- Serverless PostgreSQL means compute that autoscales, can scale to zero when idle, and is billed by usage rather than by instance size.
- Neon and Aurora serverless are the main serverless Postgres options; Aurora DSQL is serverless and PostgreSQL-compatible but a different, distributed engine.
- Supabase is not serverless compute: each paid project is an always-on instance billed hourly, although free projects pause after a week of inactivity.
- Scaling to zero saves money on idle databases but adds a wake-up delay on the next connection; providers document this as milliseconds to "slight delays", not zero.
- For a database busy around the clock, a fixed instance is often cheaper than paying serverless rates for every hour.
What a serverless PostgreSQL database is
A serverless PostgreSQL database is still PostgreSQL, but you do not choose and pay for a fixed server. Three properties usually define it:
- Autoscaling: compute grows and shrinks between a minimum and maximum as load changes.
- Scale to zero: after a period with no connections, compute is suspended and stops costing money; the next connection wakes it.
- Usage billing: you pay for compute time actually used (CU-hours, ACU-hours or processing units) plus storage.
To make this practical, serverless Postgres services separate compute from storage, so a compute node can be stopped or resized without moving the data. The trade-off is that a suspended database has to start compute again before it can answer, and some Postgres features that assume a local disk behave differently.
Serverless Postgres options compared
The table separates true serverless compute from products that only pause or stop. All facts are from each provider's documentation.
| Service | Engine | Scales to zero | Billing unit | Notes |
|---|---|---|---|---|
| Neon | PostgreSQL 14 to 18 | Yes, after 5 minutes by default | CU-hour (about 4 GB RAM per CU) plus GB-month | Branching; Free plan cannot disable scale to zero; AWS regions only |
| Aurora serverless (Aurora PostgreSQL) | Aurora PostgreSQL | Yes, with a minimum of 0 ACUs (auto-pause) | ACU-hour, billed per second (1 ACU is about 2 GiB) | Up to 256 ACUs; Multi-AZ, read replicas and Global Database supported |
| Aurora DSQL | PostgreSQL-compatible distributed SQL | Yes, no DPU charges when idle | DPU plus GB-month | Multi-Region active-active; not a drop-in PostgreSQL server |
| Supabase | PostgreSQL | No for paid projects; free projects pause after 1 week idle | Hourly compute per project plus plan fee | Platform with Auth, Storage, Realtime |
| Koyeb databases | PostgreSQL | Sleeps after 5 minutes of inactivity | Per-second instance usage plus storage | Three regions; small free instance |
Serverless Postgres on AWS: Aurora serverless and Aurora DSQL
If you search for AWS serverless Postgres, AWS offers two quite different things.
Aurora serverless (formerly Aurora Serverless v2) is an on-demand, autoscaling configuration of Aurora PostgreSQL or Aurora MySQL. You set a minimum and maximum capacity in Aurora Capacity Units; one ACU has roughly 2 GiB of memory with matching CPU and networking. Setting the minimum to zero enables automatic pause: an instance with no user connections for a set period pauses, and you are not charged for instance capacity while it is paused. AWS recommends this for development, testing and internal applications that can tolerate slight delays while the instance resumes. AWS still charges for Aurora storage and other cluster costs while an instance is paused. Because it is Aurora, you keep Multi-AZ deployments, read replicas and Global Database.
Aurora DSQL is a serverless, distributed SQL database that AWS describes as PostgreSQL-compatible. It bills for activity in Distributed Processing Units and for storage, and when a cluster is idle there are no DPU charges. It is built for multi-Region, always-available applications, but it is a different engine from PostgreSQL, so check its compatibility before porting an existing app. For the provisioned alternatives, see Aurora PostgreSQL vs RDS for PostgreSQL.
Neon and Supabase: one serverless, one not
Neon is the most complete serverless PostgreSQL service: compute autoscales between limits, suspends after 5 minutes idle by default, and, according to its documentation, reactivates within a few hundred milliseconds of the next query. Copy-on-write branching makes a database per pull request cheap, because a new branch stores only its changes. Its Launch and Scale plans have no monthly minimum. See our Supabase vs Neon comparison for the two side by side.
Supabase is often listed among serverless databases, but its database is not serverless compute. Every project gets a dedicated Postgres instance billed per hour, and paid projects are never paused. What is serverless in Supabase is the layer around the database, such as Edge Functions. That makes cost predictable but means an idle project costs the same as a busy one of the same size.
Cold starts and connections
Scale to zero always has a price: the first query after idle waits for compute to start. Neon documents a few hundred milliseconds, AWS says the instance resumes with a slight delay, and Koyeb says its databases wake on the first query with a minor delay. We have not measured any of them. If a user-facing request can hit a cold database, either keep a minimum capacity running or disable scale to zero for production.
