Serverless does not mean there are no servers — it means you do not manage them. You deploy functions or services, and the cloud provider handles provisioning, scaling and patching, billing you only for the compute you actually use. At zero traffic you pay nothing; at a spike it scales automatically.
It suits workloads that are spiky, event-driven or unpredictable, where paying for idle servers is wasteful. It suits less well workloads that run constantly at high volume, or that need very low, consistent latency, where the per-request pricing and cold starts can work against you. As with most infrastructure choices, the trade-off is specific, not universal.
Related terms