Insights Article
Get the latest insights and updates delivered to your inbox.
The most consequential earnings quality debate of the artificial intelligence capital expenditure cycle is being fought over a single line item that almost no investor reads carefully: the estimated useful life of servers and network equipment. In a series of public posts beginning November 11, 2025, the investor Michael Burry—best known for his pre-crisis short of the United States housing market—accused five public companies of having understated depreciation expense across the period from 2026 through 2028 by approximately $176 billion, with the result that, in his estimate, Oracle Corporation’s reported earnings for 2028 will be overstated by 26.9 percent and Meta Platforms, Inc.’s by 20.8 percent.¹ The companies named in those public statements were Microsoft Corporation, Meta Platforms, Alphabet Inc., Amazon.com, Inc., and Oracle.² CNBC, Bloomberg, and Fortune have each reported on the allegations and confirmed the directional movement of the underlying disclosure changes from primary filings.³
Buxton Helmsley does not adopt or endorse Mr. Burry’s specific dollar estimates, and would caution against doing so without a closed-form model that ties projected hardware vintages to amortization schedules at each named company. The forensic point of interest is different. Depreciation policy has become a material—and underanalyzed—driver of reported earnings at the largest public companies in the United States, precisely at the moment when those companies are deploying the largest concentrated capital expenditure program in the history of the public equity markets. The framework that follows is intended to assist institutional investors, audit committee members, and limited partner allocators in reading hyperscaler depreciation disclosures with the rigor the moment requires.
Under United States generally accepted accounting principles, the useful life of a long-lived asset is an estimate, not a fact.⁴ ASC 360-10-35 requires depreciation to be recognized over the period during which the asset is expected to provide economic benefit to the entity. When that estimate is revised, ASC 250-10-45-17 specifies that the revision is treated as a change in accounting estimate and applied prospectively—that is, only in the period of change and future periods.⁵ Prior periods are not restated; comparability is not preserved. ASC 250-10-50-4 requires that the effect on income from continuing operations, on net income, and on any related per-share amounts be disclosed for the current period when a change in estimate affects several future periods, such as a change in the service lives of depreciable assets.⁶
The structural consequence of that framework is straightforward. If an issuer extends the useful life of an asset class from four years to six years, reported operating income increases in the period of change and in subsequent periods, while there is no corresponding adjustment to historical results that would allow an investor to compare current earnings to a prior baseline on like-for-like terms. The change is real for the income statement and invisible to the comparative analysis that institutional investors typically perform when evaluating earnings growth. Depreciation, in this regard, is not a number—it is a position.
Get the latest insights delivered to your inbox.
The most useful exercise for the institutional reader is to set Mr. Burry’s allegations to one side and look at what the named companies have themselves disclosed.
Microsoft Corporation announced on its fiscal fourth quarter 2022 earnings call, held July 26, 2022, that effective at the start of fiscal year 2023 the depreciable useful life of server and network equipment assets in the company’s cloud infrastructure would be extended from four years to six years. Chief Financial Officer Amy Hood stated on that call that the change was expected to favorably affect fiscal year 2023 operating income by approximately $3.7 billion, with approximately $1.1 billion of that benefit recognized in the first quarter alone.⁷
Alphabet Inc. announced in its fourth quarter 2022 earnings release, dated February 2, 2023, that the company had completed an assessment of the useful lives of its servers and network equipment and was adjusting the estimated useful life of servers from four years to six years and the estimated useful life of certain network equipment from five years to six years, effective in fiscal year 2023, with an expected reduction in depreciation expense of approximately $3.4 billion for the full year.⁸ The actual full-year 2023 effect of that change, as later reported in Alphabet’s fourth quarter 2023 earnings release dated January 30, 2024, was a reduction in depreciation expense of $3.9 billion and a corresponding increase in net income of $3.0 billion, or $0.24 per diluted share.⁹
Meta Platforms has, by the public record, extended the useful life of certain servers and network assets in successive estimate revisions, culminating in an extension to 5.5 years effective January 1, 2025. Meta disclosed in its fourth quarter 2024 earnings release, dated January 29, 2025, that based on the servers and network assets placed in service as of December 31, 2024, the change in accounting estimate was expected to reduce full-year 2025 depreciation expense by approximately $2.9 billion.¹⁰ Meta’s annual report on Form 10-K for the fiscal year ended December 31, 2025, attributes a portion of the reported decrease in the depreciation growth rate to the useful-life extension.¹¹
