Nearly half of new Korean AI services based on Chinese foundation models

In South Korea, the country's burgeoning artificial intelligence (AI) industry is showing a concerning trend: a significant portion of new AI services are being developed using Chinese foundation models. A recent report by the think tank Rockbridge Network Korea reveals that nearly half of the 1,845 models published by Korean developers between January and August were built on Chinese models, with the share of domestically-developed models at just 14.9 percent.

This rapid shift towards Chinese foundation models has been a growing concern for Korean AI developers, with the share of such models increasing from a mere 1.2 percent in 2023 to 22 percent in 2025, before nearly doubling this year to 49.3 percent. Among the Chinese models, Qwen led the pack, accounting for 46.8 percent or 864 models. Google's Gemma and Meta's Llama followed with 14.7 percent and 13.8 percent respectively.

The reliance on Chinese models could potentially place Korea's AI industry in a subordinate position to China's, raising questions about the country's technological sovereignty. While domestic models are seeing growth, they still account for a relatively small fraction of new models compared to their Chinese counterparts. Kakao's Kanana, developed by the South Korean tech giant, accounted for 2.7 percent of the new models.