In 1999, the people who said the internet would reshape the world were correct. The Nasdaq still fell 78 per cent.
Both of those facts belong in the same sentence. The second one is the half of the AI conversation that keeps getting left out.
Two different things both get called a bubble.
The first is a price bubble: the claim on an asset gets bid above any plausible cash it will ever throw off. The test is one question — is the buyer buying the cash flow, or the next buyer?
The second is a capex overbuild: real resources get committed faster than the demand for them actually shows up. The money goes into real things. They are simply years too early to earn.
AI has both running at once. An overbuild in the compute, and a price on the equity claims stacked on top of it. They resolve differently, and they hurt different people.
The precedent is not reassuring, and it is not a warning either.
Railway mania, Britain in the 1840s: the direction was right, the network got built, it ran for a century. Railway shares fell roughly 85 per cent from the 1845 peak by 1850.
Telecom fibre, 1996 to 2001: the direction was right, and that fibre carried the next two decades of internet. Under 5 per cent of the laid capacity was lit by 2002. Global Crossing and WorldCom went into Chapter 11.
Dot-com equities: the direction was right, e-commerce did eat retail exactly as promised. Nasdaq fell 78 per cent. Cisco — real product, real revenue, the actual backbone of the thing — fell 89 per cent. Amazon fell 94 per cent before it went on to win.
Every one of those facts is compatible with “the direction was obvious.”
Which is the point. Clarity of direction is the fuel of a bubble, not the argument against one. You cannot overfund something nobody believes in. The obviousness is precisely what makes the money available.
Three questions get collapsed into one.
Is it real? The cheapest of the three to get right — and almost everybody gets it right. Which is exactly why it has never been an edge.
When? How many years of burn sit between here and the cash. This is the question that actually kills capital, and it is the one least often asked out loud.
Who captures it? Internet value went overwhelmingly to consumers, and to a handful of firms that either did not exist yet or were not the ones being bid up at the time.
If you own a business, three things follow.
Price the business on cash, not on the story. A buyer doing diligence in 2028 will discount stated AI intent to zero and pay for the margin line it actually moved.
Separate “we use AI” from “we are an AI business.” The first is an operating improvement and it is durable. The second is a multiple, and multiples move against you without warning.
Start the evidence trail now. Cost per unit of output, tracked quarter over quarter, is the only version of this story that survives a diligence process.
The correction, if it comes, will not destroy the utility. It will destroy the claims on the utility. The fibre stayed in the ground and got used. The shareholders did not.
Every engagement starts with a free example, built on your own accounts.