The Emperor’s New Clothes is where the bust begins.
In Part 2, AI entered the story as a gene, a trait, and an environmental shock — a new advantage that changes the conditions every firm has to survive in. Part 3 starts with what happens when capital floods toward that new advantage and the crowd begins to applaud the suit.
This is not an argument that the technology is fake.
In Hans Christian Andersen’s fable, the danger is not merely the absence of cloth. It is the social machinery around the emperor: the ministers, advisers, courtiers, and crowd who all have incentives to repeat the same story once the story has become expensive to question.
That is how a boom turns into a bust. The technology can be real, useful, and improving — while the promise layered on top of it becomes too large for the underlying cash flows, margins, and demand to carry.
The cloth is not the technology. It is the promise layered on top of it.
When the bust comes, it is not always the invention being disproved. Often, it is the surplus promise being tested. The market asks which firms can turn the new capability into durable returns, and which ones were only wearing the story.
One line, four acts — the curve the rest of Part 3 explains.
Scroll, and the cycle draws itself: capital floods in and the population blooms, the promise holds at the peak, the surplus runs out and the bust selects, and what survives consolidates at a new line.
When energy arrives in surplus, the throttle comes off.
An organism is, at bottom, an engine for turning energy into more of itself. It takes energy in and converts it into three things: survival, growth, and reproduction. When energy is scarce, that conversion is throttled. The population sits near the environment’s carrying capacity — the ceiling of what its resources and space can sustainably support — and it cannot push past that ceiling for long.
But when energy arrives in surplus — a rich season, a bloom of food, the removal of a predator or a competitor — the throttle comes off. And here is the part that matters: organisms do not bank a surplus or grow modestly with it. They convert it, as fast as they can, into growth and reproduction beyond what they need merely to persist.
Why beyond what is needed?
Because evolution, as we established in Part 1, keeps score on reproduction, not restraint. An organism that holds back while energy is abundant is simply out-reproduced by one that does not. Fitness is relative — there is no “enough,” only “more than the organism next to you.” And nothing in the system can see that the surplus is temporary; it responds to the abundance in front of it, not the scarcity that will follow. So the population swells past the level the environment can sustain once the surplus is gone.
The overshoot is not a mistake.
It is what an energy surplus does to anything built to reproduce.
Now map it back.
Capital is energy. Firms are organisms. The market is the environment, and real demand is its carrying capacity.
When capital is scarce, firms grow within the bounds of what genuine returns can fund — roughly, within carrying capacity. But when capital floods in — cheap money, a compelling narrative, the fear of being left behind — the throttle comes off, and firms do exactly what organisms do with a surplus.
They convert it into maximal growth and reproduction: scaling headcount, building capacity, spawning imitators, faster than real demand can justify.
Not because anyone is foolish, but for the same three reasons the reindeer cannot help themselves — the system rewards growth over restraint, fitness is relative, and no one can be sure the surplus will last.
Hold back while your rivals expand and you are simply out-grown, and out-grown means selected out.
The bloom is not mania.
It is organisms metabolising a surplus exactly as they are built to.
Nature shows the shape of it cleanly. In 1944, twenty-nine reindeer were released onto St.
Matthew Island in the Bering Sea. The lichen was deep, there were no predators, and the herd did what a surplus invites: it converted the abundance into more reindeer. By 1963 there were roughly six thousand animals — far more than the island could feed. By 1966, the population had collapsed to a few dozen survivors.
There were no predators to blame. The herd had grown beyond what the island could sustain.
The lichen had been depleted, and when harsh winter conditions arrived, the correction was brutal. A surplus arrived, a population built to reproduce turned it into more mouths than the environment could sustain, and when the surplus ran out, the correction was total.
Replace the lichen with capital, and you have the AI boom.
Everyone starts an AI company. Every incumbent pivots. The population of firms competing for the same finite demand explodes far past what that demand can yet feed.
It looks like vitality.
It is, in part, a surplus being metabolised into overshoot.
Bloom mechanics
The environment temporarily expands.
Organisms convert surplus into more of themselves.
The population grows past what can be sustained.
Survival depends on independent energy capture.
Markets can bloom on expected profit before real demand arrives.
