Let me put the concession first, because the rest of this piece depends on it being sincere. Artificial intelligence is a real technological revolution — very possibly the most consequential since the internet, and quite possibly larger. Nothing that follows disputes it. But the excitement runs that fact together with a second one, and prising the two apart is the beginning of financial wisdom. “Is this a transformative technology?” and “Is this, at this price, a good investment?” are not the same question. They are barely related. The internet was every bit as revolutionary as its champions claimed in 1999 — and an investor who acted on that entirely correct belief by buying the market at its peak was, for the better part of two decades, ruined by it. Value created by a technology is one thing; value captured by the companies building it is a second; and a return to the investors who financed them, at the price they paid, is a third. The three are routinely, and expensively, confused.
The confusion is not new. It is among the most reliable patterns in the whole history of capitalism — and it is worth walking through, because the pattern is the argument.
The track and the shares
Begin in Britain in the 1840s, with a technology as revolutionary in its day as anything now. Railways were about to annihilate distance, remake commerce and stitch a nation of villages into an industrial economy — and the investing public understood this, and duly lost its head. Parliament authorised some nine and a half thousand miles of new line in the single year of 1846; railway shares roughly doubled; a Yorkshire draper named George Hudson, the self-styled “Railway King”, assembled a third of the nation’s track and a seat in Parliament besides. Then, as interest rates rose and the promised traffic failed to appear, the mania broke. Bankruptcies reached a record; railway shares lost well over half their value by 1850, and many of them far more; Hudson was exposed paying dividends out of capital in the manner of a Ponzi scheme, disgraced, and died in obscurity worth almost nothing. Thousands of ordinary investors were ruined.
And yet the track remained. The lines were built, and they went on to carry the whole Industrial Revolution, exactly as the enthusiasts had promised. This is the shape of the thing, stated once so that we recognise it thereafter: the technology was real, the infrastructure was essential and endured, and the capital that financed it was very largely destroyed. The believers were right about the railways and ruined by the shares.
The most valuable company on Earth
Now to the version within living memory — and to the objection already forming in a sophisticated reader’s mind. The Nasdaq Composite reached 5,048 in March 2000 and then fell, over the following thirty-one months, to around 1,114: a decline of seventy-eight per cent, from which the index needed fifteen years merely to climb back to where it started. That much is familiar. But the standard consolation — “that was different; those were profitless fantasies, Pets.com and its kind, whereas today’s leaders make real money” — is precisely the reassurance that ought to induce vertigo, because it was equally available, and equally true, in 2000.
Consider Cisco Systems. It made the routers and switches through which the internet physically ran — the “picks and shovels” of the boom, the sober way to own the revolution without gambling on any single dot-com. It was enormously, unambiguously profitable. And in March 2000 it became the most valuable company on the planet, worth some 550 billion dollars, with analysts seriously forecasting the world’s first trillion-dollar valuation. It was also, at that price, trading at around two hundred times its earnings — a multiple that silently required it to grow into a sizeable fraction of the American economy. It did not. The stock fell by almost ninety per cent — and then, the part worth sitting with, it took twenty-five years to recover. Cisco surpassed its March 2000 share price for the very first time only last month, in December 2025; by market value it remains well below its peak even now. The company was never the problem. It grew its revenues severalfold, stayed profitable throughout, and is a pillar of the industry to this day. An investor who saw all of that clearly, who was entirely right about the company, and who bought it at the top, waited a quarter of a century simply to get their money back in nominal terms — and in real terms is behind still. The profitability of the leaders is not, and has never been, a defence against the destruction of the capital that overpays for them. The question is never whether these are good companies. It is whether this is a good price — and whether today’s obvious champion is tomorrow’s.
The over-built road
One mechanism recurs, and it bears directly on where the money is flowing now. Each of these revolutions over-builds its infrastructure, then writes much of it off. The 1840s laid redundant, competing lines no traffic could fill. The dot-com boom laid a vast surplus of fibre-optic cable across the world — so much that most of it sat unlit for years, “dark fibre” bought up later for cents on the dollar by firms that had never borne the cost of laying it. The infrastructure was needed, eventually; the capital that rushed to build it ahead of the demand was incinerated in the waiting. Today the infrastructure is data centres, graphics processors and the electricity to run them — the enormous capital expenditure an earlier piece in this series described. That this build-out is real, and much of it genuinely necessary, is not in question. Whether the hundreds of billions now pouring into it will earn an adequate return at the prices being paid is an entirely separate matter — and the historical record on that separate matter is not encouraging.
Backing the wrong horse
Even the investor who is right about everything — right that the technology is real, right to want to own its infrastructure — faces one last hazard, and it is the cruelest. The eventual winners are rarely the ones the market crowns at the outset. Return to 2000 and ask which technology company was the safe, obvious giant and which the fading also-ran. Cisco was the titan. Apple was very nearly a punchline — a niche computer-maker that had come within weeks of bankruptcy in 1997 and been kept alive, humiliatingly, by an emergency investment from Microsoft. A quarter of a century on, Apple is worth roughly ten times Cisco. The market’s confident selection of the winner was not merely wrong; it was inverted. The dominant AI names of 2026 — and the reader can supply the list as readily as I can — are not guaranteed to be the dominant names of 2036; and the “safe”, picks-and-shovels arms-dealer among them is not safe at any price.
A confession about timing
I must be careful here, because all of the above could be misread as a market call, and it is emphatically not one. Let me inoculate against that with a confession. In early 2006, on the floor of an investment bank, I argued to anyone who would listen — and to several who would rather not have — that credit markets had come dangerously unmoored, that the pricing of risk had parted company with reality, and that a serious reckoning was near. The trading desk found this hilarious. And they were right to, for a while: the reckoning I was so certain of duly arrived a full two years later, in 2008, by which time my early warning had long since curdled into a standing joke. I was entirely right about the substance and comprehensively, expensively wrong about the timing — and that is the whole of the lesson. A market can be plainly overpriced and go on rising for years; that a thing is unsustainable tells you almost nothing about when it will end. Sir John Templeton — who coined the phrase this piece has taken for its title, and who died, as it happens, in that very crash year of 2008 — grasped that the peril was never in doubting the euphoria, but in presuming to time it.
So this is not a prediction that AI shares will fall by such a figure on such a date. That would be a fool’s errand — and, as another piece in this series argued, the future is opinion, not knowledge. It is instead a structural claim, and a far more modest one: that the business cycle has not, this time, been repealed; that overpaying has never once been rescued by the truth of the underlying story; and that “this time is different” retains its long and unbroken record of being wrong.
The rhyme
The pieces assemble into something quite simple. Artificial intelligence is real, as the railways were real and the internet was real. A great deal of value will be created by it. Some companies will capture that value brilliantly and endure — and, if history is any guide, they will not all be the ones the market currently adores; a few do not yet exist. Many others, financed at valuations that assume the impossible, will go bankrupt. An immense infrastructure will be built, will prove indispensable, and will ruin a large share of the capital that raced to construct it. None of this is an argument against the technology. It is an argument against the seductive belief that faith in a technology is the same as being owed a return for owning it at any price — the belief that animates Templeton’s four most dangerous words, the words that rang true in 1845 and 1929 and 2000 and are being intoned, with perfect confidence, again. These are the rhythms of markets, treated at greater length in a standing programme elsewhere on this site; the technology changes, the human beings pricing it do not. History, as the saying goes, does not repeat itself — but it rhymes. And on the single question that matters here — whether the cycle has at last been broken, whether prosperity has finally been severed from price — the answer is the one it has always been. This time is not different. It never is.