The Return of Price Theory: An Economic Essay in the Age of Artificial Intelligence

Fears on the Old Foundation

Each era has its own foundation. People live, work, make judgments, and feel fear on it, and also imagine the future on it.

It was difficult for people in the agricultural era to imagine that a person's life could not revolve around land, seasons, and famine. When the steam engine emerged, many people first saw the unemployment of handicraftsmen but could hardly foresee railways, cities, the factory system, modern finance, and the new middle class. When electricity first appeared, people saw brighter nights but could not fully imagine refrigerators, the film industry, modern hospitals, urban nightlife, household appliances, and electronic computers. When the Internet first emerged, many people thought it was just faster mail and a larger library but did not foresee mobile payment, short - videos, cloud computing, food delivery platforms, ride - hailing services, online education, and global real - time collaboration.

When new technologies arrive, the biggest obstacle is often not the technology itself but the imagination on the old foundation.

Today, artificial intelligence is also understood within the framework of the old foundation. Many people reason like this: What used to be done by ten programmers, copywriters, translators, analysts, and customer service staff can now be done by one person plus an AI, so nine people will be unemployed. This judgment seems cold, realistic, and anti - utopian, but in fact, it is still a way of thinking from the old world. It understands the future as a cost - cutting on today's task list, technological progress as a replacement within existing positions, and economic life as a pre - written form.

But truly great technological revolutions never simply reduce a few lines in the old form; they reinvent the form itself.

The most important economic implication of artificial intelligence is not to do old jobs more cheaply but to push a large number of products and services that did not exist, were not feasible, were too expensive, too scattered, too niche, or too difficult to organize in the past into the feasible set of humanity. It does not simply replace existing labor but reduces cognitive costs, communication costs, trial - and - error costs, matching costs, and organizational costs, thereby releasing suppressed demands in the past, creating transactions that could not be maintained before, and generating industries that did not have names in the past.

Looking at AI from the old foundation, people see a reduction in jobs. Looking at AI from the boundaries opened up by new technologies, people see an explosion of product and service spaces.

This is exactly why price theory becomes important again in the era of artificial intelligence.

AI reduces production, trial - and - error, and matching costs, but it does not eliminate the problem of demand discovery. On the contrary, it expands the space of unknown demands. The more things that can be produced, the more society needs to know which things are truly valuable; the more personalized services that can be provided, the more society needs to know who needs them, when they need them, how much they are willing to pay, who is responsible, and how to form continuous transactions.

The future problem is not "Can machines produce?" but "How can humans discover what is worth producing?"

The return of price theory starts right here.

The Economic Implications of a $2 Trillion Revenue

Let's start with a seemingly exaggerated but not absurd assumption in economics.

Is it possible that in the next fifteen years, companies like OpenAI, Anthropic, or similar large - model infrastructure companies could become companies with an annual revenue of $2 trillion? Note that here we are talking about revenue, not valuation. Valuation can come from the imagination of the capital market, discount rates, risk preferences, and bubbles; revenue must come from real purchases, real payments, and real transactions.

This is not a prediction of the stock price of a particular company, nor an endorsement of a particular business model. It is a price theory exercise: If an upstream intelligent infrastructure company can obtain an annual revenue of $2 trillion in the long term, then there must be a larger - scale, higher - willingness - to - pay ecosystem of new products and services downstream.

People without economic common sense will immediately say: This means that the giants have monopolized the world and are extracting rents from all industries. This judgment may have some possibility because platform monopolies are indeed worthy of vigilance. But if we only understand the $2 trillion revenue as "rent - extraction," we will miss the more important economic logic.

For any upstream infrastructure company to continuously obtain a $2 trillion revenue, there must be a premise: The downstream is willing to continuously pay this amount. Why is the downstream willing to pay? Because AI, as an intermediate input, creates greater output, higher quality, lower costs, more new products, or stronger consumer willingness to pay for the downstream. Without a huge expansion of downstream value, the upstream revenue cannot be maintained in the long term.

This is the basic economics of intermediate inputs.

A downstream enterprise purchases model calls, intelligent agents, computing power, and automation capabilities not to do charity for the upstream, nor to simply share existing costs, but because after these inputs are combined with its own scenarios, data, processes, customers, brands, responsibilities, and organizational capabilities, they can create more value. The higher the marginal output value of AI, the higher the price the downstream enterprise is willing to pay; the more prosperous the downstream applications, the more likely the upstream infrastructure revenue will expand.

We can make a simple ledger deduction. If the cost of AI computing power, model calls, and intelligent services accounts for 10% of the cost of downstream final products, then the $2 trillion upstream revenue may correspond to a downstream final product and service market of about $20 trillion. If the cost ratio is 5%, it may correspond to a downstream ecosystem of about $40 trillion. Of course, the cost shares of different industries are different, and the future pricing structure will also change continuously, but this deduction reveals a basic logic: How much "electricity bill" the upstream "digital power grid" can collect depends on how much new value the downstream creates with this electricity.

