Arm and Meta Unveil First-Ever Arm-Designed Data Center CPU for Agentic AI
March 25, 2026
In a historic shift for the semiconductor industry, Arm Holdings plc has launched its first proprietary data center CPU, the Arm AGI CPU, developed in partnership with Meta. This move marks Arm's direct entry into the silicon manufacturing arena, diverging from its foundational business model of licensing chip intellectual property to partners like Apple, Nvidia, and Qualcomm. The launch underscores the intensifying race to build specialized, scalable infrastructure for next-generation artificial intelligence, particularly "agentic AI" systems that operate with greater autonomy.
The newly unveiled chip represents a physical product milestone in Arm's 35-year history. Co-developed with Meta, the Arm AGI CPU is engineered specifically for agentic AI infrastructure. "AI has fundamentally redefined how computing is built and deployed. Agentic computing is accelerating that change," said Rene Haas, CEO of Arm. "Today marks the next phase of the Arm compute platform and a defining moment for our company. With the expansion into delivering production silicon... we are giving partners more choices... to support agentic AI infrastructure at global scale."
Technically, the CPU is built on TSMC's 3nm process node and comprises up to 136 Arm Neoverse V3 cores across two dies, clocking speeds up to 3.7GHz with a 300W thermal design power (TDP). It supports 12 channels of DDR5 memory at speeds up to 8,800Mtps and is equipped with 96 lanes of PCIe 6.0 and CXL 3.0 for high-speed I/O. Arm claims the design can deliver more than double the performance per rack compared to certain x86 configurations, based on internal estimates.
To demonstrate scale, Arm has validated two Open Compute Project (OCP) rack designs: a 36kW air-cooled system with 30 compute blades totaling 8,160 cores, and a high-density 200kW liquid-cooled server developed with Supermicro, capable of housing 336 CPUs for over 45,000 cores. Partners including ASRock Rack, Lenovo, and Quanta Computer are also involved in system development.
Meta plans to deploy these CPUs alongside its custom Meta Training and Inference Accelerator (MTIA) chips, with both companies committing to a multi-generation roadmap. Santosh Janardhan, Meta's head of infrastructure, stated, "We worked alongside Arm to develop the Arm AGI CPU to deploy an efficient compute platform that significantly improves our data center performance density and supports a multi-generation roadmap for our evolving AI systems." Beyond Meta, launch partners include Cerebras, Cloudflare, OpenAI, and SK Telecom, signaling broad industry interest.
Early systems featuring the Arm AGI CPU are available immediately, with broader market availability expected in the second half of 2026. This strategic pivot by Arm could reshape data center economics and supply chains, offering a high-performance, power-efficient alternative to incumbent architectures and accelerating the deployment of complex AI workloads.
Source: datacenterdynamics
