Rene Haas, CEO of chip designer Arm Holdings Arm Holdings (ARM), has a message for U.S. policymakers: good luck trying to stop AI-capable CPUs from reaching China. In an interview with Reuters on Tuesday, Haas argued that restricting central processing units is fundamentally different—and far harder—than limiting Nvidia's Nvidia (NVDA) graphics cards.

The problem, Haas explained, is that CPUs are everywhere. They power everything from laptops to servers to industrial equipment, and they don't come with a handy label saying "this one's for AI." Unlike Nvidia's GPUs, which are purpose-built for AI workloads and can be regulated by setting performance thresholds and memory bandwidth limits, CPUs are general-purpose workhorses. "They would have to limit everything," Haas said, implying that any attempt to single out AI-capable CPUs would be a regulatory nightmare.

Haas's comments come as the U.S. government tightens its grip on semiconductor exports to China. Just recently, the U.S. closed a loophole that allowed AI chip shipments from Nvidia and AMD to Chinese firms operating overseas. Now, the plan is to impose license requirements on advanced chips for any entity headquartered in China, no matter where they're based. But CPUs, Haas suggests, might slip through the cracks.

CPUs: AI's New Frontier

The timing of Haas's warning is no coincidence. While Nvidia's GPUs have been the stars of the AI boom, CPUs are increasingly in the spotlight as the industry shifts toward AI inference—the process of running trained models rather than training them. Nvidia CEO Jensen Huang has already called out a $200 billion untapped market for his Vera CPUs, and he's made it clear that China is a "very important" destination despite the trade tensions.

Arm itself is riding this wave. At the Taipei Computex conference on Tuesday, Haas announced that Oracle Corp. Oracle (ORCL) and China's ByteDance have signed on as customers for Arm's AGI CPU, a chip designed to power next-generation AI systems. The CPU is gaining momentum, and Haas is betting it will be a key player in the AI infrastructure buildout.

But the broader question remains: can the U.S. effectively block China from getting the chips it needs for AI? Haas's answer is a cautious "probably not"—at least when it comes to CPUs. The genie, it seems, is already out of the bottle.