ARM Announces First Datacenter AI Chip Developed With Meta

Arm unveiled yesterday the AGI CPU, its first CPU for AI data centers designed in collaboration with Meta. After licensing its chip designs to other companies for many years, Arm believes its new AI chip can become “the foundation for agentic data centers.”

“For decades, the industry has built on the Arm compute platform through its industry-leading IP and, more recently, Arm CSS, grounded in a foundation of high-performance, power-efficient computing. The expansion into production silicon with the Arm AGI CPU marks the next phase of that evolution, extending Arm into data center silicon and bringing its power-efficient architecture to AI infrastructure at scale,” the company said in the announcement.

Arm sees the development of agentic AI as an opportunity to deliver more power-efficient chips to companies building their AI infrastructure. The chip designer says that its AGI CPU offers more than 2x performance per rack compared with x86 platforms, and it will enable “up to $10B in CAPEX savings per GW of AI data center capacity.”

In detail, Arm’s AGI CPU features up to 136 Arm Neoverse V3 cores per CPU, with with 6GB/s memory bandwidth per core at sub-100ns latency. Companies using liquid-cooled systems in their data centers can optimize their infrastructure with 45,000+ cores per rack. “These capabilities translate into greater workload density, improved accelerator utilization and more usable compute within existing power envelopes — critical advantages as AI infrastructure scales,” the company said.

Meta co-developed Arm’s fist AI datacenter CPU and will remain a strategic partner for the development of future generations of that chip. The company will also release its own board and rack designs for the Arm AGI CPU later this year under the Open Compute Project. Arm also announced that OpenAI, SAP, Cloudflare, SK Telecom, and other customers will deploy its AGI CPU in their data centers to support their agentic AI developments.