This is an essay about a nickel that didn’t arrive.
Every time a machine has taken over human work, something has gotten cheaper. The tractor made food cheaper; the container ship made goods cheaper. The savings didn’t vanish. They showed up in the family’s pocket as a nickel off the loaf of bread, a few dollars off everything on the shelf, and then they kept moving — into new spending, new jobs, new industries.
This summer the machine came for our own trade. The windfall was real, colossal even. The discount wasn’t there.
In The Last Wide Rung we laid out what we took to be a law of economic history: when the machine takes the work, a windfall appears — the gap between what a thing used to cost and what it suddenly costs — and the windfall always runs downstream. We mapped it as a five-link chain:
windfall → consumer surplus → re-spent → new demand → new production
We broke the chain at the fifth link: this time, we argued, the new production doesn’t materialize the way it used to. The first four links, we wrote, were still clicking into place on schedule.
We still say the fifth link is broken. What the summer taught us is that the second one is broken too. The savings no longer reliably become a discount. We went looking for where that nickel went.
The summer that didn’t fit
People we know lost their jobs this year, and AI had a hand in most of it. Not always as a headline — sometimes as a hiring freeze that never thawed, a contract that didn’t renew, a team of five becoming a team of two and an agent fleet. And it isn’t just our circle. Since March, AI has been the leading stated reason for American job cuts every single month, and before the summer was over the cuts attributed to it had already doubled all of last year’s1. The wide office floor is emptying.
But if the windfall runs downstream, then somewhere out there prices should be falling. The nickel should be arriving. The same machine that took our friends’ salaries should be quietly handing every household a discount — on the contract review, the tax return, the logo, the software, and eventually, through a thousand seeping channels, on the ordinary bill of an ordinary life.
And here is what made the summer so strange: the windfall itself was the easiest thing in the world to find. We are holding fistfuls of it. Working with Claude, we produce software at something like thirty times our old pace, at the same quality — this summer we shipped desktop apps for three platforms and mobile apps for two, then did it again for a second product. Over a weekend, for the joy of it, Ben wrote a sailing game. A client of ours was quoted thirty-five thousand dollars by an offshore shop for a piece of software; with some coaching, he vibecoded a complete replacement himself. And in the spring we filled an essay with operators just like him, walking away from six-figure SaaS contracts. We watch it being minted every day, and some days we’re the mint.
What we could not find, anywhere, was the windfall arriving where it has always arrived before — in the price. We want to be careful here, because these were perceptions, not proofs. Our grocery bills were flat where they weren’t up. Our software subscriptions were up. Our insurance was up. The stock market was setting records, and the gap between people who own things and people who earn things was widening so fast you could watch it move. And underneath all of it, an intuition we couldn’t shake: prices rise fast and fall slow — when they fall at all. A new cost gets priced in by Friday. A removed cost gets passed along slowly, sometimes never.
In June we had treated the windfall and the discount as one object — the surplus was the discount, delivered by the price system to everyone at once. This summer there was plainly a surplus and plainly no discount. It turned out we were not the first to notice.
Every perception has a name
Start with the intuition that prices rise fast and fall slow, because we thought we were describing a mood and it turns out we were describing a literature. Economists have studied asymmetric price transmission for decades — the finding that prices go up faster than they come down — and the nickname in the field is, literally, rockets and feathers2. Sam Peltzman published the landmark paper in 2000 under a title that requires no translation: “Prices Rise Faster than They Fall.” He looked at more than two hundred and forty markets and found the asymmetry in two markets out of three — and found, more troubling still, that standard economic theory couldn’t explain it. He tested the obvious suspect, concentrated industries, and cleared it: fragmented, competitive markets showed the same asymmetry as tight ones. The feather is not a folk myth. It’s one of the better-documented anomalies in empirical economics. It drifts in every market; what differs is whether anything ever forces it down. Our intuition wasn’t a mood; it was a measurement we hadn’t met yet.
