AMD Helios

AMD Helios is a rack-scale artificial intelligence system from AMD that packages a full rack of 72 Instinct MI400 series GPUs, AMD EPYC "Venice" CPUs, and Pensando "Vulcano" network cards so the whole rack behaves like one large accelerator rather than a row of separate servers. Built on the Open Rack Wide (ORW) double-wide standard that Meta contributed to the Open Compute Project (OCP), Helios is AMD's most direct answer to NVIDIA's rack-scale machines such as the GB200 NVL72 and the forthcoming Vera Rubin. AMD says a single Helios rack delivers up to roughly 3 AI exaflops of compute (2.9 exaFLOPS FP4 and 1.4 exaFLOPS FP8) with 31 TB of HBM4 memory, and the platform is on track to ship in the second half of 2026. ^[1][2][3][4]

AMD first previewed Helios at its Advancing AI event on June 12, 2025, showed a physical reference design at the Open Compute Project Global Summit on October 14, 2025, and put working Helios hardware on public display for the first time at CES 2026 in January 2026, where CEO Lisa Su called it "the blueprint for yotta-scale infrastructure." ^[1][3][4][5] The design connects its accelerators using open industry standards rather than a single vendor's proprietary fabric, and it marks the point where AMD stopped selling AI mostly as chips and boards and started selling it as a full rack you wheel onto a data center floor.

From chips to rack-scale AI factories

For most of the last decade, an AI accelerator was a card or a board. You bought GPUs, put eight of them in a server, and wired many servers together with whatever network you had. That model started to strain as frontier models grew. Training and serving the largest models now spreads a single job across dozens or hundreds of GPUs that have to act like one tightly coupled system, and the limit is often the links between chips rather than the chips themselves. ^[6][7]

The industry's response is the rack-scale system, sometimes marketed as an "AI factory." Instead of treating the server as the unit of sale, vendors treat the whole rack as the product. Every GPU in the rack shares a fast internal fabric, the CPUs and network cards are co-designed with the accelerators, and power and liquid cooling are engineered for the rack as a whole. NVIDIA moved first here with its GB200 NVL72 and the later GB300 NVL72, which link 72 GPUs over NVLink so they look like a single accelerator to software. Helios is AMD's version of that idea, and the company has been explicit that it now competes at the rack level, not just chip against chip. ^[6][8]

This shift changes what buyers compare. A faster GPU helps, but a model trainer cares about how much memory and compute a rack can pool, how fast the GPUs talk to each other inside the rack, and how cleanly many racks scale out into a cluster. Helios is built to be judged on those terms.

What is AMD Helios?

Helios is AMD's first rack-scale AI platform: a complete, liquid-cooled rack that combines AMD Instinct MI400 series GPUs, sixth-generation EPYC "Venice" CPUs, and Pensando "Vulcano" AI NICs into a single coherent system, all unified by AMD's open ROCm software stack. ^[4] The reference rack holds 72 GPUs and is a double-wide chassis weighing close to 7,000 pounds, with the accelerators wired together so software can treat the rack as one big pool of compute and memory. ^[3][9] AMD describes the platform as extending its open hardware philosophy "from silicon to system to rack." ^[3]

At CES 2026 AMD said each Helios rack carries more than 18,000 CDNA5 GPU compute units and more than 4,600 Zen 6 CPU cores, and delivers up to about 3 AI exaflops of performance in a single rack. ^[9][10] The accelerator at the heart of the flagship Helios configuration is the Instinct MI455X, the inference-and-training member of the MI400 family aimed at low-precision FP4, FP8, and BF16 work. ^[10][11]

What hardware is in Helios?

Helios brings together three AMD product lines that were each refreshed for this generation.

The accelerators are the MI400 series. AMD detailed the full family at CES 2026: the MI455X for flagship rack-scale AI, the MI440X, the MI450 (the base part the Helios rack is most often described around), and the MI430X, a variant that fully supports FP32 and FP64 for sovereign AI and traditional supercomputing. ^[10][11] AMD says each MI450 series GPU carries up to 432 GB of HBM4 memory with 19.6 TB/s of memory bandwidth, and that it delivers up to 40 PFLOPS of FP4 compute and 20 PFLOPS of FP8 compute. ^[1][2][12] Those memory figures are large by current standards, and AMD has leaned on capacity as a selling point because more memory per GPU lets a given model fit on fewer GPUs.

The host processors are EPYC "Venice" CPUs, AMD's sixth-generation server chips based on the Zen 6 core, with PCIe Gen6 and high core counts, fabricated on TSMC's 2 nm process. They handle the parts of a workload that do not suit the GPU, feed data to the accelerators, and run the rack's control software. ^[1][13] Venice is itself a 2026 product, so Helios lines up the GPU and CPU roadmaps in the same window.

