But Burry poses a simple question: what if the AI excitement fails to live up to expectations? What if all the spending – most of it fuelled with debt – is overblown?
Being forced to write down such a massive investment would deliver a huge blow to the profits of these companies, which have a combined valuation of around US$13t ($23t).
Yet in a worrying sign for investors, Burry’s analysis suggests these companies are on the hook for far more.
To make matters worse, he shows how they have kept these commitments off their balance sheets under controversial accounting rules.
“A company can sign a non-cancellable 20-year data centre lease, disclose it in a footnote, and carry nothing on the balance sheet until the keys turn and the data centre boots up,” says Burry.
This has allowed these five AI giants to sign US$1.2t of leases on future data centres, of which at least US$857b is non-cancellable, but keep those liabilities off their balance sheets.
Similarly, the group has made more than US$1.5t worth of promises to pay for items in the supply chain, known as purchase commitments. These do not appear on their balance sheets either but are shown in regulatory filings.
Add in some various special purpose vehicles, guarantees and other items and the total worth of their off-balance-sheet obligations climbs above US$3t. Burry says the pace of growth suggests this figure could reach US$5t by 2028.
“Let’s also remember this past year, the combined earnings of the big five are less than US$400b,” he says.
“This is a gargantuan bet on gargantuan growth. The risk here for all of this is what happens when the music stops.”
Such fears have started to take the gloss off the “magnificent seven” US tech stocks, with Oracle, Microsoft and Meta all falling from their stock market peaks in recent weeks.
“The core argument is whether all of the spending by hyperscalers can generate the necessary return on invested capital to justify it,” says Ross Mayfield, an investment strategist at Baird.
“It’s a fair argument, and one that is reflected in both the magnificent seven and Oracle’s underperformance in 2026.”
Julian Klement, an analyst at Panmure Liberum, says that even if demand for AI does live up to the hype, the effect on the hyperscaler balance sheets of these commitments will be huge.
“Once that off-balance-sheet debt comes on, the interest expense and the amortisation and depreciation charges start to explode for these companies,” says Klement.
“Once these data centres become revenue-generating in the next two years or so, this off-balance-sheet debt will have to come on to the balance sheet and will have to be serviced, including paying interest and amortising the assets acquired with this debt.”
The five hyperscalers picked out by Burry are also showing other signs of strain.
Oracle this week attempted to declare “force majeure” on a massive data centre project in New Mexico, a legal protection designed to shield parties from events that prevent them from fulfilling a contractual obligation.
In the past, there have also been examples of what has happened when the “music stopped” on major investments by tech companies. Meta took a US$5b impairment on its 2022 results when it concluded it had leased too much office space after Covid.
The same could happen with AI data centres. Too little demand will lead to an oversupply of capacity. As Burry says: “Data centres, when they are not needed, will be written down.”
So what if the worst happens and these massive AI investments are, at least partially, written off.
“There will be a fairly large hit to their earnings,” says Chris Clothier, the co-chief investment officer at CG Asset Management.
“Let’s say that we get halfway or all the way through this monumental capex build out, trillions and trillions of dollars spent.
“Most of it has been financed with debt. Then all of a sudden we see that the revenues aren’t going to be there to generate reasonable returns on those investments.
“The next thing that’s going to happen is that the valuations of the frontier AI labs, OpenAI and Anthropic, are going to collapse, whether publicly or privately. Of course, the hyperscalers and the chip companies are themselves very large investors in those businesses.
“And then on top of that, you’re then required to bring a bunch of liabilities that are currently off balance sheet onto the balance sheet.”
Balance sheet trickery
In response, Klement expects more accounting tricks from the AI hyperscalers.
“Microsoft announced it would shift future data centre leases from finance leases to operating leases,” he said.
“This means it can depreciate the leases at a lower rate, and the leases no longer show as cash out from investing activities but as cash out from operating activities in the cash-flow statement.
“The result of this accounting change is that its projected capex drops from US$190b to US$175b without the company actually changing its spending by one dollar.
“I wouldn’t be surprised if other companies followed suit, with their accounting practices in coming quarters.”
For good measure, Burry also points out in his essay that more than US$400b has been spent by the big five AI companies on construction that is already in progress.
These items do appear on balance sheets but are not shown as depreciating in value or as an expense because they have not yet been put into service.
“This is on the balance sheet, but simply not costing anything as long as they are not being used,” he says.
A Microsoft spokesman said it will “constantly assess the demand and supply environment” to inform its AI strategy while Amazon declined to comment.
Apple, Meta and Oracle did not respond to requests for comment.
Sign up to Herald Premium Editor’s Picks, delivered straight to your inbox every Friday. Editor-in-Chief Murray Kirkness picks the week’s best features, interviews and investigations. Sign up for Herald Premium here.
