In 1840s Britain, everyone invested in railways. Valuations went vertical, money poured in, track got laid at a reckless pace, capacity outran demand — and when the bubble burst, 90% of the railway companies went bankrupt. A textbook collapse. Now ask the other question: where did the rails go? Not one mile of laid track disappeared. The bubble died; the track stayed.
A century later, it was fiber. Around 2000 the world laid fiber like mad; after the crash, 90% of it sat dark, and it took ten full years before it was actually used up. The Nasdaq fell 80%. Lucent and Nortel, infrastructure giants, went straight to bankruptcy. Yet every video you stream today travels through that very cable. The technology didn’t die. What died was the money of the people who bought the stocks.
Some will say this time is different: AI’s long-term value is real. That is the strongest version of the objection, and it deserves a serious answer. Its flaw is a quiet substitution — the technology being real and the price being right are two different questions. Not one of those railway companies failed because the trains wouldn’t run. Not one meter of that fiber sat idle because light couldn’t carry data. Long-term value being true is precisely where bubbles hide best: because everyone is waiting for a ten-year story, paying a hundred times earnings today starts to feel rational.
In other words, a bubble is never a verdict on the technology — it is a verdict on the pace of the price. The pattern is fixed like physics: infrastructure first, overcapacity next, then the crash, then the technology quietly becomes a public good. Line up the two eras and the table writes itself: fiber and data centers in 2000 correspond to AI data centers, optical modules, and GPUs today; Cisco traded at over 100 times earnings then — who that maps onto now, you can decide yourself. The nastier part is the lag: the compute capacity being frantically built today all lands within one to two years, and if demand doesn’t keep up, what replays is fiber’s dark decade.
Nortel’s engineers mostly genuinely believed optical communication would change the world. After the bankruptcy, what they believed in was still here; they were gone.
Draw the boundary of this piece first: it does not predict when the AI infrastructure bust arrives or how deep it goes — such predictions have nothing verifiable behind them. And it does not say “so buy compute stocks,” or sell. Infrastructure surviving is not the same as the surviving companies making money; technology becoming a public good and shareholder returns are two different roads. This article is about the pattern, not the trade.
One question left for you: when the bubble bursts, the rails remain and the companies die. Which layer is your money on?
