Wall Street has spent the better part of this year arguing over a single spreadsheet line buried deep in corporate accounting disclosures. It sounds boring until you realize it could decide whether trillions of dollars in reported AI earnings are actually real.
Michael Burry just reached back nearly 60 years to make his point. The comparison he picked says less about computers than it does about human nature repeating itself.
Here is what his latest shot at Nvidia means for the AI trade.
The Nvidia depreciation data that set Michael Burry off
The dispute traces back to a Sept. 27 Nvidia investor presentation. The company presented a slide titled “NVIDIA AI Infrastructure Retains Value Beyond Accelerated Depreciation Schedules,” plotting retained chip values against a five-year depreciation curve.
The implication was that GPUs are worth more than their accounting suggests, Benzinga reported.
Also read: Michael Burry sends a stark warning to Big Tech stock investors
Nvidia bulls called the chart a decisive answer to depreciation critics. Burry called it something else. He has been making this argument for nearly a year, and the slide only gave him another example to point to.
To make his case, Burry pulled a passage from Adam Smith’s 1968 book “The Money Game.” In it, a skeptical critic questions how fast computers were being written off. He is mocked by a speculator called The Great Winfield and a young trader called Billy the Kid, who dismissed the concerns as the outdated thinking of an older generation, according to Benzinga.
Those 1960s traders defended their computer-leasing bets by insisting the need for computers was practically infinite, brushing off high valuations as something the older generation simply failed to understand.
Burry said the nearly 60-year-old exchange proves that human nature barely changes across market cycles.
Burry’s bigger bet against AI accounting
The 1960s comparison builds on a depreciation thesis Burry has pushed publicly since November, when he accused hyperscalers of extending the useful life of Nvidia chips and servers well beyond what he argued was a realistic two- to three-year replacement cycle.
By his estimate, that accounting choice could understate industry-wide depreciation by roughly $176 billion between 2026 and 2028. Oracle’s earnings could be overstated by close to 27% and Meta’s by about 21% by 2028, TheStreet noted.
Meta has already put real numbers behind part of the debate, disclosing that extending the useful life of certain servers to 5.5 years cut its depreciation expense by roughly $2.9 billion in a single year. Burry has cited this detail as evidence of the broader pattern he believes is understating depreciation across the industry.
On Sept. 24, Burry widened the warning to construction-in-progress assets sitting across Amazon, Meta, Alphabet, Microsoft, and Oracle. He argued the entire AI buildout is repeating a pattern seen in past capital cycles, including the telecom overbuild of the late 1990s.
Subsequent depreciation and write-downs from that cycle pushed aggregate S&P 500 net investment into negative territory for 12 consecutive quarters between mid-2003 and mid-2006.
Nvidia and Wall Street push back
Nvidia sent a seven-page memo to Wall Street analysts and named Burry directly. Four to six years is the real depreciation window, the company argued. The 2020-era A100 is still in use. Two to three years, Nvidia said, is simply wrong.
CEO Jensen Huang has also taken the fight public, posting GPU rental data on social media showing H100 prices climbing and noting that CoreWeave has booked A100 capacity through 2029.
Nvidia has cited those data points as evidence that older GPUs retain economic value longer than Burry’s depreciation thesis suggests.
More Nvidia:
- Nvidia just made a move Wall Street wasn’t ready for
- Nvidia just locked down deal that changes AI race
- Nvidia stock is doing something it hasn’t done in years
Independent data offers a mixed picture. Six-year-old A100 chips are still valued at nearly $5,000 each based on their projected future rental income. That complicates Burry’s warning, though Burry has countered publicly that strong rental demand does not prove a chip is not still losing economic value quickly behind the scenes, Benzinga reported.
Not every market voice is fully convinced either way. Today’s hyperscalers are far better capitalized than dot-com era companies were. And genuine AI monetization is already underway, which tempers some of the bubble comparisons.
What investors should expect going forward
The depreciation fight is unlikely to resolve quickly, since it hinges on accounting estimates that require significant management judgment. Neither side has produced the kind of hard evidence that would settle how quickly today’s AI chips are losing economic value.
Investors should watch upcoming hyperscaler earnings closely for any changes to stated useful-life assumptions on servers and chips. Even a modest shift could move reported depreciation and operating income at Oracle, Meta, and the other major cloud spenders, with no change to the underlying business required.
For now, Burry’s own trading record on this particular call has been mixed. His Nvidia short lagged while the stock rallied through the summer. Several of his other bearish AI positions also suffered as their underlying shares climbed.
Being early on an accounting argument and being right on timing are two very different things for anyone weighing a position on either side of the trade.
Related: Jensen Huang just answered Michael Burry’s Nvidia bear case