China continues to find ways to access advanced Nvidia chips despite strict US export restrictions, raising concerns in Washington over how the technology is being used to strengthen the country's artificial intelligence industry. From overseas data centres to suspected smuggling networks, Chinese companies are exploring alternative routes to obtain powerful chips needed to train and operate advanced AI models.

The issue has attracted renewed attention following reports involving Nvidia chief executive Jensen Huang and a Chinese executive at a Singapore-based company suspected by the United States of helping smuggle AI microchips into China. The allegations have brought greater scrutiny to the complex supply chains supporting China's rapidly expanding AI sector.

Nvidia's graphics processing units, commonly known as GPUs, are essential components in modern AI development. Originally designed to process graphics for video games, these chips can handle enormous amounts of data simultaneously, making them particularly effective for training and running artificial intelligence systems.

The launch of ChatGPT in 2022 triggered a global race to develop AI technology, driving massive demand for advanced processors. Nvidia emerged as a major beneficiary, with its market valuation reaching around $4.5 trillion, placing it among the world's most valuable companies.

US restrictions on advanced chips

Washington has restricted exports of Nvidia's most advanced Blackwell and Rubin GPUs to China, citing national security concerns and the potential military applications of advanced artificial intelligence.

The restrictions are also intended to maintain the technological advantage of US-based AI companies. However, Nvidia CEO Jensen Huang has repeatedly urged Washington to reconsider the controls, warning that limiting access could encourage China to develop its own competitive chip industry.

The United States recently eased restrictions on Nvidia's less powerful H200 chips, allowing certain sales under conditions that include a 25 percent revenue share for the US government. However, approval requirements have made shipments difficult, while Beijing has reportedly shown limited interest in purchasing the processors.

How China accesses Nvidia chips

Despite the restrictions, Chinese technology companies are reportedly continuing to use advanced Nvidia processors. Recent AI model launches by Chinese firms such as DeepSeek, Moonshot AI and Z.ai have intensified competition with American technology companies.

According to Lian Jye Su, chief analyst at Omdia, Chinese companies are still using cutting-edge Nvidia chips to train their AI models despite export controls.

One method involves accessing computing power through overseas data centres equipped with Nvidia GPUs. Chinese companies can remotely rent computing resources located in third countries, particularly in Southeast Asia.

Stephen Wu, founder of investment fund Carthage Capital, explained that US export controls generally regulate where chips are shipped rather than who uses their computing power. For example, GPUs located in Malaysia's Johor region could potentially serve customers in China's Shenzhen through remote access.

US officials have also raised concerns about Chinese AI companies accessing restricted processors abroad. White House adviser Michael Kratsios previously said that Moonshot AI had accessed high-end Nvidia chips in Thailand.

Smuggling is another suspected route. Technology publication The Information reported that DeepSeek had developed a new AI model using thousands of chips dismantled overseas and transported into China. Nvidia said it had seen no evidence supporting the alleged activity.

China's growing domestic chip industry

Although Nvidia remains a leader in advanced AI processors, Chinese manufacturers are making rapid progress. Huawei has emerged as the country's strongest domestic competitor, although its technology still trails Nvidia's most advanced products.

Huawei announced that its next-generation AI chip would be ready in the first quarter of 2027, three quarters earlier than originally planned. DeepSeek is also reportedly planning to train its next major AI model using Huawei processors.

Industry analysts say Chinese-made chips are improving quickly, although they still require more engineering work and longer training times compared with Nvidia products.

However, lower electricity costs and government subsidies could help Chinese companies overcome some of these disadvantages by using larger numbers of domestic processors.

As Washington tightens export controls and Beijing strengthens its domestic semiconductor industry, access to advanced AI chips is becoming a major factor in the growing technological competition between the two countries.