The US Department of Commerce has published a new package of export control rules aimed at cutting off Chinese companies' access to advanced Nvidia AI accelerators through foreign data centers. Until now, Washington had restricted direct supplies of equipment; however, artificial intelligence developers in China bypassed these bans in a simple way: instead of buying chips, they rented computing power from cloud providers located outside the country. The new rules extend controls to this exact access channel for the first time, turning the data center operator into a licensing target.
How Previous Bans Were Bypassed
Previously, Chinese companies did not need to purchase American accelerators directly. It was enough to sign a contract with a foreign cloud provider possessing the necessary equipment and gain access to computing within a cloud service framework. Formally, the chips remained on the territory of a third country and did not cross the border of China, allowing them to stay within existing export restrictions. Now, American authorities intend to control this method of accessing computing resources as well, closing the gray area of capacity rentals.
What Has Changed in the US Department of Commerce Rules
According to the published rules, operators of foreign data centers will be required to obtain American licenses if they provide large volumes of computing power to companies from China and other countries subject to export restrictions. At the same time, checks will have to cover not only direct clients but also organizations that actually control their activities—ruling out bypasses through front structures. Operators already providing significant computing power to Chinese customers will be given 90 days after the rules take effect to obtain the necessary permits. Restrictions will not affect a number of commercial recommendation systems, in particular product and content recommendation services.
Five Countries Under Special Control
Separate requirements are introduced for supplies of advanced AI accelerators to five countries: Singapore, Malaysia, Thailand, Indonesia, and Brazil. American authorities believe that data centers located there could be used to bypass current bans. Supplies of such processors to these jurisdictions will require licenses, but an exception is provided for companies with a US parent structure provided reporting requirements are met.
Nvidia Already Acted First
The new government requirements complement the restrictions that Nvidia itself began introducing earlier. In early November, the company demanded that its cloud partners ensure that at least 51% of the services they provide go to clients not associated with China. Thus, the US authorities now intend to regulate access to American accelerators through foreign cloud services directly, shifting measures from the corporate level to the level of state control.
Cancellation of the AI Diffusion System
Simultaneously, the US Department of Commerce is officially canceling the previous AI Diffusion export control system, adopted in January 2025. It provided for categorizing countries by access levels to advanced AI chips, but in practice it was never implemented. The new order will take effect upon publication in the US Federal Register.