Michael Burry Warns Hyperscalers Carry $3 Trillion in AI Risk

Investor Michael Burry warns that major tech hyperscalers have amassed more than $3 trillion in off-balance-sheet AI commitments, threatening Wall Street's bullish consensus as infrastructure spending pushes toward $1 trillion.

### Tape / Session Read Market breadth across major indices reflects ongoing adjustments to artificial intelligence sector valuations, with broad tape indicators showing minor moves as analysts digest infrastructure risk exposure. Broad tape indices recently registered modest movement, including the Dow Jones Industrial Average rising 0.35%, the S&P 500 advancing 0.41%, and the Nasdaq-100 climbing 0.47%.

### Infrastructure Exposure & Off-Balance-Sheet Liabilities Burry estimates that five primary hyperscalers carry nearly $1.2 trillion in uncommenced lease commitments and more than $1.5 trillion in purchase commitments. When including special-purpose vehicles, guarantees, and contingent backstops, total hidden obligations exceed $3 trillion.

Data-center infrastructure typically requires three to five years to construct and involves lease terms stretching 13 to 20 years, whereas artificial intelligence accelerators experience power density and cooling shifts within 12 to 18 months. This temporal mismatch could leave hyperscalers holding expensive, specialized facilities that are difficult to repurpose if demand slows or chip architectures shift.

Furthermore, Burry highlighted more than $400 billion in construction-in-progress assets across these five firms. Because these assets have not yet entered service, they currently generate zero GAAP depreciation, creating a risk that hardware may economically depreciate while sitting idle in warehouses.

### Why This Lane Matters The ballooning capital expenditure cycle and unacknowledged lease commitments challenge traditional valuation models across the tech sector. Analysts overlooking economic depreciation and maintenance capital expenditures risk overpricing equities heavily exposed to AI infrastructure scaling.