Michael Burry, the investor who profited from the 2008 mortgage collapse, is now targeting the largest AI infrastructure spenders in the United States. Through his advisory firm Scion Asset Management, Burry has built a short position tied to his claim that major technology companies are using depreciation accounting changes to hide roughly $176 billion in real costs. He labels the practice “a fraud,” arguing that Meta, Alphabet, Microsoft, and Amazon have all quietly extended the estimated useful lives of their servers and networking equipment, reducing reported expenses by billions of dollars at a time when AI-driven capital spending is accelerating.
Coordinated depreciation shifts across four hyperscalers
Between fiscal 2023 and fiscal 2025, each of the four largest cloud and AI infrastructure operators changed how long it expects its hardware to last, and each change cut reported costs. Microsoft moved first, increasing server and network equipment useful lives from four years to six years effective fiscal 2023, which lifted both operating income and net income for that period. Alphabet followed later that year, changing its server estimate from four years to six years and certain network equipment from five years to six years, according to its Q4 2023 earnings exhibit filed with the SEC. Amazon extended server lives from five to six years effective January 1, 2024, then partially reversed course by shortening a subset of servers and networking equipment from six years back to five years effective January 1, 2025, according to its 10-K for the fiscal year ended December 31, 2025.
Meta completed the pattern by extending estimated useful lives of certain servers and network assets to 5.5 years effective fiscal 2025. That single change, disclosed in its year-end filing, reduces full-year 2025 depreciation expense by approximately $2.9 billion based on assets in service as of December 31, 2024. Each adjustment follows the same logic: by assuming hardware lasts longer, the company spreads the same purchase price over more quarters, lowering the annual expense that hits the income statement.
How longer asset lives inflate reported earnings
Depreciation is not a cash expense. It reflects the cost of wearing out equipment over time. When a company decides its servers will last six years instead of four, the annual charge drops by a third on paper, even though the actual cash was spent years earlier. Across four firms that collectively dominate global cloud computing and AI training infrastructure, the cumulative effect is large enough to move sector-wide profit margins. Burry’s $176 billion figure, cited in secondary reporting, represents his estimate of the total economic costs being obscured by these schedule changes. No primary filing or direct Burry statement available in the public record breaks down that calculation, and the figure does not appear in any of the four companies’ SEC disclosures.
What the filings do confirm is the direction and scale of the benefit. Meta alone expects $2.9 billion in lower depreciation for 2025 from its change. Microsoft’s fiscal 2023 shift produced disclosed increases in operating and net income. Alphabet’s adjustment reduced depreciation expense and raised net income for the periods covered. These are not contested numbers. They come directly from audited or reviewed financial statements filed with the SEC.
Margin compression risk after the accounting tailwind fades
Burry’s core thesis is that these benefits are temporary and largely optical. As AI spending continues, the hyperscalers must still pay cash for new data centers, GPUs, and networking gear. Extending useful lives does nothing to slow that outflow. Instead, it delays recognition of the cost on the income statement, front-loading an earnings boost and back-loading expense recognition into later years. If AI-related revenue growth slows or pricing power weakens just as the depreciation tailwind normalizes, reported margins could contract sharply.
That risk is magnified by the scale of current capital expenditure plans. The largest platforms are committing tens of billions of dollars annually to build and refresh AI infrastructure. Under longer useful-life assumptions, each new wave of spending adds to a larger base of assets being depreciated over more years. In a benign scenario of sustained high growth, the gap between cash outlays and accounting expense may not trouble investors. In a downturn or even a deceleration, however, the same gap could expose how dependent recent earnings were on accounting choices rather than underlying economics.
From Burry’s perspective as a short seller, the setup is asymmetric. If AI demand remains explosive and cloud profits expand fast enough to absorb higher future depreciation, his thesis could take years to play out or fail outright. But if any combination of weaker enterprise budgets, higher competition, or regulatory pressure slows the top line, the reversal of these accounting benefits could amplify downside in earnings and, by extension, in share prices. The fact that four separate companies adopted similar changes within a short window strengthens his narrative that this is a sector-wide phenomenon rather than an idiosyncratic decision by one management team.
None of this means the companies have violated accounting rules. Management is allowed, and in some cases required, to revise useful-life estimates when technology or maintenance practices change. Cloud operators argue that newer server architectures, better cooling, and more efficient workloads justify longer lives. The filings themselves describe these changes as reflecting updated expectations, not as efforts to manage earnings. Auditors have signed off on the disclosures, and regulators have not alleged misconduct.
The debate instead centers on how investors should interpret the numbers. Burry calls the practice “a fraud” because he believes it masks the true economic cost of sustaining AI infrastructure at current scales. Company executives counter that the revisions simply align depreciation with reality. For shareholders, the practical takeaway is less about labeling and more about adjusting their models. Cash flow, capital intensity, and the trajectory of future depreciation charges may matter more than near-term earnings beats flattered by longer asset lives.
As AI reshapes the technology landscape, the accounting for the hardware behind it is becoming a battleground. Whether Burry’s $176 billion warning proves prescient or overstated will depend on how long the AI investment boom lasts-and how much of today’s profit strength survives once the depreciation math stops working in the hyperscalers’ favor.
Warren Cohen is a finance writer based in Phoenix, Arizona, covering personal finance topics including credit, banking, and beginner investing. He earned his degree in business administration from Arizona State University and began his career working in consumer finance, where he gained direct experience with lending and credit systems. He now writes for personal finance websites and fintech platforms, focusing on clear, practical content that helps readers make informed financial decisions.
