Oct 15, 2026 · 34 min
AI’s growth model collides with copyright and data limits
‘Creating a bubble is their whole business model’ | David Gerard
The episode examines whether AI companies can remain viable when licensing costs, legal liabilities, and deteriorating training data challenge their expansion strategy.
- 1A ruling rejecting an AI company’s fair-use defense could force companies to pay enormous licensing bills for training data.
- 2Anthropic’s revenue is set against far larger liabilities, while synthetic data offers an expensive and limited answer to model collapse.
- 3Executives understood AI’s self-undermining risks, yet investor pressure keeps the industry expanding rather than addressing its doom loops.
Don't miss
David Gerard links executives’ awareness of AI “doom loops” to an expansion strategy that cannot afford to slow down.
The brief
David Gerard argues that AI companies’ performance claims may already reflect model collapse, while their business model depends on sustaining an investment bubble.
A court’s rejection of an AI company’s fair-use defense could make licensing unavoidable, leaving companies with enormous bills for training data they already used.
The discussion contrasts Anthropic’s substantial audited revenue with vastly larger liabilities and explains why synthetic data cannot cheaply replace high-quality human-created material.
Newly unredacted documents show executives understood AI could create “doom loops,” from weakened demand to declining data quality, but expansion remains financially necessary.
Gerard’s central warning is that copyright costs, model collapse, and short-term investor pressure may leave today’s AI companies unviable in their current form.
Books & mentions
Some links are affiliate links — PodLume may earn a commission if you buy.