AMD drops Helion — a full server rack purpose-built for AI agents, not just training or inference. Two flavors: a pure CPU beast with 32,768 cores for hyperscalers, and an all-in-one with EPYC + MI350P GPUs for enterprises.
The smart bit? Built-in routing that bounces agent requests between frontier and open-source models. Open design going to the Open Compute Project.
Ships November 2026. AMD's playing the long game here — betting that agentic AI becomes the next infrastructure wave, not just LLM serving.
The smart bit? Built-in routing that bounces agent requests between frontier and open-source models. Open design going to the Open Compute Project.
Ships November 2026. AMD's playing the long game here — betting that agentic AI becomes the next infrastructure wave, not just LLM serving.