Written in collaboration with my favorite LLM. Editorial direction and final judgement are mine.
Amusing Morose Musings · sweet-kandy.blogspot.com
I am addicted to predictions that land. I have said this before and the tracker keeps me honest about it. The Gulf series is a war and I read it like a war. This is something else. This is the longest call I have ever made here, and I am making it at full confidence, in absolutes, on a ten-year clock. There is no partial credit on what follows. In 2036 these are graded pass or fail. That is the point.
So here is the whole bet, up front, no working shown yet. Four calls. Read on if you want to know why I am reckless enough to make them.
One. US real output rises by fifty percent, dollar-adjusted, against a 2026 base, by 2036 but there will be a crash before that. The World Bank settles it. There is no argument about the scoreboard.
Two. The firm shrinks and work is redefined. The median firm gets smaller by headcount. Tiny firms take a rising share of output. A company worth a billion dollars ships in this window run by fewer people than a single department used to hold.
Three. The individual is multiplied. Output per person at the frontier breaks past anything the org-chart era could produce. One person does what a team did.
Four. Freedom beats the cheque. No major developed economy adopts a structural universal basic income in answer to AI unemployment within the decade. Participation does not collapse. Self-directed and solo work rises instead. The end-of-work call misses for the fourth time, and the humane response turns out to be agency, not alms.
No priorities. No percentages. No timing windows inside the decade. One clock, and it runs out in 2036. I have left myself nowhere to hide, and that is the only kind of prediction worth making. If you want to know how I get to four absolutes off a pile of rotting computer chips, read on. If you only want the scoreboard, clip this and find me in 2036.
A note before I start. While researching this, my favorite LLM introduced me to a handful of friends I should have made long ago. I am putting them up front so you can chase them down yourself. Each is worth an afternoon.
- Carlota Perez, who wrote down the shape of these cycles.
- Paul David, who explained why the payoff always arrives late.
- Ronald Coase, who explained why companies exist at all.
- Wassily Leontief, a Nobel economist who predicted the end of work and was wrong.
- Jeremy Rifkin, who predicted it again, in a book, and was also wrong.
- Ned Ludd, who has the oldest konwn claim to being wrong about this.
Daniel Plainview turns up too. He is fictional. You will know him when he arrives.
Here is the whole claim. Infrastructure and the services built on it move in a wave. The atoms get built first, fast and overbuilt, on money that has lost its mind. The build busts. The people who built it mostly die. Then the cheap leftover atoms power a long services boom that runs two or three times longer than the build did. We are two years into the atoms phase of the largest one of these the world has seen. I am going to tell you what the services phase does, and it is not what the clever people are telling you.
The shape is not mine. Carlota Perez wrote it down properly, and I am borrowing her frame the way I borrow Hemingway for the bull. She calls the first phase installation and the second deployment. Installation is the frenzy. Deployment is the golden age. A crash sits in between and clears out the speculators. I will use her words because they are good words. But I want to earn the optimism, so I am going to walk a few cycles where the shape held, and only then tell you the one way this one breaks the rules.
Start with gold, because it is the purest version. The miners mostly went broke. The men who sold them shovels and sieves and stiff denim trousers got rich, and one of them was named Levi Strauss. You are probably wearing his idea right now. The dig is the atom. The pick and the bank that wired the money home are the services. The man who sold the shovel never once swung it.
The railways ran the same play at scale. Britain and then America laid track for half a century like men possessed. The railway mania in England was a textbook bubble. Shares in lines that would never be built. Schemes drawn over towns that did not want a station. America did it bigger, on borrowed money, and the bill came due in one panic and then another. The men who laid the track got carried out. Jay Cooke went under and took the economy with him. And then the track sat there, paid for, and the country that ran commerce over it for two generations got rich. Sears mailed a house to a farmer in Nebraska because the track existed. The builders financed the harvest. They did not eat it.
Electrification is the cycle that tells you how long the services tail really runs. Edison and Westinghouse and Insull built the grid in one long tear. The grid was made of copper, and the men who dug that copper out of Montana and Chile were not the men who got rich when the lights came on. Factories were wired early. And then, for a generation, nothing happened. Productivity did not move. Paul David worked out why. The factory did not get more productive when you dropped an electric motor where the steam engine used to sit. It got more productive when somebody finally tore the factory down and rebuilt it around the motor. The old factory was a tall building with one steam engine in the basement and shafts carrying power up through the floors. The electric motor let you put power anywhere. But you had to imagine a flat factory built around the work instead of around the driveshaft, and that took a generation to imagine. The atoms arrived. The deployment arrived twenty years later. The gap is the pattern, not an accident of it.
