AI can lift productivity, reshape the economy and still leave investors in the red

How AI evolves is perhaps the biggest question looming over the future of both the stock market and the broader economy. The ‘magnificent seven’ tech companies, all with big bets on AI, account for more than one-third of the S&P 500 Index’s market capitalization, up from about a fifth at the end of 2022.

Investment is ramping up fast: Over the past year, spending on information processing equipment and software has risen by about 20% in real terms. Together with the construction of data centres, the power generation and transmission capacity needed to operate them and the wealth effect from rising stock prices, this is making a significant contribution to economic growth.

The party can go on for quite some time. Many of the major investors (Google, Meta, Microsoft) have ample cash and considerable debt-raising capacity. The race to build the best AI is a winner-take-all game with strong network effects, so they have powerful incentives to achieve scale as fast as possible even if revenue lags. But the speed of implementation is constrained by the limited supply of chips, other specialized equipment and electricity, as well as the resources and time needed to construct large data centers.

As a result, AI spending will remain a significant exogenous growth impulse in 2026. This added investment relative to savings will also have repercussions for monetary policy. By boosting the ‘neutral’ level of short-term interest rates that neither stimulates nor hinders growth, it will push against the rate cuts that US President Donald Trump has been demanding.

In the longer term, the impact of AI will be transformational. It’ll almost certainly lift productivity as it’s applied in areas including coding, call centres, data management, onboarding clients, risk monitoring, report writing, medical care diagnostics and pharmaceutical drug innovation—and more so as it becomes embedded in other processes, such as advanced robotics and self-driving cars, buses and trucks.

This doesn’t, however, mean that investors will come out as winners in the long run. Providers of AI services must at some point generate very large revenues to achieve a sufficient return on their substantial investments.

If, for example, one assumes that outlays to build the requisite AI infrastructure will total $2 trillion to $3 trillion, then the relevant companies will need annual revenue of $1 trillion or more to deliver a decent profit—in part because the AI infrastructure will depreciate rapidly, requiring ongoing investment to remain state of the art. That’s at least 3% of US GDP, which strikes me as a high bar.

To meet such a target, AI companies will need pricing power. To that end, they’ll want to lock their customers into their proprietary infrastructure. Users, however, will push in the other direction. They’ll want to maintain the ability to switch between providers, so they can negotiate lower prices. If they succeed, AI revenues and margins will disappoint.

History suggests that the losses could be large. Earlier episodes of rapid technological innovation—the 19th-century proliferation of railroads, the 20th-century advent of the internet—proved transformational but also entailed vast overinvestment and large equity-market declines.

On a positive note, an AI bust won’t necessarily precipitate anything like the 2008 financial crisis. It’s mainly an equity market phenomenon. I don’t see the type of leverage and asset fire-sale risk that would lead to the type of catastrophe that occurred after the last big bubble, US housing. But that might not be much comfort for the investors involved. ©Bloomberg

The author is a Bloomberg Opinion columnist.