Why investors won’t know what to make of AI for a while

Confusion over AI is everywhere. Goldman Sachs has created a share-price index of firms at most risk of disruption (see chart 1). Over the past year this has fallen by more than 20%. The bank’s mirror index of “long-term AI beneficiaries”, whose earnings stand to get the biggest boost from higher AI-fuelled productivity, is down by about 5%, even though many stockmarkets are near record highs. Often investors cannot even decide whether a firm is an AI winner or loser. From May 2024 to May 2025 the share price of Duolingo, a language-educator, doubled. Since then it has fallen by 80%. Not long ago Google was apparently toast because of AI. In the past year the share price of its parent company, Alphabet, has risen by 85%.

Bond traders, for their part, think it is all much ado about nothing. In a world of higher AI-fuelled economic growth real interest rates should rise. But at 4.9%, yields on 30-year Treasury bonds are little different to where they were at the start of the year. When Isaiah Andrews and Maryam Farboodi of the Massachusetts Institute of Technology analysed bond-market moves around big AI model releases, they found that long-term yields fell.

Depending on where you look, then, AI is both everything and nothing: an existential risk for firms and a rounding error for the economy. This exasperates market-watchers. It is also, when you look to history, par for the course. For if stockmarkets are bad at pricing conflict, they may be worse at pricing technological revolutions.

For every Blockbuster, which the stockmarket started marking down two years before the video-rental darling’s revenues peaked in 2004, there is a BlackBerry or a Kodak, where investors failed to detect trouble until the business was on its last legs. Like economists and recessions, markets often anticipate a technological disruption that never happens. When Thomas Edison’s electric-lighting revolution began in the 1870s, it was the ChatGPT of its day. Smart money of the era piled into the new technology, backed by John Pierpont Morgan among other financiers, and out of gas companies, until then the providers of most artificial illumination. But gas did not become irrelevant. Investors realised that electric light would remain dearer than gas for years. Even as electric light fell in price in the following decades, gas firms found larger markets to pursue. In London, the Gas Light and Coke Company noted that in 1892 only 2% of its customers had gas cookers. By 1911 more than two-thirds did.

To complement anecdote, The Economist examined American and European equities from 2005 to 2026 and found around 80 instances where an entire industry, from luxury goods to telecoms and media, entered a sectoral bear market, with share-price drops of at least 20 percentage points over three months relative to the broader index. (We excluded the energy industry, which rises and fall with the price of fossil fuels.) When this happens, we assume that the market worries that some structural change will hurt that industry. This could be a new technology or institutional arrangement (such as globalisation).

Following such a sectoral bear market, the share price stays down half the time. In those cases, investors had correctly priced a long-term change in an industry’s circumstances. When competition from Chinese rivals permanently destroyed much of the European solar industry in the 2010s, investors priced this early. They also foresaw that telecoms firms would suffer as the internet took off (see chart 2).

In the other half of bear markets, the share price rises back above the overall market within a few years, suggesting that the early bets soured. Tobacco firms in America and Europe are a good example. Several times their share prices nosedived relative to the market, as investors worried about the effect of innovations in e-cigarettes and vaping. Time and again, tobacco stocks rose from the ashtray.

Our results suggest the stockmarket struggles to capture structural economic changes. This chimes with the work of Song Ma of Yale University. Even when companies’ technological base is becoming obsolete (measured by how cutting-edge their patents are), analysts tend to overestimate future profitability, Mr Ma finds. This props up the share prices of obsolete firms.

It would not be a surprise if today’s investors, following Mr Ma’s results, were overestimating the ai threat to some firms but underestimating the danger to others. There is, after all, even more uncertainty about the AI transition than there was about previous technological shifts. Two reasons stand out. The first concerns the technology itself. AI capabilities have improved rapidly in certain domains, notably coding. But progress is uneven across tasks. Performance on open-ended writing and idea generation is not noticeably better than it was a few months ago.

The second source of uncertainty concerns the economics of a superintelligent AI. No one knows to whom the profits of such “artificial general intelligence” would accrue. If AI reduces barriers to entry, companies’ profit margins may decline. Leading AI labs report rapid revenue growth but also enormous costs of computing power. Lots of academic research suggests that new technologies cause market bubbles as investors get excited about the future. But some, such as by Jeremy Greenwood of the University of Pennsylvania and Boyan Jovanovic of New York University, suggests that the stockmarket can actually fall because investors expect new, as yet unlisted, firms to gobble up the profits. That is certainly what many backers of OpenAI and Anthropic, two warring AI labs, are hoping.

In a world where fundamental views of AI switch quickly, today’s losers could end up as tomorrow’s winners. Much will depend on how well firms deploy AI to improve their offering to clients. Innovating your way out of supposed technological threats is a common theme in business history. Western Union, the leading telegraphy firm of its day, foolishly snubbed Alexander Graham Bell’s offer to sell it the patent for the telephone. But as phone networks grew, Western Union carved out a niche in money transfers which faced no immediate technological competition. American Express, now synonymous with credit cards, started out as a freight-forwarder. Samsung, a global technology giant, sold dried fish. Some of the software businesses being hammered today may likewise reinvent themselves.

Fans of efficient markets may be maddened by this inability to peer accurately into the future. But markets reflect only the collective wisdom of today’s investors. For as long as conversations between two Silicon Valley technologists produce three answers about AI’s impact on the world, no one will be the wiser.