Burry Warns of $3 Trillion in Hidden AI Infrastructure Debts

Michael Burry, the investor known for predicting the 2008 housing crash, has issued a stark warning about the financial exposure of major technology firms. In a recent blog post, he argues that Amazon, Meta, Alphabet, Microsoft, and Oracle have collectively accumulated more than $3 trillion in liabilities related to artificial intelligence infrastructure. These obligations are largely off-balance-sheet, meaning they do not appear as standard debt on the companies' primary financial statements.

The concern lies in how Wall Street is currently valuing these tech giants. Burry contends that market models are failing to account for the sheer scale of these commitments, treating them as minor operational details rather than significant financial risks. According to reports from Yahoo Finance, this disconnect leaves investors potentially blind to a major structural weakness in the AI boom.

Off-balance-sheet commitments exceed three trillion dollars

Burry breaks down the $3 trillion figure into specific categories of financial obligation. He estimates that the five hyperscalers have nearly $1.2 trillion in lease commitments that have not yet begun, along with more than $1.5 trillion in purchase commitments for hardware and services. When adding in special-purpose vehicles, guarantees, and other contingent obligations, the total exposure grows significantly beyond what is visible in standard balance sheets.

The investor describes these liabilities as being in hypergrowth mode, expanding at a pace that outstrips the revenue growth of the companies themselves. This rapid accumulation of debt-like obligations creates a fragile foundation for future earnings, especially if the expected returns from AI investments fail to materialize as projected by optimistic analysts.

Mismatch between infrastructure lifespan and chip cycles

A core part of the warning involves the physical mismatch between data center construction and technology evolution. Building a data center typically takes three to five years, and lease terms often stretch from 13 to 20 years. In contrast, AI hardware such as accelerators sees major changes in power density and cooling requirements every 12 to 18 months. This timeline gap means facilities built for one generation of chips may become obsolete before the debt on them is paid off.

Microsoft CEO Satya Nadella has previously expressed concern about building massive infrastructure for a single generation of hardware. Burry argues that if AI demand slows or chip architectures shift, these specialized facilities will be difficult to repurpose. The result could be a large stock of expensive, underutilized real estate that generates no return while still carrying substantial lease and maintenance costs.

Zero depreciation masks true economic cost

Burry also highlights a specific accounting issue involving construction-in-progress assets. The five companies have more than $400 billion in assets that are currently under construction. Because these assets are not yet in service, they do not generate depreciation under Generally Accepted Accounting Principles (GAAP). This creates a misleading picture of profitability, as the economic value of the hardware is declining while the books show zero cost.

The investor notes that chips may be losing value in warehouses while accounting rules keep their depreciation at zero. This disconnect means that the true economic cost of the AI buildout is hidden from standard financial metrics. Until these assets are fully deployed and start generating revenue, the market may be overestimating the financial health of these technology leaders.