Categories
Beyond

US expands AI chip restrictions beyond China

Washington targets overseas data centres to prevent Chinese companies from accessing restricted AI processors

The United States has tightened its restrictions on advanced artificial intelligence (AI) chips, moving beyond direct exports to target companies accessing American computing technology through overseas data centres. The latest measures are aimed at preventing China and other countries facing US technology restrictions from using powerful processors indirectly through cloud computing services.

The Trump administration announced the new rules on November 16, seeking to close a loophole that has allowed Chinese technology companies to access advanced American chips without purchasing or physically importing them. The move could have far-reaching consequences for Nvidia, the world’s leading supplier of AI processors, as well as international data-centre operators and cloud computing companies.

Under the interim final rule issued by the US Department of Commerce’s Bureau of Industry and Security (BIS), companies exporting, re-exporting or transferring certain advanced semiconductors and servers to overseas data centres will face additional licensing requirements if those facilities provide substantial computing capacity to businesses headquartered in restricted countries.

The regulation also introduces fresh licensing requirements for shipments of advanced AI chips to five countries: Singapore, Malaysia, Thailand, Indonesia and Brazil. Washington has identified these markets as presenting a higher-than-normal risk of technology diversion.

The measures are expected to take effect following publication in the Federal Register on November 19. Existing overseas data-centre operators covered by the new restrictions will have 90 days to obtain the necessary licences.

Until now, American export controls have primarily focused on preventing China from purchasing the most powerful US-made semiconductor chips directly. However, the rapid expansion of cloud computing has created another way for businesses to access advanced technology.

Instead of buying Nvidia processors, companies can rent computing capacity from data centres in countries where these chips are legally available. The hardware remains outside China, while customers can remotely use its processing power to train AI models and operate complex applications.

Washington believes this arrangement can undermine its existing restrictions. The latest regulation seeks to address the problem by bringing remote access to advanced computing infrastructure under closer export-control supervision.

The new rules apply to overseas data centres capable of supplying more than 12 million total processing performance (TPP), an American export-control measurement of computing capacity. This threshold is broadly equivalent to the processing power of around 750 Nvidia H100 or H200 chips.

Data-centre operators, semiconductor manufacturers and exporters will also be expected to conduct stronger checks on customers, including examining ownership structures and identifying whether restricted entities have direct or indirect control over businesses seeking computing services.

The restrictions are intended to prevent advanced AI infrastructure from being used to develop technologies with potential military, intelligence and national security applications. However, certain commercial recommendation systems used by online retailers and social media platforms are excluded from the new requirements.

Singapore, Malaysia, Thailand, Indonesia and Brazil will face additional licensing requirements for covered advanced semiconductor shipments. The decision is significant because several of these countries have emerged as important destinations for investments in data centres and AI infrastructure.

Singapore and Malaysia, in particular, have attracted substantial interest from international technology companies seeking to expand their computing capacity in Southeast Asia.

The new requirements do not amount to a complete ban on Nvidia chips entering these markets. Instead, companies will have to obtain the necessary US government approvals before exporting, re-exporting or transferring covered processors and servers.

There are exemptions for certain shipments intended for companies headquartered in the United States, including businesses whose ultimate parent companies are based there. These exemptions are subject to reporting requirements.

The broader rules also cover overseas facilities supplying substantial computing capacity to companies based in China and other restricted jurisdictions, irrespective of whether the data centres are located in the five newly identified countries.

This means the impact could extend well beyond Southeast Asia and Brazil, affecting international cloud service providers that serve customers in restricted markets.

The latest restrictions add another challenge for Nvidia, which has become one of the biggest beneficiaries of the global AI boom. Its graphics processing units (GPUs) are widely used to train large language models, develop generative AI applications and operate sophisticated computing systems.

The company has already been affected by US restrictions on sales of its advanced processors to China. The latest move could further complicate its business by placing greater scrutiny on how customers use its products after purchasing them.

The new rules also follow growing concerns over Singapore-based cloud computing provider Megaspeed International, which has come under scrutiny over allegations that overseas data-centre infrastructure was being used to provide restricted Nvidia computing capacity to Chinese businesses.

US authorities have been examining whether companies could use intermediaries and overseas computing facilities to bypass restrictions on direct chip purchases. The investigation has highlighted the difficulties of monitoring the use of advanced processors once they enter international supply chains.

Nvidia’s business model depends not only on selling processors but also on the growing demand for computing infrastructure built around them. Tighter restrictions on customers and end users could therefore affect the wider ecosystem supporting its AI chip business.

The latest decision reflects Washington’s broader effort to maintain its technological advantage over China. Advanced semiconductors have become a critical part of the competition between the two countries, particularly as artificial intelligence gains importance in defence, scientific research, industrial development and economic growth.

China has been investing heavily in its domestic semiconductor industry to reduce dependence on American technology. However, access to high-performance processors remains important for Chinese technology companies competing in the global AI market.

The new US regulation also replaces the Biden administration’s earlier AI Diffusion Rule, introduced in January 2025, which sought to establish broader country-based restrictions on advanced AI chip exports. The Trump administration had opposed that framework, arguing that it could unnecessarily restrict American technology companies and their international business opportunities.

By shifting towards controls focused on remote computing access and specific destinations, Washington is attempting to address security concerns while retaining opportunities for legitimate commercial exports.

For international data-centre operators, the immediate challenge will be to strengthen customer verification, review ownership structures and ensure compliance with the new licensing framework.

The decision signals a significant change in the way the US approaches semiconductor exports. Washington is no longer focusing only on where advanced AI chips are shipped. It is increasingly concerned about who ultimately gets to use their computing power.