AMD introduced Helion, an open rack blueprint for agentic AI infrastructure based on an Open Compute Project spec design. Helion offers CPU-only racks of up to 32,768 6th Gen EPYC “Venice” cores for hyperscale agent execution and all-in-one CPU plus MI350P racks for enterprises, with an integrated token router directing agents’ model calls. Futurum’s read is that Helion is the first rack-scale product to treat agent execution as its own infrastructure tier, the layer AMD sized but had not yet productized at Advancing AI 2026.
What Is Covered in This Article:
- The AMD Helion agentic AI rack introduced at the 2026 OCP Global Summit
- CPU-only hyperscale configurations with up to 32,768 Zen 6 “Venice” cores per rack
- All-in-one enterprise configurations pairing EPYC nodes with up to 16 Instinct MI350P GPUs at 74 petaflops FP4
- An integrated token router directing agent model calls across frontier and open-source models
- AMD’s plan to contribute the Helion specification to OCP
- Helion positioned against NVIDIA Vera racks, Arm AGI, and Intel Xeon 6+
The News: AMD (NASDAQ: AMD) introduced on October 12 the AMD Helion agentic AI rack, an open blueprint for agentic AI infrastructure based on an Open Compute Project spec design. Building on the open-platform approach of the AMD Helios rack-scale solution, Helion lets customers match CPU performance, core density, and power with GPU acceleration for the work their AI agents perform.
The blueprint supports liquid-cooled and air-cooled versions in an ORv3 44OU rack with standard modular compute boards, both native OCP spec compliant. A CPU-only configuration powered by 6th Gen AMD EPYC processors supports up to 32,768 cores for hyperscale agent execution connected to separate inference infrastructure; an all-in-one enterprise configuration combines EPYC nodes, up to 16 Instinct MI350P GPUs delivering 74 petaflops of FP4 compute, and general-purpose compute, with up to 102 Tbps of networking bandwidth per rack. Helion also includes an integrated token router that directs agents’ model calls across frontier and open-source models.
“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. “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.”
The launch arrives with ecosystem endorsements from Cisco, Dell Technologies, Hon Hai Technology Group, HPE, Lenovo, Supermicro, and Wiwynn, with Dell extending its PowerRack line to a Helion Agentic Rack Reference Design and HPE connecting Helion to its recent customer agreement to deploy AMD Helios. AMD plans to contribute the Helion specification to OCP. Configurations are under development with ecosystem partners and are planned to be available in November 2026.
AMD Helion Makes the CPU Rack a First-Class Citizen of the AI Data Center
Analyst Take: Helion is the piece of AMD’s agentic data center that was missing five weeks ago. Futurum wrote after Advancing AI 2026 that AMD had sized the agentic opportunity, raising its 2030 server CPU forecast above $200 billion from roughly $60 billion a year earlier, without showing how CPU and GPU workloads would be orchestrated across a rack. Helion is that answer in hardware. Every rack-scale AI product of the past three years placed the GPU at the center. Helion is the first rack a major silicon vendor has sold for AI in which the GPU is a configurable option, and in the hyperscale configuration absent entirely. CPU:GPU rebalancing now has a procurable unit with agentic capacity sized in cores and watts rather than accelerators.
Helion Productizes the Agent Sandbox Tier
At the AI Infra Summit in September, Forrest Norrod, executive vice president and general manager of AMD’s Data Center Solutions business, described agentic inference as an orchestration problem that has outgrown the GPU. Helion turns that sandbox concept into named products with a composable mix of dense “Venice” 256-core nodes at 400W for personal agents, 256-core balanced nodes at 600W for enterprise agents, and 96-core high-frequency nodes at 5.0 GHz for technical compute agents. The taxonomy gives hyperscalers a capacity planning language for a workload they currently assign to general purpose fleets. The qualifier is that AMD’s agents-per-watt comparisons, published in its July EPYC materials, are arithmetic from thread counts and power ratings under a theoretical workload. Agent throughput lacks a neutral benchmark yet, challenging server partners to deliver a standard product across multiple customers.
