API, infrastructure and usage-based economics

Usage-based businesses charge for what the customer consumes: API calls, tokens, gigabytes, messages, compute credits. The pitch is alignment — customers pay when they get value. The catch is that your costs often scale with usage too. That is a different animal from classic seat-based SaaS, where one more login is nearly free.

Twilio, Snowflake, Stripe, AWS and model APIs live here. AI apps that resell foundation-model tokens live here whether they admit it or not. Understanding the margin stack is the difference between a platform and a thin wrapper.

Seat SaaS often runs ~70–85% gross margin at scale. Infrastructure and AI APIs frequently land closer to 50–60% or lower early on. You have less room for S&M mistakes, and an "unlimited" plan can lose money on your best customer.

A worked margin

You charge $20 / 1M tokens. Upstream costs $8. Other variable COGS $2. Gross margin = 50%. A power user on a $20 unlimited plan who burns 5M tokens costs you $50 — negative 150% margin. A credit pack with overage would have billed that whale $95 against $40 of COGS. The rest of this section is that picture, one piece at a time.

Concepts in this section

Step	Concept	What it answers
1	Usage-based pricing	Why charge for consumption, and when to go hybrid?
2	Infrastructure COGS	Where does the 50% margin go, and what justifies a premium?
3	Developer land, enterprise expand	How does an API key become a commit?
4	AI wrapper economics	When is a markup just a reseller?
5	Commits and credits	How do you stabilize volatile usage revenue?
6	Metering and usage NRR	What do you count, and did dollars stay?

The path is usually self-serve land → usage growth → enterprise commits. Pricing pages that only say "contact sales" kill the land; pay-as-you-go only leaves enterprise money on the table. Price from customer value, then check that COGS leave acceptable margin. Cost-plus ("tokens × 1.5") teaches customers to shop the markup. What you count is the business: idempotent meters, spend alerts and a usage dashboard. Usage NRR is consumption of survivors — the same logo can cut 40% of spend without leaving the CRM.

Common mistakes

Negative-margin whales on uncapped plans.
Sales selling seats while COGS are usage.
Thin AI wrappers with no data, eval or workflow moat — a price war with the model provider.
Watching logo retention only. Usage NRR can collapse inside retained accounts.

Key takeaways

Usage pricing aligns spend with value and enables developer-led land — and revenue is volatile.
Infrastructure gross margins are often ~50–60%. Software value-add must justify the premium.
AI apps that only resell tokens face margin compression. Durable value sits above the model.
Commits, credits and hybrid packaging stabilize forecasting.
Always model power users and free tiers.