In the 1850s, whaling was one of the largest industries in America. New Bedford armed hundreds of ships that vanished over the horizon for years at a time; fortunes were made; men died doing it. And then, within a single generation, it was essentially over — and nobody attacked it. No campaign, no ban, no rival fleet sank the whalers. They were simply left holding nothing.
The mistake is to think the industry sold whales, or even whale oil. It sold light. Whale oil was just the best way anyone had found to push back the dark after sundown, and the dark was the thing people actually wanted gone. When kerosene — first distilled from coal in the 1850s, then refined far more cheaply from petroleum after Drake’s well came in in 1859 — and then Edison’s bulb in 1879 served the same want far more cheaply and safely, the want quietly took the shorter road. Other forces pushed too: the Civil War took ships, vegetable and lard oils crowded in. But the decisive fact outlived all of them — a cheaper path to light had appeared, and the harpoons did not lose a fight. They were bypassed.
This is the oldest pattern in commerce, and Theodore Levitt named the principle in 1960. A business is in danger, he wrote in Marketing Myopia, the moment it believes it sells a product rather than a benefit: the railroads thought they were in the railroad business, not the transportation business, and so they watched the trucks and the planes take their customers while they polished the rails. Clayton Christensen later sharpened it into a sentence: people don’t buy a product, they hire it to make progress on something they want. Keep the lesson, because the rest of this essay is built on it. The visible activity — the whaling, the record shop, the bank branch — is never what people wanted. The want beneath it is, and it outlives every means it ever rode.
Run that all the way down and it stops being a marketing aphorism and becomes something stranger. If the want is the constant and the means keep dying, then a value chain is not a fixed thing. It is a population of competing methods under selection — and the chain you can see at any moment is just the lineage that has survived so far. The single pressure deciding which method survives to become the next link is cost.
What a value chain is, and why it evolves
A value chain is the full path a want travels from unmet to met: every step it passes through, every hand that touches it, and every cost paid along the way. The chain for light once ran through whaling ships, try-pots, and chandlers; now it runs through a power station and a switch. Same want, different chain.
Most stories about why chains change reach for desire — consumers wanted convenience, people demanded speed. But desire is nearly useless as an explanation, because the want barely moves. A person in 1850 wanted light, warmth, to know things, to be somewhere, and to feel something about as badly as you do now. If the want were the cause of change, nothing would ever change, because the want — at the level of the category, light or knowing or belonging — is roughly fixed. (Which particular object scratches the itch shifts constantly; that is a different layer, and we will come to it.) Think of a method’s fitness as how well it serves the want divided by what it costs: the category sets the peak and barely moves, while cost is the gradient every method is pushed down. So the want is the fitness landscape — it only scores — and cost is the pressure that does the selecting.
What causes the change is variation under selection. At every moment, technology throws up new ways to cover a link in the chain. Most are worse and die unmourned. The ones that serve the same want more cheaply and faster survive, spread, and become the new normal — and the method they replaced is bypassed, exactly as the harpoon was. Joseph Schumpeter called the destructive half of this creative destruction; Simon Wardley charts the constructive half as an evolution axis, every method sliding over time from novel and expensive (genesis) toward cheap and ubiquitous (commodity). And Carlota Perez explains why the bypassings arrive in synchronized waves rather than one at a time: a general-purpose technology — steam, electricity, the computer, now machine learning — lowers one kind of cost across every chain at once, which is why a dozen unrelated industries seem to round the same corner in the same decade.
So three claims, which are the spine of everything below. The want is constant. Cost is the agent of change. And whoever owns the cost that has not yet fallen collects the rent. The trouble is that “cost” in those sentences is hiding most of the actual content — so that is where to start.
A real map of cost
“It got cheaper” is where most analysis stops, and it is where the real structure begins. A want does not cross one cost on its way to being met; it crosses several, and they fall at different times, for different reasons. The order in which they fall is most of the story.
- Production cost — and this one must be split in two, because the split is where power hides. There is the fixed cost of bringing the first unit into being, and the marginal cost of each one after. Hold onto that distinction; I will come back to it.
- Holding cost — the price of keeping a stock of the thing on hand against future need: the inventory, the full pantry, the cash balance you carry so you are never caught short.
- Search cost — the price of finding the thing, or finding who has it. George Stigler made this a formal object in 1961; the search box is its near-total collapse.