Two-Sided Silicon Leadership Makes AMD the Natural Vendor for a Two-Sided Rack
A rack that disaggregates agent execution from inference rewards the vendor with credible silicon on both sides, and AMD is the only merchant vendor claiming leadership on both: Venice at 1.8x the tokens per second of the best competitive x86 processor, 20% higher performance per core than NVIDIA’s 88-core Vera, and the MI350P, the 144 GB HBM3E PCIe card with leading enterprise MLPerf 6.1 results, as the GPU node. Helion also answers our view that AMD had to confront Intel’s 288-core Xeon 6+ rack head-on rather than benchmarking the 128-core Xeon 6980P, and had to ship Venice on TSMC 2nm to push the standard-rack ceiling past 36,000 cores. Helion’s 32,768-core ceiling lands short of that bar, and AMD’s published comparisons still anchor on the 6980P, the comparison that invites exactly the rebuttal Intel will make. What Helion supplies is deployability with the most agentic capacity per rack on hardware you can buy today, pending the Venice 2nm parts that would complete the argument on tokens per dollar per watt, the metric Futurum expects to dominate infrastructure decisions.
The Token Router and the OCP Contribution Convert the Helios Playbook Into a CPU Land Grab
The integrated token router answers the orchestration gap Futurum identified at Advancing AI, placing AMD-controlled software at the junction where inference spend is allocated. Calls steered toward self-hosted open models run on AMD GPUs, while frontier API calls leave the rack. The release offers no detail on routing policies, latency overhead, or governance, and hyperscalers prefer their own control planes, so the router reads as a statement of intent until AMD discloses architecture and adopters. The OCP contribution is the distribution strategy. Norrod noted Helios incorporates a couple of dozen OCP standards and that close derivatives are entering production from vendors beyond AMD. An open agentic rack spec invites the same proliferation, and the prize is the delta in AMD’s own forecast, since a server CPU market tripling by 2030 implies agent execution supplies most of the new dollars. An OCP spec can be populated with any compliant silicon over time, and AMD is betting portfolio breadth and a two-year head start keep EPYC the default fill while November availability puts Helion into 2027 budget cycles.
What to Watch:
- Whether named hyperscalers or neoclouds commit to CPU-only Helion deployments at or after the OCP Global Summit
- Whether Dell, HPE, Lenovo, Supermicro, and Cisco ship Helion configurations in the November window
- Whether AMD’s 14 GW of announced compute commitments extend to Helion agent capacity
- Whether AMD discloses token router architecture, model coverage, and latency overhead
- Whether a neutral body establishes an agent throughput benchmark that tests AMD’s agents-per-watt arithmetic
See the complete details about AMD Helion on the company website.
Sources
1. Introducing AMD Helion, Agentic AI Compute at Rack Scale, AMD
Declaration of generative AI and AI-assisted technologies in the writing process: This content has been generated with the support of artificial intelligence technologies. Due to the fast pace of content creation and the continuous evolution of data and information, The Futurum Group and its analysts strive to ensure the accuracy and factual integrity of the information presented. However, the opinions and interpretations expressed in this content reflect those of the individual author/analyst. The Futurum Group makes no guarantees regarding the completeness, accuracy, or reliability of any information contained herein. Readers are encouraged to verify facts independently and consult relevant sources for further clarification.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Read the full Futurum Group Disclosure.
Other Insights From Futurum:
Cerebras CS-4 Makes the Rack the New Chip by Doubling Power and Tripling Wafers
AMD Acquires Taalas to Advance AI Workload Optimization
AMD Q2 FY 2026: EPYC and Helios Fuel the Next AI Growth Phase
Author Information
Brendan is Research Director, Semiconductors, Supply Chain, and Emerging Tech. He advises clients on strategic initiatives and leads the Futurum Semiconductors Practice. He is an experienced tech industry analyst who has guided tech leaders in identifying market opportunities spanning edge processors, generative AI applications, and hyperscale data centers.
Before joining Futurum, Brendan consulted with global AI leaders and served as a Senior Analyst in Emerging Technology Research at PitchBook. At PitchBook, he developed market intelligence tools for AI, highlighted by one of the industry’s most comprehensive AI semiconductor market landscapes encompassing both public and private companies. He has advised Fortune 100 tech giants, growth-stage innovators, global investors, and leading market research firms. Before PitchBook, he led research teams in tech investment banking and market research.
Brendan is based in Seattle, Washington. He has a Bachelor of Arts Degree from Amherst College.
