Fifteen Dollars

There is a slide that appears in pitch decks now, usually somewhere around the middle, after the market size and before the team bios. You have seen this slide, or one like it, even if you have never sat through the pitch yourself.

The slide has a simple diagram. On the left, the old model: a subscription fee, maybe $99 per month, for access to the software. On the right, the new model: a percentage of value created, uncapped, aligned with outcomes. An arrow connects the two, pointing rightward, suggesting progress, suggesting inevitability.

The presenter clicks forward. "We don't charge for seats," they say. "We charge for results."

The argument goes like this: If an AI agent does something valuable for you, something you can measure in dollars, then the provider of that agent should capture a portion of that value. If the agent books you $50,000 in sales meetings, the provider should get $5,000. If the agent saves you 200 hours of legal review, the provider should get paid for some fraction of what those hours were worth. You are paying for what you get. The incentives are aligned. The customer only pays when the AI delivers.

This is called outcomes-based pricing, and there is a small but growing chorus of people who believe it will transform how software companies make money. The old model left value on the table, they argue. The new model captures it.

I think this argument is wrong. I think it misunderstands something fundamental about how prices get set, and I think the clearest way to see the error is to think carefully about food.

I.

Consider a simple negotiation about something essential.

You eat food and you drink water, and these are not optional purchases in the way that most purchases are optional. If you had to choose between food and a subscription to a streaming service, you would choose food without hesitation, because you understand at a cellular level that the streaming service is a luxury and the food is the thing keeping you alive. If you went without food or water for long enough, you would die, probably within a few weeks, certainly within a month.

So food is essential, non-negotiable, the most important line item in your budget. You prioritize it over almost everything else, and you do so automatically, without thinking, because your body has been making this argument to your brain for as long as you have been alive.

Now suppose I have made some food, a meal, and I would like to sell it to you. Given what we have just established about the importance of food, given that the alternative to eating is death, it would seem that you should be willing to pay me a great deal for this meal. The outcome I am providing is your continued survival, which is worth more than almost anything else you could name. What is that worth to you? A thousand dollars? Ten thousand? More?

And yet we both know the answer. You will pay me fifteen dollars, maybe twenty if the restaurant is particularly nice. The outcome is worth your life, but the meal is worth the market rate, and these are not the same number. The distance between them contains everything important about how pricing actually works.

II.

This is the problem with outcomes-based pricing: it confuses the value of an outcome with the value capturable by the provider of that outcome, and these are not the same thing at all. The gap between them has a name, and the name is competition.

Food is the most valuable product in human history. The demand is universal and non-negotiable. Every human being on earth is a customer, multiple times per day, for their entire life, from birth until death. The outcome being purchased is survival itself, which you would think would command a premium.

And yet farmers are not wealthy. They have never been wealthy, not as a class, not in any era or any country. In 1900, agriculture was one of the largest industries in the American economy, employing nearly forty percent of the workforce. The product these workers made was essential beyond measure, the foundation on which every other industry depended. Without farmers, there would be no cities, no factories, no universities, no anything.

The average farmer was poor. He worked brutal hours in difficult conditions to produce a product of infinite importance, and he was compensated at commodity rates, because that is what the market would bear.

This remains true today. Commercial farming is a business of thin margins and constant struggle, despite the fact that the product is more necessary than ever. The people who produce food have never figured out how to capture the value they create, and the reason is simple: too many people can farm. The outcome is valuable beyond measure. The provider of the outcome is replaceable.

III.

Let me be precise about when you can actually capture the value of an outcome, because it is possible, under certain conditions.

You need some form of scarcity, some moat, some barrier that prevents the customer from getting the same outcome from someone else. Without that barrier, the customer will simply find another provider, and you will be forced to compete on price until your margins approach zero.

