The United States’ efforts to restrict China’s access to advanced artificial intelligence chips may be far less effective than policymakers hope, according to data-center executives and industry experts who say Chinese technology companies can still train powerful AI models overseas.
Despite years of export restrictions targeting high-performance graphics processing units (GPUs) produced by Nvidia, Chinese firms are increasingly turning to data centers in Southeast Asia to access the computing power required for cutting-edge artificial intelligence development.
Experts say that while Washington has attempted to block shipments of advanced chips into China, the rules do not prevent Chinese companies from exporting data abroad for training and then importing the finished AI models back into the country.
Gary Wojtaszek, a director at Chinese data-center developer GDS Holdings, says the global nature of digital infrastructure makes strict technology containment extremely difficult.
“The US chip export rules are pretty nebulous,” Wojtaszek said in an interview with Asia Times in Dublin. “In China, you can’t import chips, but you can export your data to train your model. Then you import it back.”
“Water always finds the level playing field,” he added. “People may try to control or slow a trend, but it will eventually find a natural balance.”
Industry analysts say Southeast Asia has quietly become a critical hub in the global AI ecosystem.
Chinese companies can either rent computing power or install their own server cabinets in foreign data centers equipped with high-end Nvidia chips. These facilities allow them to train large language models outside China before deploying the finished software domestically.
Data-center infrastructure across Singapore, Malaysia and Indonesia has expanded rapidly as demand for AI computing surges worldwide.
A subsidiary of GDS Holdings, now known as DayOne Data Centers, operates facilities across several major technology hubs including Singapore, Johor in Malaysia, Batam in Indonesia, Greater Bangkok, Hong Kong, Tokyo and Finland.
The company was created in 2022 as an international arm of GDS to develop data centers beyond China. After several rounds of fundraising, the parent company retains a 25% stake.
Wojtaszek, who sits on the board of DayOne, said his view of the industry is shaped by years of global experience.
Before joining GDS, he served as president and chief executive of CyrusOne, a real-estate investment trust that builds and operates hyperscale data centers worldwide.
CyrusOne invested about US$100 million in GDS in 2017 to acquire a stake in the company, deepening ties between Western and Chinese data-center ecosystems.
Today Wojtaszek serves on multiple boards across the global industry, including advisory roles for operators in the United States, India, Europe and Brazil.
The intensifying technological rivalry between the US and China has increasingly centered on artificial intelligence.
Washington currently holds a major advantage in semiconductor technology and advanced graphics processors produced by companies such as Nvidia.
However, Wojtaszek said China is rapidly narrowing the gap.
“The US has cutting-edge GPUs and large language model algorithms,” he said. “But China is catching up quickly.”
One of China’s strengths lies in the massive volume of data available for training AI systems. With a population exceeding 1.4 billion and a huge digital economy, Chinese companies can access enormous data pools that can accelerate machine-learning development.
By contrast, US technology firms often face tighter regulatory restrictions related to privacy and data protection.
Energy supply is another factor shaping the AI competition. Running advanced AI models requires enormous electricity consumption, and China’s vast power-generation capacity could provide an advantage in building massive data centers.
The technology restrictions at the heart of this rivalry began under the administration of Joe Biden and have continued under Donald Trump.
In October 2022, the Biden administration banned exports to China of Nvidia’s advanced H100 and A100 chips, which are widely used to train large AI models.
The restrictions were expanded a year later to include slightly downgraded versions designed specifically for China, including the H800 and A800 chips.
Washington hoped the controls would slow Beijing’s ability to build powerful AI systems with potential military applications.
In early 2025, the Biden administration went further by proposing an “AI Diffusion Rule,” which sought to control the global distribution of advanced AI chips and computing power.
The rule created a three-tier system governing access to American AI technology.
Tier-one countries – including close US allies such as Japan and most European states – could freely purchase advanced Nvidia chips.
Tier-two countries, which included Singapore, Malaysia, India and the United Arab Emirates, faced tighter controls and required special licenses or validated end-user approvals.
Tier-three countries, including China, Russia, Iran and North Korea, were considered high-risk destinations for advanced AI chips due to potential military use.
However, the policy ultimately proved difficult to implement. The Trump administration later withdrew the rule after concluding its scope was too broad and enforcement would be challenging.
