Idle resources: the silent cloud budget killer
Nothing on your bill is labelled "wasted". Idle capacity looks identical to useful capacity — until you measure utilisation instead of provisioning.
Cloud providers bill for what you reserve, not for what you use. That single fact explains most of the waste we find.
Where idle capacity hides
- Non-production environments running 24/7 for a team that works 40 hours a week
- Databases provisioned for a launch peak that never returned
- Kubernetes nodes held by requests far above actual consumption
- Snapshots and volumes detached from anything, retained forever
- Old load balancers, NAT gateways and public IPs kept "just in case"
Measure utilisation, not spend
A resource costing 400 francs a month at 70 percent utilisation is healthy. One costing 120 francs at 4 percent is not. Ranking by cost alone leads you to optimise the wrong things. Rank by wasted spend: cost multiplied by unused capacity.
Make the default state "off"
Schedules beat clean-ups. Development and staging environments that shut down outside working hours cut their cost by roughly two thirds without a single architectural change, and the risk is low because the environments are re-creatable by definition.
Right-size continuously, not annually
Workload shape changes constantly. A quarterly right-sizing exercise is out of date the day it lands. Continuous recommendations, applied through the normal deployment pipeline, keep instance families and requests aligned with reality.
Close the loop
Every removed resource should have an owner who is told what changed and why. Optimization that surprises engineers gets reversed. Optimization that is visible and explained becomes a habit.