Why US is tightening AI chip restrictions amid the Nvidia-China controversy

Wait 5 sec.

The Donald Trump administration has announced sweeping new restrictions aimed at preventing Chinese artificial intelligence (AI) companies from accessing advanced American chips through data centres located outside China, as Washington faces growing questions over the effectiveness of its technology export controls.The new regulations, announced by the US Commerce Department on Monday (November 16), seek to close a significant loophole in America’s restrictions on advanced AI chips. While Washington has largely restricted the direct shipment of its most powerful processors to China, Chinese companies have been able to access the same computing capabilities by renting servers in countries such as Malaysia and Singapore.The move comes amid a widening controversy surrounding chipmaker Nvidia, following the resignation of its chief executive Jensen Huang over revelations of his relationship with a woman suspected by US authorities of having links to Chinese intelligence. The woman, Alice Huang, was previously associated with Singapore-based cloud computing company Megaspeed, an Nvidia partner under investigation over possible violations of US chip export controls.The restrictions also come as fresh research has raised questions over whether China’s leading AI developers, including DeepSeek and Moonshot AI, have already gained access to Nvidia’s export-restricted Blackwell chips.New restrictions target overseas data centres and remote computingUnder the new interim final rule, issued by the Commerce Department’s Bureau of Industry and Security (BIS), overseas data centres that provide significant computing capacity to companies headquartered in China and other countries of concern will require US government authorisation to receive certain advanced American chips.Importantly, the restrictions also cover companies whose ultimate parent organisation is based in these countries, seeking to prevent Chinese entities from circumventing the rules through subsidiaries registered elsewhere.The rule sets a threshold of more than 12 million aggregate Total Processing Performance (TPP), a measure of computing capability, for restricted remote access. It also covers computing capacity distributed across interconnected data centres.Story continues below this adAlso Read | Did China use export-banned Nvidia chips to train its most advanced AI models?Separately, Washington has introduced additional licensing requirements for exports of advanced semiconductors to five countries — Malaysia, Singapore, Indonesia, Thailand and Brazil — which it identifies as presenting a higher risk of chip diversion. Certain shipments to US-headquartered companies are exempt, subject to reporting requirements.Existing data centres providing restricted computing services will have 90 days from the rule’s effective date to obtain the necessary licences to continue doing so. The regulations exempt commercial recommendation systems, including those used by e-commerce and social media companies.The rule also formally rescinds the Biden administration’s broader AI Diffusion Rule, which was never enforced, replacing it with more targeted restrictions.Washington’s concerns extend beyond the physical movement of chipsFor several years, the US has sought to limit China’s access to advanced semiconductors, arguing that the same powerful processors used to develop commercial AI applications could also strengthen Beijing’s military and intelligence capabilities.Story continues below this adThese chips are critical to training increasingly sophisticated AI models, and Washington has sought to preserve America’s technological advantage by restricting China’s access to the most advanced hardware.Also Read | Jensen Huang resigns: Why his exit from Nvidia will be a test for the AI boomHowever, restrictions on the physical shipment of processors have created another challenge for US authorities. Chinese companies can potentially access powerful American chips without importing them, simply by purchasing computing services from overseas data centres.The Commerce Department said the new controls were necessary because advanced AI systems could be used to develop sophisticated weapons, conduct cyber-espionage, disrupt critical systems and support military operations.Nvidia’s Megaspeed controversy has intensified scrutinyThe announcement comes against the backdrop of growing scrutiny of Nvidia and the Commerce Department over their handling of Megaspeed, a Singapore-based provider of computing services with operations linked to Southeast Asia.Story continues below this adMegaspeed has faced allegations that it helped Chinese companies access restricted Nvidia processors, potentially undermining Washington’s export controls.The controversy intensified following revelations about Jensen Huang’s relationship with Alice Huang, who previously worked with Megaspeed and was suspected by US intelligence officials of having connections to Beijing.Also Read | Jensen Huang-Alice Huang affair: How the Nvidia scandal became a US security concernJensen Huang resigned as Nvidia’s CEO on November 12. However, US authorities have not concluded that Alice Huang was engaged in espionage, and Beijing has rejected the allegations.Singaporean authorities have separately investigated Megaspeed since February 2025 over possible false declarations relating to server purchases. On Monday, the US Commerce and Justice departments disclosed that Washington had initiated its own investigation into Megaspeed in 2024 over potential export-control violations.Story continues below this adMegaspeed has maintained that it complies with applicable laws, including US export-control requirements.The episode has also triggered political scrutiny in Washington, with lawmakers questioning whether the administration had done enough to prevent American technology from reaching China.Fresh evidence raises doubts over the effectiveness of export controlsThe restrictions follow research published last week by Unemployment Labs, which found evidence suggesting that several advanced Chinese AI models were developed using Nvidia’s export-restricted Blackwell processors.The researchers identified numerical fingerprints in models developed by DeepSeek, Moonshot AI, MiniMax and Ant Group that were consistent with the use of Blackwell hardware.Story continues below this adWhile the findings do not conclusively establish how or where the chips were accessed, they add to concerns that American export restrictions have not entirely prevented Chinese companies from obtaining advanced computing capabilities.Separately, research organisation Epoch AI has estimated that around two million advanced AI chips entered China during 2024 and 2025 through arrangements designed to conceal their final destination.