Summary of Key Points
The scale of China's token economy is astonishing (with an average daily transaction volume of 140 trillion tokens), but there are three critical issues:
1. The structure of usage is monolithic, with ByteDance accounting for 120 trillion tokens, primarily due to AI-driven video generation, rather than widespread adoption across various industries.
2. There is overheating on the supply side while demand remains lukewarm; policies focus on subsidizing companies' capacity expansion, but tokens, as intermediate inputs, do not translate into residents' income, consumption, or public services.
3. The distribution of benefits is uneven, with most of the gains from improved AI efficiency going to upstream companies, leaving workers out of the profit-sharing loop. If the issue of connecting demand with supply is not addressed, the token economy may repeat the same mistakes as past capacity expansions that did not improve residents' welfare.
Detailed Analysis
1. 140 Trillion Tokens: Behind the Figures, a Dominant Player
The large figure of 140 trillion tokens might suggest widespread AI adoption, but in reality, the majority comes from ByteDance's DouBao model (120 trillion tokens daily), mainly for video generation—each video creation or modification consumes tens of millions of tokens. This is more like a single company's dominant use case rather than widespread application across all industries.
For example, it's like a city's entire electricity consumption coming from one factory; this does not mean the whole economy is thriving. High token usage indicates frequent AI use in certain areas but does not necessarily reflect improved overall economic efficiency or benefit for ordinary citizens.
2. The Supply-Side "Closed Loop": Subsidies Go to Companies, Not Residents
Local governments (e.g., Yizhuang in Beijing) have introduced subsidies to promote the token economy, covering 30% of computing power rental costs and 50% of token consumption, even issuing "data vouchers." However, these subsidies target companies that use tokens for production purposes (such as customer service or content creation), which is an intermediate input, not final consumer spending.
The result is a cycle of money flowing from the government to platforms to companies, without benefiting residents through wages, social security, or public services. For instance, Yizhuang's policies aim to stimulate consumption but actually subsidize companies' token usage, with no direct benefit to residents. This approach has been used in past initiatives like the 4-trillion yuan solar energy project and photovoltaic industry, often leading to overcapacity without real benefits for residents.
3. Tokens ≠ Efficiency ≠ Profit: High Usage Does Not Equal Profit
Many assume more token usage equals higher efficiency, but:
- Tokens are an input, not the output: A single transaction may generate useless text or save time, but usage alone does not reflect value. A KPMG survey in the U.S. shows that nearly half of companies have reduced AI investments due to costs exceeding benefits (e.g., Uber exhausted its annual AI budget without product improvements).
- Revenue growth does not equate to profit: Zhispu AI's revenue increased by 132%, but its net loss expanded by 29%; the photovoltaic industry grew, yet Tongwei Corporation faced a nearly 10 billion yuan loss in 2025. High token usage does not necessarily translate into profits if it does not generate actual income or cost savings.
4. Who Gets the Benefits? Workers Often Left Out
All parties in the AI value chain (chip companies, cloud providers, model platforms, application developers) want a share of the benefits, but workers often receive little. For example:
- Food delivery riders use AI to improve efficiency but see no increase in earnings.
- Companies using AI may reduce hiring or cut wages.
- Although Ford's assembly lines increased efficiency, it had to raise wages (from $2.34 to $5 per day) to retain workers; in the AI era, companies tend to focus on cost reduction rather than wage increases.
5. The Biggest Risk: AI as a Capacity Tool Without Buyers
Some see the token economy as the next major driver after real estate, but there's a fundamental difference: Real estate creates demand for related services and products (construction, appliances), while AI is capital-intensive and labor-saving, requiring chips and servers without automatically creating many jobs or increasing residents' wealth.
If residents cannot afford the products (e.g., AI-generated services or goods), the token economy will be inefficient. This is similar to Ford's assembly lines; if workers could not afford the cars, high production would be meaningless. The token economy needs a different approach to create genuine value for consumers.
Conclusion
The core issue with the token economy is not that AI is useless, but that we are too focused on using "productionist" logic—focusing on capacity, usage, and investment while neglecting residents' income, consumption, and welfare. No matter how efficient AI is, if it does not improve people's lives, it remains a digital bubble. The key to a successful token economy lies in designing systems and policies that ensure fair distribution of benefits, similar to the principles of Fordism (stable wages, mass consumption, and social welfare).