Distillation moves beyond the lab
As the U.S. tightens export controls, Chinese AI developers are being forced to find alternative paths. According to a review covering more than 300 Chinese academic papers, patents, corporate disclosures, recruitment notices and government documents, these companies have increased their reliance on model distillation.
Distillation means using the outputs generated by a powerful model to train a smaller one. The review says the practice has moved well beyond academic experimentation and has entered mainstream industrial deployment across China.
Why it matters
U.S. chip restrictions make it harder for Chinese firms to access the most advanced hardware. Distillation offers a way to produce smaller, more efficient models with limited compute. The review notes that outputs from leading U.S. language models are being used in this process.
This raises new questions about development costs and competition: can access to the outputs of large models substitute for hardware restricted by sanctions? The review does not answer that directly, but it documents that distillation is now widespread.



