Arm’s AI Infrastructure Bet Extends Beyond Chip Licensing

Arm’s first production CPU is arriving just as hyperscalers are redesigning the data center around AI workloads, giving the chip designer a second way to participate in the spending surge.

The company’s traditional model is already benefiting. Arm reported $1.29 billion in revenue for the quarter ended June 30, 2026, up 22% from a year earlier, while royalty revenue climbed 22% to $715 million. Data-center royalties more than doubled, according to the company’s shareholder letter.

That matters because royalties are the durable part of Arm’s business. Once a customer’s processor design ships in volume, Arm collects a payment on the resulting chips, allowing revenue to compound as the installed base expands without requiring the company to manufacture every wafer or operate every server.

The AI buildout is widening that installed base. Arm Neoverse shipments have surpassed 1.5 billion cores, with the latest 500 million shipped in nine months compared with six years for the first billion. NVIDIA’s Vera CPU, which is built on Arm technology, is moving into production for AI infrastructure, while cloud companies including Amazon and Google continue deploying Arm-based processors.

Arm is also moving closer to the silicon itself. Its Arm AGI CPU, introduced in March for agentic-AI workloads, has attracted more than $2 billion of customer demand across fiscal 2027 and fiscal 2028, exceeding the initial $1 billion opportunity disclosed by management. Meta is the lead partner and co-developer, while Cerebras, OpenAI, Positron and Rebellions are integrating the processor into their systems.

The catch is valuation and execution. Arm’s royalty engine carries attractive economics, but investors are already paying for years of AI-driven expansion. The AGI CPU opportunity is also supply-constrained, meaning demand will not immediately translate into revenue. Arm’s own filings list manufacturing capacity, customer concentration and competition among the risks facing the business.

Still, the investment case is broader than a bet on one processor. If AI infrastructure keeps spreading from training clusters into inference servers, smartphones, vehicles and robots, Arm can collect across the computing stack. That is the picks-and-shovels appeal: less exposure to which application wins, and more exposure to the architecture underneath them all.