The AI profit boom: real breakthrough or mirage?

Forecasts point to sharply higher earnings, yet signs of strain are already appearing beneath the surface.

?



AI and technology shares rose early this week after Anthropic released its IPO prospectus. The filing showed that the Claude chatbot maker generated more than £16 billion in revenue during the first half, a figure that encouraged investors, according to Swissquote's Ipek Ozkardeskaya.

Cracks are showing in parts of the AI sector. Bloomberg reports that the cost of insuring CoreWeave's debt recently climbed to reflect "a more than 60% chance of default within the next five years." CoreWeave, a "neocloud" data-centre company, owns and operates assets that may lose value quickly while carrying heavy debt, according to Victor Balfour of Rothschild & Co. The shares have fallen 67% from their June 2025 peak.

That does not necessarily mean trouble for AI as a whole. The major technology "hyperscalers" appear better positioned: Microsoft shares have climbed 38% since the start of July. They are taking on substantial debt to fund AI investments, but their strong operating cash flows and balance sheets remain "reassuring."

Stijn Van Nieuwerburgh of Columbia University estimates that investment in AI data centres and the infrastructure supporting them will reach £10.3 trillion between 2025 and 2032. That amounts to 3.63% of US GDP each year. The late-1990s dotcom boom reached only 1.1% of annual GDP in capital spending, while the 1870-1890 US railroad mania had held the historical record at 2.24%.

As the credit boom continues, Van Nieuwerburgh warns that financing is shifting into opaque arrangements such as joint ventures, private credit, securitisation, special-purpose vehicles, lease commitments and loan guarantees. The result is a much murkier picture of where dangerous credit risks are building up.

AI profits bubble?

The Nasdaq 100 is trading at a forward price-to-earnings ratio just above 20 times. That is expensive, though it does not yet look like an outright bubble. Edward Chancellor makes a different point on Breakingviews: "the bubble is in profits, not valuations." Earnings at America's biggest blue-chip companies have surged. In the 12 months to July, S&P 500 earnings per share climbed 26%, compared with a long-term average of 7%. Morgan Stanley expects the 42 companies most exposed to AI to increase profits by 67% this year.

But the AI profit boom may be little more than an accounting "mirage." Microsoft can record an immediate sale when it buys a chip from Nvidia, while Microsoft records the purchase as an "asset" and spreads its depreciation over several years. If those chips fail to produce the promised earnings bonanza, the hyperscalers will eventually face the bill: "by the end of the decade annual depreciation charges will range between £500 billion and £1 trillion."

Earnings bubbles are not new. US housebuilders experienced one at the 2006 peak of the property bubble. If US profit growth falls back to its historical mean while data-centre operators and AI frontier labs begin reporting losses, the damage could "metastasise through the financial system."