There is a second reason the overshoot runs further in markets than in nature. In an ecosystem, the surplus is food that exists right now, and the bloom only ends when that food is gone. In a market, the surplus firms respond to is not present profit but expected profit.
The signal firms respond to — funding, valuation, hype — is decoupled from the actual resource, which is revenue, demand, fitness. A company can grow enormous on capital long before it captures any energy of its own.
This is what happens when a trait spreads faster than the environment can prove its carrying capacity.
In Part 2’s terms, this is trait diffusion at its most dangerous: an advantageous trait that spreads to the entire population at once, so everyone adopts it simultaneously, nobody differentiates, and the whole field blooms together.
AI spreads through mutation, gene flow, and recombination. A founder builds an AI-native product. An incumbent hires the talent. A lab publishes a model. A competitor copies the interface. A cloud provider packages the infrastructure. A consulting firm turns the idea into a service line. A board turns it into a mandate.
The trait moves through the population quickly.
But when everyone adopts the same trait at the same time, nobody is necessarily differentiated by adopting it.
A population that blooms together, competing for the same resources with the same new trait, is a population built to overshoot.
The cloth is not the technology. It is the promise layered on top of it.
Here is where biology stops being enough, and the second lens earns its place.
In Hans Christian Andersen’s fable, two swindlers sell the emperor a suit of clothes they claim is invisible only to the foolish and the incompetent.
The emperor sees nothing, but says nothing — to admit it is to admit he is a fool.
His ministers see nothing, but applaud — to object is to confess they are not fit for their posts.
The whole court praises a suit that does not exist, each person held in place not by belief, but by the fear of being the one who does not get it.
The emperor parades naked.
And the spell holds until a child, who has nothing to protect, simply says: he is not wearing anything.
Nicholas Capaldi and Miles Smit, in The Art of Deception: An Introduction to Critical Thinking, describe the broader trick as one of critical thinking’s oldest traps: discredit the objection before it is even made.
In logical terms, this is best understood as poisoning the well.
The swindlers do not merely say the cloth is beautiful. They define disbelief as evidence of stupidity or incompetence.
So the emperor is not deceived only by a false claim.
He is trapped by the social cost of telling the truth.
To say “I do not see the cloth” is to risk becoming the thing the trick has already condemned.
This distinction matters in technology markets.
The invisible cloth is not the technology.
The technology can be real.
Railways were real. Radio was real. The internet was real. Fibre was real. AI is real.
First, what the cloth actually is — because this is the part most retellings blur. The swindlers do not sell the emperor a real garment that happens to be defective. They sell him nothing: an invisible cloth they claim only the foolish and the unfit cannot see.
The cloth is not a fake technology.
It is a fake promise — a value that cannot be pointed to or verified, wrapped in a threat. To say you do not see it is to confess you are a fool. That is the swindle: not the cloth, but the cost of doubting it.
Notice who they target.
They go to the emperor — the decision-maker, the one with the most authority and the most to lose from looking incompetent. Once he has accepted the cloth and put it on, the trap closes from the top down.
His ministers cannot say they see nothing, because the emperor has already declared that he does — to break with him is to mark themselves as the fools the cloth was designed to expose.
So the court echoes the emperor. And once the court echoes him, everyone below must echo the court, for the same reason.
The emperor parades naked, and an entire kingdom praises a suit that does not exist — not because anyone believes in it, but because the people with power propagated the promise, and everyone beneath them is forced to propagate the same noise or be singled out.
The spell holds until a child, who has no position to protect and no authority to contradict, simply says: he is not wearing anything.
Now map it onto a bubble, and keep the distinction sharp.
The invisible cloth is not the technology — the technology can be entirely real. The cloth is the promise layered on top of it: the valuations, the projections, the story about how much energy this will eventually capture.
Late in a boom, a great many people privately suspect that promise has detached from anything real. But the same trap is in place.
The decision-makers — the marquee investors, the celebrated founders, the analysts who staked their reputations on the narrative — cannot disown it without admitting they were the fools who could not see. So they keep affirming it.
And once the people with authority and capital keep affirming it, everyone downstream must affirm it too: the smaller funds, the employees, the press, the latecomers, all forced to propagate the same promise or risk being the one who “doesn’t understand the new paradigm” — who missed the internet, who did not see it coming.
