- Arm announced the AGI CPU, a 136-core processor built on TSMC’s 3nm process and the Neoverse V3 platform, designed specifically for agentic AI data center workloads.
- Meta is the lead co-development partner, with OpenAI, Cloudflare, SAP, and Cerebras among the launch customers.
- The chip marks the first time in Arm’s 35-year history that the company is selling its own production silicon rather than licensing IP to other manufacturers.
- A single standard air-cooled rack can house up to 8,160 AGI CPU cores, while a liquid-cooled Supermicro configuration supports over 45,000 cores.
What Happened
Arm announced the Arm AGI CPU on March 24, 2026, a data center processor built to handle agentic AI orchestration, accelerator management, and rack-scale deployment. The chip is Arm’s first production silicon product. For over 35 years, Arm’s business model has been exclusively licensing chip architecture to companies like Apple, Nvidia, Amazon, and Google, who then manufacture their own processors. The AGI CPU marks a direct and unprecedented entry into the finished silicon market.
Mohamed Awad, Executive Vice President of Arm’s Cloud AI Business Unit, led the announcement. Meta served as the lead co-development partner for the chip, which has been designed to work alongside Meta’s custom MTIA silicon to optimize infrastructure for Meta’s family of applications. The partnership gave Arm direct access to Meta’s data center requirements during the design process.
Why It Matters
The shift from IP licensing to selling finished chips represents a fundamental change in Arm’s business model. It puts the company in direct competition with Intel and AMD in the data center CPU market, though Arm is positioning the AGI CPU specifically for AI infrastructure rather than general-purpose server workloads. Analysts at Futurum Group have estimated the addressable opportunity at approximately $15 billion.
Mohamed Awad outlined three primary use cases for the chip: head nodes that coordinate clusters of AI accelerators, standalone compute racks for inference workloads, and AI factory-level orchestrators that manage large-scale training runs. The argument is that as AI systems become more agentic, with multiple agents running concurrently and coordinating across hardware, the CPU orchestration layer becomes a bottleneck that existing general-purpose processors are not optimized to handle.
Sachin Katti of OpenAI stated that “the Arm AGI CPU will play an important role in our infrastructure as we scale, strengthening the orchestration layer that coordinates large-scale AI workloads.”
Technical Details
The AGI CPU packs 136 cores into a 300-watt power envelope, manufactured on TSMC’s 3nm process node. It is built on the Arm Neoverse V3 platform and includes 96 PCIe Gen6 lanes along with native CXL 3.0 support for high-bandwidth memory and accelerator interconnect.
Arm’s reference rack configuration supports up to 8,160 cores in a standard air-cooled 36-kilowatt rack. Supermicro has designed a liquid-cooled variant that houses over 45,000 cores across 336 chips in a 200-kilowatt enclosure. These density numbers are targeted at hyperscale operators building dedicated AI infrastructure.
The chip is designed to serve as the CPU layer in heterogeneous AI systems, sitting alongside GPUs and custom accelerators. Rather than competing on raw compute throughput, the AGI CPU focuses on orchestration: managing data movement, scheduling workloads across accelerators, and handling the control plane for agentic AI systems that coordinate multiple agents operating concurrently.
Who’s Affected
Beyond Meta as the lead partner, the launch customer list includes OpenAI, Cloudflare, SAP, Cerebras, F5, Positron, Rebellions, and SK Telecom. System integrators ASRock Rack, Lenovo, and Supermicro will build server platforms around the chip. Arm claims over 50 ecosystem partners across hyperscale, cloud, silicon, memory, networking, and software.
Intel and AMD face a new competitor in the data center CPU space, though Intel’s data center chief has publicly questioned whether agentic AI workloads require a fundamentally different CPU architecture. The naming choice of “AGI CPU” has also drawn skepticism from some industry observers who view it as marketing rather than a technical distinction.
What’s Next
Arm has not disclosed pricing, volume production dates, or specific availability timelines for the AGI CPU. The company also has not detailed how the chip will be sold, whether directly to hyperscalers or through system integrators, or how it will handle potential channel conflicts with existing Arm licensees who build their own Neoverse-based data center chips. Full details are available in the Arm Newsroom announcement.
