Every service is its own deployable unit — which means every service reasonably wants its own container image (Docker Fundamentals covers building these), its own CI pipeline (CI/CD Pipelines Explained covers what that pipeline actually does), and typically its own entry in whatever orchestrator (Kubernetes, ECS, Nomad) is scheduling and running all of them together. This is precisely the "operational overhead" cost from earlier, made concrete: a team moving from one monolith to ten services isn't just splitting code, it's taking on (or needing a platform team to already provide) ten pipelines, ten sets of infrastructure-as-usage metrics to watch, ten things that can each independently need a security patch, and ten services' worth of inter-service network configuration and access policy to keep straight.
This overhead is real, but it's also largely fixed cost that a platform/infrastructure investment (a shared deployment template, a shared observability stack, a service mesh handling cross-cutting network concerns like retries and mTLS uniformly) can amortize across many services — which is exactly why organizations with genuinely many microservices tend to also have a dedicated platform team whose job is making the marginal cost of adding the eleventh service small, rather than every team reinventing that infrastructure themselves.