You can be right about the technology and still lose the money

The most expensive mistake available to a technology optimist is being right. Not wrong — right. Right about the technology, right about adoption, right about the world it produces, and broke anyway, because transformative and profitable are different claims, and history keeps grading them separately.

This matters now because the AI thesis has reached the stage where the technology argument is effectively over. Almost nobody serious disputes that these systems are useful and getting more so. Which means the argument that remains — the only one left with money riding on it — is about who captures the value. And on that question, the base rates are grim for the people writing the checks.

Four times the future arrived and the shareholders paid for it

Railways. In the 1840s, Britain convinced itself of something true: rail would remake the economy. Parliament authorized thousands of miles of track, promotion schemes multiplied, and households across the country subscribed capital. Mid-decade, the boom collapsed. Share prices fell for years, many companies never paid the returns their prospectuses implied, and a generation of investors was wiped out — while rail traffic itself kept growing for decades afterward. The thesis was correct. The shareholders funded a gift to everyone else. The American version repeated the lesson on a bigger continent: railroad overbuilding helped trigger the panics of 1873 and 1893, and after 1893 a strikingly large share of US rail mileage went into receivership.

Airlines. Commercial aviation compressed the planet, created global tourism, and rewired trade. It was also, for most of a century, a machine for destroying capital — an industry whose cumulative profits over the jet age were famously close to nothing, thanks to brutal competition, high fixed costs, and no durable moat. Warren Buffett's running joke was that a farsighted capitalist at Kitty Hawk would have done investors a favor by shooting Orville down. Every part of the technology thesis came true. The economics never followed.

Fiber. In the late 1990s, telecom companies borrowed enormous sums to lace continents with fiber-optic cable, on the correct theory that internet traffic would explode. Traffic exploded. The companies didn't survive to bill for it: capacity outran demand by years, prices collapsed, and the sector produced some of the largest bankruptcies in American history, WorldCom and Global Crossing among them. Much of that fiber sat dark, was bought out of bankruptcy for cents on the dollar, and quietly became the cheap backbone of the modern internet. The overbuild was a catastrophe for its funders and a subsidy for everyone who came after.

Dot-com. The 1999 pitch — everything moves online, commerce, media, communication — was simply right, and more right than the optimists knew. The NASDAQ still lost most of its value from peak to trough, and even the era's great survivor, Amazon, fell more than 90% before going on to justify every word of the original story. Anyone who held it through that drawdown earned the returns. Very few did, and the median 1999 internet investor owned the companies that didn't make it.

Why the pattern repeats

It isn't stupidity. Each of these booms attracted intelligent capital reasoning correctly about the technology. The mechanism is more uncomfortable than that: when a story is obviously true, everyone can see it, so everyone funds it. Capital floods in until expected returns are competed down to nothing — and then past nothing, because the marginal dollar is chasing the narrative rather than the arithmetic. Competition transfers the surplus to customers. Cheap rail freight, cheap flights, cheap bandwidth: the value was real, it just showed up in your cost of living instead of your brokerage account.

Two things decide who escapes the pattern. One is a moat — some structural reason competition can't reach the margin. The other is not needing to be early: the durable money in most of these episodes was made after the crash, buying the surviving assets from the people who funded the buildout.

What this implies now

Notice what this argument is not. It is not "AI is a bubble," a claim that requires knowing the future. It's a narrower and better-supported claim: the probability that today's AI capital expenditure earns its cost of capital is a separate question from whether the technology works, and history says the two questions come apart more often than not.

For an ordinary investor the practical conclusion is annoyingly boring, which is usually the tell that it's correct. If you own a broad index, you already own the AI buildout at market weight — you don't need to add a concentrated bet on the most widely believed story in the world to participate in it. Expected returns come from the prices you pay, not from the vividness of the narrative, and the narrative has never been more vivid or the participation more crowded. Being right about the technology is free. Everyone gets to be right about the technology. The returns were never going to be distributed for that.