Then the sense that the price of a thing and the cost of making it had come loose from each other. That has a measurement too. Two economists, Jan De Loecker and Jan Eeckhout, tracked the markup — how much a firm charges above the cost of producing the thing — across the whole American economy. In 1980 the average markup was about 21 percent. By 2016 it was over 60 percent3. The markup nearly tripled in four decades, and almost all of the rise was in the largest firms. A markup can rise because prices go up or because costs come down and prices don’t follow. Either way the difference piles up as margin, and margin is where the rest of this essay goes looking.
Then the question the first two measurements raise: whether America is still the competitive economy it believes itself to be. Thomas Philippon — a French economist who came here because America was the land of competitive markets — spent years assembling the answer, and published it under the title The Great Reversal4. The finding is in the title. Since the late 1990s, concentration and profit margins rose across most American industries while the same measures fell in Europe; the two economies traded places. Americans invented cheap flights and cheap phone service, and Americans now pay more than Europeans for both.
Three measurements, then. The feather drifts everywhere and lands only where a rival forces it down. The margins rose. And the rivals got fewer. Whatever else is true, this all began somewhere, and the beginnings have dates.
The dates
A decline like this invites grand theories — cultural, generational, technological. Maybe some of them hold; an effect can have many causes. But one cause requires no theorizing at all, because every piece of this failure has a rule change with a date on it, and the dates cluster in one five-year window, forty-some years ago.
In 1978, Robert Bork published The Antitrust Paradox, and within a few years American antitrust had reorganized itself around his standard: block a merger only if it could be shown to raise consumer prices. Mergers that merely removed a competitor sailed through5. Startup formation — the rate at which new firms enter and challenge incumbents — peaked in the late 1970s and has declined ever since. And in 1982, the SEC adopted Rule 10b-18. Before it, a company buying back its own shares at scale risked prosecution for market manipulation. After it, buybacks had a legal safe harbor. The starting gun for the trillion-dollar buyback era was a rule, with a number, adopted in a specific year — the same little window in which the markups started climbing and the entrants stopped coming.
The tell
Last year, the companies of the S&P 500 spent roughly a trillion dollars buying back their own stock — an all-time record, in the middle of the largest cost-collapsing technology deployment in history6.
Understand what a buyback is, structurally. A company generates cash beyond its needs and faces a choice:
Cut prices and take market share from its competitors.
Invest — new capacity, new products, new people.
Hand the cash to its shareholders by buying back its own shares.
The first two are what competing looks like. The third is what not having to compete looks like — because in a genuinely competitive market, that cash never reaches the boardroom at all. Rivals compete it away before it lands. A firm running record buybacks off record margins is announcing, in the only language balance sheets speak, that nothing can force its prices down — and that after funding everything it can think of to build, this much is still left over. The objection that the cash is going into data centers doesn’t balance the ledger. The firms building the data centers are the same firms doing the buybacks, and the trillion is what was left after the data centers were paid for. What the data centers are for is a different question, and we’ll come to it.
The buyback is the tell.
And it’s the answer to the question we couldn’t put down all summer: where did the nickel go? Follow it. The AI windfall arrives at the firm as margin — costs fall, prices hold — and the buyback converts the margin into share price. The share price lands in the portfolio of whoever already owns. That’s the whole route: windfall → margin → buyback → asset price → owner. The nickel is real. It’s enormous. It just never reaches the price tag anymore. It goes upstairs.
And where a firm doesn’t already have a moat, the same cash buys one. You can hear this play described in plain language now, because it has become a genre. Over coffee last spring, an investor laid it out for us as a trade: buy a small, unglamorous business — HVAC, plumbing, industrial services — at the five-times-earnings multiple it has always sold for, from an owner who has never typed a sentence into a chatbot. Use AI to cut half a million in costs. Sell the fattened earnings at the same multiple, in under six months.
He wasn’t describing a private idea. The patient version of the same trade has a name: the search fund. A newly minted MBA raises money from investors, finds one boring business, buys it, and runs it for years rather than months. Stanford has tracked search funds since 1984. More of them launched in 2023 than in any year on record, and their returns beat most of what Wall Street offers7. Scroll LinkedIn for a week and you’ll meet a dozen of them, each with the same story about the plumbing company they bought from a retiring founder. Private equity does the same thing with a bigger checkbook, buying dozens of these businesses at a time and combining them. More than sixty percent of the fifty largest HVAC companies in America are now private-equity owned, twenty-five billion dollars has rolled up the home-services trades in under a decade, and prices at the consolidated platforms have risen faster than inflation since 20188.
