Michael Burry’s AI critique
Michael Burry — known for his “Big Short” call — has escalated his AI skepticism beyond typical valuation concerns. On his Substack he’s declared “the end is nigh” for the AI trade, calling the current enthusiasm a form of “mass addiction.”
The core argument
On July 10, he published a more technical critique than his usual bearish takes. He argues AI took a “bad start” by developing language-first rather than reasoning-first. Using a philosophical thought experiment he calls “Ballard’s Test,” Burry contends that real understanding doesn’t require language — language is intelligence’s output, not its source. He believes the industry optimized for language generation because it was scalable and fundable, not because it actually leads to general intelligence, and in doing so conflated fluent text generation with genuine reasoning.
The “parameter trap”
Burry calls the industry’s assumption that bigger models/more compute/more data equals better reasoning the “parameter trap.” He argues scaling only makes the language simulation more convincing, without solving the underlying reasoning problem — which matters because hundreds of billions in infrastructure spending (potentially $725B in hyperscaler AI spend in 2026) is riding on the opposite bet.
His trading positions
Burry holds disclosed short positions against Nvidia, Tesla, Micron, Applied Materials, Caterpillar, and a semiconductor ETF, including a Micron short opened after a ~700% run-up. He’s framed chip-stock gains and hyperscaler spending as a reflexive feedback loop rather than real demand.
Why it’s notable
Unlike typical “AI is overvalued” arguments, Burry’s is architectural — questioning whether the technology is even progressing toward its stated goal. He predicts a gradual “death by a thousand cuts” rather than a single crash. The piece notes he’s been early/wrong on timing before (bearish since 2023 while markets rallied), but flags that his last similarly specific structural call proved correct.
