Arm Launches Its First In-House AI CPU, Marking a Historic Shift From Its 30-Year-Old Licensing Model

Global leading chip design firm Arm announced Tuesday that it will begin producing its own proprietary semiconductors, a major strategic shift that abandons the company’s long-standing business model of only licensing chip intellectual property to third-party manufacturers that build and sell finished chips.

Speaking to a live audience in San Francisco, Arm CEO Rene Haas outlined how the company’s new first-party CPU will benefit the broader tech ecosystem, and laid out his case for why now is the right time for Arm to step beyond its traditional niche and compete directly with established chipmakers.

“Let me be clear: This is an entirely new line of business for Arm, and we are now supplying full CPUs directly to customers,” Haas said, while holding up one of the company’s newly finished chips. He explained that customer demand is the primary driver behind the move, but added that the widespread growth of artificial intelligence across the global economy and the accompanying skyrocketing demand for computing power has also created a major opportunity for Arm to claim a share of the fast-expanding AI CPU market.

Rumors of Arm’s in-house chip development have circulated for years, but the announcement on Tuesday finally offered a clear, public look at the company’s plans. The new chip is named the Arm AGI CPU, a reference to artificial general intelligence — the widely discussed, still hypothetical form of AI that can match human-level performance across nearly all cognitive tasks. The chip is designed to be paired with other processors in high-performance data center servers, and is optimized specifically to run complex agentic AI workloads. It is being fabricated by global leading semiconductor foundry Taiwan Semiconductor Manufacturing Company (TSMC), using TSMC’s advanced 3nm production process.

At the launch event, Arm executives highlighted the company’s decades-long track record of designing energy-efficient chip architectures, and claimed the new AGI CPU will be the “most efficient agentic CPU on the global market.” When compared to competing latest-generation x86 chips from Intel and AMD, Arm says the new design delivers far better performance per watt (the core metric measuring computing output relative to energy consumption), and could save customers billions of dollars annually on data center electricity costs.

Meta is the first major confirmed customer for Arm’s new chip, and has already received test samples of the AGI CPU. Additional companies that have agreed to purchase the chip include OpenAI, SAP, Cerebras, Cloudflare, and South Korean tech firms SK Telecom and Rebellions. Arm projects the AGI CPU will reach full mass production availability in the second half of this year.

Santosh Janardhan, Meta’s head of infrastructure, joined Haas on stage at the event, and said he expects Arm’s new chip will “expand the chip industry across multiple axes.” As Meta pushes forward with development of what it calls “personal superintelligence” — AI that will power deeply customized experiences across the company’s family of apps — Janardhan noted Meta needs access to far more semiconductor capacity, and is particularly focused on the power efficiency Arm’s design offers.

Kevin Weil, OpenAI’s vice president of science and former chief product officer, also appeared on stage alongside Haas. “One of the most common things I hear inside OpenAI is ‘I need more compute,’” Weil said. “It’s the coin of the realm for AI development.”

Pre-recorded video testimonials from other top tech leaders also aired at the event, including Nvidia CEO Jensen Huang, Amazon senior vice president and distinguished engineer James Hamilton, and Google AI infrastructure chief Amin Vahdat, all of whom praised Arm’s new hardware. None of the three executives publicly committed to purchasing the chip, but all three of their companies already integrate Arm’s intellectual property into their own in-house processors.

Arm’s corporate history dates back to the late 1970s, when it operated under the name Acorn and produced early microprocessors. In the 1990s, the company rebranded as ARM (short for Advanced RISC Machines), and its leadership at the time shifted the business to an IP licensing model that became the company’s core for 30 years. Arm has since dropped the all-caps “ARM” branding, and saw explosive growth during the global mobile technology revolution. By the 2010s, many of the world’s largest tech companies — including Apple, Nvidia, Microsoft, Amazon, Samsung, and Tesla — all built their products around Arm’s chip technology.

Arm made clear at this week’s event that it has broad backing from big-name tech partners, but the move to launch its own chips carries risks. While Arm is primarily targeting Intel and AMD (which build CPUs based on the competing x86 architecture), releasing a first-party chip could alienate some of Arm’s long-standing licensing partners. Nvidia, which is best known for manufacturing GPUs, already bundles Arm-based CPUs into its server rack systems, and announced earlier this year it would begin selling standalone Arm-based CPUs for the first time, with Meta as one of its first customers.

Ben Bajarin, CEO and principal analyst at research firm Creative Strategies, explains that as Arm’s strategy evolves, the company could increasingly be viewed as a competitor rather than a partner by its existing licensees. Right now, Arm is launching a streamlined CPU with a relatively small number of built-in processing cores, specifically designed to run AI agent workloads, Bajarin notes. Over time, Arm may expand into the broader general-purpose CPU market, while Intel and AMD continue developing their own chips tailored for agentic AI — a shift that will put all the companies in far more direct competition with one another.

Even so, Bajarin says Arm’s move is a straightforward play to capture a share of the fast-growing data center CPU market. Creative Strategies forecasts global demand for data center CPUs will grow from $25 billion this year to $60 billion by 2030, a figure that only accounts for CPUs for traditional cloud computing data centers. When demand for agentic AI-specific CPUs is added to the projection, Bajarin estimates the total market will climb to nearly $100 billion by 2030. Even if Arm captures only a small slice of that massive market, it will become a significant new source of revenue for the company.