US regulatory agency reviews Chinese AI companies' overseas access to Nvidia chips, enforcing export controls
A US government agency is conducting a sweeping review of how Chinese AI companies have been obtaining Nvidia's most advanced chips, despite years of Washington efforts to cut off that pipeline. The review, reported by Bloomberg on August 7, 2026, targets a strategy that had exploited a regulatory gap: rather than importing chips directly into China—where export controls block shipments—Chinese firms rented computing power from data centers in other countries.
US export restrictions prevent Nvidia from shipping its most powerful processors directly to China. However, those same rules, until recently, did not stop Nvidia chips from ending up in data centers in Southeast Asia, the Middle East, or elsewhere, where Chinese companies could then lease remote access to them. Industry estimates suggest that hundreds of thousands of advanced chips may have been delivered to Chinese entities through this regulatory gap before Washington moved to close it.
The urgency behind the review appears directly tied to recent Chinese AI breakthroughs that demonstrated effective utilization of Nvidia's restricted processors. On May 31, 2026, the Commerce Department's Bureau of Industry and Security issued new guidance specifically designed to close the overseas subsidiary loophole. The updated rules changed a critical variable: licensing requirements are now determined by where a company is headquartered, not where its subsidiary is located. If the parent company sits in China, every entity under its umbrella now faces the same export controls, whether it operates out of Shanghai, Singapore, or São Paulo. The guidance explicitly covers Nvidia's most advanced product lines, including the Blackwell and Rubin architectures.
Before the May clarification, a Chinese AI firm could set up a subsidiary in a third country, order chips through that entity, and avoid triggering export restrictions entirely. The new rules treat the subsidiary as an extension of its Chinese parent.
Preventing Chinese companies from renting cloud computing time on servers that already contain Nvidia hardware presents a fundamentally different challenge. Regulating such access would require either restricting who can purchase cloud computing services or mandating that cloud providers implement customer screening protocols tied to end-user nationality.
Companies like Huawei have been developing their own AI processors for years, and tighter US restrictions have only intensified those efforts. For Nvidia, the stakes are both regulatory and financial. China was once one of its largest markets, and each successive round of export controls has trimmed that revenue stream further. If the current review leads to new restrictions on overseas cloud access, it could reduce global demand for Nvidia hardware by making it less attractive for data center operators who serve Chinese clients. Companies that facilitate international chip distribution or cloud computing services may need to invest heavily in compliance infrastructure, adding costs that eventually get passed along.