China Taps US AI Models to Train Homegrown Rivals
By Ganesh Karki
Key facts
- Section
- News
- Country
- China
- Published
- 18 Nov 2026
China taps US AI models to train homegrown rivals at industrial scale, according to a Reuters review of more than 300 Chinese academic papers, patents, corporate disclosures, recruitment notices and government documents, published on 17 November.
The practice, known as distillation, uses output from leading American AI models to train domestic rivals, and it has expanded beyond academic experiments into mainstream industrial use in China, the review found.
China taps US AI models at a glance
| Detail | Information |
|---|---|
| Practice | Distillation, using US model output to train smaller Chinese models |
| Evidence base | More than 300 Chinese papers, patents, disclosures, notices and documents reviewed by Reuters |
| Example | Shanghai Jiao Tong University used about 5,000 OpenAI o1 answers to build 10,750 training examples for Alibaba Qwen2.5-72B |
| Beijing position | Government documents encourage distillation to spread AI across manufacturing |
| US position | Washington accused Chinese firms in September of industrial-scale distillation; talks held in Chengdu in July |
| Context | US said on 16 November it would further restrict China access to advanced AI chips |
What happened
Chinese entities are adopting increasingly sophisticated ways to use output from leading American AI models to train domestic rivals, documents show, as China aims to broaden its AI strategy and offset tightening US curbs on its access to chips. The records describe Chinese researchers using US models to produce answers to questions, build computer code and find solutions to difficult problems, with Chinese companies then developing software to automate the process and specialist workers preparing material for training smaller domestic models.
The documents also show that Beijing is encouraging distillation to make AI cheaper and easier to deploy across China powerful manufacturing sector. Distillation is not illegal and is prevalent in other parts of the world. While the process cannot reproduce a leading model full capabilities or eliminate the need for advanced computing power, it can help developers preserve selected skills in smaller systems that cost less to operate, potentially reducing their reliance on American technology.
In a November 2024 paper, Shanghai Jiao Tong University researchers described selecting about 5,000 answers from OpenAI o1 model, translating them using GPT-4o mini and assembling 10,750 bilingual training examples to further train Alibaba Qwen2.5-72B model. In a February 2026 paper, researchers from the same university and Chinese food-delivery company Meituan described using GPT-4o to generate plans for coordinating AI agents working together on programming problems, helping train a smaller model to organise such tasks. Alibaba researchers described a more systematic approach in an April 2025 paper, using OpenAI GPT-4 and GPT-4o to generate training material in languages other than Chinese, then refining and scoring responses before training smaller Qwen models.
The issue is now a source of friction between the world two biggest economies ahead of the November 18 to 19 Asia-Pacific Economic Cooperation summit in Shenzhen, where US President Donald Trump and Chinese President Xi Jinping are expected to meet. US authorities publicly accused Chinese companies in September of aggressive, targeted distillation on an industrial scale. At meetings in Chengdu in July, US Commerce Department officials raised concerns with Chinese regulators, warning that some activities could amount to intellectual-property theft, three people familiar with the discussions said. Chinese officials defended distillation as a standard technique and said they were open to acting where conduct constituted IP theft, but made no concrete commitments, according to a US official familiar with the previously unreported exchange. The chip curbs at the centre of the dispute are covered in the report on US AI chip curbs hitting Southeast Asia data centre shares, while the Nvidia leadership upheaval behind the scrutiny is covered in the report on the Nvidia board ruling out a Jensen Huang comeback.
Why it matters
Distillation could undercut the effectiveness of US curbs on the sale of advanced chips to China, because Chinese developers can build AI capabilities by drawing on those of their American rivals. A US official familiar with the July discussions said Washington believes industrial-scale distillation is the main reason Chinese AI models have narrowed the gap with their American rivals. Two people familiar with Chinese developers operations disputed that assessment, saying distillation was only one factor alongside advances in engineering, algorithms, data and model design.
The researchers themselves cautioned that excessive reliance on distillation could stifle original innovation, leaving developers dependent on breakthroughs made by more advanced model providers. The dispute also matters for the rules of AI development: OpenAI terms restrict using its output to develop competing models, subject to exceptions, while Anthropic similarly restricts training competing systems without permission.
What is not confirmed
Reuters could not establish the full scale of distillation inside Chinese companies, and the review rests on public documents rather than internal company records. Shanghai Jiao Tong University, Alibaba and Meituan did not respond to requests for comment, and the US Commerce Department and China Cyberspace Administration did not respond either. Whether specific cases amount to intellectual-property theft has not been determined.
Sources
Coverage: Reuters, 17 November 2026; Reuters, 17 November 2026.
Sources and references
Information checked: 18 November 2026
Related stories
Global Hiring Trends November 2026: US Rebounds, UK Vacancies Fall
Global hiring trends for November 2026: US payrolls rose 165,000, UK vacancies fell to a 12-year low, and Australia jobless rate hit a five-year high.
20 November 2026Trump Third Term: Can He Run Again Under the 22nd Amendment?
Trump third term: the 22nd Amendment bars a third elected term, so Trump cannot run in 2028 despite saying he would love another chance.
20 November 2026Las Vegas Grand Prix 2026: Race Time, Schedule and How to Watch
Las Vegas Grand Prix 2026: the race starts Saturday, 21 November, at 8:00 PM Pacific Time on the Strip Circuit over 50 laps.
20 November 2026Muslims Outnumber Christians by 2035, Pew Projects
Muslims outnumber Christians by 2035, Pew projects, with 2.57 billion Muslims against 2.52 billion Christians as birth rates drive the shift.