Amazon.com, Inc. provides the most informative case in the public record, because Amazon has moved in both directions on the same asset class within twelve months. In its fourth quarter 2023 earnings release, dated February 1, 2024, the company disclosed that effective January 2024, it was extending the useful life of its servers from five years to six years, an action that Chief Financial Officer Brian Olsavsky stated on the corresponding earnings call was expected to benefit first quarter 2024 operating income by approximately $900 million.¹²
One year later, Amazon disclosed in its fourth quarter 2024 earnings release, dated February 6, 2025, that effective January 1, 2025, it was changing the useful life of a subset of its servers and networking equipment from six years back to five years, an action that the company expected to reduce 2025 operating income by approximately $0.7 billion.¹³ Amazon further disclosed that, in connection with a decision to retire certain equipment early, the company had recorded approximately $920 million of accelerated depreciation and related charges in the quarter ended December 31, 2024.¹⁴ Amazon’s stated reason for the reversal was the “increased pace of technology development, particularly in the area of artificial intelligence and machine learning.”¹⁵ Amazon’s quarterly report on Form 10-Q for the period ended September 30, 2025, disclosed that the effect of the change in estimate, for the nine months then ended, was an increase in depreciation and amortization expense of $889 million and a reduction in net income of $677 million, or $0.06 per diluted share.¹⁶
Oracle Corporation, the fifth company named in Mr. Burry’s posts, has, according to financial press summaries of its filings, maintained a six-year depreciation schedule for servers and network equipment.¹⁷
In Buxton Helmsley’s view, the single most consequential data point in the entire debate is the divergence between Amazon and Meta in early 2025. Within the same quarter, and in the face of the same publicly available evidence regarding the pace of artificial intelligence hardware innovation, one issuer extended the useful life of its server fleet and recognized a multibillion-dollar reduction in depreciation expense, while another issuer shortened useful lives on a subset of equipment and recognized a multihundred-million-dollar increase in the same line item. Both changes were characterized by management as the product of formal useful-life studies. Both were reviewed by audit committees and concurred in by independent registered public accounting firms. Both were disclosed under the same accounting standard.
The institutional investor is not required to take a view on which company’s estimate is correct in order to recognize that the divergence is itself a disclosure. When two issuers operating substantially similar hardware fleets arrive at directionally opposite conclusions on useful life in the same fiscal quarter, the variance in reported operating income that flows from those conclusions is not driven by economics—it is driven by the elasticity of the estimate. That is precisely the condition under which institutional readers of financial statements are entitled to additional rigor.
As the named hyperscalers prepare their annual reports on Form 10-K for the fiscal year ended December 31, 2025, Buxton Helmsley believes there are five questions that institutional investors, audit committee members, and engaged shareholders should be prepared to ask.
First, does the issuer disclose the useful-life assumption applied to each material category of property and equipment, and the date on which the most recent useful-life study was completed? Disclosures that aggregate servers, networking equipment, and other technical infrastructure into a single line item without component-level useful-life ranges deprive the reader of the information necessary to evaluate the estimate.
Second, when an issuer extends or shortens a useful life, does the issuer disclose, on a quantified basis, the dollar effect on depreciation expense, on net income, and on per-share amounts, both for the current period and on a forward-looking basis to the extent reasonably estimable? ASC 250-10-50-4 requires disclosure of the current period effect; institutional readers should treat the absence of forward-period sensitivity as a meaningful omission, even where it is not technically required.
Third, what is the methodology and external evidence base for the useful-life study? Issuers that rely on internal engineering surveys without reference to secondary-market resale prices, observed retirement experience at peer companies, or reasoned consideration of next-generation hardware roadmaps should expect their disclosures to draw additional scrutiny.
Fourth, how does the issuer reconcile the useful-life assumption with the simultaneous capital expenditure trajectory? An issuer that materially extends the useful life of its existing fleet at the same time it is committing tens or hundreds of billions of dollars to deploy successor-generation hardware is making implicit assumptions about a deployment cascade—namely, that older hardware will be redeployed to lower-intensity workloads—that are themselves estimates and that should be disclosed and tested.
Fifth, what is the audit committee’s documented process for evaluating management’s useful-life estimate? Public Company Accounting Oversight Board Auditing Standard 2501 requires the auditor to obtain sufficient appropriate audit evidence regarding the reasonableness of accounting estimates, including an evaluation of management’s process and the data and assumptions underlying it.¹⁸ Boards of directors exercise a non-delegable oversight responsibility for that process. Investors are entitled to know whether the audit committee engaged in a substantive challenge of the estimate or accepted management’s representation.
