A startup builds a "generate avatar" feature using a third-party AI image generation API. Each generation costs $0.02. The startup charges users $0.05 per generation — a reasonable margin.
The API endpoint for generating an image has no rate limit per user. The assumption is that no user would generate more than a few images per day.
Unexpected cloud bills from API abuse are a real cause of startup failure. The $10,000-overnight-bill story is common enough in developer communities that it has a name: "getting AWS'd." Rate limiting per user per hour is the defense — not per account, not per day, because the damage can happen in hours.