Since the start of this year, the token economy has continued to heat up, with competition among regions reaching a white-hot stage. In June, Guangdong issued the country's first provincial-level special policy on the token economy, aiming to build a national token supply highland, a benchmark for intelligent applications, and a token overseas hub by the end of 2027.
In July, Yangzhou became the first city in Jiangsu Province to launch a special policy on artificial intelligence token vouchers, incentivizing market entities to develop and deploy large models and intelligent agents. In August, Inner Mongolia released "Token Eighteen Measures," aiming to build a national high-quality token supply base, a token overseas hub, and a pilot demonstration zone for token economy development.
Recently, Beijing released the "Action Plan for Accelerating the Development of the Token Economy in Beijing (2026-2028)" (hereinafter referred to as the "Ten Measures for the Token Economy"), covering the entire chain including token factory construction, quality evaluation, pricing and settlement, intelligent agent cultivation, financial and talent support, and token overseas expansion.
Zhao Gang, president of the Saizhi Industry Research Institute, said in an interview that compared with the computing power economy, the token economy resolves two major industry pain points. First, it breaks through the bottleneck of standardized measurement of intelligent service value. Computing power can only measure computational input and cannot measure intelligent output. As the smallest unit for large models to process information, tokens are the basic standard for billing and settlement of intelligent services, enabling large-scale continuous measurement of intelligent output and laying a solid foundation for value accounting across the industrial chain and business model innovation.
Second, it opens a path for the market-based allocation of data elements, shifting value evaluation from "how much is this batch of data worth" to "how much high-quality intelligent service can the data produce," providing a new value anchor of "usable but invisible, controllable and measurable" for the data element market. Experts interviewed generally believe that the token economy has considerable prospects, but the core contradiction in industrial development lies not in expanding token production capacity, but in establishing an institutional framework capable of releasing real value.
Forward-Looking Layout: Beijing Builds a Pilot Testing Ground for the Token Economy
More than half of domestic token production is concentrated in Beijing, and Beijing's release of the "Ten Measures for the Token Economy" reflects clear strategic considerations. Zhao Gang believes that Beijing's layout around multiple dimensions including token factories, technological breakthroughs, quality evaluation, distribution platforms, metering and settlement, and overseas services is intended to seize the commanding heights of the token economy and anchor three major development positions.
It aims to become a rule-maker for the token economy, taking the lead in establishing standard systems for token factory grading, quality evaluation, and pricing and circulation, and exporting industry rules. It seeks to build a scheduling and distribution hub for the token economy, relying on advantages in models, data, and talent to provide unified interfaces, metering and settlement, and token financial services, scheduling computing power resources upward and distributing intelligent value downward. It also aims to build a highland for the token economy application market, opening scenarios in industrial manufacturing, scientific research, biomedicine, education, and other fields to drive both token supply and consumption.
Wang Peng, associate researcher at the Management Research Institute of the Beijing Academy of Social Sciences, analyzed in an interview that this policy is a forward-looking grasp of the trend of AI industrialization. Beijing has gathered a large amount of scientific research resources, AI enterprises, and application scenarios, and needs to transform innovative resources into new industrial growth points. The "Ten Measures for the Token Economy" systematically lays out production, evaluation and distribution, intelligent agent applications, and the industrial ecosystem, essentially exploring a path for artificial intelligence to move from technological innovation to industrial economy.
By the end of 2025, Beijing's core artificial intelligence industry scale exceeded 300 billion yuan, with more than 2,400 AI-related enterprises forming a complete industrial chain. The first phase of Beijing No. 1 Token Factory has achieved a daily production capacity of 1.4 trillion tokens, putting infrastructure construction at the forefront nationwide. At the same time, it should be noted that Beijing's first-mover advantage cannot be directly copied. Basic systems such as cross-regional circulation, cross-border settlement, and data compliance in the token economy are still in the exploratory stage, and a unified national institutional framework urgently needs to be established.
Institutional Breakthrough: Building a National Rule System With Central-Local Coordination
Against the backdrop of localities competing to introduce policies while unified national rules are still being improved, how to build a national institutional framework for the token economy has become an industry focus. Zhao Gang proposed that the institutional framework should be led by the National Data Administration, in coordination with multiple departments including the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Cyberspace Administration of China, and the State Administration for Market Regulation, following a path of "standards first, industrial development, central-local coordination."
First, unify token statistical calibers, tokenization rules, and accounting methods, establish a tokenizer filing and metering audit system, incorporate response accuracy and task completion rates into evaluation, and promote a shift in billing models from purely quantity-based pricing to equal emphasis on quantity and quality. Second, improve industrial systems for token production, distribution, and trading, and support the development of token factories, evaluation institutions, and distribution platforms. Third, ensure national coordination, with the national level focusing on standards, security bottom lines, and statistical monitoring, while local levels focus on scenario opening, computing power scheduling, and financial tool innovation, forming a nationwide coordinated effort.
Wang Peng believes that national institutional construction should grasp three key points: establishing unified measurement and evaluation standards to break down barriers between different models and platforms; strengthening the data security bottom line and releasing the value of data elements under secure conditions; and exploring new trading and settlement mechanisms suited to AI characteristics. As a testing ground, Beijing's institutional exploration offers universal experience that can be promoted, while also carrying the particularity of local endowments. Institutional methods such as token quality evaluation, factory grading evaluation, treating distribution platforms as new infrastructure, and intelligent agent scenario cultivation can be referenced by other regions. However, Beijing's innovation ecosystem, with its concentration of high-end talent and leading model enterprises, is difficult for other regions to replicate in the short term. Localities do not need to fully copy Beijing's model and should instead pursue differentiated development paths based on their own resource endowments and industrial foundations.
