AI, bubbles, and creative destruction

Artificial intelligence (AI) is on everyone’s mind. In Europe, savers are reluctant to invest in equities out of fear that the AI bubble will soon burst. But things are strikingly different in the US, where retail investors have fully bought into the AI theme. Which side is right? Europeans could be missing out on an opportunity that endures, while many Americans might take a hit if they sell too late. The usual story. But even for investors, the issue of AI goes well beyond its equity-market ramifications, given the technology’s potential to make a singularly profound impact on society.

Most of us will remember the last one: the dot-com (or internet) bubble. History enthusiasts will also recall the Railway Mania bubble in the UK in the mid-19^th century.

In both of these cases, the technology bubble was accompanied by visible anomalies: in the former, Alice-in-Wonderland price/sales ratios; and in the latter, as many as four railway projects underway simultaneously – to link the same two cities!

Also in both of these cases, the bubbles served their purpose: they helped to spread the new technology quickly across the economy while preventing businesses from even asking whether they should jump on board. The only “mistake” that such businesses were not allowed to make was to wait on the sidelines. Indeed, none of the large telecom operators in the late 1990s wanted or were able to refuse to buy a 3G licence despite their exorbitant price. And today, what hyperscaler¹ would turn away from investing in data centers despite their no-less exorbitant price? Forced to make a choice, the operators all opted for the risk of death from financial ruin rather than a certain death from immediate obsolescence: stock-market bubbles follow a deterministic process.

And finally, in both of these cases, the new technology destroyed jobs and buried age-old processes in order to pave the way for new practices and professions. In other words, Schumpeter’s creative destruction was at work – enabled by entrepreneurs and financers – and eventually boosted employment and economic efficiency. As far as we’re concerned as fund managers, we need to position our investments on the right side of creative destruction. That means avoiding businesses that stand to be undermined by the widespread adoption of AI and instead focusing on those positioning themselves to benefit.

Most AI industry heavyweights are trading at comprehensible P/E ratios, albeit based on huge profits that have so far barely been eroded by capex. These profits have been made possible by the long, successful history of these companies in the various fields explored by Microsoft, Amazon, Google, Meta, and so on. Their 12-month forward P/E ratios currently range between 26 and 33, which is a far cry from those seen in the heady days of the dot-com bubble. But according to J.P. Morgan, by end-2030 these companies will have spent most of the $5–$7 trillion needed to implement AI, financed in part by the debt that until now has been used only for share buybacks. How long will it take before those AI investments pay off? And how can we evaluate the impact of the switch from a capex-light business model to a capex-heavy one?

Nvidia, the symbol of triumphant AI and the first company to reach a market cap of over $5 trillion, is the happy supplier to the hyperscalers with still-deep pockets. The company is supplying shovels and sieves to the major gold panners in the AI rush, and is trading at 32.5 times projected 2026 earnings. Yet that ratio is still justified by the firm’s impressive earnings growth.

One key unique feature of today’s AI ecosystem – and a potential source of uncertainty – is that it’s based on a company, OpenAI, which still operates as a non-profit, unlisted entity, meaning it has considerable scope for creative accounting. OpenAI has already been involved in $1.4 trillion worth of essentially circular transactions with the AI heavyweights. Time will tell.

Another unique feature, but one with more positive consequences, is that the AI revolution is coming hand-in-hand with a technology war between China and the US: “DeepSeek” AI on one side and “Nvidia inside” AI on the other. There probably won’t be a loser in this clash of the titans until both global superpowers have thrown everything they have behind “their” AI. This technology war could delay the moment of disenchantment, if such a moment comes at all. But when the fat lady sings, will we measure victory or defeat by the stock prices of the companies that spread the technology, or by the economic and societal effects of both types of AI?

Its capacity to destroy white-collar jobs is so extensive that even admirers of Schumpeter’s creative destruction process sometimes struggle to believe that the number of jobs created by AI could one day exceed the number eliminated – and here, young people are the main victim. According to research by Oxford Economics, 85% of the increase in US unemployment since the 2023 low can be attributed to a rise in youth unemployment.² US data also show that the unemployment rate for 20–24-year-olds has grown from 6% to nearly 9.5% in the past three years, although it has barely moved for other age groups. A new record for job destruction was set in October, mainly as a result of AI. How can a fresh college graduate with no business experience compete with a well-trained AI program and the benefits it can deliver in numerous areas? And looking further out, what value can future college graduates bring if they believe the only thing they need to learn is how to enter an AI prompt?

It could be that young people, owing to their capacity to adapt, will be able to find their place in an AI-driven economy, and that the destruction will once again be creative. But we can’t just sit by and hope this will happen. Instead, we need to help graduates enter into the workforce – opening doors that now seem shut – because unless young people can actively contribute to wealth creation, progress won’t be possible in our quickly ageing society. Without measures to support new job seekers, the social limits of rapid AI adoption will be reached well before the limits on rising stock prices.