Neon scales Postgres compute to zero after 5 minutes of inactivity, and storage stays put. The next query wakes the compute in a few hundred milliseconds. Your data, history, and connection strings are unchanged. You pay for the time compute is running, not for the time it sits idle.
How scale-to-zero works on Neon
Neon separates compute from storage. When the compute suspends, the Postgres process stops and CPU/RAM bills go to zero. Storage is persistent on a separate layer, so nothing is lost. When a new connection arrives, a fresh compute spins up and reattaches to the same storage. See Scale to Zero and Compute lifecycle.
Behavior by plan:
- Free: scale-to-zero after 5 minutes, can't be disabled
- Launch: 5-minute default, can be disabled
- Scale: fully configurable, from 1 minute to always-on
Cold start cost
The first query after a suspend takes a few hundred milliseconds extra while compute wakes. Subsequent queries hit normal latency. If your app is latency-sensitive, disable scale-to-zero on Launch or Scale.
What it saves you
On the Launch plan, compute is $0.106/CU-hour. A 0.25 CU database that's only active 4 hours a day uses 30 CU-hours/month, or about $3.18/month for compute. The same database left always-on at 0.25 CU would use 186 CU-hours, or about $19.72/month, more than six times as much.
Data durability stays the same
Scale-to-zero doesn't change durability. Instant restore (point-in-time recovery) keeps working through suspends. On Launch you can restore up to 7 days back; on Scale, up to 30 days.
Other cloud Postgres services with scale-to-zero
- Aurora Serverless v2 (PostgreSQL) added scale-to-zero with the automatic pause feature. You enable it by setting the cluster's minimum capacity to 0 ACUs. The engine must be Aurora PostgreSQL 13.15, 14.12, 15.7, 16.3 or later (docs). Storage persists during pause. Resume on the first connection is slower than Neon's sub-second wake.
- Supabase Free Plan projects are paused after a period of inactivity and can be restored within 90 days (docs). Paid plans do not pause projects when idle, so compute is billed 24/7 even with no traffic.
- RDS for PostgreSQL does not support auto-pause. You can stop an instance manually, but billing for storage continues and stopped instances restart automatically after 7 days.
If your workload has long idle periods and tolerates a cold start, Neon and Aurora Serverless v2 are the two managed options that bill compute only while it's running. Supabase Free Plan offers project pausing but with a different intent (free-tier abuse prevention, not low-idle cost optimization on paid usage).
Scale-to-zero is on by default for every Neon project.