Telecommunications companies capture value because building a network is expensive and heavily regulated. You cannot easily switch to a competitor if there is no competitor with towers in your area. The outcome they provide, connectivity, is valuable, but the reason they can charge for it is not the value of the outcome. The reason is that they have a local monopoly on provision.

Pharmaceutical companies capture value during the period of patent protection because no one else can legally manufacture their drug. The outcome they provide, treatment of a disease, is valuable, but the reason they can charge what they charge is not the value of the outcome. The reason is that they have a temporary legal monopoly.

Nvidia captures value because manufacturing advanced GPUs at scale is extraordinarily difficult, requiring fabrication technology that only a few facilities in the world possess. The outcome they provide, the ability to train and run AI models, is valuable, but the reason they can charge what they charge is not the value of the outcome. The reason is that the supply side is constrained in ways that cannot be easily replicated.

In each case, the value capture comes from the scarcity of provision, not from the importance of the outcome. When the scarcity disappears, the pricing power disappears with it, regardless of how valuable the outcome remains. This is why generic drugs cost a fraction of what branded drugs cost. This is why telecommunications prices have fallen as competition has increased. This is why food, despite being literally essential to human survival, is cheap enough that we take it for granted.

IV.

Now let us look at AI and ask the relevant question: where is the scarcity?

The hardware layer has scarcity. Nvidia's GPUs are genuinely difficult to manufacture, requiring expertise and fabrication capacity that cannot be conjured out of nothing. TSMC's fabs are, for practical purposes, irreplaceable, and everyone in the industry knows it. This is why Nvidia trades at the valuation it does, and why everyone is so nervous about Taiwan. The hardware companies have something that cannot be easily replicated, and they are charging accordingly.

The cloud layer has some scarcity, though less than it once did. Building data centers at the scale required to serve AI workloads requires billions in capital expenditure, and only a handful of companies have both the money and the expertise to do it well. Amazon and Microsoft and Google have moats here, though they are competing with each other, which limits how much any one of them can charge.

The model layer has what, exactly?

OpenAI makes GPT-4. Anthropic makes Claude. Google makes Gemini. Meta has released LLaMA as open source, which means anyone can run it. Mistral is open source. New models appear every few months, and the capabilities are converging. What one model can do today, the others will be able to do in six months, sometimes less.

If you build an application on Claude and Anthropic raises prices, you can switch to GPT-4 without much difficulty. If OpenAI raises prices, you can switch to an open-source model and run it on your own infrastructure. The switching costs are low because the interfaces are similar and the outputs are comparable. The lock-in is minimal because the models do not accumulate proprietary knowledge about your business that would be lost if you left.

The outcome that AI provides, cognitive work at scale, is enormously valuable. But the outcome is provided by multiple competing vendors who are all racing to match each other's capabilities and undercut each other's prices. This is not a market structure that supports outcomes-based pricing. This is a commodity market in the making.

V.

The people who argue for outcomes-based pricing in AI often reach for an analogy to make their case feel intuitive. They say that AI should be priced like an employee.

If you hired a salesperson who closed $500,000 in deals, you would pay them well, perhaps $150,000 in salary and commission combined. You are paying for outcomes, in a sense: the better the salesperson performs, the more you pay them. So, the argument goes, if an AI agent closes $500,000 in deals, the AI provider should be paid like that salesperson, taking a percentage of the value created.

This analogy is seductive, and it is wrong, because employees have scarcity that AI does not have.

There is only one of each employee, and this turns out to matter a great deal. Your best salesperson has specific knowledge of your business, your customers, your internal systems and processes. If she leaves, you cannot replace her instantly with an equivalent person, because the equivalent person does not exist. There are switching costs, relationship costs, institutional knowledge that walks out the door. This is a form of scarcity, and it is why good employees can command premium compensation.

The employee can also only work for one company at a time. Her labor is exclusive to you during the hours you employ her. She cannot simultaneously close deals for your competitor while closing deals for you. This exclusivity is another form of scarcity.

