AI was a big focus of this week’s meeting between US President Donald Trump and Chinese President Xi Jinping.
While the most advanced US models lead most benchmarks, Chinese companies such as DeepSeek, Z.ai, and Alibaba are releasing new models with significantly improved performance for tasks such as coding.
Adoption of Chinese models by businesses increased significantly globally in 2026, according to usage data shared with CNBC.
China’s AI models have gone from a relatively small share of usage to a majority on two major developer platforms that provide gateways for enterprises to access models from different providers. OpenRouter accounted for 57% to 67% of tokens spent during the week of September 14th, up from 6% to 13% in February. Vercel’s share rose from 11% in January to 55% in August.
OpenRouter’s data pertains to companies in the United States, Europe, and what the company defines as the “Global South” (82 countries in Latin America, Africa, and Asia). Vercel did not provide a geographic breakdown of the data.
Concerns are growing in Washington, with two U.S. House committees investigating the impact of increased adoption of the Chinese model.
The United States has sought to maintain its AI leadership by restricting Chinese AI companies’ purchases of cutting-edge chips through export controls.
Washington is concerned that they are gaining power by accessing NVIDIA chips remotely via overseas data centers and using “distillation,” in which new models imitate older, more established models.
Daniel Remler, a senior fellow in the technology and national security program at the Center for a New American Security (CNAS) think tank, said Chinese AI is a “real economic and security risk to the United States.”
“The ultimate concern is that the integration of China’s AI models will draw countries into China’s technological sphere of influence and strengthen geopolitical alignment,” he told CNBC.
Peter Walker, Head of Insights at OpenRouter, told CNBC that China’s open source model released this year “certainly performs in advanced agent use cases, especially when it comes to coding, in a way that just wasn’t the case in late 2025.”
It’s also “incredibly cost-effective compared to most American laboratory models,” he added.
Earlier this week, OpenAI and Anthropic both announced new models that are cheaper. Diane Penn, director of product management, research and labs at Anthropic, told CNBC that the company aims to make model answers “more efficient, thereby reducing the amount of token usage depending on effort settings.”
Harpreet Arora, head of agency infrastructure at Vercel, told CNBC that price is key to increasing adoption of Chinese models. “Chinese models are becoming more capable of doing more tasks at a much lower cost. If a model meets the quality standards for the job, the price difference becomes compelling.”
But companies still want to use the U.S. Frontier model for more complex tasks, he added.
Companies in what OpenRouter defines as the “Global South” have become the largest users of Chinese AI models on its systems in recent weeks.
More than two-thirds (67%) of the tokens used by these companies are Chinese models. About half of OpenRouter’s tokens are used by US companies
“Chinese AI models in particular have the potential for significant uptake in Southeast Asia, given the close economic and cultural ties[with China]and growing digital infrastructure,” CNAS’s Lemler said.
“From Lagos to Sao Paulo to Jakarta, wherever entrepreneurs and governments are looking for a cheap open model, they will look first to Chinese AI.”
