Arm Holdings CEO Rene Haas said that restricting exports of AI-capable central processing units (CPUs) to China would be nearly impossible, comparing their widespread use to oil. His comments, made to Reuters, underscore a strategic headache for U.S. policymakers seeking to curb China's artificial intelligence capabilities without disrupting the global technology supply chain.
CPUs are like oil
"CPUs are kind of like oil relative to the application space," Haas told Reuters. "That's a pretty hardcore cut." He explained that unlike graphics processing units (GPUs) from Nvidia, where performance and memory bandwidth thresholds can be set to control exports, CPUs are embedded in everything from servers to smartphones. "They would have to limit everything," Haas said, adding that while the U.S. could attempt such a move, policing CPU shipments would be far harder than restricting dedicated AI chips.
The remarks come as Arm lands major data center CPU deals with companies including ByteDance and Oracle. Arm, which licenses its chip designs to manufacturers globally, projects its annual data center chip revenue will reach roughly $15 billion within five years.
U.S.-China chip friction escalates
Washington has tightened semiconductor export rules over the past year, while Beijing pushes its firms to reduce reliance on foreign technology and invests in domestic chipmakers like Huawei and Cambricon. Ahead of the recent Trump-Xi summit, Chinese regulators barred companies from using throttled-down versions of Nvidia's gaming chips that had been cleared for sale by the U.S. Commerce Department.
For executives overseeing AI infrastructure, these policy shifts add another layer to supply-chain risk. Courses designed for AI for Executives & Strategy help leaders evaluate hardware sourcing strategies amid geopolitical uncertainty. Similarly, AI for Government programs provide insight into how export controls shape national AI priorities.
Why this matters for executives and strategy
Haas's assessment suggests that while GPU restrictions will remain the primary tool for limiting China's AI compute, broad CPU controls are unlikely to appear quickly. Executives planning AI infrastructure should diversify their processor architectures and monitor how licensing-based chip firms like Arm might offer more resilient supply routes than component makers subject to direct export bans. Companies heavily dependent on Nvidia should evaluate whether CPU-centric AI workloads can reduce their regulatory exposure.
Your membership also unlocks:
