Why smarter AI models could drive up compute prices 10x

The widening gap between AI lab revenue growth—projected at 10x annually—and compute capacity scaling—limited to 3x—creates a critical bottleneck in the industry. To reconcile this, labs must navigate rising inference margins, increased compute costs, and shifts in how compute is allocated between training and inference. While standard economic theory might suggest that increased AI labor supply would lower costs, the inelastic nature of compute supply and the high value of intelligence suggest that compute prices will remain high. Leading labs are already experiencing these pressures, as evidenced by premium pricing for large-scale GPU clusters. Ultimately, the strong economies of scale inherent in model development favor entities that can maximize efficiency, leading to significant power concentration and making it increasingly difficult for competitors to challenge established frontier labs in the current pre-singularity regime.