Arm Enters Merchant Silicon Market with AI-Focused "AGI CPU"

Arm Holdings plc (NASDAQ: ARM) announced a historic expansion of its business model today, March 24, 2026, revealing its first-ever production silicon product designed specifically for artificial intelligence infrastructure. The launch of the Arm AGI CPU marks the company’s transition from a pure intellectual property (IP) provider to a direct manufacturer of server chips, a move aimed at capturing the rapidly growing demand for "agentic" AI data centers.

A Historic Strategic Pivot

For more than 35 years, Arm has operated primarily as an IP licensing firm, providing the blueprints for chips used by companies like Apple, NVIDIA, and Amazon. With the introduction of the AGI CPU, the company is now offering its own Arm-designed and branded processors.

"AI has fundamentally redefined how computing is built and deployed," stated Rene Haas, CEO of Arm. "With the expansion into delivering production silicon, we are giving partners more choices to support agentic AI infrastructure at global scale".

Technical Specifications and Performance

The Arm AGI CPU is built on a 3nm lithography process and utilizes the Arm Neoverse V3 platform. It is designed to handle the orchestration and scheduling tasks required by modern AI agents, which often run continuously and coordinate thousands of parallel tasks across accelerators and networking fabrics.

Key technical highlights include:

Core Count: Up to 136 Neoverse V3 cores per chip in a dual-chiplet design.

Memory Architecture: Supports up to 6TB of DDR5-8800 memory per chip with 6GB/s of memory bandwidth per core.

Connectivity: Features 96 lanes of PCIe Gen6 and CXL 3.0 Type 3 for high-speed memory expansion.

Deterministic Performance: A 300W to 420W TDP design that assigns a dedicated core per program thread, intended to eliminate throttling under sustained workloads.

Hyperscale Partnerships and Deployment

Meta served as the lead partner and co-developer for the AGI CPU. The social media giant plans to integrate the chip alongside its own Meta Training and Inference Accelerator (MTIA) silicon for its gigawatt-scale data center deployments. Other early adopters and ecosystem partners include OpenAI, Microsoft, Google, Oracle, SAP, and Cloudflare.

To accelerate adoption, Arm is introducing a 1OU, 2-node reference server design through the Open Compute Project (OCP).

Air-Cooled Systems: A standard 36kW rack can support up to 30 blades, delivering a total of 8,160 cores.

Liquid-Cooled Systems: In partnership with Supermicro, a 200kW design can house 336 CPUs for more than 45,000 cores per rack.

Efficiency and Economic Impact

Arm claims its new silicon delivers more than twice the performance per rack compared to the latest x86 systems from Intel and AMD. According to the company’s estimates, this increased density and efficiency could lead to up to $10 billion in capital expenditure (CAPEX) savings per gigawatt of AI data center capacity.

The Arm AGI CPU is available for order now, with the company committing to a roadmap of follow-on products for the data center market.