Here's an uncomfortable question for anyone betting big on AI infrastructure: What happens when the chips you just bought for millions of dollars become worthless faster than your accountants expected?

Legendary short-seller Jim Chanos thinks he knows the answer, and it's not pretty for companies like Oracle Corp. (ORCL) and CoreWeave Inc. (CRWV). In a recent podcast interview, Chanos laid out what he calls a "massive financial risk" hiding in plain sight: data center operators are using depreciation schedules that assume Nvidia Corp. (NVDA) chips will last six years, when the reality might be closer to three or four.

The Depreciation Math That Matters

This isn't just accounting nerdery. The numbers matter a lot. If you spend billions on AI hardware and depreciate it over six years, you're spreading that cost over 72 months. But if those chips actually become obsolete in three years, you need to recognize twice the expense each year. That difference goes straight to your income statement, turning reported profits into losses.

"If the chips last for three years, you have to depreciate a third of what you spend," Chanos explained. "That's the bet you have to make if you're a CoreWeave investor."

The evidence supporting Chanos' thesis is already showing up in market prices. Rental rates for Nvidia's Hopper chips have plummeted roughly 28% year-over-year as newer models hit the market. When the thing you're renting out drops in value that fast, it's a pretty clear signal that customers think it's becoming obsolete.

Oracle's Particularly Exposed Position

Chanos singled out Oracle as the most vulnerable among major tech players. The problem isn't just the depreciation issue—it's that Oracle, unlike Microsoft Corp. (MSFT) or Meta Platforms Inc. (META), isn't currently earning its cost of capital on these massive AI investments. The company is making what Chanos described as a "bet the company" move with very little margin for error.

"If AI monetization gets pushed out… to 2030 or whenever, then Oracle will have fundamental financial problems," Chanos warned.

That's a polite way of saying that if AI doesn't start generating returns soon, Oracle could find itself drowning in depreciation expenses for hardware that isn't generating the expected revenue.

Worse Than the Dot-Com Era?

Chanos drew an interesting comparison to the late 1990s telecom bubble, and his conclusion is sobering: this might actually be riskier. Back then, the companies buying all that telecom equipment were profitable giants like General Electric Co. (GE) and AT&T Inc. (T). They had balance sheets that could absorb mistakes.

Today's landscape looks different. The primary customers driving demand for AI compute—companies like OpenAI and Anthropic—are burning through cash without proven business models. If venture capital funding dries up or investor sentiment shifts, Chanos predicts that order books for companies like CoreWeave could evaporate quickly, leaving them holding massive debts and rapidly obsoleting hardware.

It's a classic boom-cycle problem: everyone assumes the music will keep playing, so they keep building capacity for customers who might not be around in a few years.

The Market Hasn't Gotten the Memo

Despite these warnings, Nvidia (NVDA) shares have climbed 27.46% year-to-date, outperforming the Nasdaq 100's 19.51% return over the same period. On Monday, the stock closed 0.73% higher at $176.29. Over the past six months, shares have advanced 21.84%, though they've slipped 5.53% in the last month.

The stock maintains what analysts describe as a weaker price trend over short and medium terms but a strong long-term trend, coupled with a poor value ranking.

MarketDash reached out to Nvidia for comment on Chanos' claims regarding GPU depreciation schedules and the useful life of its chips, but hadn't received a response at the time of publication.

The Core Question

What makes this situation fascinating—and potentially dangerous—is that it all hinges on a single question: How long does an AI chip actually remain economically useful? If hyperscalers and neoclouds are right that it's six years, then current accounting is fine and Chanos is wrong. But if the relentless pace of Nvidia's innovation means chips become obsolete in half that time, then a lot of reported earnings are going to need serious revision.

The rental rate data suggests Chanos might be onto something. When prices for your product drop 28% in a year, that's not the behavior of an asset with a six-year useful life. That's the behavior of something that's being rapidly replaced by better alternatives.

For investors in Oracle, CoreWeave, and similar companies making massive AI infrastructure bets, this is the question that matters more than almost anything else. Because if Chanos is right, the depreciation time bomb he's warning about isn't hypothetical—it's already ticking.