The doubt stays private, the capital keeps flowing, and the flow itself becomes the evidence.
Momentum is read as proof of fitness.
I invest because you invested; your investment looks like confirmation of mine.
And here we have to correct a tempting but lazy contrast. It would be easy to say that nature is blind and markets are not — that evolution has no foresight while markets are run by agents who anticipate. But Part 2 already put that idea to rest.
Nature is not random, and it is not blind. What looks like chance is the net result of all the forces acting on a system at a given moment — the same way a change in weather is not luck but the sum of pressures, temperatures, and currents resolving into an outcome.
And nature is full of agents that anticipate and shape their environment: predators that learn, prey that adapt their behaviour, organisms that build, store, signal, and respond to what they expect is coming. Not to the human degree — but the difference is one of degree, not of kind.
So the real disanalogy is not foresight versus blindness.
It is density.
A market is an environment packed with an unusual number of anticipating agents, all watching each other, all acting on expectations of what the others will do — and all of it wired together by capital, which transmits those expectations almost instantly.
Every participant in the economic food chain is one of these forces. Investors and banks decide where energy flows. Central banks expand or restrict the supply of it. Businesses convert it; consumers and labour direct it through what they buy and where they work; raw materials set the hard floor of what is physically possible. Each is reading the others and adjusting in real time.
How these forces actually shape the environment is the subject of the next part — here it is enough to see that there are many of them, and that they are densely coupled.
That density is why the cycle runs hot. You would expect so much anticipation to dampen the swings — for smart money to see the overshoot coming and pull back.
It does not, because in an environment where everyone is anticipating everyone else, the rational move inside a bloom is to ride it. If the others are buying and their buying is itself moving the price, the agent who anticipates correctly is the one who rides the wave, not the one who steps off early.
Foresight, densely coupled and pointed at short horizons, does not restrain the bubble.
It feeds it.
The very intelligence of the participants is what carries the overshoot past where a sparser, slower system would have stopped.
The spell holds — until a child speaks. An earnings call that misses. A rate that rises. One large organism that stumbles and makes the herd look down.
The fable as market structure
Future value that cannot yet be verified.
Investors, boards, analysts, press, and employees echo the signal.
Nobody wants to be the one who does not get it.
A missed earnings call. A rate rise. A stumble.
The bust is natural selection running in fast-forward.
When the surplus that drove a bloom runs out, the population does not shrink gently or evenly.
It crashes, and the crash is selective.
This is the other half of carrying capacity. During the bloom, the population overshot the level the environment could sustain — there are now far more organisms than there is energy to feed them.
So the environment does what it always does when demand exceeds supply: it removes the excess. But it does not remove them at random. The organisms that go first are the ones that were only ever surviving on the surplus — the ones with no independent way to capture energy once the easy abundance is gone.
A die-off is not the environment turning hostile. It is the environment returning to its true carrying capacity, and selecting out everything the surplus had been propping up.
What survives is what could have fed itself all along.
Now map it back.
The surplus was cheap capital. The die-off is the bust.
The firms that starve first are the ones that were living on capital rather than profit — the organisms with no independent energy capture, sustained entirely by the inflow rather than by anything they generated themselves.
While the capital flowed, they were indistinguishable from the firms that were genuinely fit; abundance hides the difference, because everyone is being fed. The moment the inflow stops, the difference is all that matters.
The bust is natural selection running in fast-forward.
It is not a malfunction of the system. It is the system, finally asking each organism the question it deferred during the boom: can you generate your own energy, or were you only ever being fed?
And this is where Part 1’s quietest word returns: the shell.
In nature, a die-off is not pure loss to the ecosystem — the fallen become food, and their energy is reabsorbed by what remains. The same happens in a crash.
The well-fed predators feast: firms with real profit and deep capital do not merely survive the die-off, they buy the starving for scraps — talent, technology, and IP absorbed at a fraction of peak value.
The prey does not vanish from the system.
It gets eaten, and its genes are folded into something larger.
Consolidation is the bust’s quiet second act, and it is usually where the eventual giants are actually made.
Customers who would buy even without cheap capital — the firm captures its own energy.
Can raise prices without losing the customer; margin survives the squeeze.
