AI Copium follows investor Gavin Baker's conversation with Patrick O'Shaughnessy about why many people close to frontier AI have become more bullish. Baker says he looked for quantitative evidence that the boom was slowing, but instead found faster capability progress, severe compute scarcityAI compute infrastructure is the hardware, facilities, networks, power, cooling, storage, and software used to train and run AI models. and very limited adoption of existing agentic systems.
The bull case depends on more than larger models. Continual learningContinual learning lets an AI system acquire new knowledge or skills over time while retaining what it learned earlier. and better sample efficiencySample efficiency measures how much useful learning an AI system achieves from a given number of training examples or experiences. could create another capability discontinuity, while successive adoption waves could spread AI through software, physical infrastructure and labor markets. Baker expects data-center demand to remain strong because useful inference is still constrained by available compute.
The largest uncertainty is whether regulation and public consentGovernance defines decision-making authority, rules and accountability within a system or organization. can keep pace with the ambitions of frontier companies. The discussion highlights a growing gap between insiders who expect rapid transformation and a broader public that has not accepted its economic or social consequences, leaving open whether the current confidence reflects genuine insight or a technology bubble.
Watch on YouTube


