Arm Holdings foresees significant demand for its chips as swarms of intelligent agents require ever-more processing capacity.

Rene Haas, chief executive of the chip company Arm Holdings, took the stage Tuesday in a repurposed waterfront pavilion to unveil his company's latest product: a new design for the same type of computer chip that Arm has specialized in for more than 30 years.

"People thought CPUs were dead," Haas told the crowd. But as artificial intelligence rapidly develops, "You need more and more CPUs," he said. "Lots of them."

For the past few years, central processing units, or CPUs -- the core engines of most computers, from laptops to smartphones to data-center servers -- have been something of an afterthought in the world of artificial-intelligence computing. Now, thanks to how fast AI is changing, they are the belles of the ball.

The explosion of so-called agentic AI has driven a wave of demand for CPUs, and chip companies are moving quickly to capitalize on it. Nvidia, the semiconductor giant, is known best for making powerful graphics processing units, or GPUs, which accelerate the computing needed to train and run complex AI models. It announced last week that it would offer a new server rack consisting only of its Vera CPUs, with no graphics cards attached.

Meta Platforms, which has been investing heavily in its computing infrastructure and in using AI agents to target ads better and help with other tasks, recently placed a big order for Nvidia's CPUs, calling it the first significant CPU-only data-center deployment.

Close-up of the Nvidia Vera CPU tray.

"There's a rebalancing toward CPUs," said Jeroen Kusters, a top semiconductor expert at Deloitte. "The revolution was more on the GPU side. Now it's on the entire system. CPUs have always been part of the AI fabric, but now there's a massive increase in need for them."

Arm's new processors represent one of the starkest indications yet of how dramatically the rise of AI agents is changing the technology industry. For decades, Arm has made money by licensing its CPU architecture to bigger companies such as Nvidia and Qualcomm, which use it to create CPUs that they sell to others. Now, Arm is going head-to-head with them by remaking itself as a chip designer.

The new chip is the "world's most efficient agentic CPU," the company said. Long known for drawing up blueprints for chips that require less energy than others, Arm said its AGI CPU is twice as efficient as equivalent processors made by competitors.

Arm also expects its shift from selling intellectual property to actually designing and selling chips to be a major financial boon for the company. Meta has signed up as the lead partner and first big customer for the chips. Other customers include OpenAI, Cerebras, SAP, Cloudflare, SK Telecom and Rebellions.

Haas said in an interview that Arm expects annual revenue roughly to quintuple to around $25 billion over the next half-decade, about one-third higher than analysts polled by FactSet had predicted. Shares rose more than 16% on Wednesday following its chip announcements.

That projected growth reflects the way coding tools such as Claude Code and OpenAI's Codex have kicked demand for AI agents into overdrive. The release of OpenClaw -- a platform that allows users to create virtual personal assistants that can do tasks in the real world -- gave a glimpse of the mad rush for agentic AI computing power to come. Vibe-coders rushed to buy devices with high levels of memory -- such as Apple's Mac Mini -- to accommodate the computing needs of agentic AI platforms.

"As these agents are able to do more things autonomously by themselves, you want them to have their own computing environment, and you need to have a certain number of CPUs to run them," said Thomas Sohmers, chief technology officer and co-founder of Positron AI, a chip startup that has struck a deal with Arm to use its new chips in its computing platform.

"If people have multiple OpenClaws running all day, you need a lot of dedicated CPUs. A few months ago, people just weren't running this many agents," Sohmers said. "This is truly new computational demand."

Haas said: "When you go to an agentic world, where a human can generate 100 tokens in a minute or two, an agent can do 15 times that number. That's all CPU work."

Arm expects the total addressable market for data-center CPUs to grow to about $100 billion a year over the next half-decade. The company plans to ship its first orders of the chips -- which are fabricated by Taiwan Semiconductor Manufacturing -- in the months ahead.

At Tuesday's unveiling event, Arm's customers, including Meta and OpenAI, agreed that the growth of agentic AI has provoked astronomical demand for computing power.

Santosh Janardhan, Meta's head of infrastructure, said his company's goal is "to deliver personal superintelligence to billions of people," a herculean task that requires immense amounts of power, land and hardware.

But "most of all, it takes silicon. A lot of silicon," he said. "There is absolutely no sign of this slowing down."

Kevin Weil, OpenAI's vice president for science, said powering AI computing has become a matter of "system performance" rather than relying on one or two processors alone, which makes CPUs like the one Arm is launching more crucial than ever.

"GPUs kind of get top billing wherever they go, but really the CPU is playing an incredibly important role," Weil said. "'I need more compute' is the coin of the realm.... We have more demand from customers, more things we want to do, than frankly the industry can keep up with."