Signal — The New Accounting Fault Line
Michael Burry, the investor who foresaw the 2008 housing collapse, is now targeting another distortion — the way tech giants are stretching the useful life of AI infrastructure to inflate profits. Across Silicon Valley, firms are extending depreciation schedules for servers, GPUs, and networking gear far beyond their real two-to-three-year lifespan. This defers expenses, flatters margins, and conceals the true cost of scaling artificial intelligence. Burry estimates roughly $176 billion in understated depreciation across major firms, warning that this tactic masks how quickly AI hardware actually expires.
The Accounting Standards — How Time Is Being Stretched
Depreciation once measured physical wear; now it measures narrative tempo. Meta extended the useful life of its servers to 5.5 years, trimming nearly $3 billion in expenses and inflating pre-tax profits by about four percent. Alphabet and Microsoft followed with similar extensions, stretching infrastructure life to roughly six years. Amazon, by contrast, moved in the opposite direction — shortening its AI depreciation schedules to reflect the rapid turnover of GPUs and compute nodes. This divergence is not technical; it’s philosophical. Meta stretches time to protect optics. Amazon protects the truth. The first strategy buys comfort; the second builds credibility.
The Two Camps — Infrastructure Realists vs. Earnings Illusionists
Among the Magnificent Seven, two accounting cultures now define the AI era. The Amazon Category — Amazon and Apple — admits cost early, valuing transparency over quarterly symmetry. The Meta Category — Meta, Microsoft, Alphabet, Oracle, Nvidia, AMD, Intel, Broadcom, Huawei, Cambricon — extends asset lives to smooth expenses and preserve growth narratives. Their logic is simple: if infrastructure appears to last longer, profit appears to last longer too. But when hardware ages faster than spreadsheets admit, deferred truth compounds like hidden debt.
What the Numbers Conceal — The Infrastructure Mirage
AI hardware depreciates in months, not years. NVIDIA’s training cycles and chip refreshes make most GPUs obsolete within two to three years. Extending lifespan assumptions to five or six years means billions in unrealized wear are parked on the balance sheet as if time itself had slowed. The risk is cumulative: overstated assets, overstated earnings, and overstated confidence. Investors reading those filings think AI infrastructure is compounding capital — when in fact it’s consuming it. The illusion works until energy costs rise, chip generations accelerate, or revenue slows. Then, like 2008’s housing derivatives, time comes due.
Yield Distortion and Policy Risk
When depreciation is misaligned, so is yield. Pension funds, sovereign allocators, and ETF managers who rely on these inflated earnings models may be pricing their exposure on fiction. This is not a retail issue; it’s systemic. If AI-linked ETFs and staking ETPs are benchmarked against earnings that exclude the real cost of obsolescence, then the entire yield calculus becomes distorted. Regulators have not yet forced transparency in AI asset accounting. But the first audit that exposes a billion-dollar gap between reported lifespan and physical decay will trigger a new kind of contagion — one measured not in defaults, but in disclosures.
SEC, Auditors, and the Coming Reckoning
The SEC has the tools to close this gap. A review of 10-K filings shows that companies are free to define their own “useful life” assumptions for servers and networking gear, provided they disclose them. The audit process, however, often treats those numbers as internal policy, not public truth. As AI infrastructure becomes the largest capital expense class in tech, these assumptions are no longer trivial — they are material. Expect new disclosure standards, auditor scrutiny, and investor activism centered on depreciation integrity.
Closing Frame
Depreciation is no longer a footnote. It is the heartbeat of AI’s economic story — a pulse that reveals who builds truth and who buys time. Amazon’s shortening of asset lives reflects realism; Meta’s extensions reflect optimism; Burry’s warning reflects pattern recognition. Because in this choreography, infrastructure is not just hardware — it is honesty expressed in years. And when those years are stretched, the truth eventually snaps back.