![]() |
Chinese AI models are leading the pack after recording a weekly token usage of 7.359 trillion from March 16 to 22, a 56.9 percent increase from the previous week. That's in stark contrast to U.S. models, which saw a usage of 3.536 trillion tokens during the same period, with a modest weekly growth of 7.35 percent, according to the latest data from OpenRouter, the world's largest AI model API aggregation platform.
The data highlights a significant shift in the global AI landscape, with Chinese models demonstrating strong adoption and application at scale.
What is a token?
To understand the significance of this metric, an understanding of "token" is essential.
"Tokens are the basic units of text processing for large language models, which can be understood as 'word chunks' from an AI's perspective," said Ma Zhiheng, an assistant professor at the School of Computing Microelectronics, Shenzhen Institute of Advanced Technology.
Before text is fed into a model, it is broken down into tokens and converted into vectors. For example, a single Chinese character typically corresponds to one or two tokens. Every user query and the AI's subsequent response consume a certain number of these tokens.
Ou Weijie, head of the Yashan LAB at the Shenzhen Academy of Computing Sciences, provided an analogy: "If 'computing power' is seen as 'electricity,' then tokens are the 'kilowatt-hours' consumed. They are a core indicator for measuring AI activity and processing scale."
Why is China's token usage so high?
The surge in token usage is driven by a combination of factors, primarily cost-effectiveness and widespread application.
Data from OpenRouter shows that the top four models by weekly usage are all from China, namely Xiaomi's MiMo V2 Pro, StepFun's Step 3.5 Flash (free), MiniMax's M2.5, and DeepSeek's V3.2.
"In terms of pricing, domestic models represented by DeepSeek and MiniMax have significantly reduced API usage costs, stimulating demand from developers and enterprises," Ma said. Chinese companies have taken a leading role in the open-source model space, narrowing the technological gap with top-tier global closed-source models to about three months, while offering them at a much lower price, he added.
Luo Jieping, executive director and chairman of the board of Yuexiu Port Holdings Co., Ltd., pointed to the vast user base in China as another key driver. "Chinese developers have contributed a substantial amount of token consumption. Applications like WeChat, DingTalk, and Feishu can reach billions of users. These users can access AI capabilities with a simple click, which undoubtedly generates massive demand for model calls."
This accessibility makes AI a practical tool for diverse users. As cost optimization continues in areas like inference and response speed, a large number of small and medium-sized enterprises and developers are integrating AI into their business processes, triggering a long-tail effect in token consumption.