The question is not whether the depreciation policy “lies.” It is whether the institutional investor has been given enough information to form an independent judgment. Where issuers disclose component-level useful-life assumptions, quantified prospective effects, and a defensible methodology—and where audit committees demonstrate substantive engagement—the answer is yes, and the resulting earnings figures, however generous, are at least subject to informed evaluation. Where issuers aggregate, where forward effects are obscured, and where audit committees defer, the answer is no, and the reported earnings should be treated as a position rather than as a fact.
Buxton Helmsley has long maintained that the most important disclosures in financial statements are the ones that require the reader to do work. Depreciation policy in the artificial intelligence capital expenditure cycle is one of those disclosures. The 2025 annual report filing season, which began in late January 2026 and continues through April 2026, will produce the largest single body of evidence on hyperscaler depreciation policy yet released. Institutional investors who read those filings with the rigor the question deserves will be in a substantially better position to understand the earnings figures the market is presently capitalizing.
Get the latest insights delivered to your inbox.
Referenced Sources:
[1] Michael Burry, post on X (formerly Twitter), November 11, 2025, https://x.com/michaeljburry/status/1987918650104283372.
[2] Id.
[3] "Big Short" Investor Michael Burry Accuses AI Hyperscalers of Artificially Boosting Earnings, CNBC (Nov. 11, 2025); Burry's Depreciation Gripe Shines Spotlight on Big Tech Profits, Bloomberg News (Nov. 14, 2025); The "Big Short" Investor Betting $1 Billion Against the AI Bubble Says Meta and Oracle's Accounting Is Hiding the Brutal Truth, Fortune (Nov. 13, 2025).
[4] Financial Accounting Standards Board, Accounting Standards Codification, Topic 360-10-35 (Property, Plant, and Equipment—Subsequent Measurement).
[5] Financial Accounting Standards Board, Accounting Standards Codification, Subtopic 250-10-45-17 (Accounting Changes and Error Corrections—Other Presentation Matters).
[6] Financial Accounting Standards Board, Accounting Standards Codification, Subtopic 250-10-50-4 (Accounting Changes and Error Corrections—Disclosure).
[7] Microsoft Corporation, Q4 FY2022 Earnings Conference Call (Jul. 26, 2022) (remarks of Amy Hood, Chief Financial Officer); Microsoft Corporation, Form 8-K and Exhibit 99.1 thereto, filed July 26, 2022.
[8] Alphabet Inc., Press Release, Alphabet Announces Fourth Quarter and Fiscal Year 2022 Results, Exhibit 99.1 to Form 8-K filed February 2, 2023.
[9] Alphabet Inc., Press Release, Alphabet Announces Fourth Quarter and Fiscal Year 2023 Results, Exhibit 99.1 to Form 8-K filed January 30, 2024.
[10] Meta Platforms, Inc., Press Release, Meta Reports Fourth Quarter and Full Year 2024 Results, Exhibit 99.1 to Form 8-K filed January 29, 2025.
[11] Meta Platforms, Inc., Annual Report on Form 10-K for the fiscal year ended December 31, 2025.
[12] Amazon.com, Inc., Press Release, Amazon.com Announces Financial Results, Exhibit 99.1 to Form 8-K filed February 1, 2024; Amazon.com, Inc., Q4 2023 Earnings Conference Call (Feb. 1, 2024) (remarks of Brian Olsavsky, Chief Financial Officer).
[13] Amazon.com, Inc., Press Release, Amazon.com Announces Fourth Quarter Results, Exhibit 99.1 to Form 8-K filed February 6, 2025.
[14] Id.
[15] Id.
[16] Amazon.com, Inc., Quarterly Report on Form 10-Q for the period ended September 30, 2025.
[17] See, e.g., Why GPU Useful Life Is the Most Misunderstood Variable in AI Economics, Stanley Laman Group (Nov. 2025); Depreciation of GPUs: Between Useful Lives and Useful Myths, Deep Quarry (Dec. 7, 2025).
[18] Public Company Accounting Oversight Board, Auditing Standard No. 2501, Auditing Accounting Estimates, Including Fair Value Measurements.
This article is published by Buxton Helmsley USA, Inc. for informational and educational purposes only. It does not constitute investment advice, a solicitation, or a recommendation to buy or sell any security. The views expressed are those of Buxton Helmsley and are based on publicly available information as of the date of publication. Investors should conduct their own due diligence and consult with qualified professionals before making investment decisions.
Insights and updates delivered directly to your inbox.