Nvidia just reported $81.6 billion in quarterly revenue, up 85% year over year. Its data center business alone generated $75.2 billion, up 92%. The numbers are extraordinary by any measure.
Michael Burry‘s argument is that they may also be the most dangerous kind of numbers there are.
Burry argues against AI infrastructure trade
In a May 2026 Substack post titled “The Heretic’s Guide to AI‘s Stars Part III: Tracepalooza and the Bezzle,” Burry laid out his most specific argument yet against the AI infrastructure trade.
The post has been read by more than 200,000 subscribers, and the reaction from Wall Street was significant enough that Nvidia (NVDA) took the unusual step of sending a memo to sell-side analysts pushing back on his arguments on stock-based compensation and depreciation, according to CNBC.
Burry responded directly. “I stand by my analysis. I am not claiming Nvidia is Enron. It is clearly Cisco,” he wrote, CNBC confirmed.
More Wall Street:
- JPMorgan resets S&P 500 price target for the rest of 2026
- Vanguard challenges the S&P 500 as a one-stop strategy
- Goldman Sachs resets Broadcom stock forecast
“NVIDIA is benefitting from strong demand, but is selling into a concentrated set of buyers whose own demand is being distorted by a training and benchmarking phase that will not last,” Burry explained.
“That distorted demand is working like a bullwhip into NVIDIA’s own supply chain through custom supply commitments as well as downstream into data-center financing. Looming over it all is the bezzle, which once seen, cannot be unseen, and once revealed, does not exist.”
The 3 mechanisms Burry says make Nvidia specifically dangerous
Burry is not making a general market argument. He is making a specific structural argument about Nvidia’s revenue concentration, supply chain obligations, and the nature of demand driving its numbers, according to 24/7 Wall St.
The first mechanism is buyer concentration. Hyperscalers, primarily Microsoft, Google, Amazon, and Meta, account for approximately 50% of Nvidia’s data center revenue. That concentration means a relatively small change in any one customer’s capital expenditure plans can produce an outsized revenue effect.
Burry has calculated that if Microsoft alone cuts its Nvidia chip spending by 20%, the revenue impact on Nvidia is approximately 4.2%, he noted on Substack.
The second mechanism is the bullwhip effect. When buyers at the end of a supply chain over-order because they fear missing out, that distortion amplifies backward through the chain. Nvidia records record demand. It locks in custom supply commitments with TSMC, now totaling $119 billion in non-cancellable obligations. Data center financing expands to accommodate the buildout.
The entire chain bets the demand is permanent. Burry’s argument is that it is not.
The third mechanism is what he calls the bezzle, a term from economist John Kenneth Galbraith describing the gap between what people believe they own and what actually exists. In a bezzle, the money feels real and the assets feel real, until suddenly they do not.
Burry is arguing that the current AI revenue flowing through Nvidia reflects a training and benchmarking phase of AI deployment that will eventually give way to an inference and deployment phase with a fundamentally different demand profile.
Getty Images
Why Burry called this the “Cisco moment,” and how Nvidia responded
The historical analogy Burry has built his argument around is Cisco Systems during the dot-com era. Cisco supplied the networking equipment for the internet buildout of the late 1990s, and its revenue was real, its customers were real, and the underlying technology was transformative.
What was not real was the assumption that the buildout phase would continue indefinitely. When telecom companies stopped buying, Cisco’s stock fell more than 80% from its peak.
Burry wrote the comparison explicitly. “And once again there is a Cisco at the center of it all, with the picks and shovels for all and the expansive vision to go with it. Its name is Nvidia,” he wrote, according to CNBC.
He has also pointed to data showing the current top 10 AI stocks have surged 784% in 12 months. The equivalent cohort in the pre-dot-com period surged 622% at its peak. That 162 percentage-point gap is the basis for his comparison that the 2026 AI trade is running hotter than even the most extreme phase of the dot-com bubble, according to 24/7 Wall St.
Key figures from Burry’s May 2026 Substack and market context:
- Burry’s position:leveraged short on SOXX, the Philadelphia Semiconductor Index, through January 2027 put options, according to 24/7 Wall St
- SOXX performance: Philadelphia Semiconductor Index up approximately 65% in 2026 alone, 24/7 Wall St confirmed
- Nvidia Q1 FY27: Revenue $81.6 billion, up 85% year over year; data center revenue $75.2 billion, up 92%; hyperscalers approximately 50% of data center revenue, Burry noted on Substack
- Nvidia supply commitments: $119 billion in non-cancellable TSMC obligations; if Microsoft cuts Nvidia capex 20%, that equals a 4.2% Nvidia revenue hit, Burry noted on Substack
- Nvidia valuation: Trailing P/E of 43, forward P/E of 24, price-to-sales of 23; market cap approximately $5.45 trillion, according to 24/7 Wall St
- S&P 500 Shiller PE: Above 40 as of May 2026, highest since 2000 tech bubble; historical average approximately 17, 24/7 Wall St confirmed
- AI stock surge: Current top 10 AI stocks up 784% in 12 months; pre-dot-com peak cohort surged 622%, 24/7 Wall St noted
The counterargument to Burry’s Nvidia view: the timing question is complicated
Burry’s critics make a straightforward point: He has been early before. He called the market a sell in 2023, and it subsequently rose 131%. He has been bearish on Nvidia for over a year while the stock has continued to climb.
Being right about a structural argument does not tell you when the market will agree with you, and in the meantime, the cost of being short a stock this strong can be severe.
The bull counterargument is also substantive. AI is producing real revenue not just for Nvidia, but also for the hyperscalers buying its chips. Microsoft’s Azure AI growth, Google’s Gemini API revenue, and Meta’s advertising improvements from AI are all measurable and growing.
The argument that training phase demand is temporary assumes inference and deployment will generate less GPU demand, which is not guaranteed. Many researchers argue that inference at scale requires as much compute as training, just distributed differently.
What Burry is asking investors to consider is not whether AI works but whether $119 billion in non-cancellable supply commitments and a $5.45 trillion valuation can survive even a modest slowdown in the speed of that transition.
His record in 2008 was built on identifying not just that housing was overvalued, but the specific mechanism by which the overvaluation would unwind. He believes he has found the same kind of mechanism here.
Whether he is right about the timing this time is the only question that matters to investors.
Related: JPMorgan doubles down on stock market message for 2026
