Exciting inventions lead to dangerous booms, like the one occurring now in artificial intelligence. From the construction of canals and railways in Victorian England to the discovery of electricity and the arrival of the internet, the bright light of new technology attracts hordes of money. Such frenzies also tend to revive old investment misconceptions, and large-language models are no exception, Reuters reports.
Among the most egregious errors in financial reasoning is the fallacy of composition. This is the mistaken idea that what benefits one company will work for others in the same field. The misshaped logic helps explain why British entrepreneurs built three separate train lines between Liverpool and Leeds in the 1840s, and why telecom startups frantically dug up streets to install competing fibre-optic cables in the late 1990s. Each presumed it could woo enough customers to justify the outlay, but it was impossible for all to do so.
The same dynamics are at work in the race for AI supremacy. A handful of technology titans – led by Alphabet, Amazon.com, Meta Platforms, Microsoft and Oracle – are splashing out on Nvidia chips, data centres and associated energy facilities. The Bank for International Settlements reckons, spending has risen further and faster than in previous manias. Optimism nevertheless abounds, with investors counting on widespread success.
To see the self-deceit in action, consider the consensus outlook, as compiled by Visible Alpha, for the five companies known as hyperscalers. The quintet is collectively expected to devote $4.8 trillion to capital expenditures between 2026 and 2030. The splurge is already straining financial resources. This year’s outlay will amount to more than 50% of their combined revenue while the pooled free cash flow is on track to shrink to $5.2 billion by 2027, a tenth of the 2024 figure. Little wonder that Google-owner Alphabet recently issued stock worth more than $80 billion while Amazon raised almost $40 billion by selling bonds.
This cutthroat competition is unusual for several reasons. For one, it turns an industry that thrived on relatively few hard assets into a far more capital-intensive one. Second, it makes archrivals out of companies in quasi-monopolies: Google in search, Meta in social media, Microsoft in business software, and so on. “Suddenly they’re all stripping for action and they’re going to all pile in the ring together,” like some spectacle on the White House lawn, the veteran investor Jeremy Grantham remarked recently on “The Big View” podcast.
The cage fight threatens to leave lasting financial scars. Investment binges will weigh on bottom lines as the companies write down the value of new buildings and AI chips. Sure enough, depreciation and amortization costs are projected to rise as a proportion of revenue. Alphabet’s is expected to be 12% of its top line in 2030, three times the level from two years ago. Microsoft’s depreciation expense, meanwhile, will double from 11% of revenue to 21% over the same period, analysts anticipate.
A sharp increase in costs would typically squeeze profit. There’s a more upbeat attitude this time because of a commensurate surge in sales forecast. Alphabet, Microsoft, Amazon and Meta – respectively the third, fourth, fifth and eighth-largest U.S. companies by market value – are expected to roughly double revenue between 2025 and 2030, according to Visible Alpha data. Oracle’s top line is set to more than treble over the same period. This windfall, if it arrives, would ensure high profitability. Indeed, Alphabet and Amazon’s operating profit margins are seen expanding.
This bullish scenario depends on two less plausible trends. For the tech juggernauts to absorb increases in depreciation costs, they must become more efficient elsewhere. To see how it works, deduct anticipated depreciation from total costs and divide the resulting number by revenue. The results are striking: Alphabet’s non-depreciation expenses, for example, would fall to 44% of sales in 2030 from 57% in 2025, analysts calculate. Other companies are similarly projected to become more efficient.
The other questionable development is that, thanks to rising revenue and improving efficiency, analysts expect the tech fivesome to start throwing off cash again. By 2030, they are forecast to generate around $660 billion of combined free cash flow, about three times the 2023 sum, before the AI frenzy got underway.
In other words, a strange sequence of events would have to occur. For the next few years, Big Tech will throw every spare investment dollar at the AI boom in a colossal clash for dominance. But then the companies will revert to a more financially rational status quo. It sounds implausible. If AI funding pays off, the competitors will be encouraged to spend even more. Meanwhile, any setback could spark a capacity glut and a destructive price war. Consider Meta’s plan, reported by Bloomberg last week, to rent out spare data centre capacity.
Big Tech’s financial future is admittedly even harder than usual to map out. Capex budgets keep changing, and rocketing. New costs such as depreciation are opaque, especially as they can be put off until the new investment is completed. The five companies have $800 billion of unstarted leases and are on the hook for $1 trillion of future purchases, Morgan Stanley analysts estimate. Rosier outlooks point to a combined $2 trillion of revenue such contracts will provide. Much of it, however, would come from startups including Anthropic and OpenAI, which have yet to demonstrate that they are financially sustainable.
Faced with an unknowable future, equity-fund managers are behaving more like venture capitalists: Back multiple companies in one business on the premise that the eventual winners will more than offset any losses from stragglers. Jitters are already evident, though. All five of the big AI companies trade at a lower multiple of earnings than in early January.
Even after this creeping scepticism, the fallacy of composition holds. It is extremely unlikely that all five hyperscalers will be as successful as the predictions indicate. More likely is that none of them will live up to the financial hype.
