Arm launches its own AI processors, targets $15 billion in annual revenue

Arm has decided that, for the first time in 35 years, it will not limit itself to licensing its architecture and will enter the server processor market itself. The company showed the Arm AGI CPU chip for agentic AI and said that in five years it wants to earn about $15 billion a year from its sales alone.

Why the strategy is changing

Until now, Arm's business was built on a different model: the company designed the architecture and compute blocks, and made money from licenses and royalties paid by partners. Amazon, Google, Microsoft and dozens of other players have long built on this foundation. The new step changes Arm's very role in the market: it now wants to be not only a supplier of intellectual property but also a seller of finished hardware. Historically, this made Arm a more neutral player than companies that make chips directly.

The move looks logical against the backdrop of the AI infrastructure boom. Where growth once came mainly from smartphones and conventional servers, big money is now concentrated in data centers, which must serve not only models but entire chains of agentic tasks. For Arm, this is a chance to capture a noticeably larger share of revenue from each deployment than under the standard licensing scheme. In this architecture, the CPU once again becomes a critical node rather than a secondary component.

"This is a real turning point for the company," is how

Rene Haas described the launch of the new business.

Where the billions will come from

Arm puts its annual revenue for last year at roughly $4 billion and wants to raise it to $25 billion by 2031. Of that amount, about $15 billion, according to the company's plan, should come specifically from its own AI processors. In other words, Arm expects the new hardware business to account for about 60% of revenue and effectively to reshape the economics of the whole company. For investors, this means a bet not on gradual royalty growth but on a sharp expansion of the addressable market.

The market initially reacted positively: after the announcement, the company's shares rose about 6%. The bet is on growing CPU demand in the era of agentic AI. By Arm management's estimate, the need for central processors for such workloads could grow at least fourfold. The logic is simple: the more agents, schedulers, memory, networking operations and data exchange between accelerators a system has, the more important the CPU becomes as the infrastructure's coordinating layer. This is exactly where Arm is trying to earn more than before.

• Total revenue target for 2031: $25 billion
• Expected revenue from its own AI chips: $15 billion a year
• Share of the new business in revenue: about 60%
• Profit target for the first chip: about 50% net margin
• Earnings-per-share forecast: up to $9

The product itself matters too. The Arm AGI CPU is a server processor designed for agentic AI infrastructure. According to Arm, the chip is built on a 3 nm process, has up to 136 Neoverse V3 cores, supports up to 6 TB of memory, and is aimed at orchestration, scheduling, memory handling and data movement inside large AI clusters. In other words, Arm is not trying to replace GPUs but to take the management layer around them.

Who might buy it

Arm says outright that it will not force current partners to abandon their own Arm-compatible designs. Amazon, Google and Microsoft already use the Arm architecture in their server processors and are unlikely to want to change course entirely. But the new chip gives the company access to a different segment: customers who want a ready-made platform quickly, without years of in-house development. For many buyers, this may be the fastest way to launch new AI infrastructure.

Potential customers include cloud providers, builders of specialized AI systems, and companies that want to move away from Intel and AMD but lack the resources for their own CPU project. Analysts expect such processors to cost thousands of dollars. Arm also already has a strong ecosystem around the product: confirmed partners and customers named include Meta as a key co-developer, as well as OpenAI, Cloudflare, Cerebras, SAP and SK Telecom.

What this means

Arm is making the sharpest bet in its history, moving from a "we design for others" model to a "we sell our own chips" model. If demand for agentic AI really pushes the CPU market up, the company could turn from an infrastructure licensor into one of the most visible suppliers of server hardware for AI. If the plan works, the market will gain another heavyweight that competes not only on architecture but on a finished product.

*Meta has been designated an extremist organization and is banned in Russia.