Key Takeaways
- Burry describes AI’s focus on language models as a fundamental design mistake
- He presents “Ballard’s Test” — real intelligence must function without linguistic dependency
- The industry is caught in what Burry calls a “parameter trap,” endlessly expanding broken systems
- He identifies an irreconcilable tension between chip manufacturer needs and cloud provider goals
- Burry has initiated short positions targeting Nvidia, Tesla, and semiconductor ETFs
Michael Burry, renowned for his prescient call on the 2008 financial crisis, has now turned his critical eye toward artificial intelligence — and the tech giants building it.
Writing on his Substack platform “Cassandra Unchained,” Burry presented two intertwined critiques: one examining AI’s architectural foundations, the other dissecting its economic sustainability.
The Foundational Flaw in Modern AI
Burry presented a framework he terms “Ballard’s Test.” This concept proposes that authentic intelligence must demonstrate reasoning capabilities independent of linguistic structures.
According to Burry, the initial ambition of AI research centered on developing pure reasoning systems first. When that objective proved unattainable, the field pivoted toward language-centric architectures.
Burry characterizes this pivot as a “known flaw” and a “bad start.” His position is that the industry abandoned efforts to repair this foundational issue — instead choosing to build upon it.
The consequence, he argues, is what he labels a “parameter trap.” Rather than addressing the underlying deficiency, organizations are merely scaling up iterations of a structurally compromised framework.
He further emphasized that this methodology requires enormous computational resources — what he described as “zillions of power-hungry chips.”
The Economic Contradiction
Burry then shifted focus to the financial dynamics of AI, identifying what he sees as an inherent contradiction in the market.
Nvidia requires perpetual expansion in AI chip consumption. This continuous growth trajectory supports its current earnings and validates the valuation premium the market assigns to its shares.
Hyperscalers — giants like Meta, Amazon, and Microsoft — require exactly the inverse. Their business model demands that intense capital expenditure conclude within a three-to-four-year window, allowing operational costs to decline.
“The hyperscalers are promising permanent demand growth and temporary spending over 3-4 years all in the same breath,” Burry observed.
He contends these positions are mutually exclusive.
Burry also highlighted that free cash flow among leading hyperscalers is declining toward zero. While their earnings reports appear healthy, this is partially attributable to extended depreciation timelines that obscure actual expenditures.
He suggests AI optimists envision a “third door” — a scenario where demand remains robust while spending contracts, creating universal profitability.
Burry’s response is unambiguous: “There is no third door.”
He reinforced this perspective with portfolio moves, establishing short positions against Nvidia, Tesla, and the iShares Semiconductor ETF.
Regardless of whether his timing proves accurate, his thesis poses a fundamental question — who ultimately profits from AI, and is it possible for both semiconductor manufacturers and hyperscalers to succeed simultaneously?
Burry’s answer is definitive: no.





