The $176B Depreciation Trick Hiding in Plain Sight

But what happens when you change the estimate to 6 years? Your annual depreciation drops to $5,000. Your reported profit goes up by $5,000 per GPU. Multiply that by millions of GPUs across five hyperscalers, and the numbers get very large very fast.

That’s exactly what happened. Between 2023 and 2025, all five major hyperscalers — Alphabet, Amazon, Apple, Meta, and Microsoft — extended the stated useful life of their server and networking equipment. The extensions ranged from modest (4 to 5 years) to aggressive (3 to 6 years).

Michael Burry flagged this in his analysis of public 10-K filings. By his estimates, these depreciation extensions overstate combined earnings by approximately $176 billion over the 2026–2028 period. That’s not a projection based on assumptions. It’s arithmetic applied to the companies’ own disclosed accounting changes.

The GPUs don’t care about accounting. The actual product replacement cycle in AI infrastructure is 2–3 years, driven by the pace of architectural improvement. When NVIDIA releases a new generation — Blackwell followed Hopper in roughly 18 months — the old hardware doesn’t become worthless, but it becomes significantly less competitive for the training and inference workloads that justify data center economics. A 6-year useful life assumption implies that a GPU purchased today will still be commercially viable in 2032. That requires believing the pace of hardware improvement will dramatically slow down.

The financial effect is real even if you don’t believe the accounting is misleading. Longer depreciation schedules mean lower annual expenses, higher reported operating income, and stronger-looking margins. Investors pricing these stocks on earnings multiples are pricing in margins that include the depreciation benefit. If the useful life assumptions prove wrong and equipment needs replacement sooner, the write-downs will be large and sudden.

This isn’t fraud. It’s a perfectly legal accounting choice, and the companies disclose the changes in their filings. But it means that the earnings numbers you see in quarterly reports are not telling you what AI infrastructure actually costs to maintain. The real cost is higher. The real margins are thinner. And the gap between reported earnings and economic reality is $176 billion and growing.