The private-equity version adds a step the coffee version left out. Cut the costs, yes — and then buy the competitor across town, so there is nobody left to start the price war that would hand the savings to the customer. The cost-cutting and the consolidation are one motion.
Where the nickel still falls
The nickel still falls wherever competition survives. Where it does, the old story runs exactly as history said it would. The price of AI itself is the purest case: the frontier labs are in a real price war, and a unit of GPT-4-class intelligence that cost thirty dollars per million tokens in 2023 costs under a dime today9. The long version of the same pattern is decades old: televisions, clothing, electronics — the tradable, competitive categories — deflated for a generation, while healthcare, housing, and education, the moated categories, never gave back a dime. Some of that gap is labor, and economists have a name for why service prices rise as goods get cheap: a plumber’s hour can’t be shipped from Shenzhen. But the markup is measured above cost, wages included, and it rose anyway.
But there is a third case, and we mistook it for the first one for months. Freelance rates for copywriting and translation have cratered this year, and we read that as the nickel falling — thousands of sellers, no moats, textbook competition. It isn’t. Those rates aren’t falling because sellers are undercutting each other. They’re falling because the buyers left. Nobody is negotiating a lower price for the blog post; they’ve stopped ordering the blog post. That is not a price coming down. That is a market ceasing to exist, and a corpse’s asking price is not a discount. The buyer did come out ahead, but not by paying less for the blog post. The buyer stopped ordering blog posts and started making them. That is a windfall, and we’ll come to who collects it. It did not arrive through the price.
So the picture has three columns, not two.
Moated markets. Where a merger, a license, or a switching cost keeps any rival from forcing the price down, the windfall pools as margin and the price ratchets. Our grocery bills live here. So does your subscription stack, for now.
Competitive markets. Where rivals are still fighting for each sale, the nickel falls as it always did.
Replaced markets. Where the product is being made obsolete, the price doesn’t move at all — it stops mattering. The offshore shop that quoted our client thirty-five thousand dollars lives here, and we’ll come back to it.
It was carried
Go back to the tractor, because this is where we mistook the special case for a law.
The tractor made the flour cheap. But the tractor did not make your bread cheap — the grocer’s competitor did. The wheat in a loaf has never been more than a few cents of its price; the rest is everything that happens between the field and the shelf, and every hand it passes through is a place the nickel can stop. The nickel reached the family’s pocket because a thousand bakers and a thousand grocers, none of them with an ounce of pricing power, fought each other for the family’s business, and the fight carried the windfall downstream. Same with the container ship: the savings crossed the ocean as margin and became your discount only in the last hundred feet, at the shelf of a retailer locked in a price war.
The nickel never fell on its own. It was carried. Peltzman’s feather drifts in every market; it lands only where a rival forces it down. We watched two centuries of nickels arriving and concluded that nickels arrive, when the truth was that a thousand grocers had been fighting each other to hand them over — and for forty years the grocers have been bought, merged, and relieved of the need.
There’s even a number for how well this worked when it worked, and we very nearly quoted it as a law. The economist William Nordhaus famously estimated that innovators capture only about two percent of the value their innovations create — the other ninety-eight percent flowing downstream, mostly to consumers10. The measurement window: 1948 to 2001. The years when the grocers were still fighting, almost exactly.
Ask that question of today’s innovators. Do the frontier labs capture two percent of the value they create? Strangely, the two percent may still be about right — they are the one sector in a real price war, and their prices are collapsing on schedule. The other ninety-eight percent still leaves the lab. What’s changed is where it stops. Nordhaus split the value two ways, innovator and consumer, because in his window whatever sat in between passed the value along. Now the firm in between holds its price and keeps the difference. Nordhaus’s number isn’t wrong. His ledger needs a third line, and that line is where the ninety-eight percent now lands.
