A simple explanation of what Michael Burry is warning about right now
During the week Nvidia reported its Q3 earnings, Michael Burry posted a series of tweets that pointed in one direction:
the current AI boom may not be supported by real, sustainable economic demand.
Below is a straightforward breakdown of what he said and why it matters.
---
1. The AI Money Loop Is Not Real Demand
Burry highlighted a chart showing the financial relationships between Nvidia, OpenAI, Microsoft, Oracle, AMD, CoreWeave, and many AI startups.
His point is simple:
AI companies are mostly buying from each other using money that came from each other.
Examples:
Microsoft invests in OpenAI
OpenAI spends billions on Nvidia chips
Nvidia invests in GPU-cloud companies
Those GPU-cloud companies buy more Nvidia chips
Oracle signs huge cloud deals with OpenAI
Oracle then buys large amounts of Nvidia hardware
This creates a circular economy where revenue appears to grow, but the money never comes from true end customers.
Burry summarized it as:
“True end demand is ridiculously small. Almost all customers are funded by their dealers.”
This means the AI industry could be inflating itself without real outside demand.
---
2. “Suspicious Revenue Recognition”
Burry warned that many AI companies may be using accounting methods that make their revenue look larger or more stable than it really is.
This can include:
Booking multi-year deals upfront
Using credits or internal agreements as revenue
Circular transactions where companies buy from each other
His view is that if every transaction were shown clearly, the financial picture would appear far less impressive.
---
3. Misunderstanding GPU Depreciation
Another tweet targeted the idea that older GPUs retaining high utilization means they retain high economic value.
Burry argued that:
Just because older chips are still being used does not mean they are profitable.
A100 GPUs use two to three times more electricity per compute unit than newer H100 chips.
Higher energy usage means drastically higher operating costs.
He compared it to airlines keeping old airplanes around only for holiday overflow. They still fly, but they don’t make meaningful profit.
Most importantly:
Many AI companies use GPUs as collateral for loans or SPV financing.
If the market realizes these chips are economically inefficient, their value collapses.
---
4. The 1969 Buffett Letter Comparison
Burry also posted Warren Buffett’s 1969 partnership-closing letter. In it, Buffett explained that markets were overpriced, speculation was rampant, and quality opportunities had vanished.
What followed:
The S&P 500 fell sharply.
After inflation, investors lost more than half their purchasing power over the next decade.
Burry is implying that today’s AI-driven market resembles that overheated period.
---
5. Nvidia’s Buyback and Share Dilution Problem
Burry ended the week with a critique of Nvidia’s financial engineering.
From 2018 to 2025:
Nvidia earned about $205 billion in net income.
It generated $188 billion in free cash flow.
It spent $112.5 billion on stock buybacks.
Yet the total share count increased by 47 million shares.
This means the buybacks did not actually benefit shareholders.
They mainly offset dilution from stock-based compensation.
According to Burry, the real economic cost to shareholders was: $112.5 billion lost
and
owner earnings reduced by roughly half
because of how the compensation and buyback cycle worked.
---
Conclusion
Across all his tweets, Burry’s message is consistent:
AI demand is largely circular, not organic.
Revenue recognition practices may be hiding weakness.
GPU economics are misunderstood and potentially overstated.
The market resembles past speculative peaks.
Even Nvidia’s financials show signs of late-cycle strain.