04 Jun 2026
Cloud cost is an architecture problem
The invoice is a symptom. The cause is usually how data moves, how often jobs run, and what you left running overnight.
Finance sees a line item. Engineering sees a surprise. Both are looking at architecture after the fact. Queries that scan too much. Jobs that retry without a limit. Environments nobody turned off.
You cannot discount your way out of that. You change the shape of the system.
Know what each request costs
If you cannot point at the expensive path, you will guess. Tag resources. Watch the top queries. Watch the jobs that run when nobody is using the product.
A dashboard without an owner is decoration. Assign the number to a team that can change the code.
Design for idle
Products sit still more than they sit under load. Pay for what you use when you can. Scale to zero where it is honest. Do not keep a large box warm for a report that runs twice a week.
Staging is a common leak. Treat it like production for cost, even if it is not for traffic.
Fix the query before you buy more
More capacity hides a bad access pattern for a month. Then the bill returns, larger. Indexes, caches, and a store that matches the read are cheaper than a bigger instance.
We treat spend as a design input, not a later apology.
