The Conservation of Scarcity

Every technological revolution gets described the same way: as the creation of abundance. The descriptions are accurate, and they miss the point. Abundance is the visible half of the event; the invisible half is that scarcity does not die, it migrates. Value drains out of whatever a revolution makes free and floods into whatever now binds.

I think of this as a conservation law, as close to physics as economics gets. It matters now because we are making intelligence free, which is the largest test of the law there has been.

The Law

The pattern has repeated for as long as we have records.

Light was scarce for seven thousand years. William Nordhaus traced its price across that span and found it had fallen by a factor of many thousands, from a meaningful share of a day’s labor for an evening of dim flame to something too cheap to meter. Nobody got rich selling lumens after that. The value moved to everything abundant light made possible, and the scarcity moved to everything abundant light consumed.

Calories were scarce for all of human history, and for all of human history, fat was the body of a rich man. The moment calories became abundant, the signal inverted: restraint became the luxury good, and entire industries now exist to sell discipline back to people drowning in what their ancestors died for lack of.

Information was scarce until roughly 1995. Herbert Simon saw the inversion coming a generation early: a wealth of information, he warned, creates a poverty of attention. The fortunes of the internet era went to the people who owned the attention that information consumed, more than to the people who sold the information.

The mechanism is the same every time. When a good becomes free, value flows to its complements, the things you need alongside it and the inputs that now bind. The revolution relocates scarcity rather than ending it, and the relocation is where the next era’s fortunes are made.

The Abundance at Hand

The price of a given level of machine intelligence has been falling roughly tenfold a year, a cost curve that makes Moore’s law look patient. What this commoditizes is larger than “some jobs”: it is the median unit of cognitive work, the competent analysis, the workable code, the serviceable diagnosis, the reasonable plan. The middle of knowledge work, the part firms rent out by the hour, is becoming a utility.

The instinctive response is to ask which skills survive. The conservation law suggests a better question, which is where the scarcity goes.

Here the law says something uncomfortable, and it is the claim I would defend hardest: intelligence was never really the bottleneck. The world has always contained far more capable minds than deployed ones, held back by geography, credentials, permission, and capital. If raw intelligence were the binding constraint on value creation, the smartest person in any room would be the richest, and they rarely are. Artificial intelligence strips away the camouflage and shows us what the bottleneck always was.

Where the Scarcity Goes

Wanting. A frontier model can answer nearly any question. It does not, of itself, want anything; every goal it pursues is supplied from outside. Intelligence is becoming free because it can be separated from intention, and intention is an input the factory cannot manufacture.

This is where the pattern I return to most completes itself. People are extraordinary at limbo: set a bar anywhere, and they will bend to get just under it. Machines invert this. A sufficiently capable system rises to exactly the bar it is given, so output starts to depend less on talent, headcount, or budget and more on where the bar is placed. Ambition becomes a factor of production, and the gap between organizations becomes the gap between what they ask for.

Taste. When anything can be made, the job becomes choosing what to make. The scarce skill shifts from execution to selection, from “can we build it?” to “does this deserve to exist?” Taste was always the hard half of intensity, and when execution is free, a company is mostly its portfolio of questions.

Trust. An abundance of plausible output devalues output. When any document, image, or voice can be synthesized for nothing, the question stops being “is this good?” and becomes “who stakes their name on it?” Scarcity migrates from production to provenance: to the signature, the warranty, the institution that can credibly absorb blame when things go wrong. Taking responsibility is about to become a very good business.

Reality. Models are trained on recorded reality, and they are converging on everything recorded. What remains scarce is the unrecorded: proprietary data, private workflows, the texture of what actually happens inside companies and between people. Compute can be bought and algorithms leak, so the undigitized residue of reality is the raw material of the next decade.

The Most Expensive Question

We spent ten thousand years organizing civilization around the rationing of intelligence, with schools to grow it, credentials to certify it, firms to rent it, and cities to concentrate it. All of that was scaffolding around a scarcity that is now ending, and it hid the fact that intelligence was only ever the means. The point was the wanting. “What do you want?” is becoming the most expensive question in the world, because the answer is the last input that cannot be generated.

We have now built a runner that clears any bar we can articulate, which leaves one variable that matters: where you have the nerve to set the bar.

The Falsifiable Form

So that the argument can be scored rather than admired: if the conservation law holds, the great fortunes of the next two decades will not accrue to whoever builds the best model. Models are the thing becoming abundant, and selling them is selling lumens. The fortunes will accrue to whoever owns the complements: the demand, the data, the distribution, the trust, the taste to know what to ask for. If by 2035 the model layer has captured most of the value of the AI economy, this essay was wrong. I am betting it won’t be, and the history of every prior abundance is the basis of the bet.