The Chinese government is considering ways to restrict overseas access to its advanced artificial intelligence (AI) models. As the United States moves to keep China in check with measures such as export controls on advanced semiconductors, the battle for AI supremacy appears to be expanding beyond chips into the realm of AI models.
On the 7th, Reuters, citing multiple sources, reported that Chinese authorities last month discussed measures to restrict overseas access to advanced AI models with officials from big tech companies Alibaba, ByteDance Ltd., and Z.ai. The talks were said to cover not only models currently in operation but also next-generation models that have not yet been released.
The discussions were led by China's Ministry of Commerce. The meeting addressed measures to place limits on the most advanced AI models, including closed-source models and open-weight models with some technologies disclosed, and also raised options to punish leaks or theft of proprietary AI technologies under the National Security Act.
Chinese authorities are said to be concerned that domestic AI technologies could flow overseas or pose national security risks. However, the specific scope of regulation and the timing of implementation have not yet been finalized. Reuters noted that potential restrictions could apply only to models released in the future.
With this move by Chinese authorities, the AI supremacy front is widening from semiconductors to models, data, and services. Until now, the United States has sought to curb China's AI development by restricting exports of Nvidia's advanced AI Semiconductor chips. The U.S. government has also been reviewing access restrictions to prevent advanced AI models from being used by rival countries. Recently in the United States, security concerns have emerged that high-performance AI models could be used to bolster rivals' cyberattack capabilities.
Amid these circumstances, Chinese AI models have been building a presence in overseas markets by leveraging low expense and strong performance. Alibaba's "Qwen" and ByteDance Ltd.'s "Doubao" are cited as AI models widely used in China. Z.ai's latest model is also drawing attention for reportedly delivering performance that can compete with some major global models at a low expense.
Industry watchers expect that if China restricts overseas access to advanced AI models, it could ripple across the global AI market. Overseas corporations that have relied on inexpensive Chinese AI models could face higher expense burdens, and barriers between the United States and China over AI technologies could grow taller. According to Reuters, Chinese authorities have been steadily tightening measures to protect domestic AI technologies this year. Representative steps include investigations into AI startups that transferred overseas and stronger controls on overseas transactions entangled with technology, data, and national security.
The industry sees Generative AI becoming a core foundational technology that determines industrial competitiveness and national security, intensifying competition among countries to secure technologies. In particular, if the AI supply chain splits around the United States and China, corporations' AI service development expense and options are expected to be affected.