Driven by applications such as AI Agent, global word consumption has further increased. According to the latest data from OpenRouter, the total number of global AI large model calls last week (May 18 to May 24) was 28.9 trillion Tokens, an increase of 7.4% from the previous week. The number of calls has increased for five consecutive weeks, and the demand for large model calls continues to be released.
Image source: OpenRouter
Among the large AI models on the list, the weekly use volume of China's large AI models reached 9.22 trillion Tokens, a month-on-month increase of 19.89%; during the same period, the weekly use volume of large American AI models was 4.93 trillion Tokens, a month-on-month increase of 16.27%.China's weekly calls for large models have surpassed the United States for four consecutive weeks and ranked first in the world.
Up to now, DeepSeek-V4-Flash has topped the OpenRouter global AI large model call list.
OpenRouter is an AI model aggregation and calling platform that provides a transparent token-level monitoring and billing system, aiming to solve the problems of interface fragmentation, complex key management, and cost control faced by developers when calling multiple AI models. Its users are mainly overseas developers, with Chinese developers accounting for only about 6%.
in the country, the substantial increase in the number of Token calls is nothing new. According to the National Bureau of Statistics, in March 2026,The average daily Token calls in China alone have exceeded 140 trillion.; The average daily usage of bean bags doubled to 120 trillion within 3 months.
CICC estimates that in moderate usage scenarios, when the penetration rate of Agent reaches 8%, the total Token consumption of Agent is equivalent to that of Chatbot; the popularity of Agent shows a multiplier effect on Token consumption. With the synergistic improvement of single task complexity, usage time and penetration rate, it is expected to promote the average daily Token consumption to increase by more than 5 times.
As the consumption of word elements increases day by day, Token factories and Token operators have been launched one after another.
According to the official websites of each company, China Mobile launched a Token computing service product for individual users on April 21, supporting mainstream large language models such as DeepSeek and Qwen, and a sub-package can be purchased for as low as 5.99 yuan; China Telecom officially launched a series of trial commercial Token packages on May 17, with a basic version price of 39.9 yuan/month for small, medium and micro customers; China Unicom Shanghai Branch announced on May 16 that it will provide Token services to Shanghai OPC customers.
At the same time, China Telecom has issued a bidding announcement for the centralized procurement project of "Token Factory" generation capability services. Tianfeng Securities pointed out that AI data centers are evolving into “Token factories”, emphasizing the need for large-scale data processing capabilities.
CITIC Securities stated,The emergence of Token factories and Token operators marks the transformation of Token generation capabilities from ancillary to a standardized service that can be priced., will promote the computing power rental market to shift from the current fixed monthly rental model based on "bare metal" server rental time to a model billed based on actual Token usage.
The agency emphasized that when Token becomes the unit of calculation for computing power, computing power leasing service providers can fully reap the dividends brought by the continued expansion of Token demand and the rapid penetration of all AI application scenarios. The current high prosperity of the computing power leasing industry is mainly due to the mismatch between supply and demand in the domestic computing power market, which makes the advantages of leading leasing companies with high-end computing power chip resources more prominent. Combining the prosperity of the track and the current trend of the industry gradually clearing up and concentrating towards the top, we are optimistic about the growth elasticity of the top computing power rental manufacturers under the new round of growth trend of Token usage.