The AI rally is scary

Good morning. The big event this week is the first presidential debate of 2024 which, like every presidential debate, will be dreadful. But Unhedged feels obliged to watch for a sense of whether either candidate sounds serious about cutting the yawning budget deficit, which has been a major support for stock prices. What will you be listening for? Email me: robert.armstrong@ft.com. 

Inside the AI rally 

Last week Unhedged had a look at how, since the end of March, basically all the gains in the S&P 500 have come from AI stocks and AI-adjacent stocks. Below is a chart showing the dollar and percentage gains of Nvidia, the “Fab Five” Big Tech stocks (Alphabet, Amazon, Apple, Microsoft, Meta) and a basket of semiconductor stocks that have taken off in anticipation of an AI-driven digital investment boom:

I find this a bit unsettling. A rising, historically expensive market that would be falling gently were it not for a single investment theme could, it seems to me, turn into a bear market if sentiment about that theme turns sour. Imagine Nvidia cutting its outlook for revenue, or one of the Fab Five trimming its investment budget, for example.

Expectations for stocks within the AI halo are clearly high, but it is probably worth being a bit more precise about this. Consensus estimates for revenue growth for next year and for 2026 do not, in fact, seem wildly demanding. Analysts are expecting a 23 per cent annual growth rate for Nvidia over that period. This would represent something of a moderation in rate; over the past five years Nvidia revenue has grown at 50 per cent a year. Similarly, the two-year revenue growth rates pencilled in for the Fab Five are at or below the growth rates of recent history. It is only a handful of the chip stocks — Micron, Texas Instruments, Analog, and Lam — where a major acceleration in revenue is expected. 

Is the recent rally in the AI group driven by upgrades of earnings estimates? Looking at 2025 estimates, not really. Since the end of March, earnings estimates for the group as a whole have crept up in the low single digits, percentage wise. Apple, Amazon, and Micron are the only ones that have received meatier upgrades. 

What have changed radically in the AI rally are, perhaps unsurprisingly, valuations. In the past three months, the price/earnings ratios of Nvidia, Apple, Broadcom, and Qualcomm have all risen by over 20 per cent. Looking back to last October, when the rally began, the average (harmonic mean) P/E ratio in the AI group is up almost 50 per cent.

That is a lot. How to interpret it? Perhaps it just reflects momentum and animal spirits. More charitably, it could reflect the expectation that the AI business will provide an increase in profits that lasts for many years into the future. That is to say, it is a bet about the competitive dynamics within the AI industry: that it will not be hypercompetitive, and the winners in the long term will be the same as the winners now — the Fab Five and today’s leaders in the semiconductor industry.

To me, the second half of the bet, that the incumbents will keep on winning, seems like a reasonable one. Incumbency in tech is very powerful, to the extent that companies can use their strong market position in one technology to create a strong position in another (think of Microsoft moving from PC operating systems to cloud computing). The first half of the bet, that AI will not turn into a capital-intensive knife fight where no one makes high profits, I don’t know how to assess. 

To assess the stability of the rally, we also need to look in more detail at how the non-AI stocks are doing. Here is S&P 500 sector performance, once the AI stocks have been taken out:

Since March the non-AI stocks are down 2 per cent in aggregate (again on a market-weighed basis) and 9 of the 11 sectors are down. And there is a very clear pattern: defensive sectors (utilities, staples, healthcare) are doing relatively well and cyclicals (energy, materials, industrials, discretionary) are doing relatively poorly. The ex-AI market does not seem very optimistic about the economy.  

Unhedged pointed out recently that part of the reason utilities have outperformed may be that they underperformed for a long stretch starting in 2022 and became cheap, and that the prospect of rate cuts have made their yield more attractive, too. Furthermore, I have read one or two arguments to the effect that the AI boom, being incredibly power-hungry, will increase demand for power generation. In the current environment, I wouldn’t rule out AI hype reaching as far as utilities.

Still, the defensive tilt of the non-AI part of the market is clear enough. Another indicator of this might be the excellent recent performance of Apple shares, which account for fully half of the Fab Five’s increase in value since March. With its high margins, recurring revenues, and high customer loyalty, Apple looks like a safe place to be.

One internal tension within the AI rally is that the revenues of its leading company, Nvidia, are expenses for some of its biggest beneficiaries, the Fab Five. In the short term, Nvidia’s success is a drag on the cash flows of Big Tech companies, which are buying the bulk of Nvidia’s chips. Charles Cara of Absolute Strategy Research has recently made a provocative point about this. He notes that 40 per cent of Nvidia’s revenues come from Microsoft, Meta, Amazon, and Google, and that even the very large expected increase in capital expenditure at those companies is not very large relative to the expected increase in Nvidia’s revenues. His chart:

The increase in the four companies’ capital spending between the last fiscal year and 2025, at $54bn, is more than 40 per cent of the $100bn expected increase in Nvidia’s revenues, but presumably only a fraction of Big Tech’s capital spending goes to Nvidia’s GPUs. So either the Big Tech will spend more, or Nvidia will make less. “We fear either tech sector cash flows, or Nvidia sales are over estimated,” Cara writes.

It could be that other tech companies’ spending with Nvidia rises quickly next year, compensating for any shortfall at the Big Techs. That would, however, speak to how competitive the AI business may become. 

One good read

Social media is a tough business.