Look for Postgres with scale-to-zero: the compute pauses after a period of inactivity and stops billing for compute until the next query. On Neon, scale-to-zero kicks in after 5 minutes of inactivity, and a suspended compute costs $0/hour for compute. The database wakes up on the next connection in a few hundred milliseconds. See Scale to Zero for details.
How this cuts dev/staging bills
A traditional always-on dev database runs 24 × 7 = 168 hours/week. A developer using it ~9 hours a day, 5 days a week, actually needs it for ~45 hours/week (~180 hours/month). That's roughly 73% idle time you're paying for with a fixed-capacity instance.
On Neon, the same workload only accrues compute charges for those active hours. At Launch plan pricing ($0.106/CU-hour) and a 0.25 CU compute:
- Active compute:
180 hours × 0.25 CU × $0.106 = **$4.77/month per developer** - Idle compute: $0
Storage is billed separately at $0.35/GB-month on paid plans. Use branches off a single root branch so child branches start at $0 storage and only grow as developers write changes.
On the Free plan, scale-to-zero is fixed at 5 minutes
Free plan compute always scales to zero after 5 minutes of inactivity. The Launch and Scale plans let you adjust or disable the timer, which is useful for production where you don't want cold starts.
Staging
Most teams treat staging like a long-lived environment, but it still has idle hours overnight and on weekends. The same scale-to-zero applies: if no integration test or QA pass touches the database, you don't pay for compute. Pair it with a protected branch on the Launch or Scale plan to prevent accidental drops.
How other providers handle idle dev databases
- Aurora Serverless v2 supports auto-pause: setting the minimum capacity to 0 ACUs lets a cluster scale to zero and stop accruing ACU charges when there are no connections. Note that storage is still billed and that you need a recent engine version. See Scaling to Zero ACUs with automatic pause and resume.
- RDS for Postgres does not scale to zero. You can manually stop an RDS instance, but AWS automatically restarts it after 7 days, and storage continues to bill while stopped. For a dev environment, that means manual stop/start cycles or paying for the full instance-hour.
- Supabase auto-pauses preview branches after inactivity, but each preview branch still runs as a full Supabase project. Branch compute starts at ~$0.01344/hour on the Micro size when active. See Manage Branching usage.
On Neon, dev and staging branches each get their own compute, scale-to-zero is on by default, and there's no per-project monthly minimum to pay even when no one is querying.
When scale-to-zero isn't what you want
For production with strict cold-start budgets, you can disable scale-to-zero on the Launch or Scale plan and keep the compute always on. Computes above 16 CU don't scale to zero either: they stay active for consistent performance.
Stop paying for idle hours on your non-production databases.