Michael Burry has some concerns about AI accounting

Michael Burry has criticized Oracle and Meta's accounting for AI GPUs, alleging that extending the useful life of these rapidly evolving assets artificially inflates earnings by understating depreciation. While Burry argues against this practice given the quick product cycles of chips like Nvidia's, the article presents counter-evidence from companies such as Nebius and CoreWeave, which report strong continued demand and profitability from older GPU generations, often repurposed for inference tasks. This debate highlights the complexities of asset depreciation in the capital-intensive AI sector, with short-seller Jim Chanos further emphasizing the thin profitability of some AI infrastructure providers even with extended depreciation schedules.

Analysis

Michael Burry, known as "The Big Short" investor, has raised concerns regarding the accounting practices of Oracle (ORCL) and Meta (META) concerning their AI GPUs. He alleges that extending the useful life of these assets, despite rapid product cycles for chips like Nvidia's (NVDA) 2-3 year cycle, artificially inflates earnings by understating depreciation, which he views as a common modern accounting issue. This critique highlights the significant capital expenditure boom in AI and its impact on reported S&P 500 profit growth, where costs are depreciated over time rather than expensed immediately.

However, the article presents counter-evidence suggesting a more nuanced reality for GPU longevity. Industry players like Nebius (NBIS) and CoreWeave (CRWV) report strong continued demand and re-contracting for older GPU generations (e.g., Hopper, Ampere), often at favorable pricing, particularly when repurposed for inference tasks. Oracle itself benefits from revenue generated by older Nvidia chips, which appear to bolster its margins, indicating a potentially longer economic life than initially assumed for certain applications.

The debate underscores the complexity of asset depreciation in the capital-intensive AI sector, where the true "useful economic life" of rapidly evolving hardware is uncertain. While early evidence points to physical durability and demand for non-flagship compute, short-seller Jim Chanos further questions the underlying profitability of some AI infrastructure providers. He notes that even with an aggressive 10-year depreciation life for CoreWeave's estimated $20 billion in GPUs, their annualized adjusted EBITDA of $3.4 billion barely covers annualized interest of $1.2 billion and depreciation of $2.0 billion, suggesting thin margins and validating the mixed sentiment around the sector's financial transparency.

Key Decisions for Investors

• Investors should critically evaluate the depreciation schedules and capital expenditure intensity of AI infrastructure companies like Oracle and Meta, assessing whether reported earnings accurately reflect the rapid obsolescence risk of AI hardware.
• Monitor company-specific earnings calls and reports for evidence of continued demand, re-contracting rates, and pricing power for older generation GPUs, as this data can provide insights into the actual economic life and residual value of these assets.
• Consider the potential for increased scrutiny from short sellers and analysts on companies with significant AI GPU investments, as accounting practices in this evolving sector could become a source of market volatility and re-valuation.