The patient objection is that it’s early. The dynamo took forty years to show up in the productivity statistics; the computer took twenty; the consumer’s share arrives at the end, not the beginning. All true, and all of it assumes the competition that carried the earlier windfalls is still standing. We aren’t judging AI on three years. We’re reading what happened to the windfalls before it. The container ship and the personal computer did deliver, in the one column where the grocers were still fighting. Everywhere else, those same decades are the ones in which the markup tripled and the entrants stopped coming. The question for this windfall is not whether it will arrive. It’s who is left to carry it.
So here is the amendment, one sentence: the windfall becomes a discount only where competition carries it there; everywhere else it stays where it landed. And here’s why the amendment matters more now than it would have in any previous decade: AI is being created in the most fiercely competitive sector on earth and deployed into the least competitive American economy in a century. The value pours in upstream at historic volume — and the competition that used to deliver it to you was dismantled forty years before it arrived.
The cheap we didn’t question
In The Last Wide Rung we took on the optimists’ strongest objection — but everything will be so cheap — and we answered it on its own terms. We granted the cheapness. We argued only that cheap is not the same as yours: that a wage was a claim, your share of the wealth you helped make, and abundance without a claim is still dispossession. We stand by the answer. What we never questioned was the premise.
What the old windfalls delivered was never just cheapness. It was automatic cheapness — universal, passive, skill-free. The farmhand collected the tractor’s nickel by doing nothing more than buying bread. That’s the cheap the optimists are promising. And that’s the cheap that isn’t coming.
If competition no longer carries the nickel, the discount never reaches the shelf. The frontier’s miracles will keep getting cheaper. Everything moated will keep going up: housing, healthcare, the insurance on both, the service call from the platform-owned HVAC company. Those sectors collect the same AI windfall as everyone else. What they don’t do is pass it on.
Holding the price is one act, and it works in both directions. A cost shock is passed to you by Friday, and sometimes before it arrives: the Kansas City Fed found that in 2021 firms raised prices ahead of the costs they expected, and that markup growth alone could account for more than half of that year’s inflation11. A cost collapse is kept. The margin that is your inflation on the way in is the firm’s windfall on the way out, and both leave by the same door, as share price.
Inflation and the stock market have stopped being opposites. The record market isn’t happening in spite of your rising bills. The record market is your rising bills, booked on the other side of the ledger. When the inflation is margin, the inflation is the market.
Which means the summer we couldn’t reconcile was never a paradox. AI-led layoffs and record markets, record margins and no discounts — that’s not the chain misfiring. That’s the rerouted chain working perfectly:
windfall → margin → buyback → asset price → owner.
The hour that disappeared
Start with the offshore shop, because we promised to come back to it, and because it was the last great nickel most businesses ever collected. Offshoring was labor arbitrage: the same work, done by hands that cost a fifth or a tenth as much. A developer-hour in Bangalore for a fraction of one in Austin. It was a real windfall, and it reached the customer, because thousands of shops competed for the work and kept bidding the price of an hour down. For twenty-five years that is what let a small company afford custom software at all.
AI didn’t lower the wage. It removed the hour. The shop never really sold the hour. It sold the gap between two hours. Now the buyer doesn’t compare the shop’s hour to an American hour. He compares it to an agent that does the same work for a few dollars a month. Against that, no hour is cheap enough, and the American hour it was undercutting is gone the same way. The gap closed from both ends. There is no gap left to sell.
A discount is a price on something you still buy. So the shop can’t reprice its way out. And it can’t pivot to managing the agents on your behalf, because that is precisely the thing you can now do yourself, with an agent and some coaching. The thirty-five-thousand-dollar quote our client received isn’t going to fall to twenty thousand, or to five. It’s going to sit at thirty-five until the shop that gave it closes. The replaced column: the price doesn’t move. It stops mattering.
The surcharge
Now look at the software vendors, because they’re the other seller in this story, and they were built on the same bargain. For most of computing’s history, every company that needed software wrote its own, and it was the most expensive thing in the building. Then the software companies made a trade: write it once and sell it to everyone, so that no customer pays for the build, only for a slice of it. Offshoring came next and cut the cost of the hours that went into the build. Then came the subscription, and the slice got thinner still, until a company could rent for a few hundred dollars a seat what it would once have spent millions to build.
