Something is changing in the unit economics of software

Software's traditional unit economics—where distributing to millions cost nearly the same as one user—are being fundamentally disrupted by AI, as LLM inference calls introduce real, per-user computational costs that shrink margins from the historical 75-85% to around 52%. This creates a new hardware-like tradeoff where choosing cheaper models protects margins but risks product quality, forcing founders to reconsider the aggressive customer acquisition playbook that defined SaaS for the past fifteen years.