Markets and business models
Infrastructure COGS
When you sell infrastructure you often resell someone else's, plus software on top. Margin comes from the software, not the cheapest pass-through.
When you sell infrastructure, you often are a reseller of someone else's infrastructure plus software and ops on top. Twilio sits on carrier fees. Snowflake sits on cloud compute and storage. A model API sits on GPU-seconds. The winning pattern is not "cheapest resale." It is software leverage on top of COGS: reliability, developer experience, compliance, routing, observability, workflow, brand trust.
Concept
COGS are the direct costs of delivering the product. Infrastructure and AI APIs frequently land around 50–60% gross margin, not classic SaaS 80%. See gross margin and usage-based pricing.
The margin stack
| Layer | Example COGS | Margin shape |
|---|---|---|
| Telecom / messaging | Carrier fees per message | Structurally capped; Twilio-class gross margins historically ~50% band |
| Cloud data | Cloud compute/storage pass-through | Improve via efficiency, reserved capacity, multi-tenant density |
| Model inference | Tokens / GPU-seconds from a provider or your fleet | Volatile; provider price cuts help and compress your markup |
| Payments-adjacent | Interchange, network fees | Often a take-rate on volume with thin net |
Customers pay a premium over raw commodity when the premium saves engineering time or risk. If your only story is "we mark up tokens 1.5×," a competitor or the provider will undercut you.
Walking through a price cut
You charge $20 / 1M tokens. Upstream costs $8. Other variable COGS $2. Contribution = $10. Gross margin = 50%.
The provider cuts from $8 to $4. If you keep list at $20, margin expands to ($20 − $4 − $2) / $20 = 70% — until competitors pass savings through and you are forced to $12 list. Plan for pass-through races. Compete on the value layer, not the markup.
Price from customer value and alternative cost, then check that COGS leave acceptable gross margin. Cost-plus pricing ("tokens × 1.5") is easy to undercut and teaches customers to shop on markup.
Engineers at usage companies own cost observability per tenant, cache hit-rate work that is literally margin, and rate limits that are financial controls. A missing user_id on a span is a margin bug.
Common mistakes
- Competing only on pass-through price. A commodity API without DX or trust loses on price.
- Cost-plus pricing. Customers learn to shop the markup. Value-based price, then check margin.
- Ignoring support COGS that scale with confused metering.
Check yourself
List $20, upstream $8, other COGS $2. What is gross margin?
50%. ($20 − $8 − $2) / $20 = 0.50.
Upstream falls to $4. You keep $20. What happens next in a competitive market?
Your margin jumps to 70%, then rivals pass the savings through and list prices fall. Compete on the value layer.
Key takeaways
- Infrastructure gross margins are often ~50–60%, not classic SaaS 80%.
- Software value-add must justify the premium over commodity.
- Provider price cuts expand margin until a pass-through race starts.
- Price from customer value, then check COGS. Do not lead with cost-plus.
Further learning
Next, the usual motion: developer land, enterprise expand.
Related
API, infrastructure and usage-based economics
Usage-based pricing, COGS-heavy businesses and the economics of AI API products.
Gross margin and COGS, including AI
What belongs in COGS vs opex, why classic SaaS runs ~75–80% gross margin, why AI products often sit at 50–60%, and the levers that move the number.