Real-World Tests: Multiple Bottlenecks Need to Be Broken Under High Industrial Growth
The token economy connects the entire chain of computing power, foundation models, data, and intelligent agents, but the biggest contradiction in industrial implementation is not expanding the scale of supply, but establishing industrial rules for value measurement, quality control, and efficient circulation. Zhao Gang summarized the real bottlenecks into three points.
First, constraints on computing power and energy. There are risks in the supply of high-end computing power chips, and domestic computing power chip production capacity is tight. The construction of integrated computing power networks lags behind, and many computing hardware resources are difficult to connect logically. High computing power costs push up token prices, squeeze upstream and downstream profits, and suppress consumer demand.
Second, the lack of metering and pricing standards. Different models have different tokenization calibers, and the same semantic output can differ several times in token count. Counting quantity alone cannot reflect the true value of intelligence, and the transaction pricing mechanism is still in an early stage.
Third, the phenomenon of "token idling" is prominent. A large number of intelligent agents remain limited to local optimization and have not been embedded in enterprises' core businesses. Token consumption continues to rise, but business returns fall short of expectations. "The biggest challenge for the token economy is not expanding production capacity, but establishing a rule system for value evaluation," Wang Peng said. At present, token calculation and service quality vary greatly across platforms, and the output effect of the same number of tokens is uneven. The evaluation system needs to shift from "looking at quantity" to "looking at quality." Beijing's "Ten Measures for the Token Economy" sets grading indicators for token factories, covering the number of adapted models, throughput speed, first-token latency, cache hit rate, and power usage effectiveness, precisely to reverse the orientation of one-sided competition over output. At the same time, it is necessary to improve circulation and trading mechanisms, rely on distribution platforms to achieve intelligent routing and unified settlement, and govern ineffective token consumption.
Financial Innovation: Exploring Data Credit and Strengthening Credit Risk Control Defenses
The "Ten Measures for the Token Economy" proposes innovative financial tools related to computing power and tokens. At present, institutions such as Bank of China and Bank of Jiangsu have already implemented "computing power loans" and "token loans," incorporating enterprises' token consumption and computing power contracts into credit assessment and opening new financing paths for asset-light AI enterprises.
Dong Ximiao, chief economist of Zhaolian and executive director of the Shanghai Finance and Development Laboratory, said in an interview that traditional credit relies on fixed assets and financial statements, while "token loans" use dynamic operating data such as token consumption, computing power contracts, and accounts receivable as the basis for credit, shifting the risk control logic from "looking at bricks" to "looking at data and capabilities." Token invocation data can intuitively reflect the business activity of AI enterprises and is traceable, making it an important business measurement unit in the AI era.
However, supporting data infrastructure, risk control models, and regulatory rules are not yet mature, and multiple risks lurk behind innovation: token data carries risks of inflation and manipulation; relying solely on token consumption can easily misjudge an enterprise's debt repayment capacity; rapid technological iteration in the AI industry and market fluctuations may bring pressure on asset quality; and there are also hidden dangers such as gaps in regulatory guidance and leakage of enterprise operating data. Dong Ximiao believes that "token loans" are not yet ready for nationwide full-scale promotion and are suitable for pilot programs in regions with concentrated AI industries. Financial institutions cannot treat token consumption as the core credit basis, but only as a supplementary reference. They must conduct multi-dimensional cross-verification in combination with computing power orders, accounts receivable, and financial statements, focusing on identifying effective tokens that create real value, and gradually expand the scope of practice after industry standards and the regulatory system are complete.
Market Forecast: B-End to Lead and Usher in a Wave of Token Scalization
Experts interviewed unanimously judge that the first wave of large-scale growth in token consumption will occur first on the B-end, while a large-scale C-end breakout still needs to wait for technological iteration. "The key lies in the unit economics of effective tokens, that is, the measurable business value created per unit of token consumed," Zhao Gang analyzed. In C-end scenarios, free versions of large models basically cover daily needs, and mature paid business models are still relatively few. In B-end industrial and scientific research intelligent agents, a single complex task can consume millions to tens of millions of tokens, with consumption concentrated in value-creating links such as solution design and process optimization, and input-output can be quantified. B-end demand is stable and returns are calculable, which will support the token economy in completing scale-up from 1 to 10; larger-scale growth driven by C-end multimodal digital avatars and embodied intelligence still belongs to a future stage.
Wang Peng noted that China has a huge base of generative AI netizens, but enterprises have a stronger willingness to pay. Some forecasts show that the global intelligent agent market will grow from US$37.2 billion in 2025 to US$312.2 billion in 2030. Expanded token supply will empower downstream sectors in three ways: compressing unit costs through mass production and lowering the threshold for small and medium-sized enterprises; unifying interfaces through distribution platforms and reducing the cost of connecting multiple models for development; and using quality evaluation to force intelligent agents to improve task completion rates and promote application adoption. The overall industry will follow a development path of "B-end first, C-end diffusion."