When the Warning Cannot Save You
Every transformative bubble runs the same script, and the leader who survives it is the one who wrote his code before the cost arrived.
On September 5, 1929, Roger Babson stood before a business conference and said the thing nobody wanted to hear. Sooner or later a crash is coming, he told them, and it may be terrific. The market dipped about three percent that afternoon. They called it the Babson Break, and then they called him a crank. The Chicago Tribune ran the rebuttals. Economists dismissed him, and a few questioned his patriotism for talking down American prosperity. The market treated his warning as a healthy correction and went back to buying.
He was right. He was also early, and in a bull market early and wrong look exactly the same. For the better part of two months he was a punchline. Then the tide went out.
Seventy years later, Jeremy Grantham’s firm called the dot-com top. GMO was correct about 2000, and also two and a quarter years early, and in the interval it lost close to half its book as clients fled to the managers still riding the thing up. The names change. The script does not. The prophet is always credible, always early, and always punished, and the punishment looks identical to being wrong until the tide finally goes out.
Grantham is saying now that this AI market is the biggest bubble in American history, and that a seventy percent fall in the high-flyers would not surprise him. He may be early again. That is rather the point. We are living inside another version of the same story, and the only useful question is which part you intend to play in it.
Why this matters before the cost arrives
The script repeats with enough fidelity that you can cast the roles in advance, which means you can see your own role before the moment forces you into it.
Andrew Ross Sorkin’s account of the crash makes a claim that sounds financial and turns out to be about character. The crash was set in motion by specific men who rationalized their own positions while defending a structure of leverage they could not imagine failing. Confidence is the architecture of an economy, and confidence goes the way Hemingway said bankruptcy does. Gradually, then suddenly. The lesson moves from regulation to human nature. People forget. They dress hope up as certainty. They fall hardest from the height of their own conviction.
A regulatory problem you could solve with a rule. A human problem outruns the rule. The only thing that holds is a standard you chose before the pressure made the choice expensive.
We are deep inside a Fourth Turning, the stretch where the institutional debts of an eighty-year cycle come due at once. The cheap money that defined the last era is unwinding, and the technological optimism is real and also weaponized. In a season like this, complexity fails and only character scales. The question is not whether you can call the top. You cannot, and neither can anyone selling you a model that says it can. The real question is what code you have already written for the day the prophet is mocked, the optimist is believed, and you have to decide which of them you are funding.
The optimists are usually right
Here is the part most commentary gets wrong, because getting it right means conceding something painful.
The articulate optimist in these episodes is generally correct about the technology. In 1929, Irving Fisher described a coming age of prosperity built on mass production and a wave of invention the world had never seen. Fisher was telling the truth. Mass production was transformative. The American century he saw coming arrived. He was wrong about exactly one thing, and that thing was the price.
Read Fisher beside the 2026 case for AI and you cannot tell them apart. Mohamed El-Erian calls AI a genuinely transformative innovation with winner-take-all tendencies, and he is right. Didier Sornette lists the justifications that powered the dot-com mania: capital-light models, network effects, first-to-scale advantage, the real-option premium on futures nobody can yet see. Every one maps onto the foundation-model bull case. The optimists hold a coherent theory that is partly true. That is what makes a bubble seductive rather than stupid.
So the three roles are set. The prophet is punished for being early. The optimist is right about everything except valuation. The system metabolizes the warning instead of acting on it. The information is never missing; the structure is built to discount whoever carries it.
Keep one sentence from this essay: the optimists are usually right about the technology and wrong only about the price. It guards against the two errors that wreck leaders here, throwing out the technology because the valuation is absurd, and swallowing the valuation because the technology is real. A bubble is a true story told at the wrong number.
What actually ends these episodes
There is a comforting fiction that bubbles burst when the technology disappoints. It lets you watch the product roadmap and feel safe. It is also wrong.
The unifying cause is monetary. Speculative phases are most often stopped by successive increases in the cost of money. In 1929 the discount rate climbed from 3.5 percent to 6 percent; in Japan around 1990 it moved from 2.5 percent to 6 percent; the dot-coms broke the same way. The technology held. The rate rose, and the future got discounted back to a price the present could no longer pretend to afford. This is what Edward Chancellor calls the price of time. WeWork’s fall from a forty-seven-billion-dollar valuation to a fraction of it was not creative destruction but the plain kind: money mispriced long enough that a real-estate company could be sold as software.
So do not wait for the AI products to underwhelm; they may not. Watch the cost of money, the one mechanism that has ended every comparable episode. Tie your judgment to the rate path, not to the demo.
The honest counterweight, or this is just propaganda: overpaying for real growth can still work over a long enough horizon. Amazon rewarded the people who held it through valuations that looked insane. Being correct and being proved correct right away are different things, and the tension does not resolve cleanly. Distrust anyone who says it does.
Where the 1929 rhyme breaks down
In 1929 the concentrated, levered vehicle was the investment trust. Trusts owned other trusts, which owned the market, and when the cascade started the leverage ran in reverse and the structure ate itself. The temptation is to call the seven companies driving today’s index the same thing.
The logic rhymes. The circular nature of current AI revenue, where the same capital flows between the model labs and the chipmakers and the clouds and back, looks like a closed loop pretending to be a market. But the 1929 trusts ran on explicit margin leverage; today’s giants run on fortress balance sheets, and their spend is borrowed against capex, not margin. The bull’s strongest defense lives inside the Amazon playbook: take a long enough view and a company that grows fat and inefficient dies a Darwinian death, which makes the enormous AI capex read as vision rather than recklessness.
