Lakebase Postgres scales 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 don't pay for compute while it's suspended; you continue to pay for storage.
How scale-to-zero works on Neon
The Lakebase Postgres architecture separates compute from storage. When the compute suspends, the Postgres process stops and CU-hour billing stops. 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 plan: scale-to-zero after 5 minutes, can't be disabled
- Launch plan: 5-minute default, can be disabled
- Scale plan: 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 the Launch plan or Scale plan.
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 about 186 CU-hours, or about $19.72/month, more than six times as much. Storage is billed separately at $0.35/GB-month on paid plans (0.5 GB/project included on the Free plan).
Data durability stays the same
Scale-to-zero doesn't change durability. Instant restore (point-in-time recovery) keeps working through suspends. On the Launch plan you can restore up to 7 days back; on the Scale plan, up to 30 days. The Free plan includes a 6-hour history window.
Other cloud Postgres services with scale-to-zero
- Aurora Serverless v2 (Postgres) 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 Postgres 13.15, 14.12, 15.7, 16.3 or later (docs). Storage persists during pause. Resume on the first connection is slower than Lakebase Postgres's few-hundred-millisecond wake.
- Supabase Free Plan projects are paused after about 7 days of low activity and can be restored for up to 90 days after the pause (docs). Paid plans do not pause projects when idle, so compute is billed 24/7 even with no traffic.
- RDS for Postgres 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 for free-plan abuse prevention rather than low-idle cost optimization on paid usage.
Scale-to-zero is on by default for every Neon project.