The whole reason to rent a slice of somebody else’s software was that building your own cost millions. AI changed that number. If a working replacement now costs a weekend and an agent, the subscription stops being the bargain, the way the offshore hour did. We wrote an entire essay called The Cost of Software Is Now Zero.
So by the offshore shop’s logic, vendor prices should be collapsing. They are not. Salesforce has raised list prices twice since 2023, both times well above inflation, and its new AI-bundled tier lists at five hundred and fifty dollars a seat; Microsoft’s Power BI jumped forty percent in a single year; and across the industry, roughly three-quarters of vendors now charge separately for AI features12. The vendors whose costs collapsed are raising prices. The technology that gutted their costs is being sold to you as a surcharge.
So why hasn’t the vendor gone the way of the offshore shop? Not because it has a moat. Because it has a runway. The offshore shop had nothing to hold you with. The vendor has a contract with a term on it, a few integrations, and your habit. None of that is a wall. All of it is time. And a seller who can see the market closing uses the time the way you’d expect: it raises prices on whoever hasn’t crossed yet. That is what the surcharge is. Business schools have a name for it, harvesting the installed base: when an incumbent can see its terminal value falling, it stops pricing to win customers and starts pricing to extract from the ones it has, and returns the cash before the contracts lapse. The vendors aren’t ratcheting because they’re safe. They’re ratcheting because they aren’t.
The runway is shorter than it looks. We paid DocuSign for years for one thing, the signature. DocuSign presumably thought it had a moat, because it held every document we’d ever signed. We held them too, and the day Google made signing a free feature, we were gone in an afternoon.
A signature button is not a system of record, and a reader who runs one will say so: DocuSign was a feature, and Salesforce is where the company’s memory lives. Fair. A system of record has a longer runway than a signature button. But what makes it longer isn’t the code.
Permissions, audit trails, integrations, the schemas an auditor wants to see: that is the most documented, most repeated, most tested software on earth, and it sits at the exact center of what an agent does best. We have watched owner-operators replace entire ERP systems this year, not prototypes, the system the business runs on.
What’s left after the code is the compliance certifications, the indemnification, and the uptime guarantees, which a customer buys rather than builds. Those are real. But each of them is becoming something you can buy on its own, or do in-house with agents and a few people, without the vendor that used to bundle them. It is still a runway.
So the vendor is headed for the replaced column as well. It’s on a longer runway than the offshore shop was, and it’s charging for the time it has left. The price never comes down. It goes up until it stops mattering. Neither seller is going to hand you the nickel.
The firm in between
There are two more sellers in this story, and one of them is paying for the price war. The frontier labs are in the competitive column, and the price of a token has fallen three-hundredfold. Somebody is funding that, and it isn’t the labs. It’s the hyperscalers, on course to spend more than seven hundred billion dollars this year on data centers, chips, and power while owning or backing the very labs whose prices are collapsing13.
And the largest single check they write goes to one company. Nvidia sells the chips at seventy-five cents of gross margin on the dollar, to customers with nowhere else to buy at that scale, and it has collected more of this windfall than anyone, before a single token was priced14. That is the moated column, sitting upstream of the price war rather than below it. The hyperscalers are placing the bet. Nvidia is the house.
You can watch what the hyperscalers are buying from both ends at once. Take Microsoft. It sells raw tokens through Azure at whatever the market will bear. Then it sells the same intelligence inside Word and Excel at thirty dollars a seat, per month, to a user who may consume a few cents of compute. There is only one reading under which that pricing makes sense: the thing being sold isn’t intelligence, it’s the layer between intelligence and you. The bakery buys flour at the tractor’s price and sells bread at the shelf’s. The data centers aren’t the windfall. They’re the stake. Whether the layer is a moat or a runway is a question for another essay.
Four sellers, then: the shop evaporating, the vendor harvesting, the hyperscaler betting a fortune on the next place the nickel stops, and Nvidia taking the bet. None of them is going to hand it to you.
