U.S. Department of Commerce reportedly plans to restrict Chinese companies’ remote use of Nvidia Blackwell and other advanced chips
The U.S. Department of Commerce’s Bureau of Industry and Security (BIS) is in the final stages of preparing new export control rules aimed at restricting Chinese artificial intelligence companies from remotely using Nvidia Blackwell and the newer Rubin, among other controlled advanced chips, through data centers in third countries. The rules could be announced as early as this month and are still undergoing interagency coordination. Related reports, citing informed sources, said BIS plans to include Singapore, Malaysia, Thailand, Indonesia, Norway, Brazil and South Africa under a stricter chip export licensing regime.
Under the proposed rules, when data centers in these countries operated by non-U.S. companies import Nvidia Blackwell-level or more advanced controlled AI chips, they would need to apply to the United States for licenses, disclose the equipment recipient and the final use, and accept ongoing oversight. Data centers operated by U.S. companies would be conditionally exempt: as long as they prohibit remote use by countries subject to the U.S. arms embargo, such as China and Russia, they would not need to apply separately for import licenses, but would still have to report the identities of relevant country customers and the scale of computing power they use. At present, some countries, including Saudi Arabia and Vietnam, already need U.S. approval to import advanced AI chips and are subject to conditions such as restrictions on remote access.
Previously, direct U.S. exports of Blackwell and Rubin chips to China were largely restricted, and Chinese companies mainly obtained advanced GPUs by renting computing power through overseas data centers. Nvidia recently required overseas cloud service providers to allocate at least 51% of their services to non-Chinese customers; under pressure from the U.S. government, Oracle canceled a Southeast Asia computing-power leasing contract with Tencent worth about $7 billion and involving about 100,000 advanced chips. Chinese AI companies and cloud service providers are stepping up reservations for overseas Nvidia computing power to respond to possible policy tightening.
Original source:
Title: US Department of Commerce Moves to Block China's AI Chip 'Southeast Asia Detour'
Publisher: SBS News (citing The Information)
Publication date: 2026-11-11
URL: news.sbs.co.kr/english/article.do
Nature of source: Media report
Verification status: Full text read for this report (2026-11-11); the content is consistent with multiple English media summaries of The Information
Supporting content: BIS preparing new rules to restrict Chinese companies from remotely using Blackwell and other advanced chips through third-country data centers; plans to include seven countries such as Singapore under a stricter licensing regime; U.S.-operated data centers to receive conditional exemptions; rules could be announced as early as this month, among other key facts
ABAB AI Insight
U.S. controls on advanced AI chips for China began in 2022, focusing on restricting physical exports. In December 2025, the Trump administration opened conditional sales of H200 chips to China, with a 25% tariff and U.S. testing requirements, but explicitly excluded Blackwell and Rubin. In May 2026, BIS clarified that licensing requirements apply globally to entities headquartered in China, closing the physical-channel workaround of purchasing through subsidiaries. Remote computing-power leasing has long remained a regulatory gray area: the chips stay in data centers in third countries, and Chinese companies use them by logging in over the network, which does not constitute a traditional “export.” Entities such as Megaspeed, registered in Singapore and operating in Malaysia, have been accused of providing a channel for this model and are linked to the recent Jensen Huang incident.
On the capital and business side, Chinese cloud providers and AI labs have secured Nvidia computing power through data centers in Southeast Asia and Europe for model training. Nvidia itself has tried to reduce criticism in Congress by verbally requiring customers to derive at least 51% of their revenue from non-Chinese customers; Oracle’s cancellation of a large lease with Tencent shows that U.S. cloud providers are already under administrative pressure. The new rules will extend licensing and customer disclosure requirements to overseas data centers operated by non-U.S. companies, directly raising the compliance cost and visibility of Chinese companies’ access to Blackwell/Rubin-class computing power, while preserving a conditional channel for U.S.-operated data centers.
Compared with earlier controls focused only on physical exports, the proposed rules expand the regulatory target from where chips are shipped to who uses them remotely. The House of Representatives has passed the Remote Access Security Act, seeking to give BIS statutory authority over remote access, but the Senate has not yet completed legislation. If the administrative rules take effect, they could constrain specific countries’ data centers’ imports and usage disclosures before legislation is completed, creating a “licensing + ongoing supervision” mechanism. This stands in contrast to the limited opening for H200: the path for high-end chips would be further tightened, while the mid-range chip channel would still retain a conditional window.
This change directly compresses the space for Chinese AI companies to obtain top-tier computing power through overseas leasing. For non-U.S. cloud service providers, the licensing threshold for importing controlled chips would rise, and they would need to bear disclosure and ongoing oversight obligations; for U.S. cloud service providers, a conditional exemption would remain, but reporting obligations for customers and the scale of computing power used would increase. Chinese companies’ rush to reserve overseas computing power reflects their expectation that the channel will narrow. The formal text of the rules, the specific implementation of exemption conditions, and the coordination with congressional legislation will determine the actual degree of contraction in the overseas computing-power leasing market and the global distribution pattern of Nvidia’s high-end products.
ABAB News · Cognitive Law
When export controls expand from restricting the physical shipment of chips to restricting remote access to computing power, the licensing threshold for data centers in third countries rises accordingly.
Data centers operated in the United States may receive conditional exemptions, while non-U.S. operators must apply for licenses to import chips such as Blackwell and accept ongoing oversight.
Chinese companies’ rush to reserve overseas computing power reflects their expectation that remote-access channels will narrow because of the new rules.