US cracks down on AI chip exports as China faces tougher restrictions

US tightens restrictions on advanced AI chip exports to curb China's access

The US government is tightening restrictions on advanced AI chip exports to several countries, including Singapore and Malaysia, in an effort to prevent China from obtaining access to high-performance processors through indirect channels.

Under new rules announced by the US Commerce Department on Monday, exports of advanced chips to Malaysia, Singapore, Indonesia, Thailand, and Brazil will require government licences. Data centre operators using US-made AI processors will also face stricter requirements to verify their customers, targeting Chinese companies that rent computing capacity overseas rather than importing restricted hardware directly.

Washington has already banned exports of powerful AI chips to China, including NVIDIA products, citing concerns that the technology could strengthen China's AI capabilities and military applications. However, authorities believe the restrictions have been circumvented through smuggling networks and overseas data centres.

A US indictment previously described an alleged scheme in which serial-number labels were transferred from genuine NVIDIA-equipped servers to dummy machines. The deception was reportedly intended to conceal the shipment of the real equipment to China. Another concern involves Chinese customers renting computing power from data centres in countries such as Malaysia. This allows them to access servers equipped with NVIDIA processors without directly purchasing or importing the restricted chips.

Nvidia RTX Pro 6000 Blackwell Server Edition. A black and gold server-style graphics card.

The controls primarily cover items classified under Export Control Classification Numbers (ECCNs), which includes Nvidia’s H100/H200-class chips and newer generations (e.g., Blackwell), AMD equivalents (e.g., MI300/MI325 series and above), and systems or servers built around them. The 12 million aggregate TPP threshold roughly equates to capacity on the order of around 750 Nvidia H100/H200 GPUs.

Singapore-based Megaspeed has also attracted scrutiny. The company, reportedly a major NVIDIA customer, is suspected of diverting advanced technology to China. The investigation has gained further attention following reports about NVIDIA's former CEO Jensen Huang and an alleged relationship with a senior Megaspeed executive. US security officials have reportedly raised concerns about the woman's possible connections to Chinese intelligence, although these remain allegations.

This is a significant tightening that extends U.S. controls from physical hardware location to effective use by restricted parties, driven by both long-standing circumvention concerns and recent political pressure. By extending licensing requirements beyond China, Washington is attempting to close the overseas distribution and cloud-computing loopholes that have weakened its existing restrictions.

As always, for the latest news on AI-powered rendering technologies, GeForce RTX graphics cards, RTX Spark superchip, and developments across the gaming industry, be sure to follow our dedicated hardware coverage.

manhkbrady

manhkbrady

1329 Articles

A writer, and a full-time Tetris min-maxing player. Do you know that rhythm games are a form of human benchmarking?

DLCompare Web Extension Body Bottom
Discord Invite