The news is full of speculation that we are about to see a bursting of a “bubble” in AI. “Will the AI bubble pop?” asks the Sun. The New York Post predicts that the entire market is about to “burst,” with a market correction to follow. But it isn’t just tabloids getting carried away – the IMF is concerned about an AI correction, as are luminaries from the financial world such as Ray Dalio and Michael Burry. Burry, who famously predicted the 2008 financial crisis, has denounced Anthropic and OpenAI (the creators of Claude and ChatGPT) as “bubble companies.”
And there is already some evidence in the stock market of worries about AI infrastructure spend. CoreWeave, an American data center company backed by Nvidia which has managed to rack up $50 billion in debt, recently saw its stock price drop by a quarter. CoreWeave, like many companies in the artificial intelligence industry, has never turned a profit, and is instead practicing “cash burn,” betting billions of dollars on AI becoming so valuable that they can set very high leasing prices for the companies using it.
So how concerned should we be? We won’t really know until the market reaches a decision on how much Anthropic and OpenAI are worth when they have their IPOs, selling their shares to the public, soon.
Of the two, Anthropic is expected to go first, and there are already several warning signs it will miss its $2 trillion valuation target. Reuters reported last week that investors may cut Anthropic’s valuation by $500 billion due to concerns over profitability. Like CoreWeave, Anthropic has burned through billions of dollars. According to its leaked prospectus the company made an operating loss of $8 billion last year, rising from $3 billion the year before. The document also showed that Anthropic has made $518 billion in commitments to build data centers and buy services over the next decade.
This enormous investment has proved effective at increasing revenue which in the second quarter of this year was 14 times what it was a year earlier, and the company has told shareholders it is making an (adjusted) operating profit. But it will need that growth to continue for years to pay for what it has already promised to spend.
If Anthropic greatly underperforms expectations it may send a signal that AI growth, which has been sustaining the US economy for the past four years, is merely a bubble. This would be a big problem. Around half of the value of stocks in the S&P 500 index of the largest US companies is now AI-related and a third of US economic growth is now from AI. This boom has been masking decline in other parts of the American economy, which appears increasingly fragile. To give one example, total manufacturing construction spending in the United States has fallen by 30 percent since Donald Trump was elected president in November 2024.
Both pension funds and retail traders in Britain are heavily invested in the American stock market due to the historic underperformance of the FTSE 100. The benefits which AI investment is already creating in Britain – the country’s statistic bureau reported unexpected economic growth of 0.7 percent in the three months to August, thanks to a large upswing in AI – could also disappear with it.
If the market is correct, and the benefits of AI have been overstated, this reality will also have knock-on consequences for government spending. Many states, for example the British one, have increasingly pointed to AI a quick fix for unsustainable welfare states. In January last year the British government announced that it would be able to find £45 billion ($60 billion) in annual savings through technology improvements including AI. In the National Health Service (NHS), the government has been trying to sharply reduce the scale of staff recruitment over the next decade through an updated workforce plan which would rely on the adoption of technology to make clinicians more productive. The measures will mean that the NHS will have 50,000 more nurses over the next decade rather than up to 190,000. Documents leaked to the Financial Times show that one draft suggested that there will be cases where AI can “completely substitute for a role,” and will be able to take decisions autonomously.
When surveyed last year, only a quarter of British civil servants said that AI was saving them at least an hour a week
The evidence that this is working effectively is mixed. A recent study commissioned by NHS England’s London region found that adoption of AI is shifting work instead of reducing the need for it, as “the time saved in generating outputs [e.g. note writing] can be offset by the time required to verify their accuracy,” therefore suggesting that “there is a risk that AI may intensify work rather than alleviate it, potentially undermining confidence in productivity claims.” When surveyed last year, only a quarter of British civil servants said that AI was saving them at least an hour a week. The figures in Chancellor of the Exchequer John Healey’s Budget will be partly premised on efficiency savings that AI may never deliver, meaning old arguments about distribution will have to be had once more.
We can’t know yet whether the AI bubble is about to pop, but perhaps a more interesting question is whether or not we want it to. In the short run it would mean a lot of economic damage, more taxes and more pressures on public services in the medium term. On the other hand, if AI is not all it’s cracked up to be, at least we won’t have to worry quite so much about being turned into paperclips by a vengeful ChatGPT.
Comments