The United States plans to implement stricter regulations on AI GPU exports by the end of the month, aiming to close loopholes that have allowed Chinese entities to acquire restricted hardware through third-party countries.
The new measures, developed by Commerce Secretary Gina Raimondo and National Security Advisor Jake Sullivan, will establish country-specific quotas for GPU exports and introduce a comprehensive global licensing system with mandatory reporting requirements.
This regulatory update targets major transshipment hubs that currently operate outside U.S. export restrictions. Intelligence reports indicate Chinese buyers have successfully obtained advanced Nvidia GPUs, including the latest H200 models, through elaborate smuggling networks spanning the Middle East, Malaysia, Singapore, and Taiwan.
Despite Nvidia's compliance with existing regulations by not directly selling to Chinese entities, the current system has proven vulnerable. Resellers have exploited legitimate supply chains to divert hardware through intermediaries, ultimately reaching restricted Chinese buyers. By August, over 70 vendors in China were openly advertising these controlled processors online.
The proposed quota system aims to maintain adequate GPU supply for legitimate AI development in authorized countries while preventing re-export to restricted nations. Countries receiving these advanced chips will likely prioritize domestic customers over questionable resale opportunities, given growing global demand for AI infrastructure.
The Department of Justice has already launched investigations into potential export violations, including a probe into Supermicro's alleged use of Taiwan-based companies to redirect AI servers to China and Russia.
These enhanced controls reflect Washington's determination to protect sensitive AI technology and maintain effective export restrictions, particularly as Chinese entities continue seeking ways to acquire advanced computing capabilities through alternative channels.
Note: Only one link was inserted as it was the only one contextually relevant to the article's topic of AI technology regulation and control.