Oil is the one with the gusher. Spindletop blew and the world had never seen anything like it. One well in Texas produced more in a day than every other well in America combined. The derricks went up faster than anyone could drink the crude. The price fell to three cents a barrel in the fields. Three cents. Water cost more. The wildcatters drowned in their own gusher. And the men downstream, who refined it and shipped it and sold it back to you as something useful, ran the next century on it. If you have seen There Will Be Blood you have watched this happen to one man. Daniel Plainview drills the atoms, drinks everyone's milkshake, and ends up alone in a bowling alley with all the money and nothing else. He is the leveraged builder. He is every man who confused owning the well with owning the century. The refiner ate his lunch and slept fine.
Oil shows one more thing, which I will need later. The drilling moved. Pennsylvania, then Texas, then out of the country entirely, to the Gulf and the Caspian and Arabia. It went where the rock was cheap and the labor cheaper, but only after the easy domestic oil got expensive and tapped. The work left home late, not early. Hold that.
Highways are the cycle where the state laid the atoms. The American government poured the interstate system across the country in a few decades, the largest public works anyone had attempted. The road is the atom. The state laid it and the state did not get rich. What got rich was everything the road made possible. Suburbia. Trucking. The shopping mall. The motel. The drive-through hamburger that only works if a family has a car and a road and a reason to be on it. Walmart is a logistics company that happens to sell things, and it cannot exist without a road the government paid for. The builders laid the atoms. The truckers and the retailers drank the milkshake.
The car itself ran the loop on a delay. America built the car at home, at the fat margins of Detroit, while the skill was scarce and the union strong. Then the skill hardened into a process, and a process can be shipped, and the car went to Japan once Detroit got slow and sure it would always win. Premium at home, then offshore once it commoditized. Same shape, one layer up.
Glass is the one you remember. The telecom companies laid fiber across the oceans and under every American street as if bandwidth were about to become oxygen. WorldCom. Global Crossing. They borrowed to do it, they lied about the demand, and most of what they laid was dark, meaning lit by nobody. Then the whole thing fell over. Global Crossing went bankrupt. WorldCom went to prison. The greatest overbuild of atoms in living memory, and the men who did it got nothing.
And then the glass sat there. It did not rot. And over the next twenty years somebody lit it. That dark fiber is why you stream video. It is why the cloud exists. It is why a website in California serves a customer in Lagos without anyone thinking about it. Google did not lay that fiber. Netflix did not lay that fiber. They lit it. They drank the milkshake the dead men drilled.
That is the bench. Gold, rail, copper, oil, the road, glass. The shape holds every time. Atoms first, overbuilt, on bad money. A crash that kills the builders. Then a deployment phase, longer than the build, where different people get rich off infrastructure they never paid for.
Now the second act, because it is the most consistent signal in the pattern.
The services run at home first, while the margins are fat. A new industry pays its people well because the work is hard and the people who can do it are rare. Detroit paid the autoworker a wage that built a house and a boat. American programmers priced themselves like surgeons. The premium is real and it lasts exactly as long as the skill is scarce.
Then the skill stops being scarce. The work gets written down into a process. A process can be taught, and a process taught can be shipped. The moment the home version gets expensive enough, the work leaves. Cars to Japan. Software to India. Call centers, back offices, tax returns, X-ray reading, all of it, and it left late. Offshoring is not how a cycle begins. It is how you know a cycle has aged into a commodity.
And here is the loop closing. Remember the dead fiber. The same glut that bankrupted WorldCom is the reason the work could leave at all. You cannot run a Bangalore office for an Ohio insurer until transoceanic bandwidth is nearly free. The bankrupt builders made it free. The fiber glut built the Indian services economy. One cycle's overbuild becomes the next country's livelihood, and the bridge between them is the bankruptcy of the men who laid the glass.
That is the full pattern. Atoms at home. Crash. Services at home while premium. Commoditization. Offshore. Repeat. It has run for two centuries and I would have bet on it running again.
I am about to bet on it breaking.
Two years ago this started. The hyperscalers began spending money that does not look like technology money anymore. It looks like utility money. It looks like the railways. The five biggest are now guiding past six hundred billion dollars in a single year, three quarters of it AI, and the whole effort is running near five percent of American output, larger than the fiber overbuild and larger than the electrification peak. They are funding it with debt now, because the spending has run past what they earn. This is the railway mania with better balance sheets. For now.