The network cards are the Pensando "Vulcano" AI NICs, rated at 800 Gb/s, which connect each node out to the rest of the cluster. ^[1][6] Pensando is the data-center networking group AMD acquired, and Vulcano carries traffic between racks rather than inside a single rack.

Put together at rack scale, AMD's figures for a Helios rack of 72 MI450 GPUs are 31 TB of total HBM4 memory, 1.4 PB/s of aggregate memory bandwidth, 2.9 EFLOPS of FP4 compute, and 1.4 EFLOPS of FP8 compute, with 260 TB/s of scale-up bandwidth among the GPUs and 43 TB/s of Ethernet scale-out bandwidth. ^[3][6] All of those are vendor numbers tied to hardware that was not shipping when AMD announced it, so they are targets rather than measured results.

Why does Helios bet on open standards?

The sharpest difference between Helios and NVIDIA's racks is not the silicon. It is the wiring philosophy. NVIDIA's NVL72 systems use NVLink and NVSwitch, interconnects that NVIDIA controls end to end. That gives NVIDIA a tightly integrated product, and it also gives NVIDIA leverage, because a customer buying into NVLink is buying into one supplier for the fabric. ^[6][8]

AMD took the opposite route and built Helios on open standards across two different jobs.

For scale-up, meaning the fast links among GPUs inside a rack, AMD is backing UALink, the Ultra Accelerator Link standard developed by a consortium that includes AMD, Broadcom, and others as an open alternative to NVLink. The aim is that any vendor can build UALink-capable accelerators and switches that interoperate. ^[1][6] In practice the first Helios generation runs this scale-up traffic as UALink over Ethernet, layering AMD's own Infinity Fabric protocol on top of Ethernet while native UALink hardware matures. ^[6]

For scale-out, meaning the network that joins many racks into a cluster, AMD uses Ultra Ethernet, an effort under the Ultra Ethernet Consortium to tune standard Ethernet for AI traffic. The Pensando Vulcano NICs are the on-ramp to that fabric. ^[1][3] Using Ethernet here lets operators reuse familiar tooling and a broad supplier base instead of a proprietary network.

The rack itself is the most pointed part of the strategy, and it is worth being precise about who did what. The chassis follows the Open Rack Wide specification, a double-wide design (rack frames 47.25 inches wide and 94 inches tall) that Meta introduced and contributed to the Open Compute Project on October 10, 2025, for the power, cooling, and serviceability needs of next-generation AI systems. AMD did not contribute the rack standard. It built Helios on Meta's open design and aligned it with other open compute standards including OCP DC-MHS, UALink, and Ultra Ethernet. ^[3][14][15] The rack adds quick-disconnect liquid cooling, the double-wide layout for easier servicing, and standards-based Ethernet for multi-path resiliency. ^[3][16] The point of all this is to let other vendors and cloud operators build and service compatible systems rather than depending on one supplier for every part. As AMD data-center chief Forrest Norrod put it, "Open collaboration is key to scaling AI efficiently. With 'Helios,' we're turning open standards into real, deployable systems, combining AMD Instinct GPUs, EPYC CPUs, and open fabrics to give the industry a flexible, high-performance platform built for the next generation of AI workloads." ^[3]

How does Helios compare to NVIDIA's rack systems?

AMD framed Helios against two NVIDIA generations. The near-term target is the GB200 and GB300 NVL72, which also place 72 GPUs in a rack over NVLink. The forward-looking comparison is NVIDIA's Vera Rubin platform, expected in a similar 2026 window and described by NVIDIA in larger rack configurations such as an NVL144. ^[6][8] AMD said a Helios rack offers 50% more memory capacity than NVIDIA's Vera Rubin system, pointing to 432 GB of HBM4 per GPU on Helios versus 288 GB on Rubin, and it claimed up to 36 times higher performance than its own previous generation. ^[3][8][17] AMD argues the larger per-GPU memory lets Helios serve roughly 50 percent larger mixture-of-experts models on a single double-wide rack. ^[17]

Those comparisons deserve caution. They come from AMD, they were made before either company's 2026 systems were on the market, and rack-level numbers depend heavily on how each vendor counts memory, which number format is quoted, and what the real software stack achieves. Independent benchmarks of shipping Helios racks against shipping NVIDIA racks did not exist at announcement, so the honest summary is that AMD claimed parity or better on paper and will have to prove it in deployment.

Scroll sideways for more →

Why does Helios matter for AMD?

