US corporate profit boom faces its ‘Minsky moment’

Meta Platforms

Today 13:00

The author is a Reuters Breakingviews columnist.  The opinions expressed are his own.
By Edward Chancellor
LONDON, Oct 23 (Reuters Breakingviews) - The artificial intelligence generated by large language models is breathtaking. Equally wonderous is the financial alchemy that accompanies multi-trillion-dollar investments in AI infrastructure. Capital spending is instantaneously transformed into profits, even as the economic viability of the AI buildout remains questionable.
The United States is in the throes of a great profit boom. Corporate profits had reached a record $4.7 trillion for the last twelve months by mid-year. That’s a 17% increase on the previous year’s figures. Tax breaks contained in President Donald Trump’s One Big Beautiful Bill Act have led to a sharp drop in corporate tax receipts. Washington’s enormous fiscal deficit – running at around 6% of GDP – provides an economy wide fillip to corporate bottom lines.
When it comes to individual companies, however, the surge is all about AI. Earnings per share for members of the S&P 500 Index .SPX were up 26% in the twelve months to July: the long-run average annual growth rate is 7%. Expected earnings for the third quarter are 35% above estimates from a year ago.
Gerard Minack of Minack Advisors lists 19 members of the S&P 500 that are heavily involved in the AI buildout: Big Tech “hyperscalers”, chipmakers and supply-chain companies such as fibre-optic provider Corning GLW.N. Since January 2025, these “AI-19 stocks” have seen forecast earnings rise by around $700 billion, 50% more than all the other companies in the US stock market benchmark.
Analysts at Morgan Stanley predict that 42 leading companies exposed to AI will grow their earnings by 67% in 2026. The aggregate earnings growth of the other S&P 500 companies is projected at 23%. Put another way, these 42 stocks account for more than half of the earnings growth for the S&P 500 in the current year. Next year Morgan Stanley expects the basket of AI plays to contribute 78% of the S&P 500 earnings growth as AI investment in the United States rises to a forecast $1.6 trillion.
“This is the probably the most earnings accretive capex boom in history,” says Jefferies strategist Christopher Wood. AI capital spending inflates profits because, as Minack points out, one company’s capex is another’s revenue. Profits emerge because the investing company does not expense all its costs at once, whereas the seller of capital equipment immediately books the full amount as revenue. Buyers of Nvidia’s NVDA.O graphics processing units depreciate their purchases on average over around 6 years, while the chip giant books a juicy upfront profit.
What’s more, when the buyer is acquiring chips for a data centre that is under construction, the depreciation clock does not start ticking until the facility is up and running. Morgan Stanley estimates that as much as 40% of this year’s AI capex will become operational sometime between 2027 and 2029. The earnings mirage will not last indefinitely, though. By the end of the decade, the hyperscalers’ annual depreciation charges will range between $500 billion and $1 trillion, according to a recent report from Capital Gearing Asset Management.
Investments in unlisted AI companies like OpenAI and Anthropic are also super-charging profits. Tech giants reflect changes in the private market value of these investments in their earnings statements as “other non-operating income.” In the four quarters leading up to mid-2026, the four major hyperscalers – Alphabet GOOGL.O, Amazon.com AMZN.O, Microsoft MSFT.O, and Meta Platforms META.O – saw annualized earnings increase by $222 billion while "other non-operating income" rose by $224 billion. In other words, the rising value of unlisted AI investments accounted for all the increase in their reported earnings.
The fact that the AI buildout is increasingly financed with debt adds a further element of instability. The hyperscalers will plough almost all their operating cash flow into capex in the current financial year. These companies have issued around $225 billion of unsecured bonds so far this year, equivalent to around 30% of their full-year capital spending. They also have lease commitments for $1.1 trillion of AI equipment and services, which, since they haven’t yet commenced, are not reported on their balance sheets.
Next year, Morgan Stanley analysts forecast the hyperscalers will have negative free cash flow of around $550 billion. “Access to credit markets and the ability to absorb higher interest rates are fundamental for hyperscalers and neoclouds to be able to deliver against our capex estimates,” says the Wall Street bank. That shouldn’t be too much of a problem for giants like Alphabet, Microsoft, Amazon and Meta which enjoy solid credit ratings because of their dominant positions in search, software, online retailing and social media.
The economist Hyman Minsky applied the term “Ponzi finance” to businesses that do not generate sufficient cash flows to meet their outgoings. Loss-making data centre operators and AI frontier labs fit this description. Ponzi finance units, said Minsky, need rising valuations to raise additional capital. When valuations decline and the capital markets seize up, these shaky institutions collapse, causing losses to metastasise through the financial system. That is what happened in early 2000 after the credit markets lost confidence in the debt-fuelled buildout of the information superhighway.
That investors are starting to question the viability of the AI investment splurge should come as no surprise. Productivity improvements from implementing the new technology remain elusive. Rogue AI agents are causing havoc. Models continue to hallucinate. Chinese AI companies are rapidly taking market share from the leading US AI labs.
OpenAI has pushed back its planned IPO while upstart data centre operators such as Firmus Technologies have shelved their stock offerings. Prospective investors in Anthropic are weighing a lower valuation for the AI company, Reuters reported citing four people familiar with the matter. CoreWeave CRWV.O, the largest neocloud, has seen its shares and bonds tumble.
The shares of hyperscalers and chipmakers do not look expensive relative to earnings. That’s because the bubble is in profits, not valuations. US homebuilders were in the same position at the peak of the housing bubble in 2006, as were global miners when the commodity supercycle came to a sudden stop in the middle of the last decade. When the AI capex boom ends, earnings growth could revert to trend, says Minack. That moment may come sooner rather than later.
Follow @Breakingviews on X

(Editing by Peter Thal Larsen; Production by Pranav Kiran)
((For previous columns by the author, Reuters customers can click on CHANCELLO/edward.chancellor.bv@gmail.com))