Michael Burry, the investor who predicted the US housing crash, has warned that technology companies are building up more than US$3 trillion in hidden AI-related obligations that could destabilise the global economy if the boom fails to deliver.
In an essay, Burry examined regulatory filings from Apple, Alphabet, Microsoft, Meta and Oracle, arguing that the companies’ spending on data centres and related infrastructure has created a huge financial exposure concealed by accounting rules.
The five companies have signed leases worth US$1.2 trillion for future data centres, he said, including at least US$857 billion in non-cancellable commitments. They have also made more than US$1.5 trillion in supply-chain purchase commitments.
When special-purpose vehicles, guarantees and other obligations are included, Burry estimates the total could exceed US$3 trillion. He said the figure could rise to US$5 trillion by 2028 if the current pace of expansion continues.
“This is a gargantuan bet on gargantuan growth,” Burry wrote. “The risk here for all of this is what happens when the music stops.”
The commitments are not fully reflected on the companies’ balance sheets. Burry said a company could agree to a non-cancellable 20-year data-centre lease and disclose it in a footnote, without recording the liability until the facility was operating.
His analysis comes as the so-called magnificent seven technology stocks have lost some of their momentum. Oracle, Microsoft and Meta have all fallen from their stockmarket peaks in recent weeks.
Ross Mayfield, an investment strategist at Baird, said the central question was whether the spending by the large cloud companies could generate sufficient returns to justify the investment.
“It’s a fair argument, and one that is reflected in both the magnificent seven and Oracle’s underperformance in 2026,” he said.
AI data-centre debt could hit company earnings
Julian Klement, an analyst at Panmure Liberum, said the financial impact would become more visible as data centres started generating revenue over the next two years.
“Once that off-balance-sheet debt comes on, the interest expense and the amortisation and depreciation charges start to explode for these companies,” he said.
Even if demand for AI meets current expectations, Klement said the commitments would place significant pressure on the hyperscalers’ balance sheets as the debt had to be serviced and the assets depreciated.
The companies are also carrying more than US$400 billion in construction already under way, according to Burry. Those projects appear on their balance sheets, but are not yet recorded as depreciation or expenses because they have not been put into service.
Burry warned that weak demand could leave the industry with more data-centre capacity than it needs, forcing companies to write down the value of their investments.
Chris Clothier, co-chief investment officer at CG Asset Management, said a failure to generate adequate revenue after the infrastructure build-out could lead to a broad repricing of the sector.
“There will be a fairly large hit to their earnings,” he said, adding that valuations of AI companies such as OpenAI and Anthropic could also fall. The hyperscalers and chip companies are significant investors in those businesses, he said.
Microsoft has said it will continually assess demand and supply when deciding its AI strategy. Amazon declined to comment, while Apple, Meta and Oracle did not respond to requests for comment.
