The debate over the depreciable lives of GPUs and data-center equipment — a niche topic generally reserved for technical accounting teams — moved into the mainstream media after Michael Burry publicly criticized Nvidia and suggested that large cloud providers that operate hundreds of data centers – often referred to as hyperscalers – were overstating earnings by depreciating AI hardware over unrealistically long periods. Between 2020 and 2024, many large technology companies steadily extended the useful lives of their servers and networking equipment — only for the trend to diverge in 2025, when Amazon shortened the useful life of a subset of servers while Meta extended its estimate further.

Nvidia pushed back, arguing that customers consistently use four-to-six-year depreciable lives based on observed utilization and longevity, according to CNBC.¹

The reason this debate matters is simple: the CAPEX numbers are material to investors. Burry estimated that if major cloud providers shortened GPU lives from the four-to-six-year depreciation schedules currently used to something closer to the two-to-three-year hardware replacement cycle he believes reflects real-world economic life, the cumulative impact on reported earnings could exceed $176 billion for years 2026–2028.²

Even small changes in useful life assumptions may increase depreciation by billions of dollars per year for the largest companies, such as Amazon and Meta. With AI infrastructure now representing one of the fastest-growing spending categories³, the accounting treatment of servers is directly relevant in understanding the depreciation expenses and the corresponding net income numbers. 

My Substack Deep Quarry post explores the key accounting considerations and disclosure nuances that must be understood before drawing conclusions about the impact of GPU depreciation lives on financial statements.

• Extension of useful lives is not “fraud” and is not a correction of an error. Changes in depreciable lives are treated as changes in accounting estimates under ASC 250, not corrections of errors. Under GAAP, management revises useful lives when new information, suggesting that the assets will remain in service for longer, becomes available. Changes in estimates reflect updated assumptions that must be supportable and auditable.
• Useful life assessments are company-specific. GAAP requires useful life to reflect how each entity expects to use its assets, which means two companies can arrive at different, yet compliant, estimates when operating similar hardware. Differences in workloads, cooling systems, and data-center design may lead to variation in useful lives between hyperscalers.
• Straight-line depreciation is an industry convention, not a GAAP mandate, and may not always reflect actual economic consumption. While hyperscalers generally use straight-line depreciation for servers, GAAP allows any systematic and rational method, including accelerated depreciation, if it better reflects how an asset’s economic benefits are consumed. Recent reporting has noted that certain AI hardware, such as Nvidia’s H100 systems, can lose a substantial portion of their resale value within the first few years, with units trading at less than half the price of new ones by year three⁴. When value declines more steeply early in an asset’s life and then stabilizes, an accelerated method may better approximate economic reality than a straight-line schedule spread evenly over five or six years.
• Construction in progress (CIP) signals future depreciation pressure. Hyperscalers have more than $100 billion in assets classified as CIP, including those related to new data center expansion. According to GAAP, these balances are not depreciated until the assets are placed in service, meaning today’s depreciation expense reflects past investment cycles and may not fully reflect the current wave of AI-driven infrastructure spending. Given the materiality of AI-related CAPEX spending and a relatively long period required to place some of these assets in service, the CIP balances are increasingly significant for forecasting depreciation trends.

This is an abridged version of the analysis. The full Deep Quarry post — available to Substack Deep Quarry subscribers — explores the detailed history of server and GPU useful-life changes across major hyperscalers, the GAAP mechanics behind those estimates, common myths highlighted by the Nvidia–Burry debate, and how accelerating AI hardware cycles and large construction-in-progress balances may shape future depreciation and reported earnings.

For questions and data inquiries please contact olga@deepquarry.com.

Investment, Tax and Legal Disclaimer: This article is for informational purposes only and does not constitute investment, tax or legal advice. The content contained herein is not to be relied upon as the basis for any investment or other decision. Nothing herein should be construed as a solicitation, recommendation, endorsement, or offer to buy or sell any particular security, product, or service. The author has not taken into account the specific investment objectives, financial situation, or particular needs of any specific person who may read this material. Investing involves inherent risks, and there can be no guarantee that any investment or company mentioned will be suitable or profitable for any investor's investment portfolio. Readers are strongly advised to conduct their own thorough research and consult with a qualified and licensed financial professional and legal counsel before making any investment decisions. Past performance is not indicative of future results. 

Opinion Disclaimer: The opinions and views expressed in this article are those of the author and the parties quoted and not necessarily those of The National Law Review or its Guest Contributors.

Footnotes

1 “Nvidia name-checks Michael Burry in secret memo pushing back on AI bubble allegations”, CNBC, November 25, 2025, retrieved on December 9, 2025. 

2 Michael Burry on Twitter, November 10, 2025, retrieved on December 9, 2025.

3 “Is AI already driving US growth?”, JP Morgan, September 9, 2025, retrieved on December 9, 2025.

4 “The Accounting Uproar Over How Fast an AI Chip Depreciates”, WSJ, December 8, 2025, retrieved on December 9, 2025.