// training costs

All signals tagged with this topic

Generative AI's economics are finally catching up with reality

Three years of warnings about unsustainable unit economics and commodity-grade competition are vindicated: training costs have exploded (Claude 5.5 at $10 billion), inference margins are razor-thin, and a crowded field of similar models means no meaningful technical differentiation or pricing power. The trillion-dollar bet on AI rests on reliability and ROI improvements that studies increasingly question. Even optimistic progress trajectories may not arrive fast enough to justify capital already deployed, making consolidation and a reckoning on which use cases actually work likely outcomes.