Michael Burry, the investor famed for predicting the 2008 subprime mortgage crisis, warned on Monday that several tech giants are inflating profits by underestimating the depreciation cycles of AI hardware, calling it "one of the most common forms of financial fraud today."
Burry singled out major cloud computing firms—including Meta Platforms, Inc., Oracle, Microsoft, Amazon, and Alphabet—on social media platform X, accusing them of setting depreciation periods for AI computing equipment at five to six years when the actual useful life is just two to three years. He estimated that between 2026 and 2028, these companies could collectively understate depreciation by about $176 billion, with Oracle and Meta potentially overstating profits by 27% and 21%, respectively, by 2028.
Burry pointed out that while tech giants are aggressively investing in NVIDIA's rapidly evolving AI hardware, they are simultaneously extending the accounting depreciation periods for these assets in their financial statements. This accounting practice, he argued, contradicts the reality of rapid hardware obsolescence. He promised to release more details on November 25.
Under Generally Accepted Accounting Principles (GAAP), companies can spread the cost of upfront payments for large assets—such as GPUs and servers—over several years based on their estimated depreciation rates. If a company assumes a longer useful life for an asset, it can reduce annual depreciation expenses, thereby minimizing the impact on reported profits.
Burry has also previously cautioned that the current AI hype resembles the tech bubble of the late 1990s.
His warning comes as tech companies ramp up unprecedented investments in AI. Alphabet has raised its 2025 capital expenditure forecast to $91–93 billion; Microsoft spent a record $34.9 billion last quarter; and Meta expects expenditures of $70–72 billion this year, nearly double last year's figure. Bank of America projects global cloud computing capital expenditures will reach $611 billion by 2026, $145 billion higher than its forecast a month ago.
Burry's stance aligns with warnings from Jim Chanos, the short-seller known for exposing the Enron scandal, who cautioned that AI infrastructure investments could become burdens if revenue growth slows. Research from Princeton University's Center for Information Technology Policy also notes that AI chips typically become obsolete in one to three years, yet companies commonly depreciate them over five to six years. Microsoft recently shortened the depreciation period for some equipment from six to five years, estimating a $2.2 billion impact on operating profits.
However, some analysts dismissed Burry's claims. Wedbush analyst Dan Ives called his view "completely misjudging the AI trend," while Futurum Group CEO Daniel Newman argued that depreciation methods are merely a matter of financial period allocation, not fraud.
Despite the controversy, demand for older NVIDIA chips remains strong. CoreWeave CEO Michael Intrator revealed that a client recently renewed a contract to continue using H100 chips launched in 2022, with pricing only 5% lower than the original agreement—suggesting some hardware may indeed have extended lifespans.