As the token economy surges wildly, have ordinary people got a piece of the pie?

How many Tokens do you consume every month? Token is the smallest unit for AI language models to process text. Since AI services are usually charged based on the number of Tokens processed, Token has also become a common measurement unit for model usage, cost, and context length.

With the large-scale application of AI, the "Token Economy" has also become a buzzword. China is already a major Token consumption country. By March this year, the country's average daily call volume exceeded 140 trillion, an increase of more than 40% in three months.¹ Following this figure, it is easy to imagine a picture: the call volume of a certain city is growing rapidly, the previously idle computing power centers are fully occupied, enterprises connect agents to customer service, R&D, sales and office processes, and employees begin to use AI to assist work on a large scale... These things are indeed happening, and Tokens can also be counted, ranked, and written into year-end reports like GDP. But this picture does not answer a question: what does this mean for the economy and society? What can ordinary citizens get from it?

We can understand this from the perspective of supply and demand. Count Tokens into the two accounts of "supply side" and "demand side". The former covers production capacity, calls, costs, investments, corporate income and local political achievements; the latter is related to wages, employment, leisure, social security, public services and household purchasing power.

In China, the special feature of the Token economy is that it is extremely good at expanding the first account, and is used to directly regard the achievements of the first account as the achievements of the second account. As a result, supply is called demand, intermediate input is called consumption, and enterprise cost reduction is called prosperity. Precisely because AI is a truly productive force, who controls the interpretation of this value and who takes its benefits is more worthy of inquiry.

Token, a new type of "electricity consumption"

Token is indeed a technical indicator that can be measured like electricity consumption, but the rise in industrial electricity consumption will not tell you whether the current flows to a more efficient industry or inefficient excess capacity. Similarly, a Token call may generate a report that really saves you hours, or it may just repeatedly modify a paragraph of text that is not used at all in the end.

Usage is input, task is process, revenue and profit are results at the enterprise level, and productivity is a higher-level economic result. These layers are completely different distances from the lives of citizens. The improvement of corporate profits at least has a path: distribute revenue to employees, who then spend it. But if Token continuously improves the order receiving efficiency and order volume of food delivery riders without increasing their actual income, this path is broken.

This is not an abstract concern. The 140 trillion daily average call volume itself is worthy of being disassembled. Volcano Engine, under ByteDance, disclosed in April that the daily average Token usage of the Doubao large model has exceeded 120 trillion.² That is to say, the vast majority of the national total comes from this one platform, and what drives this figure up is mainly AI video generation: the generation and iteration of a single video often consumes tens of millions of Tokens. This picture is not the same as "comprehensive penetration across thousands of industries". It is more like a company, a type of scenario, and a content industry that is exploding. Taking the average daily call volume as an indicator of the overall economic health is difficult to hold up in front of this structure.

When enterprises buy Tokens, they are of course counted as demand on their own accounts. But in the entire national economy, it is likely to be only an intermediate input from beginning to end. Only when it lowers product prices, raises wages, increases leisure, improves public services, or creates new products that people are willing to buy and can afford, can Tokens truly enter the second account.

This is the division between the supply side and the demand side. The former is a typical productivism, which is probably also the default narrative when many people talk about the "Token Economy": putting production capacity, fixed assets, technical capabilities and industrial scale ahead of household income, social security and final consumption. This ordering is not accidental. Residents' sense of security is not easy to quantify, and service quality is difficult to write into year-end reports; but ten-thousand-card clusters, Token output per kilowatt-hour, daily call volume, number of parks, and investment scale are all quantifiable, planable, financeable, subsidizable, rankable and assessable. Chips, servers, power grids, liquid cooling, and data centers all have clear procurement entities, which are suitable for industrial funds, bank credit, state-owned enterprise investment, and local projects to intervene.

Conversely, increasing unemployment insurance, issuing pensions to the floating population, and improving grassroots public services require long-term and regular fiscal commitments (China's current unemployment-related expenditure is less than 0.1% of GDP, while the average for OECD countries is 1.0%)³ and will directly touch the division of responsibilities between the central and local governments. The former is a political achievement, while the latter is a trouble.

As a result, a technological revolution is first interpreted as new fixed assets, new local production capacity, and new administrative visibility. This interpretation was not completed by one party's decision. Local governments need investment and political achievements, state-owned data companies need business, computing power centers need to run at full utilization, industrial parks need to attract investment, cloud vendors and model companies need early customers — and state-owned enterprises and government departments can just act as the first batch of buyers.

