Arm Holdings has stepped into a new role. The company known for powering smartphones and now data centers just launched its first in house CPU called the AGI CPU. This processor targets AI inference in large scale data centers and Meta signed on as the lead development partner and first official customer. 🤖
The news comes after more than 35 years during which Arm only licensed its architecture. Now the firm designs builds and sells actual silicon. Meta worked closely with Arm starting in 2023 to shape the chip. Paul Saab a Meta software engineer who joined the project early said the deal brings another strong player into the ecosystem and gives his team more flexibility in both software and supply chain.
The chip design and performance edge ⚡
Each AGI CPU packs up to 136 Arm Neoverse V3 cores. Up to 64 of these chips fit into one air cooled rack for a total of roughly 8700 cores. Arm claims the design delivers two times the performance per watt versus a comparable x86 rack. That efficiency matters because power has become a limiting factor in AI data centers.
The chip sits on TSMC‘s 3 nanometer process node and production ramps later in 2026. Arm built a 71 million dollar chip lab in Austin Texas that now employs more than 1000 people to support this effort. For background on Arm‘s long history with efficient architectures read the official Neoverse platform overview.
Systems are already available for order through partners such as ASRock Rack Lenovo and Supermicro. Arm positions the AGI CPU as a drop in replacement for general compute tasks that remains transparent to developers.
Market context and customer momentum 💰
Meta plans to spend as much as 135 billion dollars on capital expenditures this year to fuel its AI infrastructure. The company already buys heavily from Nvidia and AMD and runs its own MTIA accelerators. Adding the AGI CPU gives Meta another option that works alongside those components and frees up scarce power for other parts of the stack.
Seven more customers have committed including OpenAI Cloudflare and SAP. Over 50 partners signaled support ahead of launch among them leaders from Google Amazon Microsoft and Broadcom. Chip analyst Patrick Moorhead of Moor Insights noted that even a small slice of Meta‘s spending could prove meaningful for Arm‘s top line.
This launch fits into wider trends around agentic AI where CPUs handle sequential tasks and data movement while GPUs focus on parallel training. The Futurum Group has projected that CPU market growth may outpace GPUs by 2028 because of these demands. Explore how Arm based solutions compare in the AI hardware tools directory or check the dedicated Neoverse platform profile for deeper technical context.
Arm still manufactures in Taiwan through TSMC but has left the door open to future production in Arizona if customer needs justify it. The company stresses the AGI CPU stays available to the broader market rather than locked inside one organization.
The debut raises fresh questions about how traditional licensors evolve in a world where hyperscalers build their own silicon. For now Arm has delivered a high density efficient option that Meta and others plan to put to work inside their expanding AI infrastructure.
