Michael Burry has been posting “The end is nigh” on his Substack.
He quoted the Joker from Tim Burton’s Batman: “Dancing with the devil in the pale moonlight.” He called AI enthusiasm “mass addiction” and predicted it “may die a death by a thousand cuts.” He shorted Micron at $1,051.87 on July 1 after the stock had already climbed nearly 700% over the prior year.
On July 10, he published something more specific on Substack that goes beyond valuation concerns. It’s an attack on the technical foundation on which the whole AI industry is built.
Michael Burry says AI started on the wrong path and is stuck there
The argument in the post centers on what Burry calls a “bad start.” His case is that AI development went language-first when it should have gone reasoning-first, and the industry has been paying for that choice ever since without fully admitting it.
He builds the argument around something called Ballard’s Test, a philosophical case involving a figure named Melville Ballard who achieved profound reasoning before ever acquiring language.
More AI:
- The new Chinese AI model rattling U.S. tech investors
- Anthropic restores access to Mythos 5 for select organizations
- SoftBank CEO offers stinging critique of Musk’s AI bet
Burry uses it to make a specific point: real understanding doesn’t require language. Language is the output of intelligence, not the source of it.
“We have mistaken the output of artificial intelligence (language) for the engine of it (reason),” Burry wrote on Substack.
His reading of the AI industry is that it spotted language as a tractable problem and optimized for it, not because language led to general intelligence but because it was scalable and fundable.
The models got better at generating text. Investors rewarded that. The industry kept going. And somewhere along the way, the difference between producing language and actually reasoning got lost.
What Burry means by the AI “parameter trap”
The “parameter trap” is what happens when an industry decides that bigger is the same as better. More parameters, more compute, more data, bigger models.
Each step up produces visible improvements in benchmark scores and demo performance. But Burry’s argument is that scaling language doesn’t solve the underlying reasoning problem. It just makes the simulation more convincing.
That has real financial consequences. The companies spending hundreds of billions on AI infrastructure are betting that scaling gets you there.
Hyperscaler AI spending could hit $725 billion in 2026, according to TheStreet. The Philadelphia Semiconductor Index is up 88% this year. Nvidia‘s market cap is roughly $5.45 trillion with a trailing P/E of 43. All of that is built on the assumption that the current scaling path works.
Burry is saying it might not. If the industry is spending at this scale to improve something that isn’t the actual target, the economics of the whole trade look different.
Burry’s broader bets against the AI trade right now
The Substack post didn’t come out of nowhere. Burry has been building short positions against the AI trade for months. He’s disclosed shorts on Nvidia, Tesla, Micron, Applied Materials, Caterpillar, and the iShares Semiconductor ETF. The Micron short he disclosed on July 1 came after the stock had already run nearly 700%, as TheStreet reported.
His analysis of semiconductor valuations pointed to the Philadelphia Semiconductor Index trading near peak levels within its 15-year forward P/E band. His argument is that chip stocks are rising because hyperscalers are spending heavily on AI, and hyperscalers are spending heavily because chip stocks keep rising. The feedback loop looks like demand but is partly just reflexivity.
The Magnificent 7 lost more than $2.2 trillion in market value in June 2026, according to CNBC. Burry has been short through that decline and is adding to the argument rather than easing off it.
Cho/Getty Images
Why Burry’s technical critique is harder to dismiss than valuation arguments
Most AI bears make valuation arguments. Stocks are expensive, spending is high, returns haven’t materialized yet. Those arguments are real but they’re also familiar, and the market has heard them before without much consequence.
Burry’s July 10 post is different because it’s attacking the architecture, not the price tag. He’s saying the current generation of AI might be getting better at something that isn’t actually the goal.
A system that produces more convincing language output isn’t necessarily getting closer to genuine reasoning. And if that’s true, the scaling thesis breaks down in a more fundamental way than valuation critics have described.
He predicted the AI story “may die a death by a thousand cuts.” That phrasing is deliberate. He’s not calling for a single crash moment. He’s describing a gradual erosion of confidence as the gap between what AI can do and what investors assumed it could do becomes harder to paper over with bigger models and better benchmarks.
What Burry’s warning means for investors tracking AI stocks
Burry has been early before. He called the market a sell in August 2023 and it subsequently rose 66%. He’s been bearish on Nvidia for over a year while the stock kept climbing. Being right about a structural argument doesn’t mean the market agrees with you on any particular timeline.
What’s different about the July 10 post is the specificity. He’s not just saying AI is expensive. He’s pointing to a specific technical decision made at the beginning of the current AI wave and arguing it set the whole industry on the wrong track. That’s a harder thing for the bulls to wave away with an earnings beat.
For investors holding AI infrastructure stocks, chip names, or any company whose valuation rests on the assumption that current-generation LLMs are on the path to genuine intelligence, Burry’s argument is worth reading carefully.
He might be wrong. He’s been wrong before. But the last time he made a structural argument this specific about a major market, he turned out to be right.
Related: Michael Burry doubles down on stock market, AI message for 2026
