China’s leading technology firms are quietly moving some of their most advanced AI development abroad, turning Southeast Asia into a major new hub for high-performance computing.
According to a report from the Financial Times, citing people familiar with the matter, noted that companies like Alibaba and ByteDance are leasing GPU power from foreign-owned data centers in Singapore and Malaysia, where Nvidia’s top accelerators remain available.
How China’s AI Firms Bypass U.S. Chip Restrictions
The strategy hinges on a key loophole in U.S. export rules: Washington forbids Nvidia from selling its highest-performance GPUs directly to China, including the H100 and the China-specific H20.
However, it does not prohibit non-Chinese data center operators in countries such as Singapore and Malaysia from purchasing those same chips and leasing access as cloud services to Chinese customers.
As one Singapore data center operator told the FT, hosting Chinese AI workloads is “an obvious choice” as long as current regulations allow it.
Further, training demand in offshore locations has increased since April, when the United States tightened controls on Nvidia’s H20 chip, the report said. A now-rescinded “AI diffusion rule,” which would have restricted the leasing of advanced computing to Chinese customers, was withdrawn earlier this year.
With the rule gone, leasing foreign compute has become a primary mechanism for Chinese AI companies seeking cutting-edge hardware without violating export controls.
Alibaba, ByteDance Push Ahead
Alibaba’s Qwen models and ByteDance’s Doubao systems — both of which have climbed into top-tier global LLM rankings — have been trained partly on these overseas clusters, sources told the FT.
These clusters handle the enormous bandwidth and compute required to train models with hundreds of billions of parameters.
Once models are built, companies increasingly run them inside China on domestic chips, which are seen as more cost-efficient and less geopolitically risky for inference workloads. Companies like Huawei and other local semiconductor firms are accelerating the development of homegrown accelerators to support this shift.
DeepSeek Breaks The Trend
One notable exception to the trend is DeepSeek, a fast-rising AI firm known for efficiency-focused models. The company reportedly stockpiled Nvidia GPUs well before the latest export bans, giving it enough local hardware to train inside China.
DeepSeek is also collaborating closely with Huawei, which has installed engineers inside the company to tune hardware and software for future domestically powered training runs.
Southeast Asia Becomes The New Compute Frontier
Southeast Asia — especially Singapore and Malaysia — has emerged as a natural destination for this offshore strategy, as these facilities remain owned and controlled by local or international operators.
Since Chinese companies typically do not own these centers, they instead sign long-term lease agreements with local or international operators who maintain legal control of the chips, keeping the arrangements compliant with U.S. rules. This allows them to legally use Nvidia GPUs located abroad — even as China bans foreign AI chips from its own state-funded data centers.
Global Implications
The offshore move allows Chinese companies to continue developing large-scale AI systems, but it also increases exposure to policy changes and operational risks abroad.
With export controls tightening and domestic alternatives still maturing, Southeast Asia has temporarily become a crucial part of China’s AI development pipeline, bridging the gap between ambition and capability in a rapidly shifting chip environment.
