Model cost calculators tell you what inference costs. The decision you actually face is what that forces your price to be — and payment fees, which almost every AI cost model leaves out, are part of the answer.
AI SaaS Pricing Calculator
Turn your AI cost per user into the price you have to charge to hold your margin.
$/mo
$/mo
Hosting, storage, third-party APIs, support
%
$
Payment fees are part of cost of goods sold and are routinely left out of AI cost models, which quietly overstates the margin.
You need to charge at least$48.54per user per month for a 80% gross margin
- Variable cost
- $8.00
- Payment fees
- $1.71
- Total COGS
- $9.71
- Gross profit
- $38.83
Payment fees alone are 3.5% of the price at this point — worth a look before you cut the model spend.
Minimum price at other margins
60% margin$22.37/mo
70% margin$30.63/mo
75% margin$37.56/mo
80% margin$48.54/mo
85% margin$68.60/mo
90% margin$116.90/mo
Why you divide rather than add
The intuitive move is to take an $8 cost, want an 80% margin, and charge $8 × 5 = $40. That is close, but it ignores that payment fees are themselves a percentage of the price you have not set yet.
price = (variable cost + fixed fee) / (1 - fee rate - target margin)
= (8.00 + 0.30) / (1 - 0.029 - 0.80)
= 8.30 / 0.171
= $48.54$48.54, not $40. The gap is entirely payment fees, and it grows as the target margin rises.
AI margins are not SaaS margins
| Product | Typical gross margin | Main COGS |
|---|---|---|
| Classic SaaS | 80–90% | Hosting, support |
| AI-assisted SaaS | 65–80% | Inference plus hosting |
| AI-native product | 50–70% | Inference dominates |
| Heavy generative (video, agents) | 30–60% | Inference dominates heavily |
This matters for expectations as much as pricing. An AI product at 65% margin is not underperforming — it is a different cost structure, and investors and founders alike often benchmark it against the wrong number.
Levers other than price
- Route by difficulty. Most requests do not need your largest model.
- Cache aggressively. Repeated prompts and shared context are often a large share of spend.
- Trim context. Tokens you send every call are the ones you pay for every call.
- Lower the payment rate. At a $49 price point, moving from 2.9% + 30¢ to 2.6% + 25¢ recovers about 0.4 points of margin — small, but it requires no engineering.
GOOD QUESTIONS
Frequently asked
Why divide by (1 − margin − fee rate) instead of multiplying the cost?+
Because payment fees are a percentage of the final price, which you are still solving for. Multiplying cost by a margin multiple ignores them and leaves you short. At an 80% target and 2.9% fees, the difference on an $8 cost is roughly $40 versus $48.54.
What gross margin should an AI product target?+
50–70% is realistic for AI-native products where inference dominates COGS, versus 80–90% for classic SaaS. Insisting on 85% usually means either underpricing risk or capping usage so tightly that the product stops being useful.
Should I charge per seat or per usage for AI features?+
If your costs are usage-based, some usage component protects your margin. A common structure is a flat price covering typical usage plus overage above a generous threshold — predictable for the customer, safe for you.
Do payment fees really matter next to inference costs?+
Less than inference, but they are not negligible and they are far easier to change. At a $49 price, 2.9% + 30¢ is about $1.72 — roughly 3.5% of revenue, straight out of gross margin, with no engineering work required to reduce it.