Business and economy·October 25, 2026, 07:22

Chinese AI dominates Korean developers' own models

AI-generated from the sources listed, with automated source checks.

Nearly half of the derivative AI models from Korea-linked accounts on Hugging Face are built on Chinese foundation models, according to a report from a private think tank. The figure applies to developers' models, not AI services as a whole.

When Korean developers build on other people's AI models, they choose Chinese ones more often than American ones. That is the finding of a report from the private think tank Rockbridge Network Korea, published Friday. The report was written by Choi Seong-jin, professor of business economics at Hanyang University.

The numbers from Hugging Face

Hugging Face is a major online platform where developers share AI models. From January to August 2026, Korea-linked accounts there released 1,845 derivative models. A derivative model is one built on top of someone else's foundation model rather than created from scratch.

Of those, 49.3 percent were built on Chinese foundation models, 35.9 percent on American ones and 14.9 percent on Korean ones.

The shift has been rapid. The Chinese share was 1.2 percent in 2023 and 22 percent in 2025. The American share was 79.5 percent in 2023 and has fallen since.

Alibaba's Qwen model alone accounted for 864 derivatives, or 46.8 percent of this year's total. Google's Gemma accounted for 14.7 percent and Meta's Llama for 13.8 percent. The highest-ranked Korean model, LG AI Research's Exaone, came in fourth with 171 derivatives (9.3 percent). Kakao's Kanana was sixth with 49 derivatives (2.7 percent).

Usage is hard to measure

The report examines Chinese AI in Korea based on developer activity, app usage and a company survey. According to the report, actual usage is largely invisible in the standard usage figures, because developers download models and run them offline.

An unnamed official at a Korean AI startup says in the report: "We mainly use Chinese models like Qwen and DeepSeek for high-traffic services."

In a survey of 102 small and midsize companies, 80.2 percent cited fear of data leaks as a concern about Chinese AI. 31.3 percent cited concerns about the models' performance or quality.

Questions for South Korea's AI push

South Korea's state-backed sovereign AI program, meaning the country's own AI independent of foreign players, has received roughly 200 billion won (about $151 million) in funding. Choi Seong-jin says performance alone is not enough if Korean open models are to gain traction among developers.

It is worth remembering that the report comes from a private think tank, and that it is unclear how an account is defined as Korea-linked.