How AI Accounting Practices May Be Inflating Earnings of Tech Giants Like Oracle and Meta

The rapid ascent of artificial intelligence (AI) has transformed the financial landscapes of tech giants like Oracle and Meta, but beneath the surface of their soaring profits lies a contentious accounting strategy. Investors and analysts are increasingly scrutinizing how these companies report earnings, particularly in relation to AI infrastructure. Michael Burry, the investor behind the 2008 housing crisis short, has sounded the alarm: Oracle and Meta may be inflating their profits by manipulating depreciation schedules for AI hardware, a practice that could distort earnings sustainability and mislead value investors, according to a Coinotag analysis.

The Depreciation Dilemma

Depreciation, the systematic allocation of an asset's cost over its useful life, is a cornerstone of financial reporting. However, in the AI sector, where hardware like servers and GPUs becomes obsolete in two to three years, the choice of depreciation timelines carries outsized consequences. Oracle and Meta have extended the useful lives of these assets to five to six years, significantly reducing annual depreciation expenses. For example, Oracle adopted a six-year depreciation schedule in 2024, while Meta incrementally increased its schedule to 5.5 years by January 2025, as noted in a LinkedIn post by Landman and Karny.

This practice is technically permissible under U.S. GAAP, which allows companies to adjust depreciation estimates based on "changes in accounting estimates." Yet, Burry argues that this flexibility is being exploited in a sector defined by rapid technological churn. By spreading the cost of AI infrastructure over longer periods, Oracle and Meta reduce their annual expenses, inflating net income and operating margins. Burry estimates this could lead to a 27% overstatement of Oracle's profits and a 21% overstatement for Meta by 2028, according to the Coinotag analysis.

Capital Expenditures and the Debt-Driven AI Boom

The scale of AI infrastructure investments is staggering. Oracle secured a $38 billion project-finance loan for data centers in Texas and Wisconsin, while Meta raised $30 billion in bonds and partnered with Blue Owl Capital on a $27 billion Special Purpose Vehicle (SPV) for its Louisiana data center. Total AI data center financing surged to $126 billion in 2025, a 500% increase from 2024, as reported in a Blockchain News flash. These figures underscore the sector's reliance on debt to fund growth, but they also raise questions about whether the returns from AI services can justify such massive capital outlays.

Meta's Q3 2025 earnings report illustrates the tension between revenue growth and rising costs. While ad revenue hit $51 billion, operating margins contracted by 300 basis points to 40% due to AI-related expenses, as noted in a Morningstar analysis. The company projects $100 billion in 2026 capital expenditures, nearly all directed toward AI infrastructure, according to the Morningstar analysis. For value investors, the challenge lies in discerning whether these expenditures will translate into sustainable revenue streams or become a drag on profitability.

Earnings Sustainability and Value Investing Risks

The core issue for value investors is earnings sustainability. Burry's analysis suggests that if AI hardware were depreciated over three years instead of the stated assumptions, the pre-tax profit impact for companies like Oracle could be substantial, according to the Coinotag analysis. This raises concerns about the reliability of reported metrics such as EBITDA and net income, which are critical for valuation models.

Moreover, the sector's reliance on aggressive depreciation assumptions creates a feedback loop: inflated earnings justify higher valuations, which in turn justify further capital expenditures. This dynamic is particularly risky in a sector where technological obsolescence is inevitable. For instance, Oracle's recent $18 billion project-finance loan for AI data centers in Texas and Wisconsin highlights the scale of bets being made, but also the potential for stranded assets if AI hardware becomes outdated faster than expected, as noted in the Blockchain News flash.

The Broader Implications

The debate over AI accounting practices extends beyond Oracle and Meta. Burry estimates that the AI industry as a whole could understate depreciation by $176 billion from 2026 to 2028, a figure that could reshape financial reporting standards, according to the Coinotag analysis. Regulators and auditors may need to revisit GAAP guidelines to address the unique challenges of AI infrastructure, where economic obsolescence outpaces accounting assumptions.

For value investors, the lesson is clear: traditional metrics like P/E ratios and EBITDA margins may no longer be reliable in the AI era. Investors must dig deeper into companies' capital structures, depreciation policies, and the alignment between reported earnings and economic reality. Oracle and Meta's strategies, while legally defensible, highlight the need for a more nuanced approach to evaluating tech stocks in an age of rapid innovation.

Conclusion

The AI revolution is reshaping the financial landscape, but it also introduces new risks for investors. Oracle and Meta's depreciation practices, while compliant with current accounting standards, may obscure the true costs of maintaining cutting-edge infrastructure. As the sector races to build out AI capabilities, value investors must remain vigilant, scrutinizing not just the numbers but the assumptions behind them. In a world where earnings can be as much about accounting choices as operational performance, the line between innovation and illusion grows increasingly thin.