AI has none of this. The model does not have specific knowledge of your business that another model could not acquire by reading the same documents. If you switch from one provider to another, you lose nothing but the time it takes to integrate. There is no relationship, no loyalty, no institutional memory that departs when you cancel your subscription.

And the AI can work for your competitor at the same moment it works for you. It is not exclusive. The same model that closes deals for you is closing deals for everyone else who pays the subscription fee, and this parallel service does not degrade its performance for any individual customer.

An employee is scarce in multiple reinforcing ways. AI is abundant. You cannot price abundant things like scarce things, no matter how valuable the outcomes they produce.

VI.

I want to be clear about what I am not arguing, because it would be easy to misread this as pessimism about AI.

I am not saying that AI is not valuable. It is extraordinarily valuable, perhaps the most valuable general-purpose technology since electricity. It is transforming how work gets done across every industry, and the transformation is just beginning. The outcomes it enables are real and significant and will become more so.

I am not saying that AI companies will not make money. They will make a great deal of money, probably more money than most people currently imagine. The market is enormous and growing rapidly and shows no signs of slowing down.

What I am saying is that the value will be captured through volume rather than margin. This is how commodity businesses work. The product is valuable, the demand is high, but because supply is competitive, the price gets driven toward the cost of production. You make money by selling an enormous number of units at a thin margin, not by selling a modest number of units at a fat margin.

Food is like this. A lot of money changes hands in the food industry, but the money is made through scale and efficiency, not through premium pricing for outcomes. Electricity is like this. Telecommunications, increasingly, is like this.

AI is becoming like this, whether the people building AI companies want it to or not.

VII.

The outcomes-based pricing fantasy reveals something interesting about how people in technology think about their own businesses.

There is a hope, always present in the background, that software will somehow escape the normal constraints of economics. Because software can be copied at zero marginal cost, the thinking goes, the usual rules about competition and commoditization should not apply. If you build something valuable enough, you should be able to capture that value indefinitely, regardless of what competitors do.

Sometimes this is true. When you have a genuine monopoly, when network effects create lock-in that makes switching prohibitively costly, when the product improves with scale in ways that competitors cannot match, you can maintain pricing power for a long time. This is how the great platform companies became the great platform companies.

But "I have built an AI that does valuable things" is not a monopoly. It is a description of a crowded market in which many well-funded competitors are all building AIs that do valuable things. The valuable things are largely the same valuable things, accomplished through largely the same methods, improving at largely the same pace.

In this environment, outcomes-based pricing is not a business model. It is a fantasy about escaping competition. You cannot charge a premium for outcomes when your competitors offer the same outcomes at a lower price, because your customers will switch, and you will be left with nothing but a pitch deck describing how things should have worked.

VIII.

I think sometimes about the farmer in 1900, working land in Iowa or Kansas or Nebraska, trying to make a living from wheat or corn or soybeans.

He understood something that technology founders often do not, which is that the value of his product and the price he could charge for his product were entirely different things. The value of food was immense, foundational, a prerequisite for everything else in civilization. The price was set by the market, which meant it was set by competition, which meant it was set just barely above the cost of production.

He did not spend time devising complex pricing schemes that would allow him to capture the "true value" of his contribution to human survival. He did not complain that customers failed to appreciate the importance of what he was selling. He understood that he was in a commodity business, and he made his plans accordingly, focusing on efficiency and yield and cost control, because those were the variables he could actually affect.

The AI industry will learn this lesson in time. The providers will stop talking about outcomes-based pricing and start competing on cost and reliability and ease of integration, the way that commodity providers always compete. The margins will compress, the volume will expand, and the value created will be enormous. Most of that value will accrue to the customers rather than to the providers, which is what happens when sellers have competition.

This is good for the world, even if it is disappointing for the people who hoped to capture more. The value of a product flows to the buyer when the seller faces real competition. This is how markets are supposed to work.

The meal is worth your life. You are still only paying fifteen dollars.