Track, Wire, Glass: What 180 Years of Overbuild Booms Say About the AI Build-Out

The most dangerous four words in finance are not "this time is different." They are "but the technology's real." Because it usually is. Railways were real. The telegraph was real. The internet was real. Every one of those revolutions delivered on its promise — and every one of them also produced a financing boom that destroyed the fortunes of the people who funded it. The technology and the bust are not alternatives. They are two things that happened at once, to the same asset, on different clocks.

1. The pattern, in three acts

Strip the specifics from any capital-overbuild boom and the same play runs, in the same order.

Act one — the real thing arrives. A general-purpose technology appears whose promise is not fictional. It genuinely lowers the cost of moving something the economy wants moved — goods, messages, data, computation. Capital floods toward it, correctly sensing that the world is about to change. Established

Act two — the financing outruns the demand. Building the infrastructure takes years; paying end-user demand at scale takes longer. Into that gap steps promotion. Company formation accelerates, leverage builds, and the capital structure grows self-referential — money raised on the strength of other money already committed, valuations resting on projected demand rather than delivered revenue. The build-out's own momentum becomes the evidence cited to justify the next round. Assessed

Act three — the financing breaks, but the asset does not. Demand arrives slower than the balance sheets assumed, or credit tightens, and the financing structure fails. Promoters and shareholders are wiped out; bankruptcies cluster. But the physical infrastructure that the money built is still there — laid in the ground, strung on poles, sunk in conduit. It is sold in the wreckage for a fraction of what it cost, and it becomes the cheap foundation for the technology's actual mass adoption, now owned by a second set of hands that never bore the build cost. Assessed

Hold that shape in mind. Now watch it run three times.

2. Track — the railway mania, 1840s

Britain in the mid-1840s convinced itself that the railway would remake the economy. It was right. The railway did exactly that. And in the being-right lay the trap. Parliament authorised hundreds of new railway companies in the space of a few years, a wave of speculative promotion that drew in not just the wealthy but middle-class savers buying shares on partial payment, confident the lines would be worth more than they paid. Established Much of the authorised mileage duplicated existing routes or ran where traffic could never justify the cost; the promotion had outrun any sober forecast of demand. Assessed

The break came when interest rates rose and the calls for further payment on part-paid shares fell due at once. Share prices collapsed from 1846; the mania's central figure, the "Railway King" George Hudson — who had assembled a sprawling railway empire on optimistic accounts — was exposed and ruined by the end of the decade. Established Fortunes evaporated. And yet the track remained. The lines that had been built did not un-build themselves; over the following decades they were amalgamated into fewer, larger companies and became the permanent skeleton of British rail — used, for the next century, by an industry that had shed the investors who paid to lay it. Assessed The first act was true, the second act was ruinous, and the third act handed a working national railway to the survivors at a discount.

3. Wire — the telegraph

The telegraph is the messier middle case, and it earns its place precisely because it is not a clean retail mania — it shows the same structure operating through corporate competition rather than public speculation. Through the middle and later 19th century the electric telegraph collapsed the cost of distance for information, and capital raced to string wire. Competing companies over-built parallel lines along the same routes, and the industry's history is one of repeated, ruinous competition resolving into consolidation — in the United States, ultimately under the dominance of Western Union. Established

The most instructive episode came late in the 1870s and into the early 1880s, when the financier Jay Gould built and backed rival telegraph companies that duplicated Western Union's network — not always because the routes were needed, but because a competing line was an asset one could force the incumbent to buy. The over-building was, in part, a financial manoeuvre; it ended with Gould taking control of Western Union itself in 1881, the duplicative wire absorbed into the very network it had been built to challenge. Assessed The retail investor was less central here than in the railways, but the through-line holds: capital over-built the infrastructure ahead of what any single operator's demand justified, the competing promoters were bought out or absorbed, and the wire that survived carried the messages regardless of who had been ruined stringing it. Assessed

4. Glass — the fibre glut, 1999–2001

The cleanest analogue to the present is the most recent. In the late 1990s the internet was, correctly, understood to be transformative, and capital poured into laying the fibre-optic cable that would carry its traffic. Carriers — among them Global Crossing, WorldCom, Qwest and Level 3 — raised enormous sums, much of it high-yield debt, to trench and lay glass across continents and under oceans, racing one another to build capacity for a data flood everyone was sure was coming. Established