- Transaction and coordination cost — the price of striking, trusting, and enforcing the deal, and of organizing the many hands a chain needs. Ronald Coase showed in 1937 that firms and intermediaries exist precisely to absorb this cost; the corollary is that an intermediary which existed only to provide coordination dies the moment coordination gets cheap.
- Verification and trust cost — the price of knowing the thing is what it claims to be. George Akerlof showed in 1970 that a market can collapse entirely when buyers cannot tell good from bad. Mark this one: it is the cost that rises exactly when production cost falls to zero.
- Attention and curation cost — the price of spending finite human attention across effectively infinite options. Herbert Simon put it best in 1971: a wealth of information creates a poverty of attention.
- Switching cost — the price of leaving one method for another, which is how a keeper holds you after its other advantages are gone.
Every historical mode is the death of one of these costs and the survival of the rest. Owning things outright spares you search and waiting at the moment of need, but makes you pay to hold the stock. The catalog and then the search box killed search cost — the thing comes to you. What we are living through now is the collapse of production’s marginal cost for anything made of information. When you hear that something “got cheaper,” the question that actually predicts the future is: which cost, and which costs are now left standing? Because the costs left standing are where the next chapter of the chain — and its next owner — will be.
That gives a blade sharp enough to tell evolution from mere efficiency: did the new method make a new cost cheaper, or just the same cost cheaper again? A faster truck, a cheaper ride, a meal delivered instead of cooked — these mostly shave the same old production-and-delivery cost by a margin. That is a chain getting more efficient. A chain evolves when a cost that used to be unavoidable simply stops mattering at all — the way the cost of sourcing animal oil, and the years-long holding-and-risk cost of a whaling voyage, did not merely shrink for the lighting chain but dropped out of it entirely.
Two conservation laws
Now the part nearly everyone gets backwards. Once a cost has fallen, where does the money go?
The folk answer is disintermediation: the middleman is removed and the saving passes to you. It is almost always false. The middleman is not removed. It relocates. Two conservation laws govern the move.
Law one: value migrates up to the scarcest remaining cost. When one cost collapses, the value — and the price you pay — does not vanish with it. It climbs to whichever cost is now binding. Adrian Slywotzky called this value migration; in information goods specifically, Carl Shapiro and Hal Varian showed in Information Rules (1999) that when the marginal cost of a digital good goes to zero, the scarce thing moves to attention, curation, and choosing the right version. This is why cheap generation does not make the world fall quiet — it makes it flood, and the scarce thing becomes finding the one you actually want inside the flood. We have a fresh word for the flood: slop.
Law two: intermediation is conserved — the keeper relocates, it does not disappear. A thousand record shops became one Spotify; a thousand bookstores, one Amazon; ten blue links, one box deciding which links you ever see. Distribution cost fell to nearly nothing, and instead of dispersing power it produced the most concentrated gatekeepers those industries had ever had. Ben Thompson named the mechanism Aggregation Theory: when distribution and transaction costs collapse, advantage flows to whoever owns the demand — the customer relationship, the default starting point — and that owner can then commoditize everyone upstream of it. Christensen’s law of the Conservation of Attractive Profits says the same thing from the other side: when one stage of a chain is commoditized, attractive profit is conserved by migrating to an adjacent stage that is not. The keeper is simply whoever sits on the cost that has not yet fallen. Reroute the want and you do not delete the keeper. You relocate it.
This is not abstract for this series — it is the whole series. The three essays that follow this one are these two laws worked out inside a single industry: money. When the cost of moving and converting value falls toward zero, the rent does not disappear. It relocates — into the yield a bank quietly keeps on a zero-rate deposit, into whoever ends up owning the rails, the unit of account, and the wallet that holds everything, into the settlement layer the dollar still commands. Money is just the chain where I traced the keeper’s relocation step by step. The law is general — which raises the question the next section has to answer: why does sitting on the scarcest cost harden into durable power, instead of staying contested?
Why the scarcest cost is also power
“The keeper relocates to the scarcest cost” is only a description until you can say why sitting on that cost yields durable power, and why it tends to consolidate rather than stay competitive. Four mechanisms — and one counterweight, so this does not collapse into the same fatalism it is correcting.
