Same $99/mo, $15 vs $140 in Inference: The Hidden Unit Economics of AI SaaS

startupdevrim · reddit · 2026-11-20

A developer lays out the cost-accounting problem unique to AI SaaS: subscription revenue sits in Stripe and model spend in OpenAI/Anthropic dashboards, but mapping the two per customer is hard once you have multiple models, retries, failed workflows, and wildly different usage. Two customers paying $99/month might generate $15 vs $140 in inference costs — MRR looks fine while the account economics diverge completely. They poll practitioners on four failure modes: (A) unexplained LLM cost growth, (B) agents running expensive loops without budget limits, (C) customers consuming more than their subscription covers, (D) metering/billing errors causing missed revenue, and ask whether real incidents forced pricing changes, usage limits or billing rebuilds. A useful reference for anyone pricing AI features in production.

Original post →

More from coding & agent

coding & agent channel →