Chinese technology companies and research institutions are increasingly utilising the results of work on American artificial intelligence models to train their own systems. This approach helps to reduce AI development costs, minimise the need for computing resources and partially offset US restrictions on access to advanced chips. This is according to a Reuters investigation published on 17 November.
- Reuters analysed more than 300 Chinese academic, corporate and government documents.
- Chinese developers are using the responses from American AI models to train their own systems.
- Distillation helps to reduce the need for expensive computing resources and equipment.
- Beijing supports the introduction of more affordable artificial intelligence systems in industry.
- Washington is tightening restrictions on China’s remote access to advanced chips.
How China is using American AI technology
Reuters journalists examined more than 300 academic papers, patents, corporate reports, job advertisements and government documents. The analysis showed that model distillation technology has moved beyond academic experiments and has become a widespread tool in China’s artificial intelligence industry.
Against the backdrop of US restrictions on access to Nvidia chips Chinese companies use powerful American models to generate educational materials and programme code, and to solve complex mathematical problems.
Distillation involves training a smaller model using the outputs of a more powerful system. This makes it possible to replicate certain capabilities of state-of-the-art neural networks without having to completely replicate the process of developing them.
How Alibaba and Meituan train their own models
In November 2024, researchers at Shanghai Jiaotong University used around 5,000 responses from OpenAI’s o1 model. Using GPT-4o mini, they generated 10,750 bilingual training examples to improve the Chinese model Qwen2.5-72B from Alibaba.
In April 2025, researchers at Alibaba described the use of GPT-4 and GPT-4o to develop multilingual educational materials.
In February 2026, researchers, in collaboration with Meituan, used GPT-4o to train a system capable of coordinating the work of several AI agents whilst carrying out programming tasks.
The US and China are intensifying their technological rivalry
In January 2026, eight Chinese government departments unveiled a programme to integrate artificial intelligence into the manufacturing sector. It provides for support for the refinement and optimisation of models for industrial applications.
The distillation technology itself is not illegal, but OpenAI and Anthropic restrict the use of the results of their systems’ operations to create competing products.
The US authorities regard the widespread use of this method as one of the reasons for the narrowing of the technological gap between the US and China. Chinese developers, on the other hand, point out that their own engineering and algorithmic achievements play a significant role.
On 16 November, Washington announced new restrictions on remote access to cutting-edge American chips via overseas data centres.