Helios is the centerpiece of AMD's attempt to win real share in data-center AI, a market NVIDIA has dominated. Selling racks rather than chips changes AMD's position in two ways. It raises the value of each deal, since a rack bundles GPUs, CPUs, NICs, and integration work. It also makes AMD a credible single-vendor option for an operator that wants a turnkey AI cluster, which until now usually meant going to NVIDIA. ^[6][8]

The early commercial signals were strong. On October 6, 2025, AMD and OpenAI announced a partnership for OpenAI to deploy up to 6 gigawatts of AMD Instinct GPUs over several years, starting with 1 gigawatt of MI450 in the second half of 2026, alongside a warrant that could give OpenAI up to 160 million AMD shares. ^[18] About a week later, Oracle said it would deploy 50,000 AMD MI450 GPUs starting in the third quarter of 2026, a build that AMD watchers pegged at roughly 700 Helios racks and around 200 megawatts of power. ^[6] Then on February 24, 2026, AMD and Meta announced an expanded partnership to deploy 6 gigawatts of AMD GPUs across multiple Instinct generations, with the first deployment using a custom MI450-architecture GPU on the Helios rack-scale architecture and shipments starting in the second half of 2026, again carrying performance-based warrants for up to 160 million AMD shares. ^[19] AMD also picked up systems partners: Hewlett Packard Enterprise said it would build 2026 AI systems on the Helios rack architecture. ^[20] Commitments at that scale, from buyers who also lean heavily on NVIDIA, suggest that large customers want a second credible supplier and view Helios as one. The open-standards approach reinforces that, because UALink, Ultra Ethernet, and a Meta-derived OCP rack lower the cost of running AMD gear next to everything else in the building. As Lisa Su told the CES 2026 audience, "As AI adoption accelerates, we are entering the era of yotta-scale computing, driven by unprecedented growth in both training and inference." ^[4]

There are real limits to keep in mind. As of early 2026, Helios was a roadmap and a reference design with hardware on show but not yet in volume, and the headline numbers are AMD's own projections for unshipped systems. The MI450 and MI455X GPUs, EPYC Venice CPUs, and a production UALink fabric all have to arrive on schedule (AMD reiterated a 2H 2026 timeline in February 2026) and work together at full rack scale. ^[21] AMD's software stack, ROCm, has historically trailed NVIDIA's CUDA in maturity and breadth, and rack-scale performance depends as much on that software as on the silicon. The open interconnects are also still early, so the interoperability promise is partly aspirational until several vendors ship compatible parts. The MI430X variant is already lined up for HPC duty in Oak Ridge National Laboratory's "Discovery" system and France's "Alice Recoque" exascale supercomputer, and AMD has flagged an MI500 successor for 2027. ^[10] Helios is a clear statement of direction for AMD's AI data center strategy, and its real standing against NVIDIA will be settled by 2026 deployments rather than by launch-day slides.

See also

• AMD Instinct MI400
• AMD Instinct MI355X
• AMD EPYC Venice
• NVIDIA GB200 NVL72
• NVIDIA Vera Rubin
• UALink
• Ultra Ethernet