So two disciplined companies can read the same war and write opposite codes. The hyperscalers pour cash into the buildout, betting scale is the only moat that will matter. Apple, so far, has declined the arms race, content to license what it needs and let everyone else fund the experiment. Both are defensible, and the line has to be drawn before anyone can know who was right.
I have drawn that line myself, at a smaller scale and with far less margin for error. At B:Side Capital, the small-business lending company where I serve as CEO, our balance sheet is not a casino chip and the people we serve cannot absorb our mistakes. When we built MARCUS, our internal AI system, the loudest version of the project was the expensive one: buy the enterprise platform, sign the long contract, take the vendor’s maximalist vision on faith. We chose the disciplined build instead, a small team of ASU students and recent graduates building a focused system tied to the actual work of lending. The line was simple. We would let AI do real work for us, and we would not bet the institution on a future nobody could yet price. I made that call on purpose, in the calm, and I could still defend it in the cold light of a downturn. That is the part that matters.
The question is who knew what they were betting before the bet became irreversible.
The warning cannot save you
This is the most unsettling part, and the most valuable.
The warning, once widespread, gets absorbed. Babson warns, the market dips, and then it folds the warning into the price and goes back to buying. The system holds all the information it needs and is built to neutralize whoever carries it. There is no cynicism in saying so; it is just structure. No help is coming, because the architecture metabolizes warnings the way a body metabolizes a stimulant: a brief spike, then baseline, then tolerance. You cannot outsource your judgment to the prophet, because his rightness will not arrive on a schedule you can use.
The career mechanism is brutal. You get fired for underperforming your peers in a bull market, not for being early on a bear, because by the time the bear arrives everyone is losing together. Howard Marks names the instinct it breeds: never be wrong on your own. The system punishes independent early correctness and rewards consensus late error, and the prophet who acts on the data takes on what looks, in real time, like pure downside.
This is where a code stops being philosophy and becomes operational. John Kenneth Galbraith left three tools. The twenty-year rule: ask whether the thing sold as unprecedented has any memory older than the last cycle. The intelligence test: separate the appearance of brilliance from the mere possession of money, because in a boom the two get confused and the confusion is the engine. And the this-time-is-different test, run on every claim that the old rules have been repealed, because the old rules are never repealed, only forgotten.
A leader who has internalized those has drawn a bright line before the pressure arrives, and when the moment comes the decision is execution, not deliberation. The one who has not will build a line in the moment, in a room full of people telling him the future is here, and a line built during the crisis is built too late. The warning will not save you. Only the standard you wrote down before the warning became fashionable will.
The fork that is darker than the crash
There is one more turn, the one Grantham himself flinches from. He has said he almost hopes AI fails, because the success case produces a very dangerous world. Picture it: labor’s share of output slides, and the economy generates ghost output, numbers that show up in the accounts but never circulate, because machines produce and machines do not consume. The bubble bursting is the smaller danger. The larger danger is the bubble being justified, because a justified bubble is the world where nobody ever installed the brake.
So the audit you run on your own institution has to ask not only whether the AI spend pays off, but what it sets as precedent for whoever comes after you. An organization that displaces its own people for a margin it never decided was right has made a choice about transmission, and the transmission is what outlives the quarter. This is the work George Marshall understood: building standards meant to hold long after the builder is gone. Nobody gets a bonus for the brake they installed on a machine that had not yet run away. The leaders worth following ask what their decisions teach, in a market that pays only for what they earn.
The code the moment requires
Write the line before the offer. Decide now, in the calm, what valuation, what concentration, what borrowed capex you will not fund. A line you can name in the calm is a bright line; a line you find under pressure is a rationalization wearing a line’s clothes.
Separate the technology question from the price question, and refuse to let conviction about the first contaminate judgment about the second. The technology is real and the price may be insane, both at once.
Watch the cost of money, not the demo. That is the mechanism that ends these episodes, and it will keep you sober when the launches dazzle.
Run the midnight test on the decision that feels fine. Ask what it looks like if you are wrong about your own motivation, because the most dangerous decisions are the ones that feel like governance.
Decide what you are transmitting, not just what you are earning. The quarter will pass. The standard you set for the people on the other side of your optimization will not.
None of this requires you to predict the crash. It requires you to be someone whose decisions still hold up after the prophet is vindicated and the optimist is humbled and everyone discovers, gradually and then suddenly, that the confidence was the whole structure.
The crash of 1929 eventually produced reform: Pecora, Glass-Steagall, a system that remembered for a generation. But it came slowly, and it held only as long as the memory did. We may not get even that this time, because the incentives that built the bubble are the same ones that will write the after-action report. So the reform has to be personal before it can be institutional. Write your line now, while the cost of holding it is still abstract. The prophet will be punished and the optimist will be believed and the system will metabolize every warning you might hope to get. None of that decides who you are when the tide goes out. You do, and you do it now, in the room where nobody is yet asking you to choose.
The work this essay describes, drawing your line before the pressure arrives, is a practice rather than a one-time decision. I built a new leadership app to keep that practice close at hand: a companion to Honor Under Pressure and The B:Side Way that brings the frameworks and the discipline into the moments you actually face them. Have a look at www.thefourthturningleader.com.