Scale to zero, pay per request

A database that scales to zero stops charging for compute when nobody uses it, and comes back when the next query arrives. There are two shapes. Request-priced services, such as DynamoDB on-demand, bill each read and write and never have a server to pause, so an idle table costs only its storage. Serverless versions of server databases, such as Neon or Aurora Serverless v2 with a minimum of 0 capacity units, suspend the compute after a quiet period (five minutes by default on Neon, configurable on Aurora) and start it again on the next connection.

The cost of this is the first request after a pause and the limits on growth. Neon resumes in a few hundred milliseconds, while Aurora’s resume can take up to 15 seconds, so clients need generous connect timeouts and retries. Open connections, replication or a connection proxy can keep the database from pausing at all. Request-priced services also cap how fast traffic can grow: DynamoDB on-demand absorbs up to double a table’s previous peak at once. When you compare, check the minimum bill, what counts as activity, the resume time, how storage is priced, and the point at which provisioned capacity becomes cheaper.