The AI Bubble and the Railroad

Every few decades, the world convinces itself that a new technology will change everything. And the curious part is that the world is usually right. The railways changed everything. The internet changed everything. The aeroplane shrank the planet. Yet the investors who financed these revolutions were, more often than not, left holding the bruises. Technology adoption and investment returns are two different journeys, and they rarely arrive at the same station.

The conversation around Artificial Intelligence has now reached this inflection point. Talk to friends and the word bubble surfaces within minutes, sometimes with worry, sometimes with the quiet satisfaction of having seen through the hype. But what does the bursting of a bubble actually mean? Does it mean the technology failed? Does it mean we were wrong about its promise? History suggests it means neither.

A pattern older than the stock ticker

Britain in the 1840s was gripped by railway mania. Parliament approved thousands of miles of new track, ordinary citizens mortgaged their savings to buy railway shares, and for a brief period the railways were the future itself. Then the mania collapsed and fortunes evaporated. But the tracks remained. The trains kept running. Freight kept moving for a century on rails paid for by ruined men.

America repeated the experience in the 1870s and again in the 1890s. The internet repeated it in 2000, when trillions in market value vanished almost overnight. And yet the fibre optic cables laid during that frenzy did not vanish with the money. They sat in the ground, were bought later for pennies on the dollar, and went on to carry the video calls and the streaming we now take for granted.

The economist Carlota Perez has argued that this is not an aberration but a feature of technological revolutions. A frenzy attracts far more capital than rational returns could ever justify, most of that capital is sacrificed, and what remains becomes the foundation on which society builds for decades. The bubble, in a strange way, is how the future gets financed. The investors of the 1840s never got their money back. We got the railways.

Two questions wearing the same clothes

When someone declares that the AI bubble will burst, they are usually answering one question while believing they have answered another. Will AI transform how we work and learn? Will the companies building it reward their shareholders? These sound like the same question. They are not.

Consider Cisco, the company that sold the equipment everyone needed during the internet boom. The internet won, comprehensively. And yet a shareholder who bought Cisco at its peak in March 2000 waited nearly twenty-six years just to see that price again; the stock crossed its dot-com high only in December 2025, and adjusted for inflation it sits below that summit even today. The internet kept its promise. The investor merely got his money back. One could be entirely right about the revolution and entirely wrong about the returns. A burst, if it comes, will be a verdict on valuations, not on usefulness.

The hype that fades and the tool that stays

But it is fair to ask whether AI deserves even this defence. Every few years we are told that something will change the world. A few years ago it was blockchain, and the decentralised ledger was the talk of the town. Conferences were held, white papers were written, fortunes were raised. The technology remains, quietly useful in its corners, but the enthusiasm around it has taken a back seat. Before that it was the Internet of Things, and everything, we were promised, would soon be connected to everything else. That promise too retreated into the background.

AI stands apart from these in one verifiable way. It has touched millions. One does not need to attend a conference or read a white paper to encounter it. Anyone with an internet connection can ask a chatting interface a question and receive an answer, and a significant share of the connected world already does. The more sophisticated uses, the coding and the research, have not reached everyone yet. But the ordinary user is no longer a spectator to a promise. Blockchain asked us to imagine a use case. AI handed us one.

Where the losses would land

So the looming question is not whether the bubble bursts but where the capital destruction lands if it does. If the dot-com pattern holds, the application layer startups burning cash on undifferentiated products are the candidates for oblivion, while the infrastructure of chips, data centres and power survives in some form, perhaps under new owners and at a fraction of present prices. The rails remain even when the railway companies do not.

There is one structural difference worth noting in the present cycle. The railway manias were financed by leveraged speculators and overextended banks, and when the music stopped the losses cascaded through the financial system and dragged whole economies into gloom. The current buildout is financed largely by the cash flows of already profitable technology giants. If valuations collapse, the losses will land on shareholders rather than on banks and depositors. That does not make a burst painless. It may make it less dangerous, a forest fire rather than an earthquake.

Standing at the cross-roads

So when friends insist that the AI bubble is going to burst, perhaps the honest answer is that they may be right, and that it may not matter in the way they think. The dot-com crash did not make the internet less useful in 2001 than it was in 1999. It only made it cheaper to own. The frenzy and the transformation traverse the same road, intertwined, each feeding the other, and we rarely get to choose one without the other.

What bursts is the price. What remains is the railroad.