Idle Cost
The portion of infrastructure spend incurred regardless of traffic, which determines how cost per request behaves as volume changes.
Cost per request is not constant, and understanding why is what makes it useful.
Fixed costs — reserved instances, always-on databases, load balancers, licences, monitoring — are incurred whether or not a request arrives. Variable costs — additional compute under load, data transfer, per-invocation charges — scale with traffic.
So cost per request falls as volume rises, because the fixed base is amortised over more requests. A service at 10% utilisation has a cost per request roughly ten times what it would have at full load, and that is usually the real finding when a service looks expensive.
Three consequences:
A low-traffic service is expensive per request by construction. The remedy is consolidation or a serverless execution model that scales to zero, not optimisation of the request path.
Comparing cost per request across services with different traffic profiles is meaningless unless idle cost is separated out.
Idle cost is where the easiest savings are — non-production environments running overnight and at weekends, over-provisioned baselines, and orphaned resources that serve no traffic at all.
Measuring it is simple and rarely done: spend during the lowest-traffic hour, annualised, gives the floor the architecture is paying regardless of use.