Arm Holdings is transitioning from a longstanding licensing model to manufacturing its own chips with the introduction of the Arm AGI CPU, its first in-house production after nearly 36 years. Unveiled at a recent event in San Francisco, this production-ready chip is specifically designed for inference tasks in AI data centers.
The development of the Arm AGI CPU leverages the Arm Neoverse family of CPU IP cores and a strategic partnership with Meta, which becomes the chip’s first customer. The design facilitates integration with Meta’s training and inference accelerator. Other launch partners include OpenAI, Cerebras, and Cloudflare, highlighting the significant interest in the new chip.
Arm began development of the AGI CPU in 2023, and units are now available for order. This marks a historical shift for the company, which has primarily licensed designs to competitors such as Nvidia and Apple. The pivot towards manufacturing indicates Arm’s strategy to compete more directly with its partners within the semiconductor industry.
Unlike graphics processing units (GPUs), which have dominated discussions regarding AI deployment, the Arm AGI CPU emphasizes the critical role of CPUs in handling distributed AI workloads. Arm stated that these CPUs are vital for managing tasks such as memory and storage allocation, workload scheduling, and system data movement—elements that are increasingly necessary for the efficient operation of AI infrastructure.
The timing of Arm’s chip launch is noteworthy, as there is a documented shortage of CPUs. In March, Intel and AMD informed their customers in China about extended wait times for CPU deliveries.
