked into a single vendor relationship. This kind of broad, multi-vendor backing is often a strong signal that a hardware standard will see meaningful real-world adoption rather than remaining a niche offering.
AMD’s Position in the AI Infrastructure Race
While AMD is often discussed primarily in the context of its GPU competition with Nvidia, the Helion rack underscores that the company’s EPYC CPU line remains a significant part of its data center strategy, particularly as AI workloads require substantial CPU resources for tasks like data preprocessing, orchestration, and running AI applications outside the core model computation itself.
Timing and Industry Context
Helion’s November 2026 launch target places it close behind Nvidia’s open-sourced Vera Rubin rack designs, announced around the same time, suggesting both companies are racing to establish their respective infrastructure standards as the next generation of AI data centers is being planned and built. For data center operators, the near-simultaneous arrival of both designs offers more choice, but also means infrastructure planning decisions made in the coming months could shape vendor relationships for years.
