AMD introduces Helion open rack architecture for agentic AI

AMD has introduced Helion, an open, modular rack-scale architecture designed to address the evolving computing demands of agentic AI. Unveiled at the Open Compute Project Global Summit, the architecture aims to balance the CPU resources required for complex workflow execution with the GPU infrastructure used for AI model inference.

Helion offers two primary configurations: a CPU-only rack supporting up to 32,768 sixth-generation EPYC cores, and an integrated enterprise design that combines EPYC processors with Instinct MI350P accelerators. Built on Open Compute Project standards, the system allows for independent scaling of CPU and GPU resources, providing a modular foundation for hardware customization and future upgrades.

AMD provides various configurations, including high-density, balanced, and high-frequency CPU options, to help designers match performance to specific AI workloads. Initial availability for Helion is targeted for November 2026, with commercial implementations expected through industry partners.