AMD Unveils Helion Architecture for Agentic AI Workloads

Advanced Micro Devices, Inc. introduced AMD Helion, an open, rack-scale architecture built specifically for agentic AI workloads. The platform enables customers to customize power, performance, and CPU core density, offering configurations for both hyperscale deployments and enterprises. Hyperscale options support up to 32,768 6th Gen AMD EPYC cores, while enterprise versions integrate AMD Instinct MI350P GPUs.

The architecture includes an integrated token router to manage model calls between open-source and frontier models and supports both air- and liquid-cooled designs. To foster community development, the company plans to contribute the Helion specification to the Open Compute Project.

AMD is developing the platform alongside a partner ecosystem including Cisco, Dell Technologies, Hon Hai Technology Group, HPE, Lenovo, QCT, Supermicro, and Wiwynn. The architecture is scheduled for availability in November 2026.