Michael Burry built his reputation finding the cracks nobody else wanted to look for. He just aimed that same forensic instinct at the five companies most responsible for the AI buildout. His latest target is not a stock price or a valuation multiple. It is the fine print.
In a new Substack post, the investor behind “The Big Short” argues that Amazon, Meta Platforms, Alphabet, Microsoft and Oracle have quietly built up trillions of dollars in obligations that barely show up on their balance sheets, and that Wall Street has largely chosen not to notice.
Burry says hyperscalers are hiding $3 trillion in AI debt
Burry’s math starts with two categories most investors rarely think about. He estimates the five hyperscalers carry nearly $1.2 trillion in uncommenced lease commitments and more than $1.5 trillion in purchase commitments. Once special-purpose vehicles, guarantees and other contingent backstops are added in, he puts the combined total above $3 trillion, Stocktwits reported.
“These liabilities, I say, are, in essence, in hypergrowth mode,” Burry wrote, arguing they are expanding far faster than the companies carrying them.
More AI:
- Nvidia just made a move Wall Street wasn’t ready for
- Microsoft just took sides in AI policy fight
- OpenAI just disclosed something genuinely alarming
The core problem, in Burry’s framing, is a timing mismatch. Data centers typically take three to five years to build and often carry lease terms stretching 13 to 20 years, while the AI chips running inside them can see major shifts in power density and cooling requirements within just 12 to 18 months. That gap, he argues, leaves hyperscalers exposed if hardware architecture changes faster than the buildings around it, Stocktwits reported.
To make that point, Burry cited Microsoft CEO Satya Nadella, who has previously said he did not want the company to get stuck building massive infrastructure around a single generation of hardware. Burry framed that comment as validation that even hyperscalers recognize the risk of committing too much infrastructure to rapidly changing AI hardware.
The accounting fight behind the warning
Part of Burry’s argument centers on more than $400 billion in construction-in-progress assets sitting across the five companies. While still classified as construction-in-progress, those assets do not generate depreciation under GAAP because the projects have not yet entered service.
That structural quirk fits into a broader accusation Burry first raised last November, when he argued that hyperscalers were extending the useful life of Nvidia chips well beyond their realistic two-to-three-year replacement cycle. A move he said could understate industry-wide depreciation by roughly $176 billion between 2026 and 2028, according to CNBC.
He was specific about who that would hit hardest. By his estimate, Oracle’s earnings could end up overstated by roughly 26.9% and Meta’s by about 20.8% by 2028 if the accounting assumptions behind those extended depreciation schedules hold, according to CNBC.
Burry has continued building on that thesis in the months since, reiterating in August that the resulting shortfall could still reach roughly $176 billion across the same three-year window, even as some of his short positions tied to the argument came under significant pressure.
Astrid Stawiarz / Getty Images
Burry’s track record of betting against the AI trade
This latest post did not come out of nowhere. Burry disclosed in August that he was shorting Oracle at roughly $145 a share, citing the company’s ballooning debt load and long-term commitments. He described his AI shorts broadly as “a bit like shooting fish in a barrel,” adding that AI companies had gotten very fat and very large.
He paired that Oracle position with a larger short in Nebius Group at $211.77 a share, pointing to growing leverage risk tied to off-balance-sheet commitments across the tech and cloud sector more broadly, as reported by TheStreet.
Burry has also raised concerns about circular financing, arguing that capital moving among hyperscalers, AI labs and chipmakers could artificially reinforce demand and revenue in ways that make the whole ecosystem look healthier than the underlying economics justify.
Earlier this year, Burry went further, calling AI enthusiasm a kind of mass addiction after shorting Micron following a near 700% year-over-year run in the stock. His argument: hyperscalers are spending heavily because chip stocks keep rising, and chip stocks keep rising because hyperscalers keep spending. A feedback loop that looks like demand without necessarily being one, according to TheStreet.
What investors should take from the warning
Burry has been careful to frame this as a warning about downside risk rather than an allegation of fraud, telling readers that “when the music’s over, these off-balance sheet commitments become real liabilities very quickly.” A line meant to capture how fast impairments, termination costs and unused capacity could pile up if AI demand ever slows.
That warning arrives at an odd moment for the stocks it targets. Cloud growth remained strong across the hyperscalers last quarter, capital expenditures are expected to cross $1 trillion next year, and the Roundhill Magnificent Seven ETF was heading for a third consecutive month of gains even as Burry’s post circulated.
For investors, the gap between those two realities, strong reported growth on one side and Burry’s balance sheet forensics on the other, is likely to remain the central tension in the AI trade for as long as hyperscalers keep spending at this pace without the depreciation and liability questions being fully resolved.
Related: Michael Burry’s blistering message on economic inequality
