Wall Street's Most Famous 'Cassandra' Targets Nvidia Again: Do the Three Arguments Hold Water?

Michael Burry, the fund manager immortalized in the film 'The Big Short' for accurately shorting subprime mortgages, disclosed a short list on June 30, including Nvidia (NVDA-US) (entry price: $198.09), Tesla (TSLA-US), Applied Materials (AMAT-US), and the Philadelphia Semiconductor ETF (SOXX-US). On July 1, he initiated a new short position on Micron (MU-US) (entry price: $1,051.87). On July 25, he further increased his short positions on Nvidia, Micron, and SOXX.

His reasoning is not new. He argues that AI infrastructure investment suffers from two major accounting issues: 'off-balance-sheet circular financing' and 'falsely extended depreciation periods.' According to Burry, the financial practices of Nvidia's customers may make the entire supply chain's reported profits appear healthier than they truly are. Upon this news, social media reactions were almost reflexive—another bearish narrative.

Nvidia's stock price experienced significant volatility following the disclosure. On June 30, the day Burry first revealed his position, Nvidia closed around $200, nearly matching his entry price of $198.09. In early July, after Meta announced it would rent out idle AI computing power, the Philadelphia Semiconductor Index dropped over 6% in a single day, and Nvidia briefly fell to around $194. However, it quickly rebounded and returned above $210 by mid-July.

After Burry's second increase on July 25, Nvidia dropped nearly 5% on July 27 to $196.51. The immediate trigger for this decline was a report stating that Nvidia was discussing providing OpenAI with approximately $250 billion in financing guarantees. Burry promptly commented on social media, calling it 'Nvidia guaranteeing customers to buy its own chips,' echoing his 'off-balance-sheet circular financing' argument and amplifying market concerns.

Since then, Nvidia has traded between $190 and $200. After hitting a low of $190 on July 29, it rapidly recovered and closed at $200.75 on July 31. Based on this, Burry's initial position from June 30 at $198.09 is nearly breakeven, with a slight unrealized loss. His July 25 addition at $210.28 is down approximately 4.5%.

This is almost a script Burry has followed before. Over the past few years, he has frequently voiced bearish views on the market, but his accuracy has often been questioned. This raises a more important question: when a 'professional pessimist' speaks again, how credible is their judgment? Furthermore, in Wall Street as a whole, are those actually shorting these companies a few lone gamblers, or is it a tacit collective retreat?

Looking at both his hits and misses, a clear dividing line emerges. His few successful calls were invariably cases where a 'structural problem' coincided with a 'clear catalyst'—such as the subprime mortgage default wave or the pandemic black swan. In contrast, judgments based solely on 'overvaluation' without a specific trigger have almost always been mistimed.

Underlying this is his consistent analytical framework: he disregards easily manipulated metrics like P/E ratios and ROE, instead focusing on free cash flow. He distrusts rating agencies and analyst reports, preferring to scrutinize raw financial documents like a doctor reviewing medical records.

This methodology excels at answering 'Is the asset overvalued?' or 'Are there hidden risks?' but fails to answer 'When will the risk materialize?'—which is precisely the key premise for understanding his current short position.

Consider one often-overlooked data point: Nvidia's current short interest is about 1.3% to 1.4% of its float, which is quite low (though in nominal terms, Nvidia is one of the most heavily shorted stocks in the U.S. market, so the absolute value is not small). A February 2024 report showed that institutions shorting Nvidia had collectively lost over $5 billion, making it the single most losing short position in the market at that time. Notably, while short interest shrank earlier in the year, it has since rebounded.

In reality, only a tiny minority dare to openly and aggressively short. Burry has the loudest voice, but he is neither the only one nor close to mainstream.

The three arguments Burry presents this time vary in persuasiveness.

First: Depreciation Period

This touches on a basic accounting principle: companies cannot expense the full cost of long-term equipment in one period. Instead, they spread the cost over the asset's 'useful life' across multiple accounting periods. The longer the depreciation period, the lower the annual expense, and the higher the reported profit.

Burry questions whether AI chips, which undergo rapid hardware iteration, have a real usable life of only 2–3 years. Yet, cloud providers like Microsoft, Google, and Meta are depreciating them over 6 years. The longer the period, the less annual cost is recognized, artificially inflating book profits.

Nvidia and some analysts have responded, but their rebuttals address the depreciation policies of downstream customers, not Nvidia itself. This debate will ultimately be settled by future financial reports—specifically, whether large impairments appear in coming years.

Second: Off-Balance-Sheet Circular Financing

This argument suggests that Nvidia's strong chip orders may not all stem from genuine end-user demand, but could involve a 'self-created demand' funding loop.

A specific version circulating in the market claims that Nvidia provides financing guarantees for an AI company's compute expansion. That company uses the guarantee to lease cloud services and purchase Nvidia chips. Nvidia then books this as 'real' revenue. The funds eventually cycle back to Nvidia, while many financial transactions remain off the balance sheet.

Burry worries that if such 'funding customers to buy our own products' loops are removed, Nvidia's true end demand may not be as strong as the numbers suggest. Nvidia counters that its strategic investment scale is small and represents only a tiny fraction of revenue, calling the claim baseless. This argument relies more on undisclosed information and is harder to verify than the first.

Third: Share Buyback Dilution

Burry accused Nvidia's recent share buybacks of exaggerating earnings per share (EPS). However, Nvidia pointed out that Burry's calculation incorrectly included tax payments related to employee stock compensation, a clear methodological error. Burry has not provided detailed data to counter this, making this argument the weakest of the three.

Notably, the opposite view comes from another 'Big Short' protagonist—Steve Eisman. Eisman says he currently has no plans to short Nvidia, citing continued strong revenue growth, ongoing buybacks and dividends, and expanding capital expenditure commitments from major cloud providers—all numbers he believes are currently verifiable. Still, he admits nervousness about the sustainability of this rally and has proactively reduced his position. In late July, he warned that any major tech firm cutting AI capital spending could trigger a sharp drop in U.S. equities.

Two veterans who lived through 2008 and are known for contrarian thinking have given opposite answers on the AI question.

Legendary short-seller Jim Chanos takes a more nuanced stance. He agrees with Burry on the broad direction of 'accounting mismatch': chipmakers recognize revenue immediately, while cloud providers spread massive costs over four to seven years. This, he says, closely resembles the pre-2001 dot-com bubble era. Back then, Cisco's financials looked strong, but telecom operators masked overcapacity by extending depreciation periods. WorldCom committed outright fraud, and when the scandal broke, it triggered one of the largest bankruptcies in U.S. history.

Yet, in actual positioning, Chanos has taken a completely different path. He has expressed caution on the cyclical memory chip market and largely avoids Micron. He does not short Nvidia directly, instead viewing it as more attractively valued than peers. His actual short bets are on private equity firms that have simultaneously bet on AI infrastructure and commercial real estate in low-capitalization environments—targeting the financial leverage periphery of the AI boom, not the chip stocks themselves.

Comparing these three positions reveals that while there is considerable consensus among veteran short-sellers on whether 'an AI bubble exists,' there is significant divergence on 'who to short' and 'how to bet'—almost every player acting independently.