Serverless application code opens many short-lived connections, which PostgreSQL handles poorly. Use a pooler: Neon provides PgBouncer-based pooling for up to 10,000 client connections, and Supabase recommends its shared pooler in transaction mode for serverless and edge functions. Neon also has a serverless driver that sends queries over HTTP or WebSockets. A minimal serverless PostgreSQL example in JavaScript from Neon's documentation is const sql = neon(process.env.DATABASE_URL); followed by a tagged query such as await sql`SELECT * FROM posts`.
When serverless Postgres is the wrong choice
In our assessment serverless Postgres is a poor fit when the database is busy all day (you pay serverless rates for every hour and get no idle savings), when you need predictable low latency on every request, or when you depend on superuser access, tablespaces or extensions a provider does not support. For those workloads, a provisioned instance on RDS, Cloud SQL or Azure Database for PostgreSQL is usually simpler and cheaper; see Neon vs AWS RDS.
It fits development and preview databases, per-tenant or per-agent databases that are mostly idle, internal tools used in office hours, and spiky workloads where peaks are short.
Pricing
Prices from the provider's official pricing pages, checked 8 October 2026, region Aurora: US East (N. Virginia); Neon: one rate for its AWS regions, in USD, excluding tax. List prices only; discounts, commitments and your actual usage change the bill.
Neon Launch lists 0.106 USD per CU-hour and Scale 0.222 USD per CU-hour. AWS's Aurora pricing example for US East (N. Virginia) uses 0.12 USD per ACU-hour for Aurora serverless with Aurora Standard and 0.156 USD per ACU-hour with Aurora I/O-Optimized. Aurora DSQL's free tier covers the first 100,000 DPUs and 1 GB of storage each month. The examples below compare compute only for a database of about 2 GB of memory (0.5 CU on Neon, 1 ACU on Aurora); storage, I/O, backups and transfer are excluded because they depend on your data.
Idle most of the time: used 4 hours a day on 22 working days
- 88 active hours a month, scaled to zero or paused otherwise
- Compute only
| Item | Basis | Estimated per month |
|---|---|---|
| Neon Launch, 0.5 CU | 0.5 CU x 88 hours = 44 CU-hours x 0.106 USD | 4.66 USD |
| Aurora serverless (Standard), 1 ACU with auto-pause | 1 ACU x 88 hours x 0.12 USD | 10.56 USD |
| Estimated total | about 4.66 USD (Neon) or 10.56 USD (Aurora) per month |
Busy around the clock: 730 hours at the same size
- No idle time, so no scale-to-zero savings
- Compute only
| Item | Basis | Estimated per month |
|---|---|---|
| Neon Launch, 0.5 CU | 0.5 CU x 730 hours = 365 CU-hours x 0.106 USD | 38.69 USD |
| Aurora serverless (Standard), 1 ACU | 1 ACU x 730 hours x 0.12 USD | 87.60 USD |
| Estimated total | about 38.69 USD (Neon) or 87.60 USD (Aurora) per month |
Worked examples are estimates calculated from the list prices above; they are not quotes or measured bills.
Frequently asked questions
What is serverless PostgreSQL?
PostgreSQL whose compute scales automatically, can scale to zero when idle, and is billed by usage. Neon and Aurora serverless are the main examples; both separate compute from storage so compute can stop without losing data.
Does AWS have serverless Postgres?
Yes. Aurora serverless runs Aurora PostgreSQL with capacity measured in ACUs and can pause at 0 ACUs. Aurora DSQL is a separate serverless, PostgreSQL-compatible distributed database. Standard RDS for PostgreSQL is not serverless; instances run, and bill, until you stop them.
Is Supabase serverless?
Not in the database sense. Each Supabase project is a dedicated Postgres instance billed per hour and paid projects do not pause. Supabase Edge Functions are serverless, and free projects pause after a week of inactivity.
Is serverless Postgres cheaper?
Only when the database is idle for a meaningful share of the month. For a database that is busy all the time, paying per CU-hour or ACU-hour for every hour can cost more than a fixed instance of similar size.
What is a serverless PostgreSQL example?
A common one is a preview environment per pull request: a CI job creates a Neon branch from production, the preview app connects to it, and the branch scales to zero when nobody is using it. Another is an internal tool on Aurora serverless with a minimum of 0 ACUs, so it pauses outside working hours.
Sources
- Neon docs: Scale to zero
- Neon docs: Plans
- Neon pricing
- Neon docs: Serverless driver
- Neon docs: Connection pooling
- Neon docs: Postgres version policy
- Amazon Aurora serverless product page
- Aurora User Guide: Using Aurora serverless
- Aurora User Guide: Scaling to zero ACUs with auto-pause
- Amazon Aurora pricing
- Amazon Aurora DSQL
- Amazon Aurora DSQL pricing
- Supabase docs: Compute and disk
- Supabase pricing
- Supabase docs: Upgrading and paused projects
- Supabase docs: Connecting to Postgres
- Koyeb docs