The Token economy neither grew out of the market by itself nor was created by a single order. It comes from the promotion of all parties with different reasons: some want strategic security, some want asset utilization, some want revenue, some want financing, and some want assessment results. The interface that stitches these different calculations together is Token.

Movie *Blade Runner 2049*

Stimulating demand by expanding production capacity

Beijing Yizhuang provides an almost complete sample. The local government proposes to expand the scale of the intelligent economy industry to 400 billion yuan by 2030: build four ten-thousand-card-level Token factories, the computing power scale will exceed 100,000P, the Token output per kilowatt-hour will be increased to more than six times the current level, build Token distribution and consumption infrastructure, cultivate more than 50 agent R&D collaboration platforms, and attract more than 5,000 OPC entities.

The strategy of this set of policy tools is in the same line as other Chinese industrial policies. It not only builds infrastructure, but also covers and subsidizes the entire industrial chain: 30% subsidy for computing power rental costs, 100 million yuan of data coupons per year, up to 50 million yuan for the Token aggregation and unified settlement platform, and up to 10 million yuan per year for the intelligent business reconstruction and delivery platform, all the way to subsidize enterprises to consume Tokens in actual scenarios. The government supports factories to produce Tokens, supports platforms to distribute Tokens, and also supports enterprises to use Tokens.

Subsidies have their reasonable parts. New technologies often get stuck in the early stage of promotion, such as insufficient infrastructure, enterprises are reluctant to take the lead in trials, and suppliers cannot find application scenarios. These are typical coordination failures. Computing power coupons and Token coupons can reduce the cost of the first adoption, force enterprises to open data and processes, and even generate services that did not exist before.

But it is also worth noting a "semantic transformation". The relevant chapter is titled "Comprehensively Expanding the Consumption Mode of the Intelligent Economy", but its main content is to encourage enterprises in medical health, commercial aerospace, automobile manufacturing and other fields to open high-throughput scenarios to drive large-scale and high-frequency consumption of Tokens; for enterprises that carry out actual scenario applications, 50% of the Token consumption cost will be subsidized with a maximum of 5 million yuan in financial support.⁴

That is to say, in a chapter titled "Expanding Consumption", the government subsidizes half of the consumption of enterprises' intermediate inputs. When enterprises buy Tokens, they are buying means of production, which belongs to intermediate input, not equal to residents' final consumption; policies that occur at the purchase end are not equal to demand-side policies. The government reduces the cost of Tokens for enterprises, enterprises increase calls, platforms get revenue, and the call volume in turn becomes a reason to expand production capacity — money thus circulates among the government, platforms, and enterprises, forming a closed loop. As long as this loop does not extend a branch leading to wages, social security, public services, reliable price reductions or resident transfer payments, the so-called demand side is still just the supply side buying its own products.

This is the distinction I most want to make clear in the whole article. It is not that we do not stimulate demand, but that we are too used to stimulating demand by expanding production capacity. The 4 trillion yuan stimulus in the past was like this, the photovoltaic industry was like this, and now it is the turn of Token, extending the same habit into the intelligent era.

Movie *Ghost in the Shell*

Limitations of the supply-side narrative

The supply-side narrative usually has a beautiful straight line: the model is stronger, the Token is cheaper, so it is used more; more usage leads to higher productivity; higher productivity leads to increased corporate revenue and profits, and workers' income increases accordingly, so money enters the demand side. Each link may hold true when viewed alone, but years of supply-side experience tell us that between every two links lie departmental interests, industry competition, and distribution patterns.

First of all, the most direct layer: Token is not equal to profit, it is first of all an uncontrollable bill. By mid-2026, it is this bill that has made many American enterprises hesitant. KPMG's quarterly survey shows that about half of the interviewed executives have reduced agent deployment, on the grounds that costs exceed benefits. Agents and Harness connect general-purpose models to the company's data, tools, permissions, memory and workflows, turning a single question and answer into a continuously executable task. The model is no longer just a chat box that employees occasionally open, but may become the carrier of all tasks, at least seemingly very efficient.⁵

But value cannot be judged only by consumption. From the upsurge of Token Maxxing (referring to the common practice in large technology companies around 2025: using AI as intensively as possible, measuring employees' innovation level by their Token usage, and taking consumption itself as a progress indicator. Since AI pricing rises linearly with usage, this practice directly led to the bill backlash in 2026), the myth that "Token equals efficiency" has basically been broken. Few people still doubt AI's ability to improve efficiency. The real question is: what tasks have been completed with large-scale AI use, how much time has been saved, how much income has been increased, and how much verification, rework, errors and compliance costs have been generated? If only the first three items are included in the statistics and the last four items are left for employees to digest silently, the enterprise will get a beautiful but one-sided assessment.

