The quiet cost of cloud sprawl.
Most infrastructure bills are archaeology. Notes from three rescue projects.
In the last year we inherited three cloud accounts from teams that had scaled fast and audited never. The pattern was identical each time: nobody had made a bad decision. They had made two hundred small, reasonable ones — and stopped writing them down around decision forty.
Staging environments duplicated for a demo and never torn down. Managed databases provisioned for load that arrived once, on launch day. Log pipelines shipping terabytes nobody had queried in a year. Every line item defensible in the month it appeared; indefensible in aggregate.
Sprawl isn’t a spending problem. It’s a memory problem.
What actually works
Not a cheaper provider, and not a FinOps dashboard. What worked in all three rescues was making infrastructure legible again: everything in code, everything tagged to an owner, and a standing rule that anything without an owner gets a deletion date.
The bills dropped 40–60% — but the real win was slower to show. Deploys got faster because environments were understood. Incidents got shorter because the map matched the territory again.
The takeaway
If your infrastructure can’t be explained on one whiteboard by the people who run it, you don’t have a cost problem yet. You have a comprehension problem, and the invoice is just how it surfaces.