Chinese authorities are reportedly reviewing measures to restrict overseas access to advanced artificial intelligence (AI) models developed in the country. If realized, the expansion strategy for Chinese AI models that has prioritized openness would likely require revision. Following the US government ‘Anthropic export control’ episode, access barriers to AI models in many countries have been rising, and there are expectations that the push to build ‘sovereign AI(sovereign AI)’ will also accelerate.

Reuters reported on the 7th (local time), citing local sources, that the Chinese government discussed with Alibaba, ByteDance, and Z.ai (ZhipuAI) measures to restrict overseas access to AI models. Reuters added that controls could be applied starting with next-generation models to be released in the future.

Chinese AI developers have adopted an ‘open-weight’ strategy that allows many of their models to be freely downloaded and used. Unlike US frontier AI companies such as OpenAI and Anthropic, which operate their flagship models in a closed manner, this has been interpreted as an attempt to expand market share by prioritizing accessibility.

However, after the United States issued export-control directives last month on the advanced models of Anthropic, perceptions appear to have grown in China that AI models are a core national asset requiring controls, akin to semiconductors.

The rapid improvement in the performance of Chinese AI models may also have provided justification for tighter access controls. According to the AI performance comparison platform ArtificialAnalysis, GLM-5.2 (51 points) released by Z.ai last month scored higher than the latest model from Google, Gemini 3.5 Flash (50 points).

Some analyses suggest that Chinese AI developers may adopt a ‘dual structure’ in which they maintain the existing open-weight approach for general-purpose models while imposing separate controls on cutting-edge models. In that case, it would likely be a negative for companies seeking Chinese models for ‘cost-effectiveness’. According to the AI model-routing platform OpenRouter, the models ranked 1~5 at present are all Chinese. The advanced model V4 Pro from DeepSeek is about one-tenth the cost of comparable US cutting-edge AI models.

As concern grows that border barriers for AI models could materialize, the pace of building ‘sovereign AI(sovereign AI)’ is also expected to accelerate across countries. According to Reuters, Ukraine has recently been working with Google to develop its own AI models in an ‘on-premises(keeping data on domestic servers rather than on those of the AI developer)’ configuration.

Some analyses also say this trend could create opportunities for South Korean companies. Naver Cloud said on the same day that it had signed a comprehensive partnership with ‘Mistral AI’ to drive innovation in manufacturing AI, a leading sovereign AI developer in Europe. In an AI market being reshaped around US Big Tech companies, leading technology companies from Asia and Europe are joining forces to build an independent sovereign AI ecosystem.

Lim Kyung-Tae, a professor at the KAIST Graduate School of Culture Technology, said in a phone call, “Among Chinese AI companies, a shift toward closed approaches for cutting-edge models is emerging, and this trend will strengthen going forward,” and added, “By making use of Korea being around third after the United States and China in sovereign AI capabilities, joining as a support partner in projects to build sovereign AI in major overseas countries could be a promising business model.”