To bypass strict US restrictions on advanced AI chips, Chinese companies are relying on innovative workarounds that include processing data from overseas. According to a recent Wall Street Journal report, four Chinese engineers moved from Beijing to Malaysia in early March, carrying suitcases with 15 hard drives containing 80 terabytes of data to train AI models. At a data center in Malaysia, engineers used around 300 servers equipped with NVIDIA's advanced chips to develop AI models and later brought them back to China.
Smuggling AI hardware
Since 2022, the US has been strengthening export controls of high-end AI chips to China, citing national security concerns. According to the WSJ report, these restrictions limit Chinese companies' access to cutting-edge American technology and encourage them to explore alternatives. Some have replaced American chips with household chips, while others have smuggled AI hardware in third countries. However, increasing US pressures have made smuggling more difficult, and Chinese companies are now processing non-Chinese data in regions such as Southeast Asia and the Middle East. “This was something we were consistently concerned about,” Shea Kendler, former director of export management at the Department of Commerce under the Biden administration, told WSJ, referring to Chinese companies' remote access to US AI chips.This process involves a layer of intermediaries, obscuring whether US regulations are being violated. The Biden administration proposed a country-specific cap on American chip purchases to curb such activities, but the Trump administration scrapped them in May, citing unnecessary regulatory burdens on US companies like Nvidia. Instead, they issued guidance urging businesses to prevent chips from being used to train Chinese AI models.
Opinion survey
Is Southeast Asia a strategic hub for AI development?
How and why China uses Southeast Asian countries as bases
In Malaysia, Chinese companies' operations required thorough planning. Engineers reportedly optimized dataset optimization in China for over eight weeks. Last July, the company worked through a Singapore subsidiary, but later registered a Malaysian organization to avoid scrutiny after NVIDIA and its vendors strengthened end-user audits. To avoid suspicion at Malaysian customs, engineers distributed hard drives across four suitcases and shifted to bundled ones the previous year. They returned to China with hundreds of gigabytes of model parameters leading to the output of the AI system.Southeast Asia has been reported to be emerging as a hub for such activities, and data centers are expanding rapidly. Jones Langlasar estimates that around 2,000 megawatts of data center capacity in Singapore, Malaysia, Thailand and Indonesia will coincide with Europe's largest market. Malaysian AI chip imports from Taiwan surged to $3.4 billion in March and April, totaling over 2024. Meanwhile, the Middle East is also becoming a destination for Chinese AI developers, with Nvidia recently announced significant chip sales to Saudi Arabia, Qatar and the UAE.These operations underscore the challenge of enforcing US export controls as Chinese companies leverage global data centers to access American technology. While Southeast Asian authorities are cracking down on transportation like in Singapore, the region's booming infrastructure continues to attract clients from both Western and Chinese, testing the limits of regulatory oversight.
