#

Key takeaways

• AMD introduced Helion as an open blueprint for rack-scale agentic AI infrastructure.

• The design separates agent execution from model inference in some configurations.

• AMD gives two example rack approaches: CPU-only and all-in-one.

• The blueprint supports air

• and liquid-cooled setups and uses open hardware and software.

• AMD plans to contribute the specification to the Open Compute Project.

What AMD announced

AMD introduced Helion on October 12 as an open blueprint for rack-scale agentic AI infrastructure. The design is based on an Open Compute Project spec design. AMD frames the announcement around a clear distinction between agent execution and model inference. That matters because AI-agent workloads can need different mixes of CPU cores, GPU acceleration, networking, and power.

The company presents Helion as a blueprint, not as a finished customer deployment. It says partners may tailor CPU and GPU configurations to the customer's workload. Those are design options, not measured customer results. AMD also says the rack configurations are under development with ecosystem partners and are planned for availability in November 2026.

Two example rack approaches

AMD describes a CPU-only rack for hyperscale operators. This example uses sixth-generation EPYC processors. AMD says this setup can support up to 32,768 CPU cores. It can also connect to separate model-inference infrastructure. That makes the rack suitable for agent execution while inference runs elsewhere.

AMD also gives a second example for enterprise deployments. This all-in-one approach combines EPYC CPU nodes, Instinct MI350P GPU nodes, and general-purpose compute. The announcement does not say that one model is better than the other. It instead shows that Helion can support different workload patterns.

Token routing and infrastructure choices

AMD says Helion includes an integrated token router. The router can direct agents' model calls across frontier and open-source models. This suggests a design that can manage model selection within the rack architecture. The announcement does not provide performance data for the router.

The blueprint also supports air-cooled and liquid-cooled configurations. AMD says it uses open hardware and software. That points to a modular approach. It also suggests that system builders can adapt the rack to different operational needs, within the blueprint's design boundaries.

Partner ecosystem and Open Compute Project plans

AMD says it plans to contribute the Helion specification to the Open Compute Project. That is a formal step toward broader ecosystem use. The article also includes statements of support from Cisco, Dell, HPE, Lenovo, Supermicro, and other system partners. Those statements show ecosystem interest, but they do not establish customer deployments or production results.

The source material stays careful on what is and is not proven. It does not establish installed capacity. It does not report production performance. It also does not confirm availability in Morocco. The announcement is about a blueprint and partner development, not a market rollout report.

Operational considerations

The main operational point in the announcement is workload fit. AMD separates agent execution from model inference because the two tasks may need different infrastructure mixes. That means rack design can change depending on whether a deployment needs more CPU, more GPU, or more networking and power. The blueprint also leaves room for cooling choice, which can affect how a system is built and operated.

Another consideration is integration. The token router, the open hardware approach, and the partner-led design model all suggest a system that depends on coordination across components. The source does not claim any final deployment outcome. It only shows the intended architecture and the development path.

Morocco relevance

The source reports no Morocco-specific deployment, capacity, or availability. For readers, the global lesson is simple: agentic AI infrastructure may need separate planning for agent execution and model inference, plus flexible cooling and compute choices.

Bottom line

Helion is AMD's open rack blueprint for agentic AI infrastructure. It offers two example configurations, supports different cooling methods, and is being developed with partners. The announcement focuses on architecture and ecosystem planning, not on measured deployment results.