AI cost per user: is your AI feature profitable?
Many AI products lose money on their heaviest users. This tells you your AI cost per user, your gross margin per user, and whether your pricing covers your token bill.
Per user, per month
Defaults = Claude Opus 4.8 published rates. Prices editable — verify at your provider.
These figures are before other costs — infrastructure, salaries, support, payment fees. Gross margin per user is the headroom AI inference leaves you, not your net profit.
LLM unit economics, in one number
The question is my AI app profitable? usually can't be answered from a blended monthly bill — it has to be answered per user. A flat $20/month plan looks healthy on the average user and quietly bleeds on the power user who runs ten times the requests. This cost per user calculator turns your token usage into AI cost per user and a gross margin per user, so you can see AI feature profitability before it shows up in your runway.
As a rule of thumb for LLM unit economics: keep inference well under a third of the price you charge, and you have room for infra, support and a real margin. Let it drift past half, and an AI margin calculator like this one starts flashing red — usually a sign you need model routing and prompt caching, not a price hike.
| AI cost as % of revenue | What it means | Verdict |
|---|---|---|
| Under 25% | Inference is a small share; margin scales with you | Healthy |
| 25–50% | AI is eating into margin; heavy users get risky | Watch it |
| Over 50% / negative | You lose money per user without optimization | Unprofitable |
Thresholds are illustrative guardrails, not hard rules — your right number depends on your other costs and growth stage.
Make every user profitable.
If AI is eating your margin, the fix is engineering, not a price hike. Model routing, prompt caching and right-sized infra keep cost-per-user flat as you scale. Book a free build audit and we'll measure where your spend actually goes.
Book a Build AuditQuestions about AI cost per user
How do I calculate AI cost per user?
Multiply a user's monthly AI requests by the token cost of each request — average input tokens times your input price plus average output tokens times your output price, per million tokens. Subtract that from what you charge per user to get your gross margin per user before infrastructure and team costs.
What's a healthy AI cost as a share of revenue?
It varies, but if AI inference is more than 25–30% of the price you charge, margin gets tight fast as you scale — and your heaviest users can become unprofitable. Model routing, prompt caching and batching typically cut AI cost 30–90% without lowering quality, which is what keeps per-user economics healthy.
Why do AI products lose money on power users?
Flat subscription pricing with usage-based AI cost means a small number of heavy users can consume far more tokens than they pay for. Without routing cheaper requests to cheaper models and caching repeated context, per-user cost climbs with usage while revenue stays flat.