Since the start of this year, if there is one keyword that has dominated the technology sector, it is almost certainly "Token." As tokens have risen from an early technical term to become the basic unit of measurement in the intelligent era, the token economy is accelerating into shape, spurring intelligent transformation across countless industries. And the capital city of Beijing is standing at the forefront of this tide. How is Beijing leading this new economic wave? The answer may be found in the token production chain. When policy guidance is combined with massive application scenarios, Beijing is poised to be the first in the world to ignite the token economy.

A token factory producing 500 billion tokens per day

Inside a software company, an engineer issues instructions to a large model on a computer, and within seconds a usable piece of code is generated. The foundational tokens behind this programming task come from the Beijing No. 1 Token Factory, located dozens of kilometers away, which is also the first token factory to begin operations in Beijing. As a staff member scans their badge and pushes open the door to the machine room, the reporter sees that the facility is actually running in a "lights-out" state. Liu Li, a technical expert at SoftStone Intelligent Computing, told the reporter that a token factory is a typical intelligent unmanned production facility, requiring only a very small number of personnel on duty in daily operations. The entire production system can operate autonomously 24 hours a day, seven days a week, and all processes — including machine room scheduling, equipment status monitoring, and intelligent task allocation — can be completed through remote systems. When the machine room lights are turned on, rows of computing power cabinets, each taller than a person, come into view. The reporter follows Liu Li inside the cabinets, where a wind-cooling system continuously supplies cold air from the bottom, keeping the ambient temperature of the high-speed operating equipment stable between 20 and 24 degrees Celsius. Densely arranged graphics processors and intelligent computing accelerator cards are carrying out massive inference calculations, while colorful fiber-optic cables are laid out in a crisscross pattern, continuously transmitting data. The basic production unit of the AI (artificial intelligence) world — the token — is born here in batches. "Just as electricity usage serves as a unit of measurement, the token is the smallest unit for AI to process information," Liu Li explained. As intelligent agents flourish, market demand for tokens is rising exponentially, and independent third-party token factories have emerged in response. By optimizing inference architecture, intelligently scheduling computing power, and adapting to model iterations, they proactively produce, refine, and output standardized high-quality token products, providing AI token services to the outside world. "At present, the factory has connected with leading domestic large model companies such as Zhipu, Kimi, and DeepSeek. Given the industry characteristic of rapid model iteration, we are able to follow up and adapt to updates on the same day in most cases," Liu Li said. The quality of tokens depends on the capabilities of the partner large models, while output efficiency tests the technical strength of the token factory. It is like ore mining: high-quality models produce high-quality "ore," while the factory's continuous iteration of inference architecture and optimization of software support determines mining efficiency. Since starting operations in June this year, the Beijing No. 1 Token Factory has already achieved impressive production capacity: it can currently produce 300 billion to 500 billion tokens per day, half of all tasks can be delivered within 6 seconds, 90% of task response times are kept within 10 seconds, and overall performance stability fluctuations are controlled within 2%. According to the plan, by the end of this year, the factory's daily production capacity will climb to 1.4 trillion tokens. Looking at the city as a whole, Beijing is home to leading large model companies of all kinds, and more than half of domestic token production is currently concentrated in Beijing.

