After nearly 36 years of licensing its chip architectures to giants like Nvidia and Apple, Arm Holdings has officially entered the chip manufacturing game with its own processor. The new Arm AGI CPU targets AI data centers, marking a major shift in the company’s decades-old business model.
Revealed at an event in San Francisco, the Arm AGI CPU is a production-ready processor designed specifically for AI inference workloads. It’s built on the Arm Neoverse architecture and was developed in collaboration with Meta, which is also the CPU’s launch customer.
Other launch partners include OpenAI, Cerebras, and Cloudflare, reflecting the CPU’s focus on powering the infrastructure behind large-scale AI services.
While Arm has long been known as a technology licensor rather than a chip maker, the move to produce its own silicon-work that began in 2023-signals a direct challenge to some of its traditional customers and partners.
Interestingly, Arm chose to build a CPU rather than a GPU. Although GPUs often steal the spotlight in AI due to their parallel processing power, CPUs remain vital for coordinating distributed workloads in data centers.
Arm explains that CPUs handle thousands of distributed tasks, like memory management, workload scheduling, and data transfer between systems. In modern AI infrastructure, the CPU serves as ”the key component that sets the pace of the entire system,” enabling efficient scaling across thousands of processors.
With surging demand for AI computing power, the importance of reliable CPUs is rising sharply. Industry giants Intel and AMD have already flagged supply shortages in China, causing price hikes and delays for new servers and computers.
This entrance by Arm as a chip fabricator could shift the semiconductor and AI infrastructure landscape. By producing CPUs tailored for AI data centers, Arm positions itself to reshape how the backbone of AI services is built and powered.
Note: Meta, involved as a development partner, is designated an extremist organization and banned in Russia.