By James Mackintosh

Investors have a strange relationship with risk. On one hand, they want it: Risk brings reward when it works out. On the other hand, unrewarded risk is the very last thing anyone wants.

The result is the all-too-familiar swing between fear and greed, boom and bust, as investors switch from loving risk to fleeing it. This is relevant to today's stock market because the market is riskier than it used to be on three important metrics:

   -- First, it is much more concentrated in a handful of stocks than in modern 
      times. That means investors who simply track the market are taking much 
      more single-stock risk than in the past. 
 
   -- Second, the stocks that dominate the market are heavily exposed to one 
      big bet, on generative artificial intelligence, into which they are 
      expected to pour almost $400 billion this year. 
 
   -- And third, everyone agrees that those stocks are phenomenal and bound to 
      go up, creating a form of groupthink vulnerable to a sudden reverse on 
      any setbacks. 

The concentration is well known, but that doesn't stop it being scary. Buy the S&P 500 and the top five stocks make up 27.7% of the portfolio, up from 11.7% a decade ago and the same as in 1964. Back in 1964 it wasn't a problem: The rest of the market soared as the economy boomed, and concentration dropped steadily.

But as in 1964, investors are taking on a lot more of the single-stock risk that diversification is meant to shield against. Management trouble, product problems or fraud can hit an individual company hard, but they rarely matter much to a widely spread portfolio. If Nvidia (7.8% of the S&P) or Microsoft (6.7%) has some internal difficulty, it matters to the whole market in a way that setbacks to the strategy of Kimberly-Clark (0.08%) don't.

In 1964, single-company exposure for the biggest stocks mattered, but investors were at least making a diverse set of bets. Back then the biggest stocks were a telecom operator ( AT&T), a carmaker ( General Motors) and an oil major (now called Exxon Mobil). Now the value of the top eight companies in the U.S. is driven by one big bet: AI.

You don't have to be skeptical about the benefits of generative AI to see the risks to Nvidia, Microsoft, Apple, Amazon, Broadcom, Meta, Alphabet and Tesla, the big eight.

How much will people and companies be willing to pay to use AI? How much competition will there be between AI providers? Will cheaper ways of creating the large language models behind GenAI mean big spending was wasted? How long will it take for mass adoption? How long before expensive microchips need upgrading?

At least some skepticism is warranted, too. Academics and AI workers worry that getting ever bigger isn't making AI models ever better. Businesses are finding uses for GenAI, but large-scale productivity improvement hasn't happened so far.

Made-up answers, known as hallucinations, continue to be a problem -- the latest GPT-5 model from OpenAI, for example, estimates that it makes up 10%-20% of answers when asked about things it doesn't have access to, and a few percent even where it can cite sources. That's fine for a social chatbot, but no good if real money depends on the answer. Even assuming it didn't make up its own error rate, of course.

On top of that, demands for regulation are growing after chatbots were consulted before suicides and a murder. More red tape could push up costs further and slow development.

Wall Street doesn't care. Analyst estimates for earnings in 12 months for the Big Tech stocks that dominate the market have leapt again this year, with double-digit rises for all bar Apple (its forecast earnings are up only 4%, held back by concern that it is lagging in AI) and Tesla (down 36% on poor car sales).

It isn't just that the average analyst thinks things will work out great. Agreement between analysts -- using the statistical measure of one standard deviation of forecasts as a percentage of the average -- is also extremely high. For Microsoft, where almost all analysts have a "buy" rating, forecasts have been closer together only once since 2008, and for Meta, Nvidia and Alphabet agreement is the tightest since before the 2020 pandemic.

This is odd, given the heavy investment into a new business with uncertain revenues and completely unknown profit margins. I'd expect a wider range of predicted outcomes, not a narrower one. But when markets are booming, Wall Street wants to be along for the ride.

The same goes for investors. When risk is on your side, you want more of it. And there's no doubt that risk has worked out recently, pushing the S&P to yet more new highs in the past week. Who wants to diversify when concentration has worked so well? Who wants reliable but slower-growth companies when risky but hopefully faster-growth companies are soaring?

In the past, lots of people -- but only after the market turned against risk. None of us know when that will happen again. But it's worth hedging your bets, because it will happen eventually.

Write to James Mackintosh at james.mackintosh@wsj.com

(END) Dow Jones Newswires

September 13, 2025 23:00 ET (03:00 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.