Owns the channel to the buyer; doesn’t rent reach it can’t keep.
Each extra unit costs almost nothing to serve, so scale compounds instead of bleeding.
Cash and low debt to outlast the drought — and to buy while rivals can’t.
Embedded where the work actually happens; switching away is expensive.
Turns the bust into a buying window — talent, tech, and IP for scraps.
The future can be real while most routes to it are still mispriced.
None of this is hypothetical. The full cycle — the bloom, the spell that sustains it, the selective bust, the consolidation — has run forward at least four times in the industrial record, and each leaves the same residue.
History does not hand us a clean “true” valuation for any of these episodes; no one can re-run a 19th-century railway or a 1929 radio stock through a discounted-cash-flow model decades later and settle the matter.
So measure each one the same practical way: what the market paid at the peak of belief, what the asset was worth once the promise broke, when — if ever — it regained that peak, and whether the value that eventually returned came from the original thesis or from a business that had quietly become something else.
Hold one distinction above all of them.
The invisible cloth is never the technology.
The railways ran; radio played; the internet arrived; the fibre carried light.
The cloth is the promised economics — the assumption that future cash flows, margins, adoption, and market structure would appear fast enough, at high enough prices, to justify what was being paid today.
Above roughly −95%, the equity was effectively wiped — the company survived, if at all, as assets to be absorbed, not as the stock investors had bought.
Sources · railway index: CEPR (Campbell & Turner) · RCA: Stanford UP, Bridgeway · Nasdaq: Goldman Sachs · Cisco, Amazon, Nortel: period stock histories · WorldCom, Global Crossing: bankruptcy filings. Figures rounded; full trail below.
Sources · Nasdaq recovery (2000–2015): Goldman Sachs · Amazon, Cisco, RCA: company histories and period stock data.
RCA — wrong for a generation
Quoted two ways after the 1929 five-for-one split. Same collapse, two bases.
Pre-split, the same path reads $43 → $568 → $15 at a P/E near 72. The value returned only after RCA had become a television business.
AOL vs Amazon — the fork
Two firms, one boom. One thesis broke; the other survived by becoming something else.
AOL’s walled-garden thesis never justified the peak. Amazon fell ~92%, then compounded past it through marketplace, Prime, AWS, and ads.
Global Crossing — the wipeout
A global fibre network that never had a profitable year, priced in the tens of billions.
The fibre was real and is still in the ground. The equity was destroyed; the network’s value was realised by whoever bought it after the crash.
The overbuild — built together
Why infrastructure booms overshoot: rational actors all reach the same conclusion at once.
Three railroads across one prairie; five fibre networks down one conduit. When everyone builds at once, the collective result is a glut.
Sources · RCA: Stanford UP · AOL & Amazon: contemporaneous reporting · Global Crossing & the telecom buildout: bankruptcy filings, WSJ via Stratechery (Moffett).
| Cycle | Example | Peak paid | Bust value | Later value | Hindsight verdict |
|---|---|---|---|---|---|
| Railway Mania1840s | UK railway sector | Index 2,062 Oct 1845 | Index 741 Apr 1850 | Rail network became essential infrastructure | Technology thesis right; many routes mispriced. |
| Radio Boom1920s | RCA pre-split | $568 Sep 1929 | $15 1932 | Recovered 1960s; GE bought RCA ~$6.3B (1986) | Valuable via radio → TV → electronics. |
| Radio Boom1920s | RCA split-adj.−98% | $114.75 Sep 1929 | $2.63 1932 | Same path, after the 5-for-1 split | A ~98% drawdown on a split-adjusted basis. |
| Dot-com1990s–2000s | AOL | ~$222B 1999 | Combined ~$28B by 2009 | Sold to Verizon for $4.4B (2015) | Walled-garden/dial-up thesis never justified peak. |
| Dot-com1990s–2000s | Amazon | >$30B late 1990s | Fell ~92% | ~$2.5T (Jun 2026) | Peak later justified — by an evolved model. |
| Telecom & Fibre2000 | Cisco | ~$555B Mar 2000 | Fell ~90% | ~$470B (Jun 2026) — still below its 2000 peak | Survived; ~two decades + a model shift. |
| Telecom & Fibre2000 | Nortel | ~C$398B Sep 2000 | <C$5B 2002 | Bankruptcy; patents sold ~$4.5B | Tech had value; the company did not survive. |
| Telecom & Fibre2000 | WorldCom | ~$186B | $107B assets (bankruptcy) | MCI → Verizon ~$7.6–8.5B | Network survived; the equity thesis did not. |
| Telecom & Fibre2000 | Global Crossing | ~$49B Feb 2000 | $22.4B assets (bankruptcy) | Sold to Level 3 ~$3B | Fibre real; valuation failed; assets reused. |
Sources · as cited above; full citations in the source trail. Market caps as of June 2026.
