Neon's pay-per-query Postgres pricing
Neon
This pricing model turns idle time from a cost center into a growth lever. In a traditional managed database, a team picks a server size, pays for those vCores and memory by the hour, and keeps paying even when the app is quiet. Neon only runs compute during queries, then suspends after inactivity, so a dev database, preview app, or sporadically used AI agent backend can exist for near zero compute cost between bursts.
-
The workflow difference is concrete. On Azure Database for PostgreSQL Flexible Server, billing is tied to provisioned compute and storage, and Microsoft notes the server is billed by the full hour it exists. AlloyDB likewise prices compute by the vCPU and GiB of memory configured on an instance. That means sizing decisions drive cost before workload arrives.
-
Neon can price closer to electricity than rent because compute and storage are split apart. Pages stay in object storage, while lightweight Postgres compute nodes wake up on demand, scale from fractional to larger compute unit sizes, and suspend after about 5 minutes idle. That makes short lived environments economically viable, not just technically possible.
-
This changes expansion mechanics. Instead of upselling larger reserved instances, Neon grows when customers create more branches, run more pull request previews, and let AI tools provision many small databases automatically. By May 2025, Neon had reached $25M ARR, and by April 2024 had grown to more than 700,000 databases, with over 80% provisioned by AI agents.
The market is moving toward database pricing that matches bursty software creation. As AI agents and developer tools spin up far more temporary databases than humans ever did, the winners will be the platforms that make thousands of low duty cycle databases cheap to create, cheap to keep, and easy to wake instantly when work appears.