On October 25, 2026, The Korea Times cited a report by Rockbridge Network Korea showing that South Korean developers are increasingly reliant on foundational Chinese AI models.
Among the 1,845 models published by South Korean accounts on Hugging Face from January to August 2026, 49.3% were based on Chinese models, 35.9% on US models, and 14.9% on domestic models.
The proportion of models developed from Chinese platforms surged sharply from 1.2% in 2023 to 22% in 2025, and reached 49.3% in the first 8 months of 2026.
Alibaba’s Qwen led with 864 models, accounting to 46.8%, followed by Google’s Gemma with 14.7%, Meta’s Llama with 13.8%, LG’s EXAONE with 9.3%, and Kakao’s Kanana with 2.7%.
Open-weight models like Qwen and Gemma allow developers to download, fine-tune with Korean data, and deploy locally without complete reliance on cloud services.
Qwen supports 119 languages, offers multiple model sizes, and provides a rich tool ecosystem, facilitating the development of AI applications.
The report noted that the output cost of Chinese models is about 34 times lower than comparable top models, but this advantage shrinks to about 4 times when evaluating across entire test benchmarks.
Experts warn of the risks of dependence on the Alibaba ecosystem, license changes, or operational disruptions due to US-China tech tensions.
A survey of 102 small and medium-sized enterprises showed that 80.2% are concerned about data leaks and 67.7% are worried about a lack of transparency in personal data processing.
The report proposes that South Korea develop safe, reliable domestic foundation models while building a robust AI ecosystem to ensure technological sovereignty.
📌 Conclusion: The popularity of Qwen reflects the competitive advantages of Chinese open-weight AI models in terms of cost, customizability, and development ecosystem. However, the 49.3% figure only reflects models published on Hugging Face and does not represent all AI services in South Korea. The report warns of risks related to technology dependency, licensing, and supply chains, while proposing a strategy that combines leveraging international models with developing domestic foundational AI capabilities.