The far side of the counter
And yet — our client with his thirty-five-thousand-dollar quote. The operators walking out of their SaaS contracts. Us, at thirty times our old pace. The cheap does still arrive, for the ones who go get it. But look closely at how, because it’s a different transaction in kind. Our client didn’t receive a discount. He traded his own time — his, plus an agent’s, plus some coaching — for the thirty-five thousand dollars he would otherwise have sent overseas. A friend of Ben’s from his Microsoft days, now the CTO of a developer-tools company, said it to him this summer in a single sentence: “We don’t need ICs anymore — everyone is a manager now.” He could have been describing our client, who for the length of that project wasn’t a programmer at all. He was a manager with one very fast direct report.
He escaped the price by making the thing himself. That’s the strange new channel this era has opened. The windfall no longer reliably travels down through prices — but the machine that mints it rents for a few dollars a month. So you don’t wait for the discount. You rent the machine and keep the whole gap yourself: the thirty-five thousand dollars, not five percent off it. The windfall always starts with whoever does the making, and for the first time that can be you. No earlier upheaval worked this way. The farmhand couldn’t take the tractor home. This tractor fits on your desk.
The new channel has one price the old one never did: you have to cross the counter. The old windfall paid you for shopping. The new one pays you for operating — for ceasing to be a customer at all. The gate in that counter, between doing the work and managing it, was always people skills, and an agent is a more forgiving direct report than any human has ever been. The gate didn’t get lower. It fell. Our client still needed coaching to cross, and what the coaching gave him wasn’t a skill. It was permission. The nickel is real, and it is lying in heaps, in plain sight, on the far side of a counter most people don’t know they’re allowed to walk behind.
Owners and operators
In June we wrote that the value still flows somewhere we can spend it, just not somewhere we can earn it. The spending half now needs amending. The price system was the one channel that paid everyone, automatically, and it has stopped paying. What used to arrive as a discount now arrives as share price, to whoever already owns. There is one other way to collect, and it isn’t automatic: operate the machine yourself and keep what it makes. So two kinds of people are collecting the windfall. Owners, on the rung above labor. Operators, on the far side of the counter.
We ended the first essay promising to climb to the rung above — the one made of capital, not work, where the value still pools. That essay is coming, and this one only raises its stakes. The wage is failing as a claim on the wealth. That was the first essay. The discount, the one payment history delivered to labor whether or not it asked, is being intercepted at the moat. That is this essay. With both the claim and the discount gone, nothing pays you automatically anymore. Two questions are left: who gets to own, and who can learn to operate.
Challenger, Gray & Christmas, Job Cut Announcement Report, July 2026: artificial intelligence was the leading stated reason for announced US job cuts for the fifth consecutive month (March–July), cited in 112,713 cuts year-to-date, about 24% of all announced cuts. May 2026 set the monthly record at 38,579 AI-attributed cuts (40% of that month's total). AI-attributed cuts for all of 2025 were 54,836. Note the denominator: total announced cuts through July 2026 were 477,033, down 41% from the same period in 2025. Layoffs overall are not at a record. AI as the stated cause is. Stated is the operative word: executives have reasons to attribute cuts driven by overhiring or interest rates to AI, and some of this is surely that. But the fact that AI has become the reason companies want to be seen giving is itself information.
The term traces to Robert Bacon's 1991 study of UK petrol prices ("Rockets and Feathers"). Sam Peltzman, "Prices Rise Faster than They Fall," Journal of Political Economy (2000), examined over 240 consumer and producer markets and found the asymmetry in roughly two-thirds of them, with the gap persisting for months — and concluded standard theory offers no good account of it.
Jan De Loecker, Jan Eeckhout, and Gabriel Unger, "The Rise of Market Power and the Macroeconomic Implications," Quarterly Journal of Economics (2020): average markups across US firms rose from about 21% above marginal cost in 1980 to about 61% by 2016, with the rise concentrated in the largest firms.
Thomas Philippon, The Great Reversal: How America Gave Up on Free Markets (2019). Representative findings: US broadband and mobile prices roughly double comparable EU rates; US airline markups rising while EU fares fell after low-cost entry — reversing the positions the two economies held in the 1990s.
