AMD has introduced the Helion agentic AI rack, describing it as an open blueprint for rack-scale AI infrastructure based on an Open Compute Project design.
The launch extends AMD's rack-scale AI strategy beyond its existing Helios platform and targets workloads tied to so-called agentic AI, in which software agents carry out tasks such as code execution, workflow management and enterprise automation.
Helion is designed to let customers choose between CPU-only systems and mixed CPU-GPU systems, depending on the AI work being run. The blueprint supports both liquid-cooled and air-cooled configurations and is intended to work with open software.
One version of the rack uses 6th Gen AMD EPYC processors and supports up to 32,768 CPU cores. That setup is aimed at hyperscale users seeking dedicated capacity for agent execution while linking to separate model inference systems.
Another version combines EPYC CPU nodes, Instinct MI350P GPU nodes and general-purpose compute in a single rack. AMD said the design is intended for enterprise users who want one platform for different agentic AI tasks.
At the centre of the pitch is AMD's argument that agentic AI will not run as a single standard workload. Instead, the company is positioning Helion as infrastructure that can be tuned for different mixes of processing, power use and cost, while fitting into existing IT environments.
Dan McNamara, Senior Vice President and General Manager, Compute and Enterprise AI, AMD, said the company sees a need for a broader set of infrastructure choices as AI deployments become more varied.
"Agentic AI is not a single workload. It spans personal agents, enterprise applications, code execution and complex workflows, each with different compute requirements," said Dan McNamara, Senior Vice President and General Manager, Compute and Enterprise AI, AMD.
"To do that takes optimized compute and infrastructure that can adapt to different workloads. With AMD Helion, we are bringing AMD 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," McNamara said.
Open design
AMD plans to contribute the Helion specification to the Open Compute Project, placing the architecture into a wider industry effort around shared data centre hardware designs. That mirrors its message that AI infrastructure should be developed through a more open hardware and software ecosystem rather than tightly controlled proprietary stacks.
Helion also includes an integrated token router intended to direct model calls from agents across frontier and open-source models. This points to a model in which customers may not rely on a single provider, but instead route workloads according to task, cost or other operational requirements.
Support statements from server makers and infrastructure suppliers suggest AMD is trying to build a broad manufacturing and integration network around the design. Comments came from Cisco, Dell, Hon Hai Technology Group, HPE, Lenovo, QCT, Supermicro and Wiwynn.
Jeremy Foster, Senior Vice President and General Manager, Cisco Compute, said: "Agentic AI is changing what customers need from their infrastructure. It's not just about compute performance, but how the entire system is connected, managed and secured. Architectures like AMD Helion, combined with Cisco's full-stack infrastructure capabilities, can give customers greater flexibility to deploy, operate and scale agentic workloads into their environments as their needs evolve."
Dell also framed the launch around customer choice and validation. "Dell PowerRack gives customers choice without compromise, and our support for AMD Helion Agentic Rack Reference Design extends that promise to the next generation of agentic AI. By combining AMD's rack-scale architecture with Dell's proven engineering expertise, along with the flexibility to extend deployments with Dell PowerRack for Networking and Storage, we're giving customers a complete, validated path to deploy agentic compute at scale with the manageability and reliability they expect from Dell," said Arun Narayanan, Senior Vice President, Compute and Networking, Dell Technologies.
Broader push
The announcement comes as AMD seeks to widen its role in AI infrastructure beyond selling individual processors and accelerators. In related remarks on its rack-scale portfolio, AMD said Helios is moving into production, with customer deliveries, deployments and workload validation under way, and early systems running at OpenAI, Meta, Microsoft and Oracle.
That matters because it shows AMD trying to compete at the level of complete systems design as AI buyers increasingly focus on power, cooling, integration and model orchestration alongside chip performance.
For hardware partners, Helion offers a reference design that can be adapted for different customer environments. For enterprise and hyperscale users, AMD is presenting it as a way to deploy agentic AI using existing infrastructure investments while adjusting the mix of CPUs and GPUs to suit particular tasks.
"AMD Helion provides an open, flexible foundation for agentic AI infrastructure. Wiwynn is excited to collaborate with AMD, bringing our expertise in rack-level integration and advanced cooling to help hyperscaler customers tailor systems that meet the demanding workloads, power requirements, and data center environments of the agentic AI era," said William Lin, President and CEO, Wiwynn.