So the installation phase is here and it is enormous, and on that everyone agrees. The argument is only about what comes after. I am going to give you the after. But first, the one rule this cycle does not obey.
Every cycle I walked built atoms that lasted. That is the reason deployment worked.
The canal lasts a century. The railway lasts longer. The grid runs fifty years on the same copper. The glass laid in the boom was still carrying your video twenty years on. The atom waited patiently for demand to arrive and find it. That patience is what let deployment run two and three times longer than the build. The infrastructure could wait because it did not decay while it waited.
The AI atom does not wait. A GPU loses about a fifth of its value every year and is economically dead in three or four. The chip you buy today is a paperweight before deployment has properly begun. The hyperscalers are about to carry depreciation larger than their profit. They are not building a railway. They are building a railway that rusts to nothing every four years and has to be re-laid from scratch, forever, while they are still waiting for the towns to appear along the line.
There is a flip side to the rot, and it is the reason I am putting a ten-year clock on what comes next. Every prior deployment ran slow because the rewiring was physical. Electrification waited a generation because somebody had to tear the factory down and pour a new floor around the motor, and concrete takes a generation. This time there is no concrete. The rewiring is software. You do not demolish a building to drop a model into a workflow. The thing that made every prior lag long is the thing this cycle removes. So the payoff that took twenty years after the lights came on does not get twenty years this time. That is a claim, and I am about to bet on it.
This breaks the pattern in two places.
Break one. The build never stops. In every prior cycle there was a moment when building ended and harvesting began. You finished the track. You finished the grid. You let the fiber sit. A clean handoff from atoms to services. There is no handoff here. Because the atoms rot, the builders are trapped re-buying them in perpetuity. There is no year when the capex ends and the milkshake starts. The men who own the wells re-drill them every four years just to stand still, and the man downstream still drinks the milkshake. NVIDIA sells the shovels forever. The hyperscalers re-buy them forever. The value still leaks downstream to whoever builds the thing that uses all this, which does not exist at scale yet, the way Google did not exist when the fiber went in.
Break two. The offshoring valve closes. Every prior cycle, when the services got expensive at home, the work moved to cheaper people somewhere else. The arbitrage was geographic. Find a cheaper human in a cheaper country and ship the work over. That is how India got the back office and Japan got the cars. The work left, but it went to other people, and it built the next economy somewhere else.
AI changes the arbitrage from geography to automation. When the AI services get expensive at home, the cheaper alternative is not a human in another country. It is the model. The model is already cheaper than the cheapest offshore human, and it sits nowhere and everywhere at once. So the work that, in every prior cycle, would have left home and built the next India, does not leave. It gets eaten in place. The valve that used to read send it to Manila now reads give it to the model.
Put the two together and you see why this cycle is genuinely different, and I do not use that phrase lightly, because that phrase is how people lose money. The fiber glut built the Indian services economy. The AI deployment removes the reason to offshore at all. Same shape, opposite sign. The previous overbuild created a services class in a poorer country. This deployment eats that class and does not seed another one anywhere cheaper, because there is nowhere cheaper than a machine.
Every infrastructure before this was a platform for labor. The rail carried goods people made and moved. The grid powered factories full of people. The fiber carried work done by people in Bangalore. Each one amplified human work, so each one eventually needed more humans, in more places, doing more things. AI is the first infrastructure that is a substitute for labor rather than a platform for it. That is the difference.
Because a substitute for labor is also a multiplier of the person who holds it.
Here is where I leave the consensus, and the consensus is loud, so let me be exact about what I am rejecting. The clever and the well-meaning have decided that AI ends work, and that the humane response to the end of work is a cheque. Universal basic income. A salary for existing, paid by the machines, handed out by the state. I think that is wrong in the way the loudest predictions are usually wrong, and I have history on my side, so let me show the work before I make the call.
The end of work has been called three times, loudly, by serious people, and it has missed three times.
The Luddites broke the looms because the looms would end weaving. Weaving employed more people two generations later than it ever had. The work changed. It did not end.
Leontief said automation would do to human labor what the tractor did to the horse. He meant it literally. He thought we were the horse. His era produced government commissions on what to do with the permanently unemployed. The permanently unemployed never arrived. Employment hit records.
Rifkin wrote a book called The End of Work, and it sold, and it was wrong. The decades since produced more work, more kinds of work, work nobody could have named at the time.
Every time, the deployment phase of a new infrastructure produced more human capability and more human work, not a permanent jobless class on a stipend. The doom call is the oldest call there is, and it has the worst record in economics. I am betting it misses a fourth time, and I am betting against the cheque.