Robert Bork, The Antitrust Paradox (1978), supplied the "consumer welfare standard" that narrowed enforcement from the 1980s on. On entry: Decker, Haltiwanger, Jarmin, and Miranda document the US firm entry rate declining from roughly 13–14% in the late 1970s to under 9% by the 2010s. SEC Rule 10b-18, adopted November 1982, created the safe harbor that made large-scale open-market buybacks legally routine.
S&P Dow Jones Indices: S&P 500 buybacks for the twelve months ending September 2025 reached a record 1.02 trillion dollars, with Q1 2025 (293.5 billion) the largest single quarter on record and full-year 2025 the first calendar year above one trillion. The pre-pandemic exhibit: Bloomberg calculated that the major US airlines spent about 96% of their free cash flow on buybacks in the decade before 2020 — shortly before receiving a taxpayer bailout.
Stanford Graduate School of Business, 2024 Search Fund Study: 681 search funds formed in the US and Canada since 1984, with a record 94 launched in 2023; aggregate pre-tax internal rate of return of 35.1 percent and 4.5x return on invested capital across the sample; median purchase price of an acquired company 14.4 million dollars. The study attributes the growth partly to the wave of retiring baby-boomer owners putting small businesses up for sale.
Industry tallies put more than 60% of the top 50 US HVAC companies and over half of the top plumbing companies under private-equity ownership, with more than 25 billion dollars deployed into home-services roll-ups over the past eight years; trade and press coverage (including the WSJ) reports service prices at consolidated platforms rising faster than general inflation since 2018.
OpenAI's GPT-4 launched in March 2023 at 30 dollars per million input tokens. By mid-2024 GPT-4o mini delivered comparable benchmark performance at 15 cents per million, and by 2026 several trackers put GPT-4-class capability at 6 to 10 cents per million across vendors — a decline of roughly 300- to 500-fold in three years, or about 10x per year for a fixed capability level. See DeepLearning.AI's The Batch on falling token prices and the TokenCost AI price index
William Nordhaus, "Schumpeterian Profits in the American Economy: Theory and Measurement" (NBER Working Paper 10433, 2004): producers captured an estimated ~2.2% of the social value of innovations, 1948–2001.
Andrew Glover, José Mustre-del-Río, and Alice von Ende-Becker, “How Much Have Record Corporate Profits Contributed to Recent Inflation?” Federal Reserve Bank of Kansas City Economic Review (Q1 2023): markups grew 3.4% in 2021 against PCE inflation of 5.8%, so markup growth could account for more than half of that year’s inflation. The authors read the pattern as firms raising prices in anticipation of future cost increases rather than as a rise in monopoly power — which is the rocket half of rockets and feathers stated in a central bank’s own words.
Salesforce raised Enterprise and Unlimited list prices 9% in August 2023 and a further ~6% in August 2025, framing the second increase around its AI (Agentforce) capabilities; Sales Cloud Unlimited went from 300 to 330 to 350 dollars per user per month, and the new Agentforce 1 Sales bundle, with agentic actions included, lists at 550. Microsoft raised Power BI Pro from 10 to 14 dollars per user per month in April 2025, a 40% increase and its first in nearly a decade. Industry pricing surveys in 2025–26 find roughly 73% of SaaS vendors now charging separately for AI features, with median year-over-year price increases near 8%.
Calendar-2026 capital-expenditure guidance from the four largest hyperscalers, as of their July 2026 earnings: Amazon about 200 billion dollars, Alphabet 175 to 205 billion, Meta 115 to 145 billion, Microsoft on a run-rate near 190 billion — roughly 725 billion dollars combined, against about 410 billion in 2025. See Futurum, "AI Capex 2026" and the companies' own guidance
NVIDIA, second-quarter fiscal 2027 results (quarter ended July 26, 2026): revenue 96.2 billion dollars, Data Center revenue 89.0 billion, GAAP gross margin 75.0 percent; third-quarter revenue guided to 108 billion. Analyst estimates put NVIDIA at 75 to 80 percent of AI-accelerator revenue in 2026, with the four hyperscalers roughly 40 percent of its sales, all four of them shipping custom silicon of their own.