China Increasingly Relies on Copying American AI

The Economist examines the growing dependence of Chinese AI labs on "distillation"—training models on the outputs of other models rather than independently: experts estimate that Chinese models lag the US frontier by 3–6 months, and without distillation, the gap would have nearly doubled.

Compiled by Sergey KostenchukPublished 2026-10-30Updated 2026-10-30

2026-10-30 Business OpenAI

🇨🇳 China Increasingly Relies on Copying American AI

The Economist examined "distillation": Chinese labs are training models on the outputs of other models instead of independent training. The lag behind the US frontier is estimated at 3–6 months; without distillation, the gap would have nearly doubled.

🌍 In September, Anthropic calculated that five Chinese companies collected over 180 million messages from Claude in five months. After closing the leak of step-by-step reasoning and billing for rejected requests, the volume collected by distillers dropped by 90–99%, at the cost of 5–7% of revenue.

👤 A distilled model repeats the teacher rather than generating something new: Chinese models sometimes call themselves Claude or ChatGPT. ByteDance abandoned distillation—it "at best allows the gap to be constantly narrowed."

Source 1: https://www.economist.com/science-and-technology/2026/10/25/china-increasingly-depends-on-copying-american-ai Source 2: https://news.ycombinator.com/item?id=50299507

Sources