Bubble, in the context of financial markets, signifies that the price of certain assets exceeds their intrinsic value — some rational estimate of future returns — and, by definition, implies some mania, euphoria, or irrationality. Global financial markets over the centuries have seen multiple bubbles, which eventually burst. The Tulip mania in the eighteenth century is a classic example of this phenomenon.
Despite widespread talk of bubbles, such irrationality is still hard to diagnose in the current wave of market enthusiasm for artificial intelligence. That does not mean all the capital going into the sector must earn good returns — far from it. But for now, at least, AI is best thought of as a boom — one that may turn into a bust — rather than as a bubble. Those who fear a bubble have plenty of exuberance to point towards.
There is the spectacular performance of AI stocks, with Nvidia briefly becoming the first company worth more than $5tn; the huge share of US output going into tech investment; the AI start-ups instantly valued in the billions of dollars; the increasing use of debt to fund data centres; and the dubious, circular deals such as OpenAI’s partnerships with Nvidia and AMD, whereby tech suppliers invest in AI companies that immediately use the money to buy the supplier’s product.
Exuberance, however, need not imply irrationality. It is essential to distinguish between two situations that resemble a bubble, but are not. One is over-optimism. Whenever a radical new technology, such as AI, comes along, there is considerable uncertainty about its value. Does it work? What are the applications? Will it continue to improve exponentially, etc., etc.? Investors have to make judgments with minimal information, and as the utility of the technology becomes more apparent, it may turn out that their initial assessments were wrong.
That will manifest as an investment boom that turns to bust — not a bubble that deflates. An overestimate of intrinsic value is not a departure from it. Three years after the launch of ChatGPT heralded the arrival of generative AI, its ultimate utility is still unclear. Many companies have found that chatbots do not take them very far. Still, by 2025, AI has seen some serious applications, such as in computer programming, and the technology continues to evolve rapidly. There are still grounds for optimism about its value.
A second, closely related situation is a different kind of error, where investors correctly assess the value of a new technology but mistake the winners. One of the most remarkable things about the 1990s dotcom boom, in retrospect, is how rational it was. The investors of the time were correct about the internet’s enormous value. They quite naturally placed early bets, buying up the leading companies of the day — Yahoo and Lycos, Amazon and AOL — but Google was then based in a garage and Mark Zuckerberg was still at school.
It may be that the winners of the AI era have not yet been founded, but again, error does not imply irrationality. The biggest reason to call this a boom rather than a bubble, however, is the driving force behind it: a small group of established technology giants with coldly rational reasons to spend hundreds of billions on AI.
One of the sacred texts of Silicon Valley is Only the Paranoid Survive, a 1996 book by the late Intel chief executive Andy Grove. When generative AI arrived, the tech giants — sitting atop some of the most valuable quasi-monopolies in human history — had plenty to be paranoid about.
The ChatGPT interface was an immediate and obvious threat to internet search. Algorithms, filtering, and content creation with generative AI affect social media. With a bit of imagination around AI agents and voice interfaces — and the executives who run these companies have highly developed imaginations when it comes to competition — the technology could also disrupt the smartphone and e-commerce even before one gets to Microsoft and the rest of the computing industry.
Protecting these enormously valuable businesses is easily worth spending a fortune just as an insurance policy, even if AI does not, in the end, create much new value. “If we end up misspending a couple of hundred billion dollars, I think that is going to be very unfortunate, obviously, but . . . I actually think the risk is higher on the other side,” Zuckerberg said in September. He may be wrong. He does not sound delusional. OpenAI, the biggest new company to emerge in this area, demonstrates the point rather than contradicts it. The most plausible reason for it to be worth hundreds of billions is the potential to monetise its more than 800 million weekly users at the expense of the existing tech giants.
Meanwhile, if someone is spending hundreds of billions a year on AI, that supports a lot of investment in data centres and buys a lot of Nvidia semiconductors, regardless of how well the technology ultimately works out, it shouldn’t be considered part of a bubble. With Meta and Alphabet trading on 25-30 times earnings, their valuations look optimistic, but not euphoric.
To repeat, this is not a claim that AI will triumph, that market expectations are correct, or that the boom will not turn to bust. Given the uncertainty, it would be more surprising if the market’s current beliefs were correct than if they were wrong. Still, investors need to wrestle with this technology’s actual potential rather than dismiss it as a bubble.
