The AI stock market wobble shows that this time is not different for investors. While profits for chip makers and suppliers rose, experts warn that this spending boom is temporary. Rob Almeida said, “To me the market doesn’t have a P problem, it has an E problem.” Investors remain cautious.
The profits aren’t a mirage. But they should only reassure us if we can rely on what legendary financier John Templeton said were the four most dangerous words in investing: “This time it’s different.”
The AI build-out hasn’t yet shown it can avoid the problems of past massive capital projects, such as railways and the internet. At least so far, this time isn’t different.
Start with the profits. Capital expenditure, or capex, by one company generates revenue for other companies. The data-center boom is creating massive revenue elsewhere. Makers of everything AI needs have been able to jack up prices for chips, cooling equipment, cables, backup generators, turbines and a host of other supplies, boosting profit margins as sales soar.
But this is temporary. Once the build-out is complete in a few years, capital spending will drop drastically, as maintenance, repair and upgrades are cheaper than a whole new building.
Investors understand this, and the debate is primarily over how long profits can stay high, not whether they can stay high. This can be seen in the share prices. Yes, they’ve gone up, but the gains have been modest when you consider the huge surge in profits. Investors know much of the profit is temporary.
The result is that while prices have gone up, valuations, measured as the price-to-forecast-earnings ratio, have gone down. Investors are paying less per dollar of the earnings coming up in the next year, because they know it won’t last.
The valuation of Nvidia, the most valuable U.S. company, has dropped from 23 times predicted earnings to 15 times, even as its shares rose 15%, and memory-chip stock Micron trades at a forward PE ratio of just under five times after its shares tripled.
“To me the market doesn’t have a P problem, it has an E problem,” says Rob Almeida, global investment strategist at MFS Investment Management.
Some think this is so extreme it is a bubble in earnings, the E part of the PE ratio. If earnings deflate, elevated valuations will become obvious.
Gerard Minack of Minack Advisors points out that forecast earnings for the eight biggest AI stocks—Amazon, Alphabet, Broadcom, Meta, Microsoft, Nvidia, Oracle and Palantir—plus what he calls the “next 11” AI players, have increased by $700 billion this year, against an increase of less than $500 billion for the rest of the S&P 500.
Bulls can take comfort that investors understand that amazing earnings won’t last. Bears can say froth in earnings explains why PE ratios haven’t soared the way they usually do in a bubble, and look at measures that try to smooth cyclical highs and lows, such as the Shiller price-to-10-years’-earnings ratio. There’s also a parallel to 2007, when the bubble in bank stocks (usually valued by comparison to balance sheet, rather than earnings) was hidden by overstated book values.
This past week the threat to capex-related profits became obvious, as higher-than-expected inflation drove up financing costs and so threatened the AI construction boom. Shares in CoreWeave, an independent data-center operator, plunged 24% to wipe out all gains this year, those in memory-chip maker Micron dropped 11% and its South Korean rivals Samsung Electronics and SK Hynix tumbled twice as much. Other AI suppliers also fell hard.
The moves show just how much it will hurt if the AI capex doesn’t materialize.
Yet, if the falls prove merely another brief interlude in the raging AI bull market and construction continues as expected, it will be just as risky—in a different way. Spending $9 trillion on AI by 2032, as Columbia Business School professor Stijn Van Nieuwerburgh estimated in a Brookings paper recently, would make it the biggest U.S. capital project ever, at 3.2% of GDP a year.
For this to work out for investors, we have to get past the temporary capex-related profits to generate sustainable AI profits. That means finding worthwhile uses that increase productivity enough that customers are willing to pay the full cost of developing and using the technology. If it disappoints, much of that $9 trillion will have been wasted.
The lessons from historical capital-spending booms offer both good and bad news. The good news is that mostly they delivered as much as or more than expected, with huge benefits for society. The bad news is that, from the canal mania of 1790s England, the British railway bubble of the 1840s, its American equivalent in the 1880s, electrification, radio, television and up to the fiber build-out of the dot-com bubble, investors were sorely disappointed.
It took too long for the benefits to be realized, profits were smaller and came later than hoped or competition meant excess profits flowed to the users of the new technology, not its owners. AI is vulnerable to each of these. Investors really need this time to be different.