Break-Even Price Calculator for an AI Product

Works out the lowest price per user that covers your inference, your fixed costs and the payment fee — and what a heavy user does to it.

Requests per user per monthBillable model calls, not page views

Input tokens per requestPrompt, system prompt and tool schemas

Output tokens per request

Input rateUS dollars per 1,000,000 tokens. Example figure — replace it with yours.

$/1M

Output rateUS dollars per 1,000,000 tokens. Example figure — replace it with yours.

$/1M

Other variable cost per userStorage, logs, vector search — per user per month

$/user/mo

Fixed costs per monthAnything you pay whether or not anyone calls

$/mo

Paying customersThe fixed cost is spread evenly over these

Heavy-user multiplierHow much more a power user calls than average

Payment feePercent of the price the processor takes

Fixed fee per chargeThe flat part, per monthly charge

Target gross marginWhat you want left after the cost to serve

Break-even price per user per month

$12.91

Below this you lose money on every subscription. To clear 70% gross margin you need $42.33.

Token cost of one request: $0.0096
Token cost per user per month: $3.84
Other variable cost: $0.40
Fixed cost share ($4,000 ÷ 500 customers): $8.00
Cost to serve one customer: $12.24
Payment fees at the break-even price: $0.67
Price for 70% gross margin: $42.33
Break-even for a 3× heavy user: $20.82
Gross margin if you charged the break-even price: 0% — by definition

What this assumes: a month is 30 days and one charge per customer per month, so the fixed payment fee is counted once. Gross margin here means (what the processor pays you − the cost to serve) ÷ what the processor pays you: it excludes salaries, marketing and everything else below the gross line. Fixed costs are spread evenly over paying customers, which flatters you while you are growing and punishes you when churn bites. Free users are not modelled — add their inference cost to the fixed figure if you carry any. The pre-filled rates are round example numbers, not anyone's price list.

What this number is, and what it hides

The break-even price is not a price. It is the floor under one: the point where a customer stops costing you money, computed from usage you can measure rather than from what a competitor charges. Its job is to tell you when a pricing idea is arithmetically impossible before you spend a quarter discovering it.

Two terms move it more than anything else. The first is usage concentration. The break-even above is computed on the average customer, and averages are the wrong statistic for a metered cost, because the distribution of AI usage is not symmetric — a small share of accounts call ten times as much as the median. That is what the heavy-user multiplier is for: if your price sits below the heavy-user break-even, you are solvent only as long as heavy users stay rare, and your best customers are the ones you lose money on. The second is the payment fee, which is why break-even is a fixed point rather than a sum. You do not need to cover the cost to serve; you need to cover it out of what the processor hands over, which is why the formula divides by (1 − fee) rather than adding the fee on.

What it leaves out: engineering time, support, and the cost of the requests that produce nothing — retries, aborted streams, trials that never convert. It also assumes today's rates. Per-token prices have mostly fallen, but a model upgrade you want for quality reasons can move your cost to serve by a multiple overnight, and a price you announced is much harder to move than a model you deployed.

Break-Even Price Calculator for an AI Product

What this number is, and what it hides

The reasoning behind this

Related tools