AnswersPublished26 sources
AMD Helion Adds CPU-First Racks to Its AI Infrastructure Lineup
AMD introduced Helion on October 12, 2026, as an open rack blueprint for agentic AI: its CPU only hyperscale configuration is reported to scale to 32,768 cores, while enterprise racks combine CPUs and Instinct GPUs. AMD says Helion is based on an OCP specification and plans to contribute the design to OCP; reports d...
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- AMD introduced Helion on October 12, 2026, as an open rack blueprint for agentic AI: its CPU only hyperscale configuration is reported to scale to 32,768 cores, while enterprise racks combine CPUs and Instinct GPUs.
- AMD says Helion is based on an OCP specification and plans to contribute the design to OCP; reports differ on the rack generation label and on when partner systems will be available.
- Helios pairs 72 Instinct MI455X GPUs with EPYC CPUs and Pensando networking, with AMD rating the rack at up to 2.9 exaflops of FP4 compute.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did AMD announce on October 12, 2026, about Helion, its open rack-scale reference design for agentic AI, including its CPU-only hypersc. Article summary: AMD announced Helion on October 12, 2026, as an open rack-scale blueprint for agentic AI, offering CPU-only hyperscale racks and integrated CPU/GPU enterprise racks—not a replacement for its GPU-heavy Helios platform. [3. Topic tags: general. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake numbers, clickbait thumbna
AMD introduced Helion on October 12, 2026, as an open rack-scale blueprint for agentic AI infrastructure. It gives hyperscalers a CPU-only option and enterprises an integrated CPU/GPU option, positioning CPU capacity for agent execution alongside GPUs used for model computation. Helion extends AMD’s rack-scale lineup; it does not replace the more GPU-focused Helios design. 8
9
17
What AMD announced with Helion
Helion is a reference design rather than one fixed server configuration. AMD says customers can tailor CPU core density, performance and power to their workloads, with CPU-only configurations for hyperscale deployments and integrated CPU/GPU systems for enterprises. 8
The largest reported CPU-only configuration supports up to 128 sixth-generation EPYC 9996 “Venice” processors, for a total of 32,768 Zen 6 cores. That figure describes a maximum configuration, not every Helion rack. 3
4
For enterprise deployments, AMD describes a combined CPU-and-GPU rack using EPYC processors and Instinct accelerators. Reports identify different Instinct models, so the available material does not establish one definitive accelerator SKU for all configurations. 3
13
15
OCP basis—and a rack-standard caveat
AMD describes Helion as an open blueprint based on an Open Compute Project (OCP) specification. AMD also says it plans to contribute the Helion specification to OCP, which is different from saying that contribution is already complete. 8
49
The physical rack details are less consistent across reports: one identifies a 21-inch Open Rack v4 foundation, while another describes ORv3. The provided reporting therefore does not settle which rack-generation label is correct. 1
3
Why add CPU capacity for agents?
AMD’s strategy is to add CPU-server capacity alongside GPU infrastructure as agentic workloads grow. 17 In this model, CPUs can handle parts of agent execution—such as coordinating tasks and interacting with tools—while GPUs continue to serve workloads such as model inference and training. This is a division of infrastructure roles, not a claim that all planning or reasoning happens on CPUs; model-based reasoning can still require GPU computation.
The practical distinction is clearest in the two Helion configurations: hyperscalers can scale CPU execution separately from GPU infrastructure, while enterprises can put CPU execution and GPU inference in the same rack. 6
9
Helion versus Helios
Helios is AMD’s more GPU-intensive rack-scale design. AMD specifies 72 Instinct MI455X GPUs, EPYC “Venice” CPUs and Pensando Vulcano networking. Its stated peak FP4 performance is 2.9 exaflops; that is a rated compute figure, not a promise of application throughput. 33
Helion serves a complementary role: it gives AMD and its partners a blueprint for scaling CPU-heavy agent execution, either separately from GPU infrastructure or in an integrated enterprise rack. Both designs are reference platforms for system builders rather than finished racks sold directly by AMD. 33
AMD said Helios shipments to customers, including Microsoft, would begin in the second half of 2026. Microsoft’s expanded partnership with AMD includes planned Helios deployments on Azure as well as new Azure instances powered by EPYC Venice processors. 38
40
Availability remains unclear across reports
The available reports give different Helion timing signals. Some describe partner systems or initial configurations as expected in November 2026, while AMD’s social post says the new agentic AI rack is planned for the first half of 2027. The supplied information does not establish a definitive partner-by-partner shipment schedule. 10
11
12
15
The broader strategy is a portfolio of CPUs, GPUs, networking and rack-scale designs: Helion emphasizes agent execution, while Helios concentrates on GPU-intensive AI computation. The key caveat is that rack standards and availability details remain inconsistent across the reporting.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "AMD Helion Adds CPU-First Racks to Its AI Infrastructure Lineup"?
AMD introduced Helion on October 12, 2026, as an open rack blueprint for agentic AI: its CPU only hyperscale configuration is reported to scale to 32,768 cores, while enterprise racks combine CPUs and Instinct GPUs.
What are the key points to validate first?
AMD introduced Helion on October 12, 2026, as an open rack blueprint for agentic AI: its CPU only hyperscale configuration is reported to scale to 32,768 cores, while enterprise racks combine CPUs and Instinct GPUs. AMD says Helion is based on an OCP specification and plans to contribute the design to OCP; reports differ on the rack generation label and on when partner systems will be available.
What should I do next in practice?
Helios pairs 72 Instinct MI455X GPUs with EPYC CPUs and Pensando networking, with AMD rating the rack at up to 2.9 exaflops of FP4 compute.