Arm Builds Its First In-House Chip in 35 Years — and Meta Is the Launch Customer
Arm Holdings has announced the Arm AGI CPU, its first self-designed production chip after 35 years of exclusively licensing its architecture to other chipmakers. Meta is the chip's first customer, and OpenAI, Cerebras, and Cloudflare are among the launch partners.
Arm Holdings has spent 35 years licensing CPU designs to other companies. On March 24, at an event in San Francisco, it announced the Arm AGI CPU — its first in-house production chip, built for running AI inference in data centers.
This is a significant structural shift. Arm’s entire business model has been built on IP licensing: design the architecture, let Nvidia, Apple, Qualcomm, and others build the silicon. The Arm AGI CPU breaks that pattern. For the first time, Arm is a chipmaker, not just a chip designer.
The chip was developed in partnership with Meta, which is also its first customer. The Arm AGI CPU is designed to work with Meta’s own training and inference accelerators — it’s a CPU, not a GPU, purpose-built for the data center inference workloads that have exploded with large language model deployment. Meta’s involvement from development through purchase signals this is a serious production chip, not a reference design.
Other launch partners include OpenAI, Cerebras, and Cloudflare. The chip is already available to order.
Why a CPU? While GPUs dominate AI training headlines, inference is a different problem. Running a trained model at scale — responding to millions of API calls — requires efficient, cost-effective compute that CPUs can often handle better than power-hungry GPUs, especially for lighter workloads. Arm’s Neoverse family has been the go-to CPU architecture for cloud inference already; the Arm AGI CPU is a purpose-built extension of that lineage.
The competitive implications are significant. Arm’s architecture is already inside chips from Nvidia (for its CPU-GPU systems), Apple, Qualcomm, Amazon (Graviton), and Microsoft. Those companies have all been paying Arm licensing fees. Now Arm is entering the market as a direct competitor for data center CPU sales — the same market it helped enable for all of those customers.
Arm is majority-owned by SoftBank, and this move fits the conglomerate’s AI infrastructure push. SoftBank has been investing aggressively across the AI stack, including its $500 billion Stargate commitment with OpenAI in the US. Having Arm actually manufacture and sell chips — rather than just collect royalties — is a natural extension of that strategy.
The chip started development in 2023, per earlier CNBC reporting. The fact that it’s already production-ready and available to order suggests Arm has been disciplined about this launch: no paper announcement, no “coming soon.”
For the broader AI infrastructure market, the Arm AGI CPU adds a new competitor to a space currently dominated by Nvidia’s H100/H200 for training and AMD/Intel for inference. The combination of Arm’s architectural efficiency, Meta’s validation, and a roster of cloud-native launch partners makes this worth watching closely.