What AMD announced
AMD CEO Lisa Su announced Helion in her OCP Global Summit keynote in San Jose on October 12. The company said it builds on the "open-platform approach" of the Helios rack-scale solution. Helion is aimed at hyperscale customers and is powered by 6th Gen EPYC "Venice" server CPUs.
Tom's Hardware reports that Helion is a standardized, 21-inch rack reference design that complements the Helios rack AMD revealed at the OCP Global Summit last year.
Helion is a blueprint rather than a product AMD will sell. Tom's Hardware says AMD won't sell racks itself. It is contributing the design to OCP, and OEMs and ODMs will bring it to market. AMD's own release says it plans to contribute the Helion specification to OCP so the community can develop and adopt it further.
Two configurations
Helion comes in two flavors, according to the reports:
- CPU-only. A rack with 6th Gen EPYC processors supports up to 32,768 CPU cores. It is designed for hyperscale customers and provides dedicated capacity for agent execution. Tom's Hardware ties that figure to 128 EPYC 9996 CPUs at 256 cores each, and 128 × 256 does equal 32,768. That is a design maximum, not a benchmark.
- All-in-one. This rack combines EPYC CPU nodes, Instinct MI350P GPU nodes and general-purpose compute on one platform for enterprise agentic AI. Tom's Hardware adds that it supports up to 16 MI350P GPUs and 102 Tbps of networking, at the cost of CPU capacity. No source says what the 102 Tbps figure measures, such as aggregate or per-direction bandwidth, so treat it as a headline number.
Why CPU-heavy? Tom's Hardware's explanation is that personal AI agents each run in a sandbox, which gets a slice of host compute. CPU-only servers handle tasks like building software, while heavy inference runs on separate servers. That is the publication's interpretation of the design's role. AMD's quoted framing is that agentic AI spans personal agents, enterprise applications, code execution and complex workflows, each with different compute requirements.
Rack format and standards
Tom's Hardware says the rack is based on OCP's ORv4 specification and supports 44 OpenU trays. It also says the design is DC-MHS compliant and supports both air-cooled and direct liquid-cooled deployments. AMD's release also mentions liquid- and air-cooled versions.
One detail needs caution. Tom's Hardware describes the OpenU trays as 21 mm wide, which conflicts with its own description of a 21-inch rack. That is probably a typo. Confirm tray dimensions against the final OCP specification before planning around them.
OCP's own description helps with the standards. Its Rack & Power project covers both 19-inch EIA-compatible racks and 21-inch Open Rack systems. DC-MHS aims to provide consistent interfaces and form factors among modular building blocks. Standards alignment makes interoperability easier. It does not guarantee that every vendor's trays, power and cooling will drop into every other vendor's rack.
Who is building it
Tom's Hardware said AMD did not name partners in the press pre-briefing. AMD's release later filled that gap. Per Investing.com's coverage, supporting partners include Cisco, Dell Technologies, Hon Hai Technology Group, HPE, Lenovo, QCT, Supermicro and Wiwynn. AMD's release quotes Supermicro as bringing Helion to market as an open, ORv3-compliant rack-scale architecture through its Data Center Building Block Solutions. Customer deployments have not been named.
Conflicting availability dates
The sources disagree on timing:
- Tom's Hardware reports Helion will be available starting in the first half of 2027.
- AMD's release says that Helion agentic rack configurations are under development with ecosystem partners and are planned to be available in November 2026. Data Center Dynamics also reports a planned November launch date.
AMD's own statement is the stronger source, so November 2026 is the more likely target. Partners may still ship at different times, and "available" is not defined. Treat any date as a partner-by-partner question until OEMs publish orderable configurations.
Why Windows and enterprise IT readers might care
Most readers will not buy a 32,768-core rack. The release does matter for infrastructure planning in a few ways:
- Agent hosting is CPU-hungry. If agent sandboxes, code execution and build tasks drive demand, CPU capacity matters alongside GPUs. That affects capacity planning for any enterprise running agents or developer workloads at scale.
- Reuse of existing infrastructure. AMD says Helion enables customers to deploy agentic AI in the IT infrastructure they already have. That is vendor marketing, and it is untested. Check rack, power and cooling compatibility before assuming it applies to your facility.
- Open racks. An OCP-based design with many OEM partners could reduce lock-in. Whether the partner implementations stay interchangeable is the open question.
What's still missing
AMD and its partners have not published several details:
- Power and cooling requirements per rack.
- CPU and memory configurations.
- Pricing.
- Tested workload performance.
- Final tray dimensions.
Until those arrive, Helion is best read as a credible design direction with a long partner list. It is not yet a validated product.