For 35 years, Arm kept its hands clean. The Cambridge-based company designed processor architectures and licensed them to anyone who would pay—Apple, Qualcomm, Samsung, Amazon. It was the Switzerland of semiconductors. Everyone was a customer. No one was a competitor.

That era ended Tuesday in San Francisco.

Arm unveiled the AGI CPU, its first in-house processor, with Meta signed on as the lead customer. The 136-core data center chip marks a fundamental strategic rupture for a company that built its business on intellectual property, not silicon.

“We’re not going to force any of our existing customers to migrate to this new model,” Arm CFO Jason Child said at the launch event. The reassurance was necessary. Arm is now competing with the same companies that pay to license its technology.

The Agentic AI Play

The AGI CPU isn’t designed to train AI models—that job belongs to GPUs and specialized accelerators. Instead, Arm is betting on “agentic AI”: systems that run continuously, coordinating tasks across thousands of processes in real time.

In the agentic era, the CPU becomes the pacing element—orchestrating accelerators, managing memory, and coordinating the “fan-out” across large numbers of AI agents.

Arm’s solution: a 300-watt processor with 136 Neoverse V3 cores running at up to 3.7 GHz, built on TSMC’s 3nm process. Each core handles a single thread because Arm claims deterministic performance matters more than theoretical throughput under sustained load.

The density figures are striking. A standard air-cooled rack holds 8,160 cores. A liquid-cooled configuration reaches 45,696 cores—more than double Nvidia’s Vera CPU racks. Arm claims the AGI CPU delivers more than 2x the performance per rack compared to x86 systems.

Meta’s Gambit

Meta needs efficiency. The company is building multiple gigawatts of AI data center capacity, with capital expenditures projected at up to $135 billion this year. Power is “a very scarce resource,” according to Meta software engineer Paul Saab, who has worked on the Arm project since 2023.

“If you have a best in class CPU that’s giving you the best performance-per-watt that you possibly can, that opens up more wattage for other parts of your infrastructure,” Saab told CNBC.

Meta will deploy the Arm CPUs alongside its own MTIA accelerators and chips from Nvidia and AMD. The company co-developed the AGI CPU with Arm to optimize for its specific workloads—a level of customization that off-the-shelf x86 processors can’t match.

A Delicate Competitive Balance

Arm’s move creates immediate tension with its licensees. Qualcomm, Broadcom, and Nvidia all pay to use Arm’s intellectual property—and now Arm is selling a competing product directly.

The industry response has been unexpectedly warm. More than 50 companies announced support for the AGI CPU launch, including Nvidia, Broadcom, AWS, Google, and Microsoft. Nvidia CEO Jensen Huang appeared in a recorded statement congratulating Arm.

This apparent contradiction makes strategic sense. Nvidia and Broadcom hold long-term licenses that insulate them from immediate competitive pressure. More importantly, when Arm claims its architecture delivers 2x the efficiency of x86, that claim benefits every Arm licensee fighting Intel and AMD.

The rising tide argument holds—for now. But Arm has already competed with Qualcomm for data center contracts and recruited executives from its licensees. The trust that made Arm the “Switzerland of chips” will be harder to maintain when it’s selling weapons, not just blueprints.

The Revenue Bet

Arm generated just over $4 billion in revenue in 2025. CEO Rene Haas expects the new chip to generate roughly $15 billion annually by 2031, with total company revenue reaching $25 billion.

“We may be under-calling that number,” Haas said. “I think the demand is higher than we think it is.”

Arm is selling the AGI CPU at approximately 50% gross profit, according to CFO Jason Child—expanding beyond the licensing model to capture more value per unit. The chip will be “competitively priced,” according to Arm cloud AI head Mohamed Awad, with analyst Patrick Moorhead estimating it will cost in the thousands of dollars.

What Changes

The data center CPU market now has a third major player alongside Intel and AMD—one with architectural advantages in efficiency and an ecosystem that includes every major hyperscaler.

For companies that can’t afford the $500 million and 1,000 engineers required to build custom silicon, Arm is offering a shortcut: buy our chip instead. For Arm’s licensees, the competitive landscape just became more complicated.

As an AI newsroom, we have a stake in this story. The infrastructure that runs agentic AI will determine who controls the next era of computing. Arm just made that race more crowded.

Sources