Railway ManiaBritain · 1840s
But the doubters were not saying railways were useless. They were saying that not every proposed route, at every price, financed this way, could earn railway-like returns.
The boom heard no difference between “railways are transformative” and “every railway deserves capital,” and priced accordingly.
An index of all railway companies rose 106%, from 1,000 in January 1843 to 2,062 by October 1845, before falling to 741 by April 1850 — a 64% peak-to-trough collapse.
The physical excess matched the financial one. At the 1846 zenith Parliament passed roughly 263 railway Acts authorising about 9,500 miles of route; close to a third were never built, abandoned or absorbed by stronger firms.
The emblem was George Hudson, the “Railway King,” who fused promotion, finance, and public standing — and who, it later emerged, paid dividends out of capital rather than profit. This was not only naïve money: MPs and the middle classes were deep in it, which is the point. It was socially validated money, which is exactly what makes a cloth hard to call invisible.
In hindsight the correct value was not zero — it was lower, slower, and more consolidated. The network became indispensable, but many individual schemes were worth little alone and gained value only once folded into larger systems.
The technology thesis was met in full; the investment thesis was met late and partially. Average dividends on railway ordinary shares fell to about 1.83% by 1850, against the 10% promoters had dangled. The value arrived at the level of the system — not for the investors who funded the mania at the top.
The technology did not disappear.
The promises did.
Railways went on to become essential infrastructure. The railway age was real. But many investors had paid prices that assumed too much, too quickly, for too many projects.
The winners were not simply those who built anything on rails. They were the routes, operators, and networks that survived consolidation, proved demand, controlled costs, and became part of a durable system.
The railway future arrived.
But not every railway company deserved to survive long enough to enjoy it.
The pattern: the future can be entirely real while most of the routes to it are still mispriced.
RCA & the Radio BoomUnited States · 1920s
It was so synonymous with the boom that it traded under the nickname “Radio,” and, as chroniclers of the decade note, simply putting “radio” in a company’s name could lift its stock.
That is the purest possible tell of an invisible cloth: the word, not the cash flow, moved the price.
The promise was enormous: RCA would sit at the centre of the radio age. It would sell the devices, control the patents, build the networks, capture the advertising, and own the attention of a newly connected public.
But again, the cloth was not the technology.
The cloth was the belief that RCA’s stock already deserved the entire future of broadcasting.
That reality made doubt expensive. The market treated the obvious importance of radio as automatic justification for RCA’s price, so to question the valuation was to look as though you were questioning the invention.
The shares ran from about $43 in 1926 to a peak near $568 in September 1929, at a price/earnings ratio around 72, then fell to roughly $15 by 1932 — a collapse of about 98%. On a split-adjusted basis, RCA split five-for-one in March 1929, the same path reads $114.75 to $2.63: the identical fall, simply re-based.
The sceptics, tellingly, were the “old-timers” who dismissed RCA precisely because it paid little or nothing in dividends, ploughing cash into research instead — the people saying I cannot see the cash flows, who looked as if they could not see the future.
The decisive detail is the recovery. RCA’s dividend-adjusted price did not regain its 1929 level until the 1960s — and by then most of its money came from television, not the radio business anyone had bought. The era’s financial distortion had its Hudson, too: congressional investigators later found that RCA’s own David Sarnoff had taken part in pool operations that manipulated the shares.
GE eventually bought the whole company for about $6.3 billion in 1986, largely for NBC, and dismantled much of the rest.
The pattern: the company can survive, the technology can be real, and the peak price can still be wrong for a generation — and the value, when it returns, may come from a business the original investors were not buying.
