Arm Holdings has introduced its first self-developed chip, the Arm AGI CPU, signaling a major shift from its long-standing model of licensing designs to other chipmakers.
Arm has spent nearly 36 years licensing chip designs to companies such as Nvidia and Apple instead of producing any hardware itself.
The company revealed the Arm AGI CPU at an event in San Francisco. The processor is production-ready and designed for AI data centers, specifically for inference workloads. Arm developed the chip using its Neoverse family of CPU IP cores in partnership with Meta.
Meta is also the first customer for the new chip, which is built to integrate with its training and inference accelerators.
Arm said it is working with several partners for the launch, including OpenAI, Cerebras, and Cloudflare. The processors are already available to order, with development reportedly beginning in 2023.
By producing its own silicon, Arm is now entering direct competition with some of its existing partners. The company is majority owned by SoftBank Group.
Unlike many recent AI hardware developments focused on GPUs, Arm’s new chip centers on CPUs. The company said CPUs handle critical tasks in data centers, including memory and storage management, workload scheduling, and data movement across systems.
Arm described the CPU as a key component in maintaining efficiency across distributed AI systems.
The launch comes as CPU supply constraints continue to affect the market. In March, Intel and AMD informed customers in China of longer wait times due to shortages, according to Reuters. Rising demand has also contributed to increasing computer prices.