Arm just crossed a line it’s never crossed before. After decades of licensing chip designs to other companies, the UK-based semiconductor giant is building its own CPU, according to The Verge AI. The first customer? Meta.
The chip is called the Arm AGI CPU, and it’s built specifically for AI inference, the cloud processing that powers AI tools and agents capable of spawning tasks on the fly. Meta isn’t just buying these chips. The company is both the lead partner and co-developer, with plans to collaborate on “multiple generations” of data center CPUs.
This is significant for two reasons.
🔍 Why Arm Is Making This Move
Arm has always been the architect behind the scenes. Its designs power everything from smartphones to AWS Graviton chips. But selling blueprints is a different business than selling silicon. By producing its own chip, Arm is stepping into direct competition with the companies it supplies.
The target market: companies that can’t afford to design processors in-house. Arm cloud AI head Mohamed Awad told CNBC the goal is to offer an alternative for organizations that lack the resources of a Google or Amazon.
🧠 What the Chip Actually Does
The specs are worth paying attention to:
- Built on the Neoverse platform (same foundation as AWS Graviton and Nvidia Vera)
- Up to 136 cores per CPU
- 64 CPUs per air-cooled server rack
- Claims twice the performance per watt compared to traditional x86 CPUs
- Designed to reduce memory bottlenecks, a real pain point for inference workloads
That efficiency advantage matters. AI inference is quickly becoming the dominant workload in data centers, and power consumption is the biggest constraint on scaling.
🤝 The Customer List Tells a Story
Beyond Meta, the lineup includes Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, SAP, and SK Telecom. Amazon AWS, Microsoft, Google, Marvell, Nvidia, and Samsung all sent congratulatory notes.
One name conspicuously absent: Qualcomm. The company declared “complete victory” over Arm in a licensing dispute last fall. No love letter from San Diego on this one.
Financial terms weren’t disclosed. Neither was the volume of chips Meta plans to deploy.
📡 What This Means Going Forward
Meta has reportedly struggled to launch its own AI chips. Partnering with Arm gives it a co-developed solution that works alongside existing Nvidia and AMD hardware. For Arm, owned by SoftBank, this is a bet that the company can capture more value from the AI infrastructure boom by selling finished products, not just designs.
The shift from licensor to manufacturer is a fundamental change in Arm’s business model. If it works, Arm becomes a direct player in the AI chip race. If it doesn’t, the company risks alienating the partners that built their businesses on its architecture.
Chips are expected to arrive in Meta’s data centers later this year. More details at The Verge AI.