Instead, Washington adopted a narrower policy that allows China to import certain Nvidia chips, including the H200 model, but still blocks access to the company’s most advanced Blackwell-generation GPUs.
Despite these restrictions, industry insiders say Chinese firms have found practical workarounds.
One example reported by The Wall Street Journal illustrated how companies can exploit regulatory gaps.
According to the newspaper, four Chinese engineers traveled from Beijing to Kuala Lumpur carrying hard drives filled with AI training data. They rented roughly 300 Nvidia servers at a Malaysian data center to process the information.
Because the chips never entered China and the data was processed abroad, the operation did not appear to violate US export laws.
The case highlighted how global cloud infrastructure can circumvent hardware restrictions.
A separate report by the Financial Times said major Chinese technology companies, including Alibaba and ByteDance, have been training AI models in Southeast Asian facilities equipped with Nvidia chips.
Once the training process is completed, the resulting models can be transferred digitally back to China.
For many analysts, this demonstrates a fundamental weakness in export-control strategies focused solely on hardware.
At the same time, Beijing has encouraged technology firms to prioritize domestic alternatives to American chips.
Chinese authorities have asked local companies to consider processors produced by Huawei Technologies and other domestic semiconductor developers before relying on imported hardware.
The push reflects China’s long-term strategy to reduce dependence on foreign technology in strategic sectors.
One example of Chinese innovation is the emerging AI system developed by the company DeepSeek, which industry observers say delivers competitive performance while using less computing power.
Wojtaszek described the system as a sign of China’s ability to innovate under pressure.
“DeepSeek is ingenious,” he said. “They did something interesting at significantly lower cost and with lower power.”
He added that China has historically excelled at refining and improving existing technologies.
“They will continue to reinvent other people’s inventions and make improvements to them,” he said.
Ironically, the geopolitical struggle over AI chips has done little to slow Nvidia’s commercial success.
The company reported revenue of US$215.9 billion for the fiscal year ending January 25, 2026 – a 65% increase compared with the previous year.
Singapore alone accounted for about one-fifth of Nvidia’s revenue, making it the company’s second-largest market after the United States.
Analysts say that even if direct exports to China remain restricted, demand from international cloud providers and overseas data centers continues to fuel Nvidia’s growth.
Concerned about these loopholes, US lawmakers are now considering new legislation that would expand export controls beyond physical chip shipments.
In January, the US House of Representatives passed the Remote Access Security Act, a bipartisan proposal designed to prevent foreign adversaries from accessing sensitive technologies through remote cloud computing services.
If enacted into law, the measure would allow Washington to regulate access to AI chips even when they are used through overseas data centers rather than physically shipped to another country.
Such a move could significantly reshape the global AI infrastructure landscape by bringing remote computing under export-control rules.
Meanwhile, Southeast Asia’s rise as an AI data-center hub has been reinforced by a decline in Hong Kong’s position as a regional connectivity center.
According to Doug Adams, chief executive of NTT Global Data Centers, many international companies now prefer to host their computing infrastructure in Southeast Asian locations rather than Hong Kong.
“Hong Kong is a challenging market,” Adams said. “Most customers in that region, given the choice, prefer to be in Singapore and other parts of that area.”
Cities such as Johor Bahru in Malaysia and technology hubs in Indonesia and Thailand are seeing rapid expansion in data-center construction.
Changes in global telecommunications networks have also contributed to the shift.
Several US technology companies have canceled or modified plans to connect new undersea fiber-optic cables directly to Hong Kong following national-security concerns raised in Washington.
Projects such as the Pacific Light Cable Network were forced to reroute connections toward alternative hubs including Taiwan, the Philippines and Singapore.
At the same time, Hong Kong faces additional obstacles in adopting generative AI services.
Major Western AI platforms – including ChatGPT, Gemini and Claude – are not officially available for consumer use in the city, limiting its appeal as a technology development hub.
Taken together, these trends illustrate how the global AI race is reshaping digital infrastructure.
While the United States continues to lead in semiconductor technology, China is developing new strategies to maintain progress in artificial intelligence despite restrictions.
Southeast Asia has emerged as a crucial bridge in this competition, offering the computing power and connectivity that allow companies to bypass geopolitical barriers.
