AMD introduced Helion on 12 October 2026, describing it as an open blueprint for agentic AI infrastructure based on an Open Compute Project spec-design. In its press release, AMD says one example configuration, a CPU-only rack powered by 6th Gen AMD EPYC processors, can support up to 32,768 CPU cores.
AMD says Helion configurations are under development with ecosystem partners and are planned to be available in November 2026, and that it plans to contribute the Helion specification to OCP.
Business Pill · COMPLEMENTARY INPUTS
A one-minute explainer of complementary inputs: things that only work together. It teaches the general idea only and says nothing about any person or organisation in this story.
The key insight: As we read it, AMD is splitting agent work into two jobs and offering a rack for each: the release describes CPU-only capacity for agent execution, connected to separate model inference, and an all-in-one rack that puts CPUs and MI350P GPUs together.
The release gives two examples. The first is the CPU-only rack, which AMD says is designed for hyperscale customers, providing dedicated capacity for agent execution and connections to separate model inference infrastructure.
The second is an all-in-one rack that, in AMD’s words, combines AMD EPYC CPU nodes, AMD Instinct MI350P GPU nodes and general-purpose compute on one platform. AMD says it is designed to give partners and customers an all-in-one concept for enterprise agentic AI.
AMD’s product page puts the difference this way: hyperscalers deploy agents at very large scale, optimise for the number of agents per watt and already operate separate accelerated infrastructure, while enterprises typically want a single rack that delivers both agentic compute and acceleration.
AMD Helion as AMD describes it on 12 October 2026: one example configuration, a CPU-only rack of 6th Gen AMD EPYC processors, can support up to 32,768 CPU cores; the all-in-one rack can hold up to 16 Instinct MI350P GPUs in two eight-GPU servers; AMD plans availability in November 2026. These are AMD’s figures and plans.
What AMD Says Is in the Rack
According to the product page, the enterprise configuration uses AMD Instinct MI350P, eight GPUs in a 5U server, and two GPU servers can be added to a rack for sixteen GPUs total. AMD says the two nodes operate independently, with no scale-up GPU fabric between them, because the configuration is designed for enterprise agentic workloads and acceleration rather than frontier-scale training.
The page describes a 21-inch ORv3 44OU rack in direct-liquid-cooled and air-cooled versions. It lists the OCP standards the design follows: ORv3 for the rack and busbar, DC-MHS for the host processor module, DC-SCM for management and security, HPRv2 for power shelves, OCP NIC 3.0 for networking and Caliptra for the silicon root of trust.
The release also says Helion combines open hardware and software with an integrated token router that directs agents’ model calls across frontier and open-source models.
How It Differs From Helios
AMD says Helion builds on the open-platform approach of its Helios rack-scale solution. Its product page says the two serve different needs: Helios is built for frontier-scale AI training and large-scale inference with a unified scale-up GPU domain, while the Helion enterprise configuration is for organisations that want agentic infrastructure with inference in a single rack.
Dan McNamara, AMD’s senior vice president and general manager for Compute and Enterprise AI, said in the release: “Agentic AI is not a single workload. It spans personal agents, enterprise applications, code execution and complex workflows, each with different compute requirements.”
Who Is Building It
The release carries statements from eight companies: Cisco, Dell Technologies, Hon Hai Technology Group, HPE, Lenovo, QCT, Supermicro and Wiwynn. Dell names its PowerRack line in its statement, and Supermicro says it is bringing Helion to market through its Data Center Building Block Solutions.
HPE’s statement refers to its recent customer agreement to deploy AMD Helios. Rack-scale design was also the subject of Microsoft’s Maia 300 announcement the same day. Separately, AMD agreed on 28 September to acquire World Labs, the world-model start-up.
The Structural Read
As we read it, the CPU-only rack treats agents as a CPU workload: AMD says it provides dedicated capacity for agent execution, and its product page says hyperscalers optimise for the number of agents per watt.
As we read it, the integrated token router is the software half of the same idea: the release says it directs agents’ model calls across frontier and open-source models.
As we read it, the date is the thing to watch: AMD says configurations are still under development with partners and are planned to be available in November 2026.
Dan McNamara, AMD, in the 12 October 2026 release
“Agentic AI is not a single workload. It spans personal agents, enterprise applications, code execution and complex workflows, each with different compute requirements.”
Three Implications
HYPERSCALERS AMD’s product page says the CPU-only rack is designed for hyperscalers, neoclouds and sovereign AI operators deploying agents at large scale.
ENTERPRISES According to the release, the all-in-one rack combines EPYC CPU nodes, Instinct MI350P GPU nodes and general-purpose compute on one platform.
SERVER MAKERS The release carries statements from Cisco, Dell Technologies, Hon Hai Technology Group, HPE, Lenovo, QCT, Supermicro and Wiwynn.
The framework puts it this way: “The agentic surface — the harness — is becoming where value pools in the entire AI stack.”
As we read it, Helion is AMD designing hardware for that layer: AMD says the CPU-only rack is built for agent execution, with model inference kept on separate infrastructure.
The eight companies with statements in AMD’s 12 October 2026 Helion release. These are statements of support; the release names no customer orders for Helion.
What Is Not Established
AMD gives no price, no performance figures for agent workloads and no named customer for Helion itself. The 32,768-core figure describes one example configuration as AMD states it, and the November 2026 date is a plan, with the configurations still under development with partners. We have seen no independent test.
Business Engineer Framework
The Agentic Harness War
A Business Engineer framework on why the agent layer, not the model, is where value pools in the AI stack.
AMD has set out Helion as an open, OCP-based blueprint with two example racks: a CPU-only rack that AMD says can support up to 32,768 CPU cores for agent execution, and an all-in-one rack that adds MI350P GPUs. AMD says the configurations are planned to be available in November 2026.
A note on sourcing. On 12 October 2026 we read AMD’s announcement in full: its press release introducing Helion and its Helion product page. Every configuration and capability here is AMD’s own description, and November 2026 is AMD’s planned date; we have not seen independent tests. Nothing here is investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur, HackerNoon and Search Engine People. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