Thus there is the most basic ladder: call does not equal task completion, task completion does not equal increased revenue, increased revenue does not equal improved profit, and improved profit does not mean that macro productivity and social welfare rise synchronously. Each level requires new evidence to support. This is not a threshold specially set for AI, any intermediate input has to pass the same financial common sense test.

Even if an AI project does improve enterprise efficiency, the matter is not over. Call volume, platform revenue, enterprise profit and downstream customer value are four different things. According to Zhipu's 2025 annual report, API calls and revenue are growing rapidly: annual revenue increased by 132%, and the annual recurring revenue of the MaaS platform increased by 60 times within twelve months, while the adjusted net loss still expanded by 29% in the same period.

Moreover, this is not achieved by reducing prices to increase volume. Zhipu raised the API price by 83% in the first quarter of 2026, but the call volume increased by 400% instead.⁶ With pricing power, usage, and revenue all rising together, profits still have not kept up. These figures cannot prove that the model business has no value. High-growth enterprises may of course exchange losses for R&D and market, but they can prove that we cannot use a first-level indicator to impersonate the next-level result. The photovoltaic industry went through the same path, and the loss did not narrow with the expansion of scale, but grew larger: Tongwei Co., Ltd. recorded a net loss of 7 billion yuan in 2024, and forecasted a loss of 9 billion to 10 billion yuan in 2025.⁷

Further on is the distribution problem. If AI creates 100 yuan of incremental value, chip companies, cloud providers, model providers, application platforms, using enterprises, workers and consumers will all come to share this cake. From the logic of the supply side, most of this 100 yuan will most likely be taken away by enterprises upstream of the chain. Technology has not eliminated distribution politics, it has just put on a new coat of AI and reappeared.

The macro structure makes this distribution problem more obvious. The IMF estimates that China's total capital formation accounted for about 40% of GDP in 2024,⁸ and *Qiushi* stated that the household consumption rate in the same year was about 39.9%.⁹ The two figures have different calibers and cannot be directly compared precisely, but they point to the same structure: the investment scale is extremely large, and the household sector occupies relatively limited economic resources. This is the basic common sense of the Chinese economy.

The World Bank's estimate of the 2025 fiscal impulse is 1.6% of GDP, of which the part directly facing households is only about 0.5 percentage points, and the rest is mainly public investment. The bank also pointed out that the link between growth and employment is weakening.¹⁰ These phenomena cannot be attributed to AI, but they constitute the initial conditions for AI to enter the Chinese economy. When enterprises generally face weak demand and price competition, the easiest thing to do first with AI is not to invent a new consumption world, but to reduce costs, reduce recruitment, speed up supply, and improve export competitiveness.

For a single enterprise, this is very rational; when all enterprises are added up, it may become a typical fallacy of composition: suppressing wages and youth employment, making household expectations and consumption weaker, making price competition more fierce, and making enterprises more dependent on automation to continue to reduce costs. I do not intend to assert the final outcome of AI, but this cycle has already repeated in many industries. If productivity gains mainly stay in fixed assets, platform rents and upstream corporate profits, the efficiency improvement at the micro level may not be transformed into income growth at the macro level.

Movie *Modern Times*

A thousand Henry Fords, and the missing half

In the discussion of the Token economy, the problem is no longer just how many Tokens are consumed, but that the entire production model is being reshaped. In an exclusive interview with Huxiu, Sun Tianshu, a professor at Cheung Kong Graduate School of Business, said:

The greater opportunity for this round of AI in China is not to replicate a new consumer Internet, but in AI To B: the industrial reconstruction of thousands of industries by agents; every vertical industry can have its own "industrial brain" with the help of AI, and every enterprise needs to build its own "agent system"; to complete this transformation, a group of "Henry Fords" who understand industrial tacit knowledge, business architecture and organizational change and have AI-native thinking are needed — they are not adding an AI function (+AI) to the old process, but redesigning the production mode, business model and industrial order (AI+) from the first principles, just like Mr. Ford in the electrical age 100 years ago, at the Highland Park plant in Detroit, designed the "assembly line mode" centered on electricity, reconstructing productivity and production relations, and creating a new era of "electricity-native" for the automobile industry.

The metaphor of "Henry Ford" has insights, but also blind spots. The insight lies on the supply side: stuffing a chat box into the old process usually only brings partial improvement, and what enterprises really need to do is to redesign