The first is the production split from earlier. A method with near-zero marginal cost but enormous fixed cost — a search index, a frontier model trained for nine figures, a planet-spanning logistics network — is the textbook recipe for natural monopoly. The first copy costs a fortune; every copy after is almost free; so the largest player can always undercut, and the market tips toward one. The collapse of marginal cost is precisely what builds the fixed-cost fortress, which is why the keeper so reliably ends up owning the biggest cost that didn’t fall. The second is the network effect: each user makes the thing more valuable to the next, so whoever leads pulls further ahead. The third is subtler — the moat is not control of supply but ownership of demand, being the place attention starts; this is the core of Ben Thompson’s Aggregation Theory.
The fourth decides whether any of the others bite, so it is worth slowing down on: appropriability. Not every scarce cost is ownable. Rent can be charged at a bottleneck only when something makes it defensible — network effects, model weights and compute, exclusive rights or data, control of distribution. Where the residual cost is not ownable, where anyone can stand on it, the saving really does pass to the public and no keeper forms at all. This is the clause that keeps the law from predicting giants everywhere: it names the cases where the law loses.
The counterweight keeps the law honest: consolidation is contingent, not guaranteed. Two-sided-market theory says tipping depends on whether users multi-home and on relative concentration across the chain. Where the suppliers are more concentrated than the aggregator, the rent sits upstream — in music, three major labels out-concentrate any streaming service, which is exactly why Spotify, unlike Netflix, cannot dictate to its catalog, and earns thin margins while the labels do well. The keeper lands wherever concentration is highest, and that is not always the shiny new layer.
And the cost lens forces one moral distinction the reflexive “middlemen are bad” story misses: the difference between rent and return. The whaler displaced by kerosene is a story of efficient substitution — light got cheaper for everyone, surplus flowed to the public, and the new keeper earned a return on real investment. Amazon’s Buy Box fee, or a self-preferenced clone of your own bestseller, is extraction — rent on a chokepoint with no matching gain to the buyer (Lina Khan’s 2017 Amazon’s Antitrust Paradox is the canonical argument that our tools quietly lost the ability to tell the two apart). Both are “the keeper relocating.” They are not the same event, and a theory that scores them identically is missing the part that matters.
The want, and human nature
We have kept the want constant this entire time — the fixed landscape that cost climbs around. But constant does not mean simple, and if you want to know where a chain can end — whether it can keep evolving or tops out at a faster truck — you have to look at the structure of the want itself. This is where the chain finally reaches human nature, and where the cheapest theory of wanting fails most usefully.
The folk model is that we want things because they will feel good, and that every want reduces to that one currency. If that were true, the last scarce cost — the thing left over when everything else is free — would be trivial: just generate whatever feels good. It is not trivial, and the reason is one of the sturdiest findings in the science of motivation. Wanting and liking are two different systems. Kent Berridge and Terry Robinson showed that the brain’s dopamine circuitry does not manufacture pleasure; it manufactures wanting — the pull toward a cue, the salience that makes you pursue something before you have even decided whether you will enjoy it. Liking, the actual hedonic hit, is a separate and smaller system. They usually move together, which is why we conflate them, but they come apart cleanly: addiction is enormous wanting with little liking.
That dissociation is the key to the flood. When the cost of generating a thing falls to zero, you can manufacture wanting without limit — infinite cues, infinite scroll, infinite plausible outputs — while doing almost nothing for liking. Slop is precisely this: output engineered to pull (it is engaging, it is endless) that starves the thing it pretends to serve (none of it is what you actually wanted). And it explains why the last cost is so stubborn. Once making is free, the residual scarcity is liking — landing on the one that is genuinely right for you — and liking is first-person, mood- and context-bound, and adaptive. Generic generation cannot supply it: one output for everyone is the wrong shape for a target that is this particular person at this particular moment. Liking is not un-automatable — it is exactly the prize a personalization keeper races to capture, which is the contest the final section returns to — but it is the wrong thing to ask cheap, one-size-fits-all generation to produce. Taste is a liking problem wearing an economics costume.
Two engines keep the want from ever dying, which is why there is always a next cost to climb to. The first is hedonic adaptation — the treadmill Philip Brickman and Donald Campbell described in 1971: we acclimate to whatever we get, satisfaction decays back toward baseline, and the want re-arms. A static catalog goes stale on its own; the treadmill guarantees demand for the next thing. The second is mimetic desire — René Girard’s observation that we do not want in isolation but want what others want, through models. Taste is borrowed; status and belonging are wants in their own right; and so desire is socially inexhaustible. This also names the most troubling keeper of all: one that does not merely sit on a cost but manufactures the want itself, by owning the feed of models we copy from. That keeper lives one level above cost, and the cost lens alone will not catch it. This is the one place the constant-landscape claim shows a seam, so it is worth naming precisely. The want category — light, knowing, belonging — is the near-fixed landscape that only scores. The want object — which song, which model, whose life to envy — is what re-arms after adaptation and what a feed can manufacture. Want-manufacturing bends the landscape only at the object level; the cost lens is scoped to the category, which is exactly why it goes blind to this particular keeper.
