Chinese artificial intelligence companies have turned to an unusual workaround after U.S. export controls tightened around advanced Nvidia chips. Instead of shipping hardware directly, engineers are physically carrying vast amounts of training data to Malaysia in checked luggage. This method lets firms harness high‑end processing power on rented servers without tripping export restrictions.
In early March, a team of four Chinese tech workers flew from Beijing to Kuala Lumpur, each carrying 15 hard drives loaded with 80 terabytes of data. In total they moved roughly 4.8 petabytes—enough information to train multiple large language models. They split the drives among passengers so customs agents would not single out any one traveler. Upon landing, the group drove to a Malaysian data center where their company had secured 300 Nvidia AI servers. There, they fed the drives into the machines, letting powerful chips crunch the numbers needed to refine new AI algorithms.
This strategy relies on the gap in U.S. export rules. The government heavily restricts sending cutting‑edge chips overseas but places fewer limits on cloud‑based processing. By leasing servers in Malaysia, Chinese firms comply in letter with U.S. regulations while sidestepping their spirit. In practice, American‑made chips still power the AI models; only the data arrives by airplane.
To further shield their work, firms have set up complex legal structures. One company formed a Singapore‑based subsidiary to sign rental contracts with Malaysian centers. When Singapore tightened its own export rules, operators in Kuala Lumpur asked the Chinese clients to register locally. This shift helped data center owners claim they had no direct ties to Singapore companies under scrutiny. In this way, firms juggle regulations across borders to keep AI development on track.
The suitcase tactic is just one part of a wider effort by Chinese developers to access restricted tech. Major cloud platforms such as Amazon Web Services and Microsoft Azure still offer computing power under existing laws. Chinese firms can tap these services remotely to run their AI experiments. Others have struck deals with overseas brokers. In Australia, for example, entrepreneurs have arranged server farms loaded with Nvidia’s latest H100 chips for use by Beijing‑based teams. Meanwhile, some companies stockpile chips when they can and pour resources into homegrown chip design to reduce future exposure to foreign limits.
Malaysia has become a key node in this network. Investment, Trade and Industry Minister Tengku Zafrul Abdul Aziz told reporters that U.S. officials have asked Malaysia to monitor all incoming shipments that involve Nvidia hardware. In response, the government formed a special task force dedicated to overseeing data center growth and export compliance. Malaysia’s data center industry has surged, drawing over $25 billion in investments from tech giants like Nvidia, Microsoft and ByteDance within the last year and a half.
Local officials insist they will not let their facilities serve as a backdoor for smuggling chips into China. They stress that the centers must follow Malaysian law. Investigators have begun spot checks and tightened permit processes for foreign clients. Companies that fail to register properly now risk fines or suspension of service.
The episode underscores the challenge of controlling technology flows in a global market. Export rules can slow down official chip transfers, but they cannot stop data from crossing borders in passengers’ bags. As regulators scramble to plug loopholes, companies on both sides of the Pacific will keep adapting. In the end, the race to lead in AI may hinge less on physical hardware and more on the ingenuity of those who move data around the world.
