Arm Built Its First Ever CPU. Meta Gets It First.

For decades, Arm played a very specific role in the chip world. The UK-based company designed chip blueprints and licensed them to other companies to manufacture. Everyone from Apple to Nvidia used Arm’s designs. But Arm itself never actually made a chip.

That just changed. And Meta is first in line to plug it into its AI data centers.

The Arm AGI CPU Finally Arrives

Arm’s debut processor is called the Arm AGI CPU, and it’s built specifically for AI inference workloads. Inference is the part of AI that actually runs in real time — the processing that happens when you ask an AI assistant a question, or when an AI agent spawns dozens of tasks at once to get something done.

The specs are genuinely impressive. Each chip packs up to 136 cores. You can fit 64 of these CPUs into a single air-cooled server rack. Plus, Arm claims the AGI CPU delivers twice the performance per watt compared to traditional x86 chips from Intel and AMD. That efficiency gap matters enormously when you’re running hundreds of thousands of chips around the clock.

The chip runs on Arm’s Neoverse platform — the same architecture powering AWS Graviton, Nvidia Vera, and Microsoft’s own AI chips. So it slots naturally into existing cloud infrastructure.

Meta Steps Up as Lead Partner

Meta is more than just a customer here. The company calls itself both the lead partner and co-developer of the Arm AGI CPU, with plans to collaborate on multiple generations of the chip going forward.

That’s a significant commitment. And the timing makes sense. Meta has reportedly struggled to get its own in-house AI chips off the ground. Rather than keep waiting, the company is betting on Arm’s new silicon to handle AI workloads alongside hardware from Nvidia and AMD in its data centers.

The announcement drew congratulatory notes from a long list of Arm’s biggest customers, including Amazon AWS, Microsoft, Google, Marvell, Nvidia, and Samsung. Notably absent from that list was Qualcomm, which is still navigating a legal dispute with Arm over licensing terms.

A Chip for Companies That Can’t Build Their Own

Beyond Meta, Arm has already lined up several other customers for the AGI CPU. Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, SAP, and SK Telecom are all in the queue.

Arm’s cloud AI head Mohamed Awad explained the broader vision to CNBC. The goal is to give companies a strong alternative when they can’t afford to develop custom processors themselves. Building a chip from scratch costs billions and takes years. Most companies simply don’t have that runway.

So Arm is positioning the AGI CPU as a practical option for the middle ground — organizations that need serious AI performance but aren’t Google or Amazon with dedicated silicon teams.

What This Means for the Chip Race

The AI chip market is intensifying fast. Nvidia still dominates GPU-based AI training and inference. But a growing number of companies are betting that specialized CPU architectures can handle inference workloads more efficiently and at lower cost.

Arm entering the field as an actual chip producer — not just a licensor — is a genuinely new development. SoftBank, which owns Arm, has long pushed the company toward more direct involvement in the AI hardware boom. This CPU represents that ambition turning into a real product.

Financial terms of the deal with Meta weren’t disclosed. Neither was the exact number of chips Meta plans to deploy. But Meta’s willingness to sign on as co-developer suggests this is a serious, long-term commitment rather than a trial run.

The Arm AGI CPU is expected to land in Meta’s AI data centers later this year. Considering how quickly AI agent workloads are scaling up, that timeline can’t come soon enough for companies racing to keep up with demand.