Neon charges for compute by the second, in CU-hours (compute-unit hours). When your database is idle, it suspends after a configurable timeout and stops accruing compute charges entirely. There's no per-instance monthly fee, no minimum, and no separate charge to keep a small database around.
How CU-hours work
A Compute Unit (CU) is approximately 4 GB of RAM plus matching CPU and local SSD. Sizes range from 0.25 CU (≈1 GB RAM) up to 56 CU (≈224 GB RAM, Scale plan only). Compute usage is compute size × hours running = CU-hours.
Pricing by plan:
- Free: 100 CU-hours included per project per month. That's 400 hours of a 0.25 CU compute, enough to keep a small project always-on.
- Launch: $0.106/CU-hour, autoscaling up to 16 CU.
- Scale: $0.222/CU-hour, autoscaling up to 16 CU or fixed sizes up to 56 CU.
A 0.25 CU database running for 5 hours costs 0.25 × 5 = 1.25 CU-hours, or about $0.13 on Launch. If the same database stays online but idle for the other 19 hours, scale-to-zero kicks in and those hours don't count.
Scale-to-zero in practice
After 5 minutes of inactivity (the default), Neon suspends the compute. On the next query, it wakes up. On the Launch plan you can disable this; on Scale it's configurable from 1 minute up to always-on.
# Sample Launch-plan bill for a small side project
# Active 4 hours/day on 0.25 CU compute
# 0.25 × 4 × 30 = 30 CU-hours × $0.106 = $3.18/monthCold starts
Suspended computes wake on the first connection. The wake is typically fast but isn't free latency. For a latency-sensitive API, set a higher minimum compute size or disable scale-to-zero on the Launch and Scale plans.
Storage is billed separately at $0.35/GB-month, and instant-restore history at $0.20/GB-month. The Neon usage-based cost examples page walks through full bills at different usage levels.
How other managed Postgres services bill compute
| Provider | Billing unit | Scale to zero |
|---|---|---|
| Neon | CU-hours, billed by the second | Default, suspend after configurable idle window (1 minute to never on Scale) |
| Aurora Serverless v2 | ACU-hours (1 ACU is ~2 GB RAM) | Supported via auto-pause, requires Aurora PostgreSQL 13.15+, 14.12+, 15.7+, or 16.3+ |
| RDS for PostgreSQL | Per-instance hour at the chosen instance class | Not supported. Stopping a DB instance pauses for at most 7 days before AWS restarts it |
| Supabase | Compute add-on hour at the chosen size | Not supported on paid plans. Free-tier projects pause after inactivity |
Aurora Serverless v2 supports scale-to-zero through the auto-pause feature when you set the minimum ACU to 0. It requires Aurora PostgreSQL 13.15, 14.12, 15.7, 16.3 or later. See Scaling to zero ACUs with auto-pause. On resume, capacity may come back at a lower ACU than at pause.
RDS for PostgreSQL is billed per hour for the chosen instance class. There's no auto-pause; you can stop a DB instance manually but AWS restarts it after seven days. See DB instance classes.
Supabase bills a flat hourly rate per compute add-on size. A Micro project costs $0.01344 per hour even when idle. See Supabase compute usage.
See worked examples of monthly costs for Launch and Scale workloads in the Neon plans documentation.