The latest data from OpenRouter, the world's largest artificial intelligence (AI) model API aggregation platform, shows that from March 16th to March 22nd, the total amount of AI models deployed globally was 20.4 trillion tokens, an increase of 20.7% compared to the previous period. The reporter noticed that among the top ten AI models on the list, the weekly survey usage of Chinese AI models was 7.359 trillion tokens, an increase of 56.9% from the previous week; The weekly usage of AI big models in the United States was 3.536 trillion tokens, up 7.35% month on month. This is China's AI big model weekly survey usage surpassing the United States for three consecutive weeks. Why has the number of tokens called by Chinese AI models continuously surpassed that of the United States? The reporter conducted interviews around the relevant issues. First question: What is Token? Token is the basic unit of text processing in large language models, which can be understood as' blocks in the eyes of AI '. ”According to Ma Zhiheng, assistant professor at the School of Computational Microelectronics at Shenzhen University of Technology, before inputting the model, the text will be divided into tokens and converted into vectors. For example, in Chinese, each character usually corresponds to 1 to 2 tokens, and each question and AI answer consumes a certain amount of tokens. Ou Weijie, the head of Yashan LAB at Shenzhen Institute of Computing Science, said that if "computing power" is regarded as "electricity", then Token is the "electricity" consumed, which is the core indicator for measuring AI activity and processing scale. Ma Chaoliang, Executive Director of the Token Digital Economy Research Center at the Comprehensive Development Research Institute (Shenzhen, China), believes that behind tokens, there is a larger trend where humans are "breaking down" the world into the smallest units that can be understood and processed by machines. Second question: Why is the call volume in China large? According to OpenRouter data, the top four global call volume rankings last week were all Chinese AI models, including Xiaomi MiMo V2 Pro, Step 3.5 Flash (free), MiniMax M2.5, and DeepSeeker V3.2. In terms of price, domestic models such as DeepSeek and MiniMax M2.5 have significantly reduced the cost of API usage and stimulated the demand for calling from developers and enterprises. ”According to Ma Zhiheng's analysis, "Chinese companies dominate the field of open source models, and the technology gap with the world's top closed source models has been shortened to about three months, and the price is much lower than the latter, which has become an important attraction for widespread use." "Chinese developers have contributed a large amount of token consumption, and applications such as WeChat, DingTalk, and Feishu can reach billions of users. These users can easily invoke AI capabilities with just a click, which undoubtedly brings a massive demand for model invocation. ”According to Luo Jieping, Executive Director and Chairman of the Board of Directors of Guangdong Port Holdings Limited, "When building AI applications, most companies are very cost sensitive. Domestic models have turned AI into a necessity like firewood, rice, oil, and salt with lower training costs, and have gained the favor of global developers through price advantages. ”Ou Weijie believes that with the continuous optimization of domestic large models in terms of inference cost, response speed, API cost, etc., a large number of small and medium-sized enterprises and developers have begun to integrate AI into business processes, triggering the long tail effect of call volume. Question 3: What does this data mean? Last week, Alibaba announced the official establishment of the Alibaba Token Hub business group, aiming to build a complete AI ecosystem around "creating tokens, delivering tokens, and applying tokens"; Huang Renxun, founder and CEO of NVIDIA Corporation in the United States, also proposed "Token Economics" at GTC 2026, defining data centers as factories that produce AI intelligent tokens, emphasizing that computing power is revenue. In Ma Zhiheng's view, Token usage is a "thermometer" that measures the actual implementation and scale of AI usage. China continues to lead the way in terms of deployment volume, indicating that the focus of AI development is shifting from "model release" to "large-scale application", and the industrialization process has entered a stage of acceleration. Luo Jieping admitted that this record breaking data means that China's AI industry is entering a positive cycle of "technology iteration cost reduction application explosion", shifting from "following" to "leading". Through the open source model and rich application scenarios, China has taken a different path from foreign closed source models, forming the advantage of cluster based rise. Ou Weijie reminds that behind the massive amount of tokens, there are larger scale data throughput and more complex data governance challenges. Every call to the big model relies on the precise management and millisecond response of the underlying database system to real-time data, historical knowledge, and user interaction. Ma Zhiheng said, "We must also be aware that the United States still maintains significant advantages in areas such as original model innovation, high-end chips, and computing infrastructure