CPUs have not been killed off by AI, but humanity is going to need lots more of them, according to Arm chief executive Rene Haas.
CPUs have spent the last few years living in the shadow of AI hype, but with agentic AI tearing through workloads, the CPU is suddenly the belle of the ball.
The surge in so-called agentic AI is driving a fresh wave of demand, and chip outfits are scrambling to cash in. Nvidia, best known for GPUs, said last week it will sell a new server rack made up only of its Vera CPUs, with no graphics cards attached.
Meta Platforms, which is piling money into infrastructure and using AI agents to sharpen ad targeting and other chores, has placed a big order for Nvidia’s CPUs. It called it the first significant CPU-only data-centre deployment.
Deloitte semiconductor expert Jeroen Kusters said that there is a rebalancing toward CPUs.
“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 are one of the clearest signs that AI agents are warping the old pecking order. For decades Arm licensed its CPU architecture to bigger players such as Nvidia and Qualcomm, which then built and sold their own chips.
Arm is stepping on toes by remaking itself as a chip designer that wants to go head-to-head with its own customers. The firm is pitching the new part as the “world’s most efficient agentic CPU,” and you can practically smell the marketing department.
Arm claims its AGI CPU is twice as efficient as rival processors, leaning on its long-running reputation for low power. Meta has signed up as lead partner and first big customer, with other names on the list including OpenAI, Cerebras, SAP, Cloudflare, SK Telecom and Rebellions.
In an interview, Haas said Arm expects annual revenue to roughly quintuple to around $25 billion over the next five years, about a third higher than analysts polled by FactSet predicted. Investors lapped it up, and Arm shares jumped 16.38 per cent on Wednesday.
The frenzy is being fuelled by coding tools such as Claude Code and OpenAI’s Codex, which have kicked demand for AI agents into another gear. The release of OpenClaw, a platform for virtual assistants that can perform real-world tasks, has pushed vibe-coders to buy memory-heavy boxes like the Fruity Cargo Cult Apple’s Mac Mini.
Positron AI chief technology officer and co-founder Thomas Sohmers said: “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.”
“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. This is truly a new computational demand,” he said.
Haas framed it as raw arithmetic: “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.” Haas said.
Arm estimates the total addressable market for data-centre CPUs could reach about $100 billion a year within the next five years. It plans to ship the first orders in the months ahead, with the chips fabricated by TSMC.