Power companies have revenue because electricity drives factories, households, cities, hospitals, and entertainment systems. Cloud computing companies have revenue because downstream digital enterprises use cloud services to build search engines, social media, video platforms, finance, logistics, and enterprise software.

If the AI infrastructure one day reaches a revenue of $2 trillion, it is not because the world gives it taxes out of the blue, but because countless downstream enterprises, individuals, and organizations use it as a basic input for cognition, reasoning, design, matching, and automation to create a larger - scale new economic activity.

This fact in turn reminds us that what is really worth paying attention to may not be how big OpenAI or Anthropic itself will be, but what kind of downstream application layer may emerge behind them.

In the next fifteen years, the overall scale of application - layer companies is likely to be much larger than that of the infrastructure layer. The infrastructure layer provides general intelligence, and the application layer completes demand discovery. The former sells capabilities, and the latter sells specific values. The infrastructure layer is like the power grid, water network, and roads in the digital age; the application layer is closer to consumers, organizes scenarios, industry processes, trust relationships, and specific needs. Industries such as healthcare, education, law, insurance, finance, entertainment, psychological services, elderly care, enterprise management, scientific research tools, urban services, household services, cultural experiences, and personal growth may all produce large - scale application - layer companies.

What consumers ultimately buy is not "model parameters" but cured diseases, understood anxieties, improved learning abilities, saved time, improved lives, reorganized workflows, created experiences, and solved problems. What enterprises ultimately buy is not "tokens" but lower inventories, higher conversion rates, better risk control, faster R & D, more detailed customer service, more stable supply chains, and clearer organizational decisions.

Therefore, the fact that the application layer is larger than the infrastructure layer is not a miracle but a normal result after the diffusion of general - purpose technologies. Electricity is important, but the industrial system driven by electricity is larger; cloud services are important, but the digital economy growing on the cloud is larger; large models are important, but the new products, new services, new organizations, and new lifestyles supported by large models may be much larger.

If an AI infrastructure company with a $2 trillion revenue really appears in the future, we should not first understand it as the end of the world, nor simply regard it as a technological cult. We should first ask a price theory question: Where does such a huge willingness to pay come from? What kind of downstream innovation does it support? What demands that could not be traded in the past does it release? What products and services that did not exist in the past does it make possible?

This is economic common sense.

Scarcity Will Not Disappear, but Only Change Its Form

Many people mistakenly think that the end of technological progress is the disappearance of scarcity. As long as AI is strong enough, there are enough robots, computing power is cheap enough, and goods are abundant enough, the price mechanism will withdraw, and the market will become redundant.

This is a misunderstanding of scarcity.

Scarcity is not simply a lack of physical quantity. Scarcity is relative to people's desires, time, knowledge, location, relationships, opportunity costs, and future uncertainties. As long as people's desires are heterogeneous, changeable, and context - dependent, scarcity will not disappear. It will only shift from "availability" to "suitability," from "sufficiency" to "being exactly what is needed at this moment," and from material shortage to structural shortage.

In the industrial era, much of the scarcity was manifested as a lack of quantity: not enough food, clothes, housing, doctors, schools, or transportation. The task of mass production and modern organization was to replicate these basic products and services in large quantities.

But in a more prosperous and intelligent era, many key scarcities are no longer simply about quantity. A person needs not just any lunch but a lunch that suits his physical condition, blood sugar fluctuations, exercise consumption, emotional needs, and aesthetic preferences today. A child needs not just any math class but a learning path that suits his current understanding obstacles, attention state, family environment, and self - esteem structure. An old person needs not just any health advice but a service relationship that he can really believe in, really implement, and really stick to. An enterprise needs not just any AI system but a specific solution that can be embedded in its own processes, incentives, organizational structure, and customer relationships.

This is the scarcity in the era of differentiation.

More specifically, at least three types of scarcity will be intensified in the AI era.

The first type is the scarcity of adaptability.

Whether a product or service is suitable for a person, an organization, a moment, or a situation will become increasingly important. In the era of standardization, the important question was "Is there enough supply?" In the era of differentiation, the important question is "Is this supply exactly suitable for me?" AI makes personalization possible, but it also makes the adaptation problem more complex. Because people's bodies, minds, relationships, work, and preferences are all changing. What is truly scarce is not any supply but appropriate supply.

The second type is the scarcity of trust.

AI can give advice, but whether the advice can be believed, adopted, and implemented is another matter. Just because a patient knows he should take medicine does not mean he will take it in the long term; just because a student knows he should study does not mean he will persevere; just because an enterprise knows it should transform does not mean the organization will accept it; just because an old person knows he should control his diet does not mean he is willing to change his decades - long lifestyle. The value of many services lies not in the information itself but in the trust relationship that turns information into action. In the future, trust, reputation, responsibility, and companionship will become important economic assets.

The third type is the scarcity of direction.