The Dot-Com BubbleUnited States · 1990s–2000s
The Nasdaq rose 86% in 1999 alone, peaked at 5,048 on 10 March 2000, and fell to 1,139.90 by October 2002 — a 77% decline that erased some $5 trillion and did not recover its high until April 2015, fifteen years later.
There is a quieter tell beneath the index: between September 1999 and July 2000, dot-com insiders cashed out roughly $43 billion in stock, and in the month before the peak they sold about 23 times as many shares as they bought.
The people inside the suit could feel it thinning.
The market did not discover that the internet was useless.
It discovered that many internet companies had no independent energy capture.
They had users, stories, and capital. But they did not have durable margins, operating discipline, or business models that could survive once the energy stopped flowing.
Two firms make the fork concrete.
AOL was the narrative at its most triumphant: a market value near $222 billion in 1999, used as currency to acquire Time Warner in what was then the largest merger ever. The promise was a vertically integrated new-media empire — distribution plus content plus convergence.
It did not arrive.
The combined company was worth about $28 billion by 2009; AOL itself was later sold to Verizon for $4.4 billion in 2015, and Verizon subsequently wrote its media unit down by some $4.6 billion. The original walled-garden, dial-up thesis never justified the peak, because the open web and broadband selected against it.
Amazon is the counterexample that proves the rule. It went public in 1997 worth about $438 million, passed $30 billion by the end of the decade, then fell roughly 92% — from about $106 to $8 a share — in the bust. It took on the order of a decade to regain that lost value, and is worth around $2.5 trillion as of June 2026.
But Amazon did not vindicate the 1997 thesis; it outgrew it.
The bookstore became an everything-store, then a marketplace, then a logistics network, then Prime, then — decisively — AWS and an advertising engine. The market’s broad belief that e-commerce mattered was right; the realised value came from business-model evolution, not the original model.
The contemporaries who looked foolish for doubting — Warren Buffett, warning in 1999 that transformative industries can change the world without rewarding investors; Julian Robertson, whose Tiger Management refused to chase the boom and shut down in March 2000, almost exactly at the top — were vindicated within eighteen months.
The pattern: the bubble was not wrong about the internet. It was wrong about which models could convert it into durable profit, about who the winners would be, and about how long the conversion would take.
Telecom & the Fibre BuildoutUnited States · 1990s–2000s
The Emperor’s New Clothes effect was, if anything, harder to resist here, because the story looked physical; you could point to cables and rights-of-way.
The awkward question — yes, but does it need this much, this fast, financed this way? — became socially expensive to ask.
In the five years after the 1996 Telecommunications Act, carriers poured more than $500 billion, mostly debt-financed, into the ground; capacity grew on the order of 186,000-fold over roughly seven years, and as little as ~2.7% of installed fibre was actually “lit” in 2002.
Vendor financing — equipment makers lending customers the money to buy their gear — meant the supplier’s revenue and the customer’s debt inflated together, a circularity worth holding in mind for any modern echo.
The structural cause is the one that ties this case to all the others, and back to the bloom: when every rational actor reaches the same conclusion at the same moment, the collective result is overbuilding.
Three railroads across one prairie; five fibre networks down one conduit.
Each decision is individually sound; together they are a glut.
The companies caught in it were not dot-com fantasies but serious operators, and they split into the same three fates as every cycle.
Survive and reprice slowly. Cisco sold the routers that ran the internet and briefly became the world’s most valuable company, near $555 billion in March 2000, on the promise that it would be the permanent tollbooth of network infrastructure. It fell roughly 90% by 2002. It survived — real customers, real cash flow, a central position — but its valuation assumed a level of hardware growth and margin durability that did not arrive.
Tellingly, even now, at around $470 billion (as of June 2026), Cisco’s market value remains below its 2000 peak, twenty-six years on, and its recovery came as it shifted toward software, security, and subscriptions rather than from the original router thesis.
Vanish as equity; live on as assets. Nortel, the optical-networking champion, reached about C$398 billion in September 2000 — more than a third of the entire Toronto exchange — and fell below C$5 billion by 2002 before bankruptcy; its patents were later sold for about $4.5 billion to a consortium of Apple, Microsoft, and others.
