US moves to restrict China's remote access to AI chips after Nvidia row
The US will require companies in five countries, including Singapore and Malaysia, to obtain export licences for advanced AI chips under new rules taking effect from November 19
)
The new rules are scheduled to take effect from November 19. (Image: Bloomberg)
Listen to This Article
The United States (US) on Monday introduced fresh restrictions to prevent Chinese companies from accessing advanced artificial intelligence (AI) chips through data centres located in other countries. The move comes amid growing concerns in Washington that existing export controls are being bypassed through overseas computing facilities, according to a report by Nikkei Asia.
The new rules, issued by the US Commerce Department's Bureau of Industry and Security (BIS), require companies in Singapore, Thailand, Malaysia, Indonesia and Brazil to obtain US export licences for advanced chips. The department identified these countries as having a higher risk of such chips being illegally diverted to China, the news report said.
The regulations also target overseas data centres that provide large amounts of computing capacity remotely to companies in countries facing US chip restrictions, including China. The new rules are scheduled to take effect from November 19.
"The rule is designed to ensure that existing restrictions on access to sensitive American technology cannot be circumvented through remote computing arrangements involving data centres in third countries," BIS said in a statement.
Why is the US targeting overseas data centres?
The latest move follows reports that the Commerce Department failed to prevent the illegal movement of advanced Nvidia AI chips into China. There have also been concerns that Chinese companies are using data centres in countries such as Singapore to access computing power that they would otherwise be unable to obtain directly.
Also Read
Since Washington introduced export controls on advanced chips to China in October 2022, data centres in Southeast Asia have increasingly rented computing capacity to Chinese customers. This allows Chinese companies to run AI models on servers located outside mainland China, the news report said.
The news report, citing data from the US-based think tank Institute for Progress, says remote access could account for a quarter of China's total AI computing capacity by the end of 2026. The share could increase to one-third by the end of 2027.
Nvidia controversy puts export controls under scrutiny
The issue has drawn greater attention following allegations involving former Nvidia Chief Executive Officer (CEO) Jensen Huang and the possible diversion of advanced chips to China.
Huang has repeatedly argued that allowing Nvidia to sell more chips in China would benefit the US. However, reports linking him to alleged Chinese spy Alice Huang, a former executive at Singapore-based data centre company Megaspeed, have raised questions about whether Nvidia's advanced processors reached China through illegal exports or remote access arrangements in Southeast Asia, the news report said.
US Commerce Secretary Howard Lutnick has also faced criticism over the department's decision not to place Megaspeed on the Entity List, the US government's trade blacklist. The company is suspected of helping divert Nvidia chips to China.
Will DePue, a former technical staff member and researcher at OpenAI, said in a post on X last week that his research found Chinese companies DeepSeek, Moonshot, Minimax and Ant Group had developed their latest AI models using Nvidia's advanced Blackwell chips. These processors are currently barred from export to China.
Chinese AI companies face another possible restriction
Apart from access to advanced chips, Chinese AI companies are also concerned that Washington could restrict or ban Chinese open-weight and open-source AI models.
A source at a leading Chinese AI laboratory told Nikkei Asia that developers are worried about such a move.
US policymakers have discussed restricting Chinese AI models on national security grounds since Moonshot launched its Kimi K3 model in July. The model was seen as a major step towards narrowing the gap between open-weight AI systems and leading proprietary models developed by companies such as Anthropic and OpenAI, while offering services at a much lower cost.
Huang has been among the prominent voices opposing restrictions on open-weight models. Nvidia has also signed open letters urging Washington not to impose such curbs and has supported industry efforts to promote open-source AI, the news report said.
Chinese open-weight models have gained popularity among US businesses, particularly small and medium-sized companies, because they are relatively inexpensive.
Although Nvidia has lost access to China's domestic chip market, the company could still benefit from the wider use of Chinese AI models in the US. Businesses deploying these models would continue to need computing infrastructure.
Completely removing Chinese AI models from the internet could be difficult. However, the Chinese AI laboratory source said US lawmakers could consider making it illegal to download and deploy these models, which could discourage their use, the news report said.
Meanwhile, Chinese companies are gradually shifting towards domestic semiconductor solutions. DeepSeek is among the companies pushing ahead with this transition. It plans to use Huawei's Ascend 950 chips as the main computing platform to train its next-generation V5 model, with the project targeted for 2027, the news report said.
Other large Chinese technology companies, however, continue to depend heavily on Nvidia's CUDA software ecosystem for mainstream AI model training.
More From This Section
Don't miss the most important news and views of the day. Get them on our Telegram channel
First Published: Nov 17 2026 | 11:21 AM IST