The mechanism runs through Ronald Coase, who asked the only good question anyone has asked about why companies exist. A firm exists only because doing things through the open market is expensive. Finding people, negotiating, coordinating, checking the work. When coordination is cheaper inside a hierarchy than out in the market, you get a firm. The size of the firm is set by the cost of coordination. Drop that cost and the firm shrinks.
AI is the largest single drop in the cost of coordination in human history. So the firm shrinks. Not as a metaphor. The median firm gets smaller. The share of everything made by tiny firms rises. And somewhere in this window somebody ships a company worth a billion dollars run by a handful of people, because one person now wields what used to take a department, and a handful now wields what used to take a corporation. That is not the end of work. That is the redefinition of the firm, and through it the redefinition of work.
I want to be honest about the shape of this, because uplift is never evenly spread. Most of humanity does not rocket. Most of humanity does what it has always done, which is grind slowly upward along the path of progress, a little better off each generation than the one before. That grind is real, and it is most of the story, and it is not nothing. But every cycle a few people hit escape velocity. The ones who caught the railway. The ones who caught the oil. The ones who lit the fiber. They did not inch upward. They left the ground. This cycle will mint its own, and the multiplier is the launch pad. The only thing different this time is that the launch pad is portable, and it fits in one pair of hands.
The person at the center of this is not a horse being retired. The person is a foreman who used to run ten people and now runs ten thousand instances of a capability. Individual capacity expands by an order of magnitude. That expansion is the deployment phase. It is the milkshake, and for once it does not all pool in one company downstream. It distributes, unevenly, the way it always has, broadly to the patient and steeply to the few who seize it.
So I do not want the cheque. The cheque is what you offer the horse. It assumes the person is finished and needs feeding. I am claiming the opposite. The person is about to be more capable than any person in history, and the right answer to a more capable person is more freedom, not more welfare. Agency, not alms. You do not put a population whose capability just multiplied tenfold on a stipend. You get out of its way. The UBI consensus is the most pessimistic reading of the most empowering tool, and in ten years I think it will look the way the automation commissions look now. A serious answer to a question that turned out to be wrong.
These two things look like they cannot both be true. They can, and they fit together the oldest way in the pattern. The treadmill is the grim part. The atoms rot, the builders are trapped re-buying them, somebody gets carried out. The fifty percent more global output and the far more capable individual are the sunny part. Both are true at once.
The pain concentrates on the builders. The gain distributes across everyone else. The men who laid the GPUs eat the depreciation and the debt and the write-down, the way WorldCom ate the fiber and Plainview ate the wells. The broad economy and the individual harvest the deployment surplus, the way the world got cheap bandwidth and cheap oil and a wired factory floor off the backs of dead builders. Concentrated pain, distributed gain. That is how global output can rise by half in a decade while the people who built the thing that caused it get carried out of the building. There is no contradiction. There never was. The milkshake was always drunk by someone other than the man who drilled.
The atoms always come first. They came first for gold and rail and copper and oil and the road and the glass, and they are coming first now, six hundred billion dollars a year of them, rotting as fast as they are laid. The men building them will mostly not be the men who get rich, because they never are. Plainview drilled the well and died alone with the money. Google lit the fiber it never paid for. The pattern is two centuries old and on most of it I am simply reading the meter.
What is new is the one rule this cycle breaks. Every infrastructure before this was a platform for human work, so each one made more room for people, somewhere, eventually. This one is a substitute for human work, so the work does not move to a cheaper country this time. It does not move at all. The valve is closed.
The consensus reads that closed valve and reaches for the cheque. I read the same closed valve and reach for the opposite, because a tool that substitutes for labor multiplies the person who holds it, and you do not put a multiplied person on welfare. You set them free. Most will grind upward as they always have. A few will leave the ground entirely. The cheque is what you offer a horse, and we are not the horse. We are the foreman who just got handed ten thousand hands.
Four predictions, full confidence, no exits. In ten years the tracker tells me whether I read the mechanism or fell for the oldest mistake there is, which is believing that this time is different. I believe it is different. I have shown the one way it is, and the many ways it is not. If I am wrong, the tracker will say so plainly, the way it always does. I would rather be wrong in public on a real claim than right in private on a hedge.
Armchair strategist. No formal qualifications in economics, technology, or finance. Opinion and analysis only. Not investment advice. I have been wrong before and will say so when I am, and the tracker exists so there is nowhere to hide when I am.
Written in collaboration with my favourite LLM. Editorial direction and final judgement are mine.
Amusing Morose Musings · sweet-kandy.blogspot.com