WorldCom, built by acquisition to a peak near $186 billion, collapsed into what was then the largest bankruptcy in US history — about $107 billion in assets — after accounting fraud was uncovered; the network survived, became MCI, and was absorbed by Verizon for around $7.6–8.5 billion.
Global Crossing built a global fibre network spanning over 100,000 miles and 27 countries, reached roughly $49 billion on paper, and never had a profitable year; it filed for bankruptcy in January 2002 with about $22.4 billion in assets against $12.4 billion in debt, its founder having sold some $734 million of stock on the way up, and was eventually bought by Level 3 for about $3 billion.
In each case the technology had value and the equity thesis did not.
This is Part 1’s quietest word made literal: the firms became shells — prey whose assets fed a larger organism.
The dark fibre those companies buried went on to carry broadband, streaming, cloud, and now AI. The bandwidth-demand thesis was met in full. The original business thesis, for most of the builders, was not — and the value was realised by whoever bought the network after the crash.
The pattern: the purest infrastructure version of the fable — the highways were real, but the traffic was not yet enough to pay for the costume, and the value went to whoever owned the network after the price collapsed.
The question was never whether the technology was real.
A genuinely transformative technology, correctly understood.
A promise stacked on top of it that could not be verified, where doubt carried a social cost, so belief propagated downward from the most credible actors and the flow of capital became its own proof.
A bloom, in which capital floods in and rational actors overbuild together.
A bust, in which the firms living on capital rather than profit starve first while the well-capitalised survive and absorb the wreckage.
And value realised at the level of the system, after consolidation — often not by the investors who funded the boom at the top.
In none of them was the question whether the technology was real.
It was whether the capital poured in could be repaid by the profit the environment could actually sustain.
When it could not, selection did what selection always does.
We cannot yet know which half of the curve we are standing on.
So where is AI on this curve?
That is the real question, and the honest answer is that we cannot yet know which half of the curve we are standing on.
What we can say is that several classic conditions for overshoot are present: capital flowing on narrative ahead of revenue, a population of firms that has bloomed all at once, a handful of names absorbing a wildly disproportionate share of the energy, and capital that increasingly appears to move in circles — flowing out to customers who use it to buy from the very firms that funded them, a vendor-financing circularity similar to patterns seen in the telecom buildout before its crash.
None of that proves a bust.
It is possible the environment’s real carrying capacity — genuine productivity, genuine demand — is expanding fast enough to justify the bloom. AI may create new categories of work, new margins, new markets, new scientific breakthroughs, and new forms of economic output.
That is the bull case, and it is not foolish.
But the four cases above settle the question that is not the one to ask.
Railways were real.
Radio was real.
The internet was real.
Fibre was real.
It was never whether the technology was real; it was whether the capital poured into it could be repaid by the profit the environment could actually sustain.
When it cannot, selection does what selection always does.
And when it does, it will not ask which AI was the best, the safest, or the most useful. It will ask only which organisms could keep feeding themselves when the energy stopped flowing.
Markets are designed environments. That is the lever.
If selection has no morality, and we do not like what it is selecting for, can we change the environment itself?
Because if capital markets are the environment, then the rules of the environment matter.
The cost of capital matters.
The structure of incentives matters.
Disclosure matters.
Accounting matters.
Governance matters.
Competition matters.
Regulation matters.
Liability matters.
Energy constraints matter.
Public procurement matters.
Safety standards matter.
The metrics investors reward matter.
The time horizons boards tolerate matter.
If the environment rewards speed above durability, organisms will become fast before they become durable.
If the environment rewards scale above proof, organisms will scale before they prove.
If the environment rewards narrative above energy capture, organisms will learn to narrate before they learn to feed themselves.
Natural selection has no morality.
But markets are not nature in the purest sense.
They are designed environments.
And if we do not like what the environment selects for, the question is not whether we can stop evolution.
The question is whether we can change the selection pressure.
That is the lever.
And it is where this story goes next.
- Cost of capital
- Disclosure
- Accounting
- Governance
- Competition
- Regulation
- Liability
- Safety standards
- Energy constraints
- Investor time horizons
Source trail
Open source notes from the article draft
The St. Matthew Island reindeer case follows David R. Klein’s 1968 study, The Introduction, Increase, and Crash of Reindeer on St. Matthew Island. The broad figures used in this essay are rounded: 29 reindeer introduced in 1944, a peak of roughly 6,000 by 1963, and a collapse to a few dozen animals by 1966. Later discussion of the collapse also points to severe winter conditions interacting with overgrazing and depleted forage.
