Chinese-developed value AI models are spreading globally, prompting concern from US policymakers about competitive pressure and potential dual-use applications.
Chinese AI models are spreading across global enterprises far faster than anticipated. In 2026, Chinese models now account for more than half of enterprise usage on major developer platforms, driven by significantly improved performance and lower costs.
On OpenRouter, Chinese models represented 57 to 67 percent of total token usage for the week of September 14—a steep climb from just 6 to 13 percent in February. Vercel saw an equally dramatic shift, with Chinese models rising to 55 percent of platform usage in August from 11 percent in January.
While top U.S. models continue to lead most benchmark tests, Chinese AI companies have made substantial strides. DeepSeek, Z.ai, and Alibaba have released new models that significantly improve performance on demanding tasks such as coding.
Peter Walker, head of OpenRouter Insights, noted that Chinese open-source models released this year have matured to the point where they can reliably handle advanced agent-type use cases, particularly in coding applications. He emphasized that Chinese models offer superior cost efficiency compared to many U.S. AI labs' offerings.
Price has emerged as the decisive factor in adoption. Harpreet Arora, head of agent infrastructure at Vercel, explained that once Chinese models meet the quality thresholds required for business use, the cost advantage becomes the primary decision driver. However, for more complex tasks, companies still prefer cutting-edge U.S. models.
The rapid advance of Chinese AI is triggering alarm in Washington. Two U.S. House committees are investigating the broader implications of widespread Chinese model adoption. Though the United States has restricted Chinese AI companies from purchasing cutting-edge chips through export controls, U.S. officials worry that China may close the technological gap by remotely accessing Nvidia chips through overseas data centers or by using distillation techniques to replicate existing models.
Geographically, the most significant growth in Chinese model usage has occurred in countries classified as the Global South. Companies across 82 countries in Central America, South America, Africa, and Asia used Chinese models for 67 percent of their tokens, while U.S.-based companies accounted for roughly half of all OpenRouter token usage. Daniel R. Lemer, a senior fellow at the Center for a New American Security, predicted that Southeast Asia could experience a sharp surge in Chinese AI model adoption, given its deepening economic and cultural ties to China and its expanding digital infrastructure.