So You Want to Sell Inference

Reselling inference at cost yields zero margin because customers compare prices to raw API costs and route around providers. Cost-plus pricing caps revenue at inference cost; value-based pricing charges per resolved task or generated report, decoupling price from compute. Optimization reduces inference cost to $0.70 via routing, caching, and distillation to small models.