AMD Unveils Helion Rack Scales for Agentic AI Infrastructure at OCP Summit 2026

Partners expected to deliver CPU-only and hybrid CPU/GPU setups by November

AMD is broadening its rack-scale AI portfolio with a new architecture designed for AI agents, supporting both CPU-only configurations and integrated CPU/GPU deployments in liquid-cooled or air-cooled form factors.

Unveiled at the Open Compute Project (OCP) Summit 2026 in San Jose, California, the company’s new rack reference design is called Helion. This follows its existing Helios system, a naming choice that some may find overly similar.

Helion provides an OCP-compliant blueprint for agentic AI infrastructure anchored by AMD CPUs and accelerators.

AMD introduced Helios earlier this year targeting frontier-scale AI training and large-scale inference. Helion, conversely, merges agent execution and inference into a single rack for enterprise configurations.

Clients may select a CPU-only hyperscale configuration or an enterprise configuration featuring a mix of CPU and GPU blades.

Depending on the build, AMD advertises up to 32,768 “Zen 6” CPU cores delivered through compute nodes equipped with 256-core Venice EPYC processors, up to 16 Instinct MI350P GPU accelerators, and up to 102 Tbps of networking bandwidth.

The CPU-only hyperscale variant dedicates capacity to agent execution with links to separate model inference infrastructure, whereas the enterprise configuration integrates agent execution, GPU inference, and supporting workloads within a single rack.

All configurations support dense, balanced, or high-performance CPU blades. Dense deployments rely on EPYC 9006 chips rated at 400 W, performance blades feature a 96-core variant clocked at 5 GHz, and balanced blades utilize 256-core processors rated at 600 W.

The design also includes GPU inference blades with eight Instinct MI350P GPUs housed in a 5U server, alongside general-purpose blades intended for database, storage, or technical compute workloads.

AMD SVP for Compute and Enterprise AI, Dan McNamara, stated that agentic AI is not a single workload, spanning enterprise applications and complex workflows with varying compute requirements.

“To do that takes optimized compute and infrastructure that can adapt to different workloads. With AMD Helion, we are bringing Epyc performance, open software and flexible rack designs together so our partners can build systems around customer needs and make efficient use of available power,” he added.

AMD plans to contribute the Helion specifications to OCP, allowing the community to further develop and adopt them. Partner-built Helion configurations are scheduled for release in November 2026.

AMD cites Cisco, Dell, HPE, Foxconn, Lenovo, and Supermicro as partners on board with the new hardware.

Source link