Sort wants by their structure and the structure predicts how far the chain can evolve — which is the real payoff of taking the want seriously:
- Terminating vs non-terminating. A “have-it” want is over the instant you have it, so its chain tops out at speed — a faster truck — and never needs to become anything richer. A “be-in-it” want is ongoing; you stay in it and return to it, so its chain can keep evolving toward something you inhabit rather than receive.
- Rivalrous vs not. Rivalry is whether satisfying one person uses up another’s access: a meal or a seat has a hard allocation floor; a song or a fact does not. That is a separate axis from the cost of copies — a song is non-rivalrous yet its first instance can be hugely expensive, while a haircut is rivalrous with no cheap copy at all. A chain’s cost falls toward zero only where reproduction is cheap, not merely where the good is non-rivalrous.
- Solitary vs social. A social want pulls the chain toward shared, visible, status-bearing forms, and hands leverage to whoever owns the social layer.
- Bodily vs abstract. Some wants are answered in the body, and here the science is concrete: the human pull to move with rhythm — entrainment, the “groove” response studied by Petr Janata and Maria Witek — is a cross-cultural near-universal, which is why a few chains end in doing rather than receiving.
One axis deserves its own name, because it decides the shape of the endpoint rather than its reach. When making becomes free, the cheap thing can arrive in two forms: a slot you type a wish into and receive a finished result, or a space you move through and steer. These are economically almost identical and psychologically opposite. Self-Determination Theory — Edward Deci and Richard Ryan — identifies autonomy, competence, and relatedness as needs in their own right, things we want intrinsically rather than as routes to pleasure. A slot satisfies the surface want and starves all three: no autonomy (you only wish), no competence (re-rolling a prompt builds no skill), no relatedness. A space feeds them. This is why “cheap through a slot” so often feels hollow even when it technically works — it delivers the liking-object while quietly violating wants you did not know you were paying for. Agency is not a luxury feature bolted onto the cheap thing. It is a want that the cost-collapse threatens, and the chains that win the long game are the ones that hand it back.
The open question
So here is the shape of the whole thing. A value chain is a population of methods under selection; the selector is cost; the want is the constant landscape it climbs around; and when a cost falls, the rent does not vanish — it migrates to the scarcest cost still standing and finds an owner there. Run it forward and the costs fall from the outside in. We collapsed the cost of keeping, then the cost of finding, then the cost of making — and each time, value climbed to the next cost left.
Look at what is left standing now, in industry after industry, and it is no longer out in the world. It is in us. When generation is free, the binding costs are liking (taste — the one that is right for you), verification (trust — what is real in the flood), attention (what is worth your finite time), and agency (the sense of having acted rather than wished). These were always costs. They were simply never the binding ones, because cruder costs sat in front of them. And they do not all settle in the same place. Some stay in the world but move to its edges — verification migrates to provenance and institutions, much of attention to whoever builds the funnel. Others migrate the whole way into the person: liking and agency cannot be relocated anywhere but you. The sweep of the chain, seen end to end, is that the binding cost keeps moving inward — until the last thing anyone pays for is something only a human can feel.
Which leaves the question this series keeps circling and will not pretend to settle. If intermediation is conserved — if the keeper relocates to the scarcest cost wherever that cost can be owned — then the moment the scarce cost becomes a human one and someone finds a way to fence it, the next keeper is whoever owns that. The optimistic reading is that some costs cannot be owned: my liking is mine, my attention is mine, and a chain that ends in them ends in something no toll-taker can fence. The pessimistic reading is that we are already learning to own them — the feed that shapes what you want, the model that learns your taste, the interface that decides what you attend to — and that the final relocation of the keeper is not onto a cost out in the world, but onto the wanting machinery inside you.
That is the contest worth watching, and it is the one the whalers’ customers never had to think about, because no one could own the dark. The dark just lost to a cheaper light. The open question, for every chain still evolving, is whether the last scarce thing — now that it lives inside us — loses as cleanly, or whether someone owns it first.