AI’s great expectations problem

Almost exactly a year ago, I wrote that the artificial-intelligence boom looked suspiciously bubbly. The technology was real, the investment was vast, but the profits were expected to arrive at some more convenient point in the future. At the time, this felt mildly contrarian but, since then, the warning signs appear to have multiplied.

After the US midterm elections, the Nasdaq fell 5.5% in one session. Nvidia, whose chips power much of the AI industry, lost around 9%, while rival chipmaker Advanced Micro Devices did even worse. Investors, having spent several years worrying about missing the boom, seem to have abruptly woken up to the fact that they could also lose money in it. Since then, shares have rebounded somewhat, but there is a lingering question as to whether the spell has been broken for some investors.  

Amongst all of this is a communications lesson, and not a particularly novel one. Expectation management is hardly an exciting new discipline. It may be the oldest (respectable) activity in corporate communications. Just like suggesting we wear seatbelts, familiar advice is not unimportant advice.  

Expectations are useful until they become obligations. The AI industry has promised a transformation so comprehensive that almost any level of investment might be justified. Markets have duly priced in extraordinary growth, rising productivity and abundant future customers. The difficulty with describing the future as inevitable is that people become impatient when it fails to arrive on schedule.

There is another contradiction not working in their favour. AI companies, on the one hand, tell us they have created something so powerful that its development may need to be slowed to save humanity, while simultaneously scouting venues for the IPO party. The two positions appear to sit in direct opposition to each other. One of the more cynical interpretations might be that the largest companies, sensing the threat from rapidly improving open-source models, hope to push governments into regulation that only they can satisfy.

The problem with the great expectations trap is that companies must sound ambitious enough to attract the vast amount of capital, customers, and talent required. Governments want to hear that the technology will deliver prosperity, productivity, and strategic advantage. The public is told that working life, healthcare, education, and almost everything else are about to improve dramatically. Each claim may be defensible on its own. Together, however, they create a standard against which even remarkable progress can look slightly disappointing.

The worrying part is that the scale of the wager is considerable. The OECD calculates that AI companies attracted 61% of global venture-capital investment in 2025, while Goldman Sachs expects worldwide AI investment to exceed $1 trillion this year. Should the market decide the mammoth valuations are not realistic, we all have a problem.

The largest technology companies remain immensely profitable. But investors are no longer asking only whether AI will matter; they are increasingly asking when the expenditure will produce returns, how large those returns might be, and which companies will capture them. This is an altogether less comfortable conversation.

Corporate communicators therefore face a delicate task. If every new product is revolutionary, every investment transformational, and every market opportunity unprecedented, ordinary success begins to resemble failure.

History offers some comfort, though perhaps not for today’s shareholders. AI may still transform economies, just as railways and the internet did, though both endeavours also ruined lots of investors along the way. AI companies must now manage three sets of expectations: investors want returns, governments want strategic capability, and communities want to know what it will cost them. Telling each audience a different version of the future may work briefly, but eventually, the versions will meet.