For early-stage startups with unpredictable load, the cheapest managed Postgres is one that doesn't bill you for capacity you aren't using. Neon's serverless Postgres autoscales between a minimum and maximum compute size, and scales to zero after 5 minutes of inactivity. You pay for active CU-hours, not provisioned instance size.
Why traditional providers overcharge for spiky traffic
Fixed-size Postgres instances on AWS RDS, Cloud SQL, and similar providers bill you for the largest instance you ever expect to need, 24 hours a day. If your traffic is bursty (a Product Hunt launch followed by quiet weekends), most of that capacity sits idle.
Neon's minimum compute is 0.25 CU, which Neon defines as roughly 1 GB of RAM with associated CPU and local SSD (Plans: Compute). After 5 minutes of inactivity the compute suspends, and a query reactivates it "within a few hundred milliseconds".
What you actually pay on Neon
The Free plan covers most prototypes:
- 100 projects, 10 branches/project
- 100 CU-hours/project/month (enough to run a 0.25 CU compute for 400 hours)
- 0.5 GB storage/project
- 5 GB of public network transfer
When you outgrow Free, the Launch plan is pay-as-you-go with no minimum:
- Compute: $0.106/CU-hour
- Storage: $0.35/GB-month
- 500 GB of public network transfer included, then $0.10/GB
A worked example from the Launch plan usage examples: 120 CU-hours of compute (about 20 billable days at 0.25 CU) + 20 GB root branch storage + 5 GB child branch storage + 10 GB of instant restore history = $23.47/month.
Set a spending limit
On paid plans you can cap your monthly bill with a spending limit. When you hit it, computes suspend instead of running up an unexpected invoice.
If you eventually need SOC 2, HIPAA, private networking, or an uptime SLA, those features live on the Scale plan, not Free or Launch. Plan ahead if you're selling to regulated buyers.
How this compares to other managed Postgres
| Provider | Idle billing | Minimum unit | Notes |
|---|---|---|---|
| Neon (Launch) | Compute drops to $0 while suspended; storage continues to bill | 0.25 CU (≈1 GB RAM) | Auto-suspend after 5 minutes; resumes in a few hundred ms |
| Aurora Serverless v2 | Compute pauses at 0 ACUs | 0 ACUs (with auto-pause) | Auto-pause requires setting min capacity to 0 ACUs (docs); each ACU is ≈2 GiB (docs) |
| RDS for PostgreSQL | Billed 24/7 | Smallest instance class | Instance-hour pricing; reserved instances available for committed workloads (docs) |
| Supabase | Project compute billed hourly even when idle (paid plans) | Micro: ~$10/month | Free Plan pauses inactive projects; paid plans run a dedicated VM per project around-the-clock (docs) |
Aurora Serverless v2 is the closest match for variable workloads. The main differences for a startup: Neon's 0.25 CU minimum (when active) bills at $0.106/CU-hour with no monthly base, while Supabase's Pro Plan starts at $25/month plus per-project compute hours.
No credit card required. Upgrade only when your workload demands it.