The demand forecast was the load-bearing error. A claim circulated widely — that internet traffic was doubling every hundred days — and it was cited to justify almost unlimited capacity. It was not true; traffic was growing fast, but nothing like that fast, and the claim was later traced and debunked. Assessed By the time the fibre was in the ground, only a small fraction of it was actually "lit" and carrying traffic — the rest sat dark, capacity built for a demand curve that had been imagined. Assessed

Then the financing broke. The telecom sector's collapse in 2001–2002 produced some of the largest bankruptcies in history to that point — Global Crossing early in 2002, WorldCom later the same year — and vaporised an immense quantity of shareholder and bondholder capital. Established The glass, however, stayed in the ground. Dark fibre and the failed carriers' networks were bought out of bankruptcy for cents on the dollar, and that cheaply-acquired capacity became the physical substrate of the broadband internet, streaming video and, eventually, the cloud — the platform on which the 2010s were built, owned by companies that never paid to lay it. Assessed Act one, real. Act two, ruinous. Act three, the wreckage got used.

5. What the pattern claims — and what it does not

The temptation, reading three cases that rhyme, is to run the tape forward and announce the ending. The desk will not, and the reason is a matter of editorial doctrine, not caution for its own sake. A pattern that has repeated three times over 180 years is a template for judgement, not a mechanism of prophecy. It tells you what to watch and what a bust would look like; it does not tell you that one is coming, or when.

What the pattern does claim is narrower and sturdier than a forecast, and it lands directly on today's lead. The single most repeated mistake in each of these episodes was to treat the reality of the technology as evidence about the soundness of the financing. They are different questions on different clocks. Railways genuinely remade commerce and the railway financing still collapsed. The internet genuinely remade everything and the telecom financing still collapsed. The revolution and the bust were not rivals; they co-occurred. Assessed So when the AI build-out is defended with "but the technology is real" — and it plainly is; the models work and the compute is genuinely consumed — the history says: agreed, and irrelevant to the question the lead actually asks. Whether demand is real enough, soon enough, to service the financing that is being raised against it is a separate matter, and the reality of the technology has never once settled it.

That is where the template earns its keep as a companion to the lead. The Navigator piece describes financing that has turned partly self-referential — the seller of the chips helping fund the buyers of the chips — and notes that the IMF and BIS have flagged it. Set against the three acts, that is a textbook act-two symptom: the capital structure feeding on its own momentum, the build-out's scale cited as its own justification. Assessed The reported figures make the scale of the current wave impossible to wave away — hyperscaler capital spending on the order of some $700bn or more in 2026 as reported, atop the circular arrangements the lead details. Assessed None of that is a crash. It is the part of the movie where, in the earlier films, the audience already knew how it ended — and the characters did not.

6. Where the analogy breaks — the honest caveat

A template used without its limits is propaganda, so here is the place the AI case may genuinely differ, and it cuts against the pattern's reassurance rather than for it.

In all three historical booms, the consolation prize was durable. Railway track, telegraph wire and fibre-optic glass are long-lived physical assets; they sat in the ground for years or decades waiting to be useful, and their build cost was largely sunk into things that did not decay on any short horizon. That is why the "the wreckage gets used" third act was available: the asset outlived the financing that created it, so someone could buy it cheap and wait. Established The most expensive component of the AI build-out is not like that. High-end graphics processors depreciate on a horizon of a few years, not a few decades — driven out by successor chips and by the sheer energy cost of running older silicon. Assessed

The implication is uncomfortable and worth stating plainly. If an AI financing bust arrives, the part of the infrastructure that behaves like track and glass — the data-centre shells, the power interconnections, the fibre and networking, the land — would likely survive it and be bought cheap in the classic way. But the chips, which are where much of the money is going, may not offer the same salvage: a GPU fleet that is two product generations old at the moment of the bust is a far weaker asset than a strung telegraph line or a lit fibre pair. Assessed The pattern's happy ending — the technology diffuses cheaply on the over-built base — is contingent on the base surviving. For the concrete, copper and power, it should. For the silicon, it is an open question. That is not a reason to expect a bust. It is a reason to know which parts of the wreckage would be worth salvaging if one came, and which would not.

Bottom line: Three times in 180 years — railway track, telegraph wire, fibre-optic glass — a real technology drew in more capital than its demand could service, the financing broke, the promoters were wiped out, and the over-built infrastructure survived to be bought cheap and used by whoever came next. The lesson is not that the AI build-out will follow suit; it is that the technology being real has never once been evidence that the financing was sound. Judge the AI wave by the demand that services its financing, not by the brilliance of the models. And note the one break in the pattern: this time the most expensive part of the infrastructure — the chips — may not survive the bust long enough to become anyone's bargain.