Complexity, network calls instead of function calls, distributed data consistency, harder debugging, and operational overhead. You need a real DevOps setup to run them.

A single request can hit multiple services. Tracing it requires distributed tracing and central logs. You cannot just set a breakpoint in one process and step through everything.

What was a transaction in a monolith becomes a saga across services. You need to handle partial failures, idempotency, and eventual consistency because no single database owns all the data.