AMD unveils Helion racks for AI agents at OCP summit

AMD is broadening its rack-scale AI lineup with a style to run AI representatives, providing CPU-only setups or incorporated CPU/GPU implementations, and liquid-cooled or air-cooled variations.

Exposed at the Open Compute Project (OCP) Summit 2026 in San Jose, California, the chipmaker’s brand-new rack recommendation style is called Helionwhile its existing system brings the name Helios, which we can’t assist sensation is welcoming confusion a little.

Helion uses an OCP-compliant plan for agentic AI facilities developed around AMD’s CPUs and accelerators.

AMD introduced Helios previously this year for frontier-scale AI training and massive reasoning. Helion’s business setup rather integrates representative execution and reasoning within a single rack.

Clients can go with a CPU-only hyperscale setup or a business setup with a mix of CPU and GPU blades.

Depending upon setup, AMD promotes as much as 32,768 of AMD’s “Zen 6” CPU cores, provided through calculate nodes fitted with 256-core variations of its Venice Epyc processor, as much as 16 of its Instinct MI350P GPU accelerators, and as much as 102 Tbps of networking bandwidth.

The CPU-only hyperscale variation provides committed capability for representative execution and connections to different design reasoning facilities, while the business setup integrates representative execution, GPU reasoning, and supporting work in one rack.

Both setups use thick, well balanced, or high-performance CPU blades. For thick releases, EPYC 9006 chips ranked at 400 W are utilized, while efficiency blades sport a variation of the chip with simply 96 cores however clocked at 5 GHz. Well balanced blades utilize 256-core processors ranked at 600 W.

There are likewise GPU reasoning blades, with 8 Instinct MI350P GPUs in a 5U server, plus general-purpose blades, planned for database and storage servers or technical calculate work.

AMD SVP for Compute and Enterprise AI Dan McNamara stated agentic AI is not a single work, covering business applications and complicated workflows, each with various calculate 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 included.

AMD states it prepares to contribute the Helion requirements to OCP, making them readily available to the neighborhood for more advancement and adoption. Partner-built Helion setups are set up for November 2026.

In its statement, AMD mentions Cisco, Dell, HPE, Foxconn, Lenovo, Supermicro, and others as on board with the brand-new package. ®