Microservices Are a Tax (And Most Teams Can’t Afford It)

Microservices impose significant costs in latency, infrastructure spend, and operational overhead that most small-to-medium teams cannot justify. Network calls add 1–10ms per hop versus nanosecond in-process calls, service meshes like Istio can consume up to 90% of pod CPU, and observability tooling alone can cost $50K–$500K/year. Real-world examples show teams cutting API response times by 93% and infrastructure costs by 95%+ after consolidating back to monoliths. The 2025 CNCF survey shows 42% of organizations actively consolidating microservices. The core argument: tight coupling is the actual failure mode, not monoliths. A modular monolith with hard internal API boundaries delivers the same DORA metrics as microservices teams without the network tax. Microservices only pay off at 1,000+ engineers or with extreme traffic variance requiring independent scaling. Teams should default to modular monoliths, use tools like ArchUnit or Spring Modulith to enforce boundaries, and extract services only when a specific, measurable scaling or ownership problem demands it.