Meta is partnering with Arm, the British chip architecture company, to co-develop a new class of data centre central processing units (CPUs) designed specifically for artificial intelligence workloads.
The collaboration will produce multiple CPU generations intended to deliver higher performance per rack more efficiently than legacy chips, and to complement Meta's existing custom MTIA silicon, its in-house AI accelerator.
The first product, the Arm AGI CPU, is Arm's first data centre processor designed for the AI era, and Meta will serve as lead partner and co-developer.
Arm plans to make the chip available to the broader AI ecosystem beyond Meta.
"Delivering AI experiences at global scale demands a robust and adaptable portfolio of custom silicon solutions, purpose-built to accelerate AI workloads," said Santosh Janardhan, head of infrastructure at Meta.
Janardhan said the chip would improve data centre performance density and support a multi-generation roadmap for Meta's AI systems.
Related reading
- From Vera Rubins, Blackwell, Nemotron, NemoClaw, to Isaac and Cosmos - here's a whistlestop guide to NVIDIA's…
- Samsung and AMD sign memory partnership deal to power next generation of AI chips
- NVIDIA hands over DRA to open source
Rene Haas, chief executive of Arm, said the partnership expands Arm's compute platform into production silicon for large-scale agentic AI deployments, where AI systems take autonomous actions across multi-step tasks.
Meta also said it will publish its board and rack designs through the Open Compute Project, an initiative that promotes open-source hardware standards for data centres, later this year.
The recap
- Meta and Arm will co-develop data center CPUs for AI workloads
- Arm AGI CPU is Arm’s first data center CPU designed for AI
- Meta will publish board and rack designs under Open Compute Project
