AMD Unveils Helion Agentic AI Rack Design
AMDがAIエージェントの実行と推論を分離するHelionを発表。CPU専用とGPU統合構成、提供計画を整理。
AMD introduced Helion on October 12, 2026, as a rack-scale design for agentic AI infrastructure, separating agent execution from model inference where appropriate.
AI agents run tools, execute code and coordinate workflows in addition to querying models. These operations can place substantial demand on CPUs.
One example CPU-only rack using sixth-generation EPYC processors supports up to 32,768 CPU cores. This is a design configuration, not a claim about every deployment.
An integrated configuration combines EPYC CPUs and Instinct MI350P GPUs, aiming to serve enterprises that prefer agent execution and inference in one rack.
AMD plans to contribute the specification to the Open Compute Project and targets availability from November 2026, subject to partner development.
Buyers should evaluate concurrent agents, power use, operating costs and connectivity to inference infrastructure rather than GPU performance alone.