References

1. Michael Larabel. "AMD Previews Instinct MI400 Series & Helios AI Rack." Phoronix, June 12, 2025. https://www.phoronix.com/news/AMD-Instinct-MI400-Preview ↩
2. AMD. "AMD Unveils Vision for an Open AI Ecosystem, Detailing New Silicon, Software and Systems to Power the Next Era of AI." AMD newsroom, June 12, 2025. https://www.amd.com/en/newsroom/press-releases/2025-6-12-amd-unveils-vision-for-an-open-ai-ecosystem-detai.html ↩
3. AMD. "AMD Showcases 'Helios' Rack-Scale Platform Built on the Open Compute Project Open Rack for AI, Introduced by Meta." AMD newsroom, October 14, 2025. https://www.amd.com/en/newsroom/press-releases/2025-10-14-amd-showcases-helios-rack-scale-platform-built-o.html ↩
4. AMD. "AMD and its Partners Share their Vision for 'AI Everywhere, for Everyone' at CES 2026." AMD newsroom, January 5, 2026. https://www.amd.com/en/newsroom/press-releases/2026-1-5-amd-and-its-partners-share-their-vision-for-ai-ev.html ↩
5. The Next Platform. "AMD Contemplates And Engineers Yottascale AI Compute." January 6, 2026. https://www.nextplatform.com/2026/01/06/amd-contemplates-and-engineers-yottascale-ai-compute/ ↩
6. Timothy Prickett Morgan. "Oracle First In Line For AMD 'Altair' MI450 GPUs, 'Helios' Racks." The Next Platform, October 14, 2025. https://www.nextplatform.com/2025/10/14/oracle-first-in-line-for-amd-altair-mi450-gpus-helios-racks/ ↩
7. The Register. "AMD taking AI fight to Nvidia with Helios rack-scale system." November 5, 2025. https://www.theregister.com/special-features/2025/11/05/amd-taking-ai-fight-to-nvidia-with-helios-rack-scale-system/1208044 ↩
8. Anton Shilov. "AMD debuts Helios rack-scale AI hardware platform at OCP Global Summit 2025, promises easier serviceability and 50% more memory than Nvidia's Vera Rubin." Tom's Hardware, October 14, 2025. https://www.tomshardware.com/tech-industry/amd-debuts-helios-rack-scale-ai-hardware-platform-at-ocp-global-summit-2025-promises-easier-serviceability-and-50-percent-more-memory-than-nvidias-vera-rubin ↩
9. Techloy. "Everything AMD Announced at CES 2026: Helios Racks, MI455X GPUs, and Ryzen AI 400 Chips." January 2026. https://www.techloy.com/everything-amd-announced-at-ces-2026-helios-racks-mi455x-gpus-and-ryzen-ai-400-chips/ ↩
10. Anton Shilov. "AMD touts Instinct MI430X, MI440X, and MI455X AI accelerators and Helios rack-scale AI architecture at CES." Tom's Hardware, January 2026. https://www.tomshardware.com/tech-industry/artificial-intelligence/amd-touts-instinct-mi430x-mi440x-and-mi455x-ai-accelerators-and-helios-rack-scale-ai-architecture-at-ces-full-mi400-series-family-fulfills-a-broad-range-of-infrastructure-and-customer-requirements ↩
11. ServeTheHome. "AMD's EPYC Venice, Instinct MI455X, & Helios Hardware On Display for First Time at CES 2026." January 2026. https://www.servethehome.com/amds-epyc-venice-instinct-mi455x-helios-hardware-on-display-for-first-time-at-ces-2026/ ↩
12. Hassan Mujtaba. "AMD's Next-Gen Instinct MI400 Accelerator Doubles The Compute To 40 PFLOPs, Equipped With 432 GB HBM4 Memory at 19.6 TB/s and Launches In 2026." Wccftech, June 12, 2025. https://wccftech.com/amd-instinct-mi400-accelerator-doubles-compute-40-pflops-432-gb-hbm4-memory-2026-launch/ ↩
13. Wayne Williams. "Helios rack design to support MI400 GPU and PCIe Gen6 EPYC Venice chips." TechRadar Pro, 2025. https://www.techradar.com/pro/amd-gets-ready-for-nvidias-vera-rubin-and-2026-with-432gb-mi400-gpu-monster-paired-with-256-core-epyc-venice-and-i-cant-wait-to-see-the-sparks-fly ↩
14. InsideHPC. "AMD Showcases 'Helios' Rack-Scale Platform on the OCP Open Rack Wide Spec." October 2025. https://insidehpc.com/2025/10/amd-showcases-helios-rack-scale-platform-on-the-ocp-open-rack-wide-spec/ ↩
15. AMD. "AMD 'Helios': Advancing Openness in AI Infrastructure Built on Meta's 2025 OCP Open Rack for AI Design." AMD blog, October 2025. https://www.amd.com/en/blogs/2025/amd-helios-ai-rack-built-on-metas-2025-ocp-design.html ↩
16. DataCenterDynamics. "AMD launches Instinct MI350 GPUs, unveils double-wide Helios AI rack-scale system." June 2025. https://www.datacenterdynamics.com/en/news/amd-launches-instinct-mi350-gpus-unveils-double-wide-helios-ai-rack-scale-system/ ↩
17. Futurum Group. "At CES, NVIDIA Rubin and AMD 'Helios' Made Memory the Future of AI." January 2026. https://futurumgroup.com/insights/at-ces-nvidia-rubin-and-amd-helios-made-memory-the-future-of-ai/ ↩
18. AMD. "AMD and OpenAI Announce Strategic Partnership to Deploy 6 Gigawatts of AMD GPUs." AMD newsroom, October 6, 2025. https://www.amd.com/en/newsroom/press-releases/2025-10-6-amd-and-openai-announce-strategic-partnership-to-d.html ↩
19. AMD. "AMD and Meta Announce Expanded Strategic Partnership to Deploy 6 Gigawatts of AMD GPUs." AMD newsroom, February 24, 2026. https://www.amd.com/en/newsroom/press-releases/2026-2-24-amd-and-meta-announce-expanded-strategic-partnersh.html ↩
20. Anton Shilov. "HPE adopts AMD's Helios rack architecture for 2026 AI systems." Tom's Hardware, 2026. https://www.tomshardware.com/tech-industry/semiconductors/hpe-adopts-amd-helios-rack-architecture-for-2026-ai-systems ↩
21. Timothy Prickett Morgan. "AMD Says 'Helios' Racks And MI400 Series GPUs On Track For 2H 2026." The Next Platform, February 23, 2026. https://www.nextplatform.com/compute/2026/02/23/amd-says-helios-racks-and-mi400-series-gpus-on-track-for-2h-2026/ ↩