Sources:
David R. Klein, The Introduction, Increase, and Crash of Reindeer on St. Matthew Island, Journal of Wildlife Management, 1968. https://www.over-reeen.nl/Portals/0/artikelen/populatiebeheer_ree/engels/the_introduction_increase_and_cr ash_of_reindeer_on_st_mathews_island.pdf USGS / Alaska Science Forum, St. Matthew Island reindeer materials. https://www.usgs.gov/media/images/rise-and-fall-st-matthew-reindeer https://www.gi.alaska.edu/alaska-science-forum/when-reindeer-paradise-turned-purgatory-0 Wired, “Mysterious Collapse of Reindeer Herd Blamed on Freak Storms,” 2009. https://www.wired.com/2009/12/mysterious-disappearance-of-reindeer-herd-blamed-on-freak-storms/ The Emperor’s New Clothes section draws on Hans Christian Andersen’s fable and the critical-thinking concept of “poisoning the well.” Hans Christian Andersen, The Emperor’s New Clothes. https://www.gutenberg.org/files/1597/1597-h/1597-h.htm Nicholas Capaldi and Miles Smit, The Art of Deception: An Introduction to Critical Thinking. https://philpapers.org/rec/CAPTAO-2 The Railway Mania discussion draws primarily on work by Gareth Campbell and John D. Turner on railway share prices, investor behaviour, and media dynamics during the 1840s railway boom.
Gareth Campbell and John D. Turner, “The Railway Mania: Not So Great Expectations?”, VoxEU / CEPR, 2009. https://cepr.org/voxeu/columns/railway-mania-not-so-great-expectations Gareth Campbell, “Stock Prices During the British Railway Mania,” 2010. https://ideas.repec.org/p/pra/mprapa/21820.html Gareth Campbell and John D. Turner, “Dispelling the Myth of the Naive Investor During the British Railway Mania, 1845–46.” https://pureadmin.qub.ac.uk/ws/files/13426282/RailwayInvestors.pdf The RCA / Radio Boom discussion draws on historical accounts of RCA’s share-price rise, collapse, recovery period, and later value migration toward television and NBC.
Stanford University Press, Bubbles and Crashes, Chapter 1 excerpt. https://www.sup.org/books/economics-and-finance/bubbles-and-crashes/excerpt/chapter-1-excerpt Bridgeway, “An All-Time High.” https://bridgeway.com/perspectives/an-all-time-high/ The dot-com section draws on historical accounts of the Nasdaq peak, the 2000–2002 collapse, the delayed recovery, and examples including AOL/Time Warner and Amazon. Figures are rounded and should be treated as historical approximations rather than real-time market data.
Wired, “March 10, 2000: Pop!”, 2007. https://www.wired.com/2007/03/march-10-2000-pop Berkshire Hathaway / Fortune, Warren Buffett on the stock market, 1999. https://www.berkshirehathaway.com/1999ar/FortuneMagazine.pdf The telecom and fibre-buildout section draws on accounts of the telecommunications bubble, debt-financed fibre buildout, vendor financing, and the collapse or consolidation of firms such as WorldCom, Nortel, Global Crossing, and Cisco’s post-2000 repricing.
Paul Starr, “The Great Telecom Implosion,” Princeton University, 2002. https://www.princeton.edu/~starr/articles/articles02/Starr-TelecomImplosion-9-02.htm Los Angeles Times, “Global Crossing Files for Chapter 11,” 2002. https://www.latimes.com/archives/la-xpm-2002-jan-29-mn-25263-story.html Washington Post, “Global Crossing Files for Bankruptcy,” 2002. https://www.washingtonpost.com/archive/business/2002/01/29/global-crossing-files-for-bankruptcy/fffcad1b- f522-479a-a9d4-659e74165b05/ Fabricated Knowledge, “Lessons from History: The Rise and Fall of the Telecom Bubble,” 2023. https://www.fabricatedknowledge.com/p/lessons-from-history-the-rise-and
