Major Chinese technology firms, including Alibaba and ByteDance, have reportedly shifted their AI model training operations to data centres in Southeast Asia. This strategic move comes after U.S. export restrictions, imposed in April 2025, limited China’s access to Nvidia’s advanced AI chips—particularly the H20—to curb China’s progress in developing large-scale AI models.
The offshore data centres are leased from non-Chinese entities, which — according to industry insiders — allows Chinese firms to legally bypass export restrictions while still leveraging high-performance computing hardware. This approach has reportedly become more common since the export ban, with companies increasingly relying on overseas infrastructure to train new large language models.
There are exceptions: for instance, the AI firm DeepSeek had stockpiled a substantial supply of Nvidia chips before the ban took effect, allowing it to continue training models domestically. DeepSeek is also collaborating with local chip-makers such as Huawei to develop domestically produced AI chips — part of a broader long-term effort by Chinese firms to reduce reliance on foreign hardware.
In broader terms, this trend underscores how geopolitical tensions and export controls are reshaping the global AI development landscape. It shows that technology constraints — even when imposed via policy — can drive firms to find creative workarounds. For observers of AI and geopolitics, it highlights the growing interdependence of global supply chains, the strategic value of computing hardware, and how multinational operations are adapting to shifting regulatory environments.
