Establishing "token factories" to enhance resource utilization and provide computational power support for tokens; improving data quality and model capabilities, constructing operational service platforms, and promoting token distribution and commercial operations... As artificial intelligence application scenarios continue to deepen, tokens are linking up new economic chains. Enterprises are now deploying across the entire chain surrounding token production, scheduling, commercial operation, and application. From users generating an AI video with a large model to real-time AI scheduling on production lines, every scenario where AI applications land involves massive token calls behind the scenes. The "National Data Resource Survey Report (2025)" released by the National Data Administration shows that in 2025, the national daily average token call volume grew from over one trillion at the beginning of the year to 100 trillion by year-end, with the cumulative annual token call volume reaching approximately 21,100 trillion. The exponential growth in token consumption relies on massive, convenient computational power supply. At the AI infrastructure level, relevant companies are focusing on "token factories" to promote efficient and intelligent large-scale token production. For instance, Wu Wen Xin Qiong recently announced securing an additional over 700 million yuan in financing. The funds will be primarily allocated in three directions: strengthening the technological layout of software-hardware collaboration to improve production efficiency "from electricity to tokens"; building AI infrastructure with autonomous evolution capabilities, providing solutions for different industry scenarios to enhance value conversion "from tokens to productivity." Wu Wen Xin Qiong CEO Xia Lixue stated that currently, token demand exhibits explosive characteristics akin to "mobile data traffic in the 3G era," and the core upgrade path for the infrastructure foundation is moving toward Agentic Infra (autonomous artificial intelligence infrastructure). Enterprises are currently fully committed to building efficient, large-scale, intelligent "token factories," integrating dozens of domestic chips and hundreds of models to achieve heterogeneous diversity and software-hardware collaboration, maximizing resource utilization. Data quality and model capabilities jointly determine the value content of tokens. At the large model manufacturer level, KNOWLEDGE ATLAS CEO Zhang Peng noted that since the beginning of this year, intelligent agent applications represented by OpenClaw have ignited token consumption. In the first quarter, KNOWLEDGE ATLAS's API (Application Programming Interface) pricing increased by 83%, while call volume grew by 400%, confirming that high-quality tokens are currently a scarce resource. Facing a supply-demand imbalance, KNOWLEDGE ATLAS will continue to increase investment in the integrated optimization of domestic chip software and hardware, continuously improving inference performance. Lu Feng, President of the Beijing Frontier Future Technology Industry Development Research Institute, stated that in response to the surging trend in token demand, enterprise deployments reflect their efforts to reduce token production costs through extreme optimization of computational power. Meanwhile, computational power is no longer merely about stacking but involves precise investment around the conversion rate "from tokens to productivity." CICC Research believes that since 2026, Agentic AI has accelerated its penetration globally, driving exponential growth in large model API consumption. "Token factories" have also become the business model with the fastest short-term growth and the most certain commercialization prospects for model manufacturers. Regarding the commercial operation of tokens, companies represented by the three major telecom operators have been frequently active recently, accelerating the transition from "traffic management" to "token management." Recently, China Mobile launched the AI-eSIM "1+3+9" multi-ecosystem intelligent service system—one AI-eSIM chip entry point, three core engines, applicable to nine major scenarios including toys, home appliances, wearables, and office equipment. A China Mobile representative stated that they will use AI-eSIM as a key entry point for the token economy, providing integrated operations of "traffic + tokens + intelligent agents (Byte+Token+Agent)" to promote the integration of token applications into various industries. China Telecom introduced the Xingchen TokenHub operational service platform to support integrated token operations, covering model calls, capability scheduling, scenario services, and ecosystem operational needs. Lu Feng believes that telecom operators can leverage this opportunity to deeply integrate computational power with networks, break through the "traffic pipeline" dilemma, obtain value-added shares in the AI industry chain, and upgrade from passive transmission to active scheduling. Additionally, leveraging their network infrastructure advantages, telecom operators are poised to become "token distribution hubs," providing developers and enterprises with nearby, efficient token production and access services, thereby building new competitive barriers. Looking ahead, the token economy is expected to unlock greater development space. J.P. Morgan predicts that China's AI inference token consumption will achieve explosive growth, increasing from approximately 10 quadrillion in 2025 to about 3,900 quadrillion by 2030. Zhu Guang, Senior Engineer at the Policy Research Institute of the China Academy of Industrial Internet, believes that tokens link the core elements of the intelligent economy—the "iron triangle" of data, algorithms, and computational power—transforming computational power consumption, model operations, and service supply into measurable, tradable value units. This supports the transformation of intelligent services from customized "luxuries" to universal "necessities" and gives rise to a series of native new business formats such as model services, computational power scheduling, data operations, and intelligent agent applications. Tokens will become the core link for value conversion and commercial closure in the intelligent economy.
Comments