Michael Burry's next 'Big Short': An inside look at his analysis showing AI is a bubble (www.cnbc.com)

Michael Burry — fresh from deregistering his hedge fund but still actively investing — is publicly backing research by former Scion analyst Phil Clifton arguing that AI is becoming a speculative bubble. Clifton’s notes, cited by Burry, claim generative-AI demand economics don’t justify the massive infrastructure buildout: OpenAI may approach $20B in revenue, but hyperscalers have pushed capex toward ~$400B annually with forecasts of ~$3T over five years. Scion likens the rush to the early-2000s telecom overbuild (where capacity vastly outstripped demand and prices collapsed), and points to early signs of strain — Microsoft cancelling large data-center projects and Alibaba warning of a bubble — to argue that supply could outrun genuine economic use. Key technical and market implications target the compute stack and valuation models central to AI/ML deployment. Scion highlights a mismatch between rapid GPU product cycles (Nvidia’s chips refresh yearly) and extended depreciation schedules for servers (often six years), which could leave datacenters holding obsolete, inefficient hardware before it’s fully written down. Nvidia counters that software (CUDA) preserves value, but critics say both claims can’t hold simultaneously. For the AI/ML community this raises practical risks: rising capital intensity and hardware obsolescence could push teams toward efficiency-focused model design, edge/heterogeneous deployments, or open-source alternatives — and may compress margins or valuations for companies predicated on perpetual, high-growth GPU demand.