Is artificial intelligence the 21st-century version of the dotcom crash? Last week, Andrew Bailey, the governor of the Bank of England, spelled out the fear that is keeping global policymakers awake at night.
Writing an open letter in his capacity as chair of the global Financial Stability Board, Bailey warned about the risks to cybersecurity posed by frontier AI models. But alongside this, he highlighted the signs of an old-fashioned bubble. Markets, he wrote, “remain vulnerable to a potentially disorderly correction that could spread across borders” with one particular concern being “stretched asset valuations (particularly artificial intelligence-related investments)”.
Central bankers often speak this way, but this particular warning was not especially hard to unscramble. It is quite easy to see those stretched valuations that the governor was writing about. By most measures, AI stocks look expensive relative to their longer-term history, which is usually a sign of excessive optimism.
Suggested Reading
Ukraine is 3D-printing its way to victory
And that optimism is looking extremely excessive. When it put its shares up for sale, SpaceX claimed that the market for AI services would grow to $28.5tn over the coming years. It was a preposterous figure.
Most worryingly of all is the scale of the investment splurge. Last year, the big US tech firms spent $450bn on the chips, data centres, power and other infrastructure required to drive the AI rollout. This year they plan on spending another $900bn and, according to their forward looking plans, another $1.2tn in 2027. These are, by any definition, huge sums of money, all being spent now in the hope of immense profits in the future.
The most worrying comparison is with the dotcom crash, which took place a quarter of a century ago. But the point is not that the technology itself wasn’t transformative. As we can now see, the internet changed everything. But even though the tech eventually lived up to the predictions of its wildest promoters, the dotcom crash was still an immense, global disaster.
Between the spring of 2000 and the autumn of 2002, the Nasdaq index of leading tech shares fell by almost 80% and didn’t regain its previous peak until 2015. This decline dragged everything down with it – the MSCI World Index, a broad measure of global share prices across sectors, dropped by 50% over the same period.
In the aftermath of the bubble bursting, consumer spending fell and investment dried up. The US spent most of 2001 in recession while growth slowed sharply across Europe.
For students of economic history, it was a familiar pattern. Even when a new technology offers big economic benefits – whether that was the internet in the 20th century or the railways in the 19th – investors get overexcited, bid values up to unsustainable levels and overinvest in any firm involved in the new technologies.
So is the same thing happening with AI? Is it all a massive bubble? There is no doubt that generative AI really does work. The enormous advances in the first part of this year triggered large share price falls for software companies, many of which now look vulnerable to AI.
Suggested Reading
Why Trump could order Microsoft to kill Office 365
The worry now is less to do with the actual technology itself and more that, even if it works extremely well, there is no way it’s going to generate revenue and profit fast enough to cover all of that spending. In other words, like the railway and internet booms before it, the AI boom could result in big losses for people who’ve been swept along in the excitement.
The world economy is already wobbling, due to the ongoing conflict in the Gulf, which is pushing up interest rates – the last thing it needs now is a market crash. Plunging share prices tend to be associated with weaker household spending and delayed or cancelled corporate spending. They weaken tax receipts, which puts further pressure on already stretched public finances. And those major investment binges by big tech firms might be excessive, but they have provided a useful source of additional economic demand over the last two years.
The end of any boom, especially one as large as AI, tends to be painful for everyone. Not just those who benefited. Andrew Bailey is right to be concerned.
