Introducing AMD Helion, Agentic AI Compute at Rack Scale

What AMD announced

AMD introduced AMD Helion, described as an agentic AI rack and an open blueprint based on an Open Compute Project spec-design and powered by AMD compute. AMD positions Helion as a flexible, rack-scale approach that lets partners and customers configure CPU performance, core density, and power alongside GPU acceleration for agentic AI workloads. The release states AMD plans to contribute the Helion specification to OCP so the community can adopt and extend the design. These descriptions and intentions are presented as AMD statements in the company press release.

Sources: S1

Configurations and target use cases

AMD provided examples of configurable rack types. One example is a CPU-only rack using 6th Gen AMD EPYC processors described as supporting up to 32,768 CPU cores and aimed at hyperscale customers to provide dedicated agent execution capacity with separate model inference infrastructure. Another example is an all-in-one rack combining AMD EPYC CPU nodes and AMD Instinct MI350P GPU nodes, framed as an enterprise-focused option that consolidates agent execution and inference on a single platform. AMD emphasized that Helion can be delivered in liquid- and air-cooled versions and that partners can tailor CPU and GPU allocations to match agent execution, model inference, and supporting compute to deployment requirements.

Sources: S1

Ecosystem and partner statements

AMD named multiple partners and included partner quotations endorsing the open, rack-scale approach and expressing support for integrating Helion into broader infrastructure offerings. The press release lists Cisco, Dell Technologies, Hon Hai Technology Group, HPE, Lenovo, QCT, Supermicro, and Wiwynn among contributors or supporters and notes Helion configurations are under development with ecosystem partners.

Sources: S1

Availability and forward-looking language

AMD states that AMD Helion agentic rack configurations are currently under development with ecosystem partners and are planned to be available in November 2026. The release contains a cautionary statement that forward-looking statements about features, functionality, performance, availability, timing and expected benefits are subject to risks and uncertainties and that actual results could differ materially.

Sources: S1

What remains uncertain

• All technical descriptions, configurations, timelines and partner statements are reported from AMD's press release and partner quotes; the press release itself is an issuer source and not independent verification.
• Planned availability in November 2026 is a forward-looking statement and may be affected by risks and uncertainties listed in the press release, including supply chain, manufacturing, and regulatory factors.