Why Michael Burry Sees a 1960s Computer-Leasing Parallel in the AI Boom

Michael Burry’s comparison between today’s AI infrastructure buildout and the late-1960s computer-leasing boom is about how investors value and finance expensive equipment. His concern is that spending may rely on optimistic assumptions about future demand, GPU rental income and resale value. If those assumptions fail while debt and accounting schedules remain in place, the fallout could reach chip owners, lenders and investors. It is a warning about risk—not proof that AI demand will collapse. 11923

The disagreement: useful life versus economic value

Nvidia’s investor presentation argued that A100, H100 and B200 infrastructure can retain value beyond an accelerated five-year depreciation schedule. The chart addresses the value of the equipment over time, but Burry disputes what that value means for recovering the cost of buying it. 6

Depreciation is not simply a forecast of when a chip stops working. In Burry’s framing, it is about recovering the capital invested over the period when the asset can earn money competitively, while accounting for any value left at the end. A GPU might continue to run—and even find a renter—yet generate less income as newer chips compete for the most lucrative work. 16

That is why Burry challenges the inference that continued operation proves a long economic life. He argues that residual-value estimates based on projected future cash flows are not the same as evidence from actual resale prices. The practical test is whether a chip’s lifetime earnings and eventual resale value justify its original cost. 216

Why he invokes the 1960s leasing boom

In Burry’s analogy, the similarity is not that computers and GPUs are identical. It is the possibility that a surge of investment rests on expectations of sustained demand and valuable equipment, while financing models depend on those assets continuing to produce returns. If demand, rental income or resale values disappoint, a costly buildout can become harder to support. Reports of GPU-backed borrowing and private-credit financing are part of the parallel he draws. 11923

The analogy should not be mistaken for a prediction that history will repeat exactly. It identifies a vulnerability: when a business case depends on equipment holding value, a change in how much that equipment can earn can matter as much as whether it still functions.

What shorter GPU economics could mean for Big Tech

The accounting dispute matters because a longer useful-life assumption spreads equipment costs over more years, reducing annual depreciation expense in the meantime. Burry argues that some hyperscalers’ schedules are too long: he sees two to three years as closer to the most profitable competitive life, while reports describe companies using schedules around five or six years. Those estimates are contested, not settled measures of every GPU’s lifespan. 3

If Burry’s view is right, companies could be recognizing too little depreciation expense today, which would make reported profits higher than they would be under a shorter schedule. They might also need to replace equipment sooner or record write-downs if expected values do not hold. Burry has estimated a large cumulative depreciation gap for cloud providers through 2028; that figure is his estimate, not an established loss. 3

There is a counterpoint: older GPUs can remain in service and earn rental revenue. That supports the case that hardware does not become worthless as soon as a newer generation arrives. But it does not, by itself, settle how much income an older chip can earn or whether that income recovers its purchase cost. 916

Why debt-financed GPU operators could be more exposed

An operator that borrows to buy GPUs faces a mismatch if equipment income or collateral value falls while loan obligations remain. In that scenario, weaker rental economics could make it harder to service or refinance debt. This is a conditional risk, not evidence that every GPU-backed loan will fail; Burry’s broader warning specifically points to GPU financing and private credit as part of the historical comparison. 223

The key factors to watch are therefore not just chip performance or utilization. They include what customers will pay to rent older hardware, what that hardware can sell for, and whether those cash flows cover the costs of owning and financing it.

What Burry’s Nvidia puts do—and don’t—show

Reports say Burry covered Nvidia common-stock shorts and shifted to put options, including contracts expiring in September 2027. He has described moving his expected timeline for trouble forward. 1722

A put is a time-limited bearish position: it gives its holder the right to sell at a specified price before a set expiry. So the reported position indicates a view that Nvidia shares could fall within the option’s lifetime; it does not establish that the shares will fall or that Burry’s broader thesis is correct. 29

The central question remains empirical: can older GPUs keep earning enough, for long enough, to justify their cost and the financing behind them? Continued use is relevant evidence, but it is not the whole answer. Burry’s warning will depend on how GPU rental rates, resale values and real-world returns develop—and those outcomes are still uncertain.