Overview
Arm Holdings' chief executive, Rene Haas, said on Tuesday that efforts to block the export of CPUs capable of running advanced AI workloads into China would face substantial practical hurdles. Haas argued that the widespread use of CPUs across applications and the blurred line between general-purpose processors and those tuned for AI make export controls especially difficult to design and enforce.
Why CPUs are harder to police
Haas likened CPUs to a kind of utility in application space - a fundamental building block that underpins many workloads - and said that treating them like a restricted AI component would require sweeping measures. He noted that, compared with graphics processing units from companies such as Nvidia, drawing a clear performance-based distinction for CPUs is more complex because determining specific thresholds for compute performance and memory bandwidth is not straightforward.
On the practicalities of imposing limits, Haas said: "They would have to limit everything." He added that while the U.S. could pursue restrictions, CPUs present deeper control challenges than dedicated AI chips.
Context on U.S. restrictions
The U.S. has stepped up efforts to restrict Chinese companies' access to advanced semiconductors and supercomputing systems that support AI development, citing national security concerns. In line with that approach, Washington has also moved to prevent shipments of Nvidia AI chips to Chinese entities operating outside China.
Arm's commercial progress
Separately on Tuesday, Arm announced that ByteDance and Oracle have become customers for the company's AGI-focused data center CPU. Haas said demand for the Arm data center CPU has strengthened relative to the situation eight weeks ago, allowing the company to confirm two substantial customers.
Arm projects that the data center chip could generate roughly $15 billion in annual revenue within five years, marking a new line of business for the British firm.
Implications
The comments underscore the technical and regulatory complexity of distinguishing between general-purpose CPUs and processors optimized for AI, a distinction that matters for trade policy and national-security-driven export controls. At the same time, Arm's customer announcements and revenue target highlight the commercial opportunity the company sees in data center AI processing.
Summary
- Arm CEO Rene Haas warned that banning AI-capable CPUs from reaching China would be difficult because CPUs are widely deployed and hard to classify strictly as "AI" chips.
- Haas said creating enforceable performance and memory bandwidth limits for CPUs is more complex than for GPUs, and that attempted restrictions would need to be very broad.
- Arm named ByteDance and Oracle as customers for its AGI data center CPU and expects the chip to approach $15 billion in annual revenue within five years.