China’s IT heavyweights relocate AI model training offshore to get Nvidia chips, according to the Financial Times.

(Reuters) – According to the Financial Times, top Chinese enterprises are training their artificial intelligence models abroad in order to gain access to Nvidia processors and evade US sanctions aimed at limiting their growth in sophisticated technology.

According to the article, Alibaba and ByteDance are among the tech companies training their newest huge language models in Southeast Asian data centers, citing two sources with firsthand knowledge of the topic.

Reuters could not immediately confirm the allegation. Nvidia has declined to comment on the story.

According to the research, training in offshore locales has steadily increased since the US attempted to prohibit sales of the H20 chip in April.

According to the publication, Chinese enterprises rely on leasing agreements for overseas data centres owned and controlled by non-Chinese organizations, with the exception of DeepSeek, which acquired a big stock of Nvidia chips before to the US export limits and trained its model locally.

Deepseek is also working with local chip manufacturers, including Huawei, to optimize and build the next generation of Chinese AI processors, according to the Financial Times.

Alibaba, ByteDance, DeepSeek, and Huawei did not return requests for comment.