AI may survive the next crash. A lot of the companies building it may not.

The dot-com bubble did not collapse because people suddenly stopped using the internet. It collapsed because the money financing the expansion became more expensive, investors stopped funding companies with no clear path to profitability, and businesses that had been building infrastructure ahead of demand discovered they could not keep paying the bills.

There is a similar vulnerability in the AI boom. Everyone is talking about how much demand there will be for AI over the next decade. Far less attention goes to a simpler question: how much money will it cost to build everything, and who is going to keep financing it?

Look at what is happening. Meta is raising billions through stock offerings to fund its AI expansion. Oracle just sold new shares at $116 when its previous closing price was $130.67, while borrowing another $10 billion at interest rates reaching 8.65%. CoreWeave’s bonds have been trading at yields around 15% to 16%. These companies are not all in the same financial position, but the cost of funding this infrastructure is becoming harder to ignore.

And the problem goes beyond borrowing costs. AI infrastructure requires enormous upfront spending on chips, data centers, electricity, cooling and networking. Those investments need years of revenue to pay for themselves. If financing gets more expensive, investors demand faster returns, or customers slow their spending, projects that looked profitable under cheap financing can suddenly become very difficult to justify.

Character.AI has already filed for Chapter 11, although its problems also involve lawsuits, regulatory pressure and operating costs. It is an early example of how having millions of users does not guarantee a sustainable business.

The internet did not disappear after the dot-com crash. The companies that survived eventually built some of the most valuable businesses in the world. But plenty of companies that spent billions building the infrastructure of the future never made it to the other side.

AI could follow a similar path. The technology may become even more useful while a large number of today’s AI companies, infrastructure providers and speculative investments get wiped out. That creates an unusual opportunity for investors. The next winners may become obvious only after the weakest companies have failed, valuations have come down, and the market stops treating every company with an AI connection like a guaranteed winner.

The real test of this boom may not be how quickly AI adoption grows. It may be what happens when the people supplying the money start demanding their money back.