Neon suspends a Lakebase Postgres compute after a period of inactivity and resumes it on the next connection. The Free plan and Launch plan suspend after 5 minutes, and the Scale plan lets you set anything from 1 minute to always on. Amazon Aurora Serverless v2 also auto-pauses at a minimum capacity of 0 ACUs, on Aurora PostgreSQL 13.15, 14.12, 15.7, 16.3, or later.
How scale to zero works
A provisioned Postgres instance keeps running when no one is connected, and you pay for it by the hour either way.
The lakebase architecture separates storage from compute. After 5 minutes of inactivity (by default), Neon suspends the compute. Your data stays in storage, which bills separately: $0.35/GB-month on paid plans, or up to 1 GB per project on the Free plan.
Per-plan settings from Neon plans:
- Free plan: 5-minute idle timeout, can't be disabled
- Launch plan: 5-minute idle timeout, can be disabled
- Scale plan: 1 minute to always-on, fully configurable
Cold starts and connection limits on resume
When the next query arrives, the compute resumes, typically within a few hundred milliseconds. If that latency matters to your application, disable scale to zero on the Launch plan or Scale plan.
A traffic spike right after resume can also run into connection limits. Neon sets max_connections by compute size, and a 0.25 CU compute (≈1 GB RAM) allows 104 connections, with 7 reserved for Neon's superuser. That leaves 97 direct connections for your app, and a burst of serverless function invocations can use them all.
Neon's PgBouncer-based connection pooling accepts up to 10,000 client connections per compute and routes them through a smaller pool of Postgres connections. The pooled connection string has -pooler in the hostname.
postgresql://alex:AbC123dEf@ep-cool-darkness-123456-pooler.us-east-2.aws.neon.tech/dbname?sslmode=require&channel_binding=requireUse pooled connections from serverless
Runtimes that open a connection per request (Vercel Functions, AWS Lambda) should use the pooled string. Use the direct string for schema migrations, pg_dump, logical replication, and session features like LISTEN/NOTIFY.
How other Postgres providers compare
| Provider | Pauses when idle | Notes |
|---|---|---|
| Neon | Yes, after 5 min idle by default | Can be disabled on Launch; 1 minute to always on with Scale; storage still bills |
| Aurora Serverless v2 (Postgres) | Yes, when min ACU is 0 | Requires Aurora PostgreSQL 13.15, 14.12, 15.7, or 16.3+; pauses per DB instance |
| Amazon RDS for Postgres | No | Instance pricing is hourly regardless of activity; you can stop a database instance manually, but RDS auto-starts it after 7 days |
| Supabase | Free plan only | Free projects pause after a week of inactivity and must be resumed from the dashboard; paid projects don't pause |
With Aurora Serverless v2, you set a cluster minimum of 0 ACUs and an idle timeout, and Aurora pauses a DB instance when it has had no user connections for that long. The writer and readers with failover priority 0 or 1 pause and resume together, while readers with priority 2 to 15 can pause independently. See Scaling to Zero ACUs with automatic pause and resume.
Amazon RDS for Postgres doesn't auto-pause. Instances are billed by the hour for the chosen instance class whether or not connections are active.
Supabase pauses Free plan projects that don't get enough database activity over a week, and a paused project stays paused until you resume it from the dashboard. That's a way to reclaim unused Free projects, not an idle timeout between requests.
Vendor details verified on 2026-09-23 against the linked pages.
On the Free plan, computes suspend after 5 minutes of inactivity.