Will your AI SaaS actually make money?
Token costs vs subscription revenue, end to end. See your real gross margin, breakeven users, and the lever that's killing your unit economics.
Gross margin
93.8%
$13,605 on $14,500 MRR
Token cost / user
$0.15
1% of price
Conservative
83.8%
Realistic
93.8%
Aggressive
95.0%
Bands swing on token-mix drift (longer answers), power-user concentration, and model price changes.
Healthy SaaS margin (70%+). You have room to invest in growth and survive a model price hike.
Infra is the drag at this scale. Re-evaluate vector DB tier, idle compute, and observability tools — most teams overspend 2–3x.
Breakeven users
15
At $28.01 contribution per user
- Token cost / user = (input tok / 1k × input $) + (output tok / 1k × output $)
- Revenue = users × price
- Total COGS = (token cost / user × users) + monthly infra + (revenue × payment fee %)
- Gross margin = (revenue − COGS) ÷ revenue
- Contribution / user = price − token cost / user − (price × payment fee %)
- Breakeven users = monthly infra ÷ contribution / user
Common questions
What gross margin does a healthy AI SaaS need?▾
75%+ is the SaaS norm. AI wrappers often run 40–60% because the model is the COGS. Anything under 50% means you're either a thin wrapper or pricing too low — usually both.
How do I estimate tokens per user?▾
Look at your average request size × avg requests per user per month, both input + output tokens. A chat-heavy app is often 50k–500k tokens/user/month. A copilot or agent can hit millions.
What pricing model works for AI SaaS?▾
Usage-based or hybrid (low base + metered overage) is safer than flat-rate. Flat-rate works only if your power users are <5% of accounts and your model cost stays predictable.
Should infra include vector DB and embeddings?▾
Yes. Add Pinecone/Weaviate/pgvector hosting and embedding generation costs to the fixed monthly. They scale separately from chat tokens but they're real COGS.