Chinese entities use US AI model outputs to train domestic systems
Chinese companies and universities are using distillation techniques to train domestic artificial intelligence models using outputs from American AI models. This practice utilizes systems such as OpenAI models to generate training data for domestic systems like Qwen. This method allows Chinese developers to mitigate the impact of United States chip export curbs.
Industrial scale distillation practices
The practice of distillation has expanded from academic research into mainstream industrial use across China. Chinese researchers use US models to generate answers, computer code, and solutions to complex problems. These outputs are then used to create training data for smaller domestic models. This process aims to preserve selected skills in smaller and less expensive AI systems.
The use of distillation could potentially reduce reliance on advanced computing power and American technology. The Chinese government is encouraging distillation to reduce costs and facilitate AI deployment across its manufacturing sector. This strategic move comes as the United States government announced on November 16, 2026, that it would further restrict China's access to advanced chips from companies such as Nvidia.
US authorities previously accused Chinese companies in September 2026 of aggressive, targeted, industrial-scale distillation of American AI models. This follows earlier concerns raised by US Commerce Department officials. During meetings in Chengdu in July 2026, officials raised concerns regarding developers collecting model output for these purposes.
Regional economic context
The increased use of distillation is occurring ahead of the Asia-Pacific Economic Cooperation (APEC) summit. The summit is scheduled to take place from November 18 to November 19, 2026, in Shenzhen. The technological competition between the United States and China remains a central theme in the region.
Evidence of these practices is documented across a wide range of sources. A review of more than 300 Chinese academic papers, patents, corporate disclosures, recruitment notices, and government documents shows the scale of this activity. These documents confirm that the use of American model outputs is a growing component of the domestic Chinese AI development strategy.
Technological implications and restrictions
The distillation process allows for the creation of efficient models that do not require the same level of hardware investment as larger systems. By using the intelligence of existing American models, Chinese developers can accelerate the development of their own domestic technology. This approach directly addresses the limitations imposed by US export controls on high-end semiconductors.
The ongoing tension between the United States and China regarding technology exports continues to shape the global AI landscape. As the US government tightens restrictions on companies like Nvidia, the reliance on distillation techniques is expected to increase among Chinese developers seeking to maintain their technological trajectory.
Glossary
- Chengdu — Chengdu is a major city located in southwestern China and the capital of Sichuan Province. It serves as a major economical and transportation hub in the region.
- distillation — Distillation is a technique used to train domestic artificial intelligence models using outputs from other models. This process allows developers to create efficient models that do not require the same level of hardware investment as larger systems.
- Nvidia — Nvidia is a company that produces advanced chips used in AI servers.
This article was written with AI assistance from multiple news reports and is reviewed by our editors under our editorial principles. Editorial Standards · Tips & Contact
Comments 0