Chinese Tech Giants Divert AI Model Training Abroad To Access Nvidia Chips

Illustration of Alibabab Cloud (photo: x @alibaba_cloud)

JAKARTA - A number of major Chinese technology companies have reportedly started training their foreign artificial intelligence models to secure access to Nvidia chips, according to a Financial Times report on Thursday, November 27. This step was taken to circumvent US strict rules that limit China's advances in advanced technology.

Alibaba and ByteDance are among the companies now training their latest generation large-language models at data centers in Southeast Asia. Two sources familiar with the operation said that the trend of shifting training to offshore locations increased after the US restricted sales of the Nvidia H20 chip in April 2025.

The report said Chinese companies relied on lease agreements with overseas data centers owned and operated by non-Chinese entities. DeepSeek is an exception, as the company had time to collect large stock of Nvidia chips before the export ban was imposed so that it remained able to train AI models in the country.

DeepSek is now also working with domestic chipmaker Huawei to optimize and develop a new generation of Chinese AI chips, according to the FT. However, Alibaba, ByteDance, Nvidia, DeepSek, and Huawei also did not immediately comment when asked for a response.

This development reflects the geopolitical pull of technology that is heating up, with Chinese companies navigating US restrictions while trying to maintain their AI innovation rate.