AI can generate countless solutions, but the capital, time, organizational attention, and experimental opportunities in the real world are still limited. An enterprise cannot implement a hundred strategies at the same time, a laboratory cannot build a hundred reactors at the same time, a hospital cannot restructure all processes at the same time, and a city cannot test all governance schemes at the same time. When the possibilities explode, what is truly scarce is the ability to choose a direction: choosing which path to take, taking which risks, and giving up which tempting possibilities.

Therefore, the power of AI is not to turn the world into a completely homogeneous abundance but to make it possible for scale and personalization to occur simultaneously for the first time. In the past, only a few wealthy people could enjoy private doctors, private teachers, private consultants, private assistants, private psychological companions, private designers, and private research teams. In the future, these may enter the lives of ordinary people with a new cost structure. But once they enter ordinary life, the problem is no longer "Can it be produced?" but "How to adapt, how to trust, and how to choose a direction."

This means that the market will not disappear. On the contrary, the market will become more active, more detailed, and more deeply involved in the micro - level of life. Because as products and services become more and more specific, society needs a mechanism to discover the real value of different people, at different times, and in different scenarios.

This mechanism is price.

Price Is a Discovery Mechanism, Not Just a Distribution Mechanism

Price is often misunderstood as a cold - blooded distribution tool. It seems that only when things are in short supply do we need price to decide who gets them and who doesn't; once technology is advanced enough, price can be abolished, and distribution can be left to algorithms.

But the most profound function of price is not to distribute known goods but to discover unknown information.

How much a person is willing to pay for a service contains a lot of information that others cannot know in advance: his intensity of preference, time cost, income constraint, degree of urgency, alternative choices, risk judgment, degree of trust, and emotional state. This information is not simply written in a database, nor can it always be obtained through questionnaires. Many times, people themselves do not fully know what they want until a product appears, a price appears, a comparison occurs, or an experience takes place.

Price does not come into play only after the demand is completely given. Price participates in the formation and discovery of demand.

This is especially important in the AI era. Because AI will significantly expand the set of producible products and also significantly reduce the cost of generating new product prototypes. In the past, many product ideas died before they had a chance to be tested in the market due to R & D costs, organizational costs, and communication costs. Now, more people can quickly make prototypes, more small teams can enter the market, and more niche demands can be tried to be met. The problem then changes: It is not a lack of ideas but a lack of a mechanism to screen ideas; it is not a lack of possibilities but a lack of a mechanism to judge which possibilities are worth investing real resources in.

Price is the core of this screening mechanism.

When an entrepreneur proposes a new product, he is actually making a conjecture about future demand. Whether consumers buy or reject it is a test of this conjecture. If the price is too high and the product cannot be sold, it means that the value is insufficient, the positioning is wrong, the cost is too high, or the target group is incorrect. If the product is still bought at a relatively high price, it means that a certain demand is stronger than bystanders imagine. Profits attract imitation and expansion, and losses force exit and correction. This process is not a simple transaction but a large - scale distributed experiment conducted by society under uncertain conditions.

Without price, society loses this experimental feedback.

This is especially true in the era of differentiation. Suppose AI can generate a thousand new education services, ten thousand new health management models, and one million personalized entertainment experiences. Which are real demands and which are just technological shows? Which consumers are willing to pay continuously, and which will only try once? Which services can be scaled up, and which can only stay in the niche market? Which require human participation, and which can be fully automated? Which are worth capital investment, and which should be quickly abandoned?

These problems cannot be solved only by expert judgment, nor can they be determined by a central algorithm at once. They need price, transaction, profit, loss, and competition to continuously screen.

Price is also a mechanism for compressing local knowledge.

Whether a consumer is willing to pay at a certain moment is not just a mechanical function of income and price. It may include his physical feelings today, yesterday's experiences, family relationships, work pressure, future expectations, social identity, and aesthetic preferences. Whether an enterprise is willing to pay for an AI system is not just a matter of technical indicators but a comprehensive reflection of its internal processes, employee capabilities, customer structure, regulatory risks, and competitive pressures. This local knowledge usually cannot be fully transmitted to a center. Market prices transform these scattered judgments into observable action signals.

Therefore, the price mechanism is not a remnant of the old era but a discovery device in an open future.

The stronger AI is, the more possibilities there are; the more possibilities there are, the more important screening is; the more important screening is, the more important the price mechanism is.

Incentives: Why New Products Won't Appear Automatically

Merely having technology will not automatically produce new products and services.

There is a long and complex process between the capabilities in the laboratory and the products in the market: Who will identify the scenarios? Who will bear the risks? Who will organize the team? Who will transform the processes? Who will educate the consumers? Who will handle the responsibilities? Who will face the failures? Who will turn a technological possibility into a stable service delivery?

This requires incentives.

Price theory is not only about price levels but also about incentive structures. Why do people invest time, capital, reputation, and organizational capabilities to explore new products? Because they believe that if the exploration is successful, they can get rewards. If all new services are immediately copied for free