AI News

US Commerce Department Plans to Restrict Chinese Firms' Remote Access to AI Compute via Overseas Data Centers

cnBetaUpdated Nov 17 00:22

AI summary

On November 16 local time, the US Commerce Department introduced an interim rule requiring companies to apply in advance for a license when exporting, re-exporting, or transferring AI chips and servers to data centers worldwide if the relevant facilities may provide compute to firms from restricted countries such as China. The applicable threshold is more than 12 million total processing performance (TPP), which according to TechCrunch estimates is roughly equivalent to the maximum compute of about 750 Nvidia H100 or H200 chips; overseas data centers that already provide compute meeting the threshold must apply for a license within 90 days after the rule takes effect. The new rule also adds Malaysia, Singapore, Indonesia, Thailand, and Brazil to the scope of advanced semiconductor export licensing, and formally rescinds the Biden administration's "AI Diffusion Rule" introduced in January 2025.

Why read this

SelectedAI score 80

The new rule extends export controls from physical chips to remote compute services at overseas data centers, and sets a threshold of 12 million TPP and a 90-day application window.

Topics

Tags

Body · AI translation

On November 16 local time, the U.S. Department of Commerce introduced an interim rule requiring companies, when exporting, re-exporting, or transferring AI chips and servers to data centers around the world, to apply in advance for a license if the relevant facilities may provide computing power to enterprises in China or other restricted countries. This means that the scope of regulation is no longer limited to whether chips and servers are actually shipped into restricted countries, but also begins to cover situations in which chips are deployed overseas and then provide computing power to restricted customers through cloud computing services.

The new rule targets the D:5 country group in the U.S. export control system. This group includes China, Russia, Iran, North Korea, and other countries subject to U.S. arms embargo restrictions. Previously, the United States had already required companies to obtain licenses when selling advanced AI chips to companies in these countries, or to companies held by enterprises from these countries. This adjustment further targets the situation of overseas data centers providing remote computing power services. In the interim rule, the U.S. Department of Commerce set the applicable threshold at more than 12 million total processing performance (TPP). According to TechCrunch's estimate, this scale of computing power is roughly equivalent to the maximum computing capability of 750 NVIDIA H100 or H200 chips. Even if Chinese companies do not sign contracts directly with data centers but instead obtain computing power indirectly through intermediaries, agents, or other third parties, they may still be subject to the new rule.

To implement regulatory requirements, chip exporters and data center operators must strengthen customer due diligence, verify customers' equity structures, actual control relationships, and business activities, and prevent enterprises from restricted countries from circumventing restrictions through complex corporate structures or intermediary institutions. If an overseas data center has already provided computing power services meeting the threshold to enterprises from relevant countries before the new rule takes effect, it must also apply for a license within 90 days after the rule takes effect, otherwise it will face subsequent export restrictions.

The new rule also adds Malaysia, Singapore, Indonesia, Thailand, and Brazil to the scope of advanced semiconductor export license requirements. The U.S. Department of Commerce believes that these countries have above-normal chip transshipment risks. However, if the chips are ultimately delivered to companies headquartered in the United States, or to companies whose ultimate parent company is headquartered in the United States, they may be exempt from this license requirement if the corresponding reporting requirements are met.

It is worth noting that the rule explicitly excludes commercial recommendation systems, such as systems on e-commerce platforms or social media that recommend goods, services, and content based on user behavior. But the scope of this exemption is quite limited: if a system can also generate new text, images, audio, video, or software content based on user input, it cannot automatically obtain an exemption merely because it also has recommendation functions.

This policy adjustment also formally rescinds the "AI Diffusion Rule" introduced by the Biden administration in January 2025. That rule originally planned to divide countries and regions into different tiers, impose limits on the export quantity of advanced AI chips, and restrict the transfer of some powerful AI models. The Trump administration announced in May 2025 that it would abolish this framework on the grounds that it imposed unnecessary restrictions on U.S. companies and allies. The U.S. Department of Commerce said this time that the new rule will replace the previous global framework with more targeted destination controls and end-use reviews.

Before the policy was introduced, the United States had already begun to pay attention to the phenomenon of Chinese AI companies obtaining computing power from NVIDIA advanced chips through overseas cloud computing facilities. Reuters previously reported that some Chinese AI companies are stepping up bookings for overseas NVIDIA chip computing power to cope with the risk that the United States may further tighten restrictions on remote access. Data centers in Southeast Asia have therefore become a focus of attention for relevant companies.

At the same time, the investigation surrounding Singapore data center operator Megaspeed has also increased regulatory pressure on Washington. The U.S. government suspects that the company purchased large quantities of NVIDIA chips and provided computing power to Chinese companies through overseas data centers in order to circumvent existing export restrictions. The United States has also considered adding it to the special entity list. However, the relevant allegations still need to be distinguished from confirmed facts.

The U.S. Department of Commerce explained in the rule that advanced computing power reaching a certain scale can not only be used to train and run large AI models, but may also be used for military and intelligence-related activities, including developing more advanced weapons, carrying out complex cyberattacks, and disrupting critical infrastructure. Therefore, the U.S. government believes it is necessary to extend regulation of physical exports of advanced chips to remote computing services provided by overseas data centers.

This rule means that U.S. AI chip export controls are further extending from traditional regulation of cross-border hardware flows to computing power services themselves. For Chinese AI companies that rely on overseas data centers to rent NVIDIA high-end GPUs, even if they do not directly import chips, they may face stricter customer scrutiny, service license restrictions, and uncertainty in computing power supply. At the same time, overseas data center operators will also bear heavier compliance responsibilities.

Source

cnBeta