"Bridge builders" bring tokens into thousands of industries

Between the tokens flowing out of factories and their landing in industry, there are still barriers to cross — how to efficiently and precisely deliver standardized tokens to thousands of industries. In Beijing, a group of professional "bridge builders" has emerged. In the Zhongguancun AI North Latitude Community in Haidian, Beijing, Yuan Lijiang, an entrepreneur running a "one-person company," begins each workday by turning on his computer and checking the working status of his intelligent agent "employees." Riding the AI wave, Yuan has developed multiple intelligent agents to provide services to companies with overseas expansion and intelligent transformation needs. However, busy as he is, he occasionally runs into small hiccups. "Sometimes when a task assigned to an intelligent agent reaches a certain step, the system suddenly prompts that the server is busy and the model request fails. The underlying reason is still tight computing power," Yuan said. To obtain a suitable token package, he sometimes even has to set an alarm to snap one up. To solve the "sweet trouble" in Yuan's entrepreneurial journey, a brand-new profession has emerged. In the office area of Qingcheng Jizhi in the Jinqiu International Building in Haidian District, a group of computing power optimization engineers is helping enterprises "run more smoothly and spend less" when using domestic computing power, making token services more efficient. "No matter how good a model is, if the computing cost is too high, it can only stay in the laboratory," Shi Tianhui, co-founder of Qingcheng Jizhi, told the reporter. In the past, when the industry discussed large models, it was keen on comparing parameter scale and leaderboard rankings. Today, as large models enter the industrial implementation stage, what enterprises care about most is whether the model is "affordable" and whether tokens are too expensive. Based on this pain point, his team developed a one-stop token service platform called "AI Ping," which works like a computing power price-comparison navigation platform. It can monitor in real time the performance of more than 500 models and more than 30 computing power service providers on the market, and automatically match users with the optimal computing power scheduling route. "Whether users want their tasks to run fast or want to consume fewer tokens, the platform can use algorithms to meet the corresponding needs," said Liu Dongsheng, a computing power optimization engineer. Computing power optimization engineers, frontier deployment engineers, model tuning specialists… As demand for tokens grows vigorously, a batch of new professions and new models is emerging. They connect computing power, models, and industrial needs more tightly, becoming "bridge builders" for AI's journey into thousands of industries and opening up the "last mile" of AI implementation. The meteorological team of large model company Zhipu, together with organizations including the Public Service Center of the China Meteorological Administration, jointly built the hundred-billion-parameter "Fenghe" large model and established an intelligent agent matrix covering public users and key industries such as transportation and energy, promoting the intelligent, scenario-based, and proactive transformation and upgrading of meteorological services. Researchers at the Zhongguancun Institute of Artificial Intelligence went deep into the front lines of metal manufacturing production such as annealing and electroplating, turning manual inspection experience into a data-driven quality early-warning system. They initially achieved online perception of electroplating anomalies and intelligent screening of bonding sphere and coating defects, and revealed the relationship between current, annealing tension, and mechanical properties, providing a basis for process optimization in traditional industries. After Beijing University of Technology introduced the self-evolving AI intelligent agent "Famou" into its research process, the fault diagnosis accuracy of its hydrogen production system rose from 92.26% to 95.04%, the research cycle was compressed from "weekly" to "hourly," the parameter count of the rotating machinery fault diagnosis model dropped by 34%, and the prediction error fell to 1% of its original level. From meteorological services to industrial manufacturing, from scientific research to livelihood support, tokens are integrating into thousands of industries in a subtle, nourishing way. An intelligent ecosystem linked by "tokens" is accelerating its formation in Beijing, injecting strong momentum into the high-quality development of the digital economy.

Forward-looking planning to ignite the token economy first

While many places across the country are still focused on building computing power hardware, Beijing has already systematically laid out the token economy across multiple dimensions, including supply, demand, standards, and finance. Recently, the city's first municipal-level special policy on tokens, the "Beijing Action Plan for Accelerating the Development of the Token Economy (2026-2028)," was released, marking the beginning of standardized development of the token industry at the city level. "Beijing has a precise and forward-looking vision in its token economy layout," said Liu Yi, senior vice president of SoftStone Power Group and president of SoftStone Intelligent Computing, when discussing why the country's first benchmark token factory was located in Beijing. Beijing gathers many leading domestic large model companies, and its industrial ecosystem advantage is irreplaceable. More importantly, the policy does not focus only on the supply side of building token factories, but also lays out the demand side in a forward-looking manner, providing corresponding support measures for token distribution and industry applications. "Only by fully activating market demand can token supply be continuously expanded in reverse, forming a positive industrial cycle," Liu Yi said. In addition to top-level guidance from municipal policy, the Beijing Economic-Technological Development Area took the lead in issuing the "Ten Token Measures," becoming the city's first district-level special policy covering the entire token industry chain. The policy anchors a complete closed loop of "factories supplying tokens, platforms distributing tokens, and communities using tokens," clearly supports the landing of ten-thousand-card-level token factories, builds a "super entrance" for token scheduling, and issues 100 million yuan in model vouchers or token vouchers each year. It also sets industrial development goals, planning for the intelligent economy industry scale in the area to exceed 400 billion yuan by 2030, greatly improving token output efficiency per unit of energy consumption, and cultivating a large number of intelligent agent innovation entities to gather and develop. In addition, the city's first batch of token loans has also landed in the development area. Enterprises' actual token consumption is included in credit assessment dimensions, opening a new financing path for the digital industry and injecting financial vitality into the vigorous development of the token industry. According to estimates in the "Research Report on AI Infrastructure Development in the Intelligent Agent Era (2026)" by the China Telecom Research Institute, domestic token consumption in 2026 will reach the level of 1 billion trillion. Strong demand opens broad growth space for the token supply track. In the view of Wang Zhongyuan, president of the Beijing Academy of Artificial Intelligence, Beijing is forward-lookingly encouraging the development of the token economy, cultivating new business models such as Token as a Service (TaaS), Agent as a Service (AaaS), and Result as a Service (RaaS), and exploring the establishment of token service quality evaluation, billing standards, and evaluation index systems to promote the transformation of intelligent agent services toward value creation. "At present, the comprehensive cost of token inference calls for China's foundational models is only one-tenth to one-sixth of the international advanced level, giving it a very significant cost advantage. Combined with massive application scenarios, this is expected to make Beijing the first in the world to ignite the token economy," Wang Zhongyuan predicted.