Arm’s inaugural CPU set to power Meta’s AI data centers this year.

After years of licensing its chip designs, UK-based Arm has unveiled its first self-produced chip, the Arm AGI CPU, along with its inaugural customer. This new chip is tailored for inference tasks, powering cloud processing for AI applications, including AI agents that can handle multiple tasks simultaneously. Meta is the first company to adopt this technology, having faced challenges in developing its own AI chips.

Meta is not just a customer; it’s also a lead partner and co-developer, intending to collaborate on several generations of these datacenter CPUs, which will work alongside hardware from other manufacturers like Nvidia and AMD. Other notable Arm customers, including Amazon AWS, Microsoft, Google, Marvell, Nvidia, and Samsung, expressed their support with congratulatory messages. However, Qualcomm, which recently claimed a legal victory over Arm regarding licensing terms, did not join in the celebration.

Details about the financial aspects of the deal and the quantity of chips Meta plans to utilize have not been disclosed. Arm, currently under Softbank’s ownership, states that its new chip operates on the Neoverse platform, which is also used by AWS Graviton, Nvidia Vera, Microsoft, and other AI chips. The AGI CPU boasts up to 136 cores per CPU and can support 64 CPUs in a single air-cooled server rack. Arm claims that this chip delivers double the performance per watt compared to traditional x86 CPUs while minimizing memory bottlenecks, leveraging its design’s longstanding efficiency.

Additional companies interested in Arm’s chip include Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, SAP, and SK Telecom. Mohamed Awad, head of Arm’s cloud AI division, mentioned that the goal is to provide a viable option for companies that cannot afford to develop their own processors.