LONDON—It is a truth universally acknowledged that transformative new technologies rewrite the rules of finance. Or so we are told in numerous commentaries, social-media posts, and regulatory filings relating to the current AI investment boom, which may well prove to be a bubble.
Fundamentals can be overlooked for only so long. For those too young to remember the dot-com crash of the early 2000s—which, I suppose, includes just about every AI industry leader right now—the late-1990s investment frenzy holds important lessons.
The dot-com boom grew out of the early internet. Working as an engineer at the infrastructure level (designing the hardware and networks over which revolutionary new apps like Google, Napster, and Lastminute.com would run), I well remember the mood in the air. It was a frantic, explosive, inspiring roller-coaster, much like the AI boom, but without any talk of human extinction. One of the companies I worked for had an investor presentation with a picture of the Sierra Nevada mountain range. The text said: “In 1848, you didn’t want to be panning for gold. You wanted to be selling shovels.”
The reference was to the California Gold Rush, and the lesson was that, though some would succeed and many more would fail, you would be fine if you were the one building the infrastructure. Selling shovels was a sure thing, and it justified valuations that departed from normal accounting metrics. The company I was working for sought to persuade investors that latter-day shovel-mongers need not worry about EBITDA (earnings before interest, taxes, depreciation, and amortization).
The only problem with this metaphor is that it wasn’t true. The telecoms infrastructure company I was working for went under. I lost my job, my visa, and my right to live in the United States. Many other infrastructure providers lost far more: Global Crossing and WorldCom filed for Chapter 11 bankruptcy, while Nortel Networks and Lucent lost approximately 99% of their stock value. As the vast majority of players were wiped out, a handful of deep-pocketed companies—Microsoft, Google, Amazon, and later Facebook—bought out would-be rivals, grew, and consolidated into the monopolistic giants we know today.
Infrastructure providers were hit hard during the dot-com bust because they had relied heavily on debt to finance large capital expenditures. Then as now, a significant share of this financing was circular, with companies offering “vendor financing” to help their own customers pay for their products.
For example, the company I worked for made boxes designed to boost high-speed-internet capacity, which it sold to infrastructure providers. Often, the company gave the customer the money to buy the box. Everyone made money and got paid back as long as the boom continued. But when the bust came, the virtuous circle turned vicious.
We are seeing the same circular financing in the AI boom. Anthropic buys compute capacity powered by Nvidia chips from a company in which Nvidia is a major investor. Microsoft has invested more than $13 billion in OpenAI while also “selling” it cloud services through its Azure division. Is this a case of financial innovation? Or is it an unsustainable boom-and-bust engine?
At the macro level, the figures speak for themselves. It is estimated that justifying the trillions of dollars being spent on AI infrastructure will require more than $2.5 trillion in annual revenues by 2030. Those cashflows must come from paying customers who are using AI services to improve their own revenues, which will generally mean removing costs, primarily salaries, from their bottom lines.
But removing trillions of dollars from the West’s salary bill would cause a societal upheaval, and given the current volume of private debt, a financial crash may well follow. Alternatively, markets will conclude that AI firms are unable to repay their debts, and the order of consequences will be reversed: a potentially unprecedented financial crash, followed by societal upheaval.
I say unprecedented for two reasons. First, the scale of debt exposure today is considerably larger than in 2001, making the situation more like the run-up to the 2008 global financial crisis. Second, wealth today is far more concentrated in the technology, media, and telecommunications sector. During the 2008 crisis, TMT accounted for one-fifth of the broader US large-cap equity market; now it represents nearly half.
Booms stop when financiers become unwilling to lend into them any longer. The first big sign that my sector was going under in 2001 was American Express’s decision to suspend our company credit cards, which came some months before we filed for Chapter 11 bankruptcy protection.
If the financial industry decides to stop funding AI, would that leave more money to invest elsewhere—such as in Newcastle’s productive economy, or in unemployed young people? Or will investors flee to traditional safe assets like gold?
I do not know. But with the UK putting AI at the center of its ambitions during its 2027 G20 presidency, its agenda must include security, standards, and planning for any crash that may be coming.
Copyright: Project Syndicate, 2026.
