Recently, token economies have become a hot topic of interest in the AI industry. As the smallest unit through which large models process information, the exponential growth in the number of word tokens has become a driving force for breakthroughs and large-scale implementation in terminal agent technology, propelling AI from the cloud to the terminal and from a supporting tool to an autonomous agent. Data shows that as of March, China's average daily word call exceeded 140 trillion, compared with 100 billion at the beginning of 2024 to achieve a super thousandfold increase. The head of the National Data Bureau said that the significant increase in the average daily word usage marks that China's artificial intelligence development has entered a rapid growth stage. At the same time, China's high-quality data supply system continues to improve. By the end of 2025, China has built more than 100 thousand high-quality data sets, with a total volume of 890 PB, which has laid a solid data foundation for the development of meta-economy and terminal intelligence. Every time an intelligent person thinks, plans, or performs, it consumes words. Facing the pain points such as high word consumption cost, strong cloud dependence, and privacy and security risks, terminal manufacturers have taken the lead in cracking the difficulties with end-side intelligence technology, and promoting the evolution of word economy toward "high efficiency, low cost, and secure."Recently, Honor unveiled its end-device-based “lobster” AI entity, the YOYO Claw technology.The technology relies on intelligent routing algorithms to enable the end-cloud coordinated and precise scheduling. Simple, high-frequency tasks are executed at 100% end-side, achieving zero word consumption; Complex tasks are called to the cloud on demand, improving efficiency through context optimization and memory matching, and significantly reducing user costs. As the intelligent cockpit evolves to a higher level, soaring demand for computing power, high cloud costs, and complex development are the three major challenges. Shang Tang Juying launched the Sage Box (Thousand Wit Box) gives the crack. Shang Tang CTO (Chief Technology Officer) Xiao Feng introduced that the device is equipped with a large base model that realizes the ability of complex agents at the end of the car, and realizes "Token zero cost." According to estimates, a single car can save about 30 yuan per day in the cloud, after a large-scale deployment, "10,000 cars can save more than RMB annually," and the system response time can be reduced to 0.5 seconds. The development of artificial intelligence cannot be achieved without the coordinated empowerment of computing networks and infrastructure. China Mobile High Speed Data Network takes "high speed circulation, security endogenous, compliance traceability" as its core, enabling second-level interconnection across networks and provinces based on business level, ensuring the traceability of data circulation throughout the whole chain, and greatly improving data transmission efficiency and business response speed. Liu Hong, vice general manager of China Mobile's digital intelligence division, said that the first high-speed data network solutions were in Tianjin, Suzhou, Ganzhou and other cities have carried out construction practices, successfully promoting the construction of key scenarios such as the integrated application of new energy vehicle data, banking financial risk control, efficient flow of meteorological data, and data flow operation services, helping to transform data transaction delivery from offline to online. Lu Peng, a researcher at the Institute of Sociology at the Chinese Academy of Social Sciences, believes that the explosion of word economy and the development of end-intelligence are two-way enabling relationships.The paradigm shift in the field of artificial intelligence is forcing end-user intelligent entities to move towards “universal accessibility and efficiency.” This has established a value anchor for AI services, enabling intelligent entities to transition from mere technical concepts to quantifiable productivity.In order to reduce the user's word usage cost, end vendors have invested in end-side compute power and system-level optimization to improve the word conversion rate through soft and hard collaboration. This has enabled intelligent agents to serve as user-friendly assistants that are ready to be used out of the box, thereby facilitating the implementation of AI in the consumer market.As end-point agents continue to penetrate various fields such as PCs, vehicles, and smart homes, a new AI ecosystem driven by word units as “fuel” and edge computing power as “engine” is rapidly becoming a reality.Lu Peng said that word development still faces three shortcomings: first, cost and energy consumption pressures, and computing power supply and electricity costs remain hard constraints limiting word scale adoption. Second, there are security and privacy risks, and the risks of data breaches and abuse of rights brought about by high-frequency calls of smartphones need to be systematically guarded. Third, there is a lack of standards and ecology. There is no uniform standard for the measurement and pricing of terms, and the industry ecology is not mature. "Only by closing these shortcomings can the word economy grow steadily." Lu Peng said. (Looking ahead to a new era)