Software investors are used to high gross margins, but AI products introduce variable costs that can scale directly with usage. Revenue growth therefore tells only part of the story.<\/p>\n
Inference, third-party model APIs, media generation, usage-linked data services and human review may belong in cost of revenue when they are required to deliver the paid product.<\/p>\n
A highly engaged customer is not automatically a profitable customer. If usage grows faster than subscription revenue, gross margin can fall as engagement rises.<\/p>\n
Usage-based credits can protect margin for expensive features such as image or video generation. The trade-off is more pricing complexity.<\/p>\n
Routing, caching, smaller models and better prompts can reduce cost without changing headline features. Investors should ask how cost per successful task has changed over time.<\/p>\n
Some AI workflows rely on manual review, labeling or teleoperation. These costs should not disappear into general operating expenses if they scale with delivery.<\/p>\n
Text, image and video features can have very different economics. Blended margin may hide an unprofitable fast-growing feature.<\/p>\n
Healthy AI companies can show that quality and retention are improving while unit inference cost declines or monetization rises. The direction of the curve matters as much as the current margin.<\/p>\n