In a widely shared post on X, Atreides CIO Gavin Baker argued that the strongest long-term case for AI infrastructure is a shift away from high-margin frontier labs and toward cheaper models, whether open or closed. His case is simple: If intelligence gets cheaper per dollar, customers use more of it, token demand rises, and more profit flows to infrastructure providers with the lowest per-token costs. Baker also wrote that this is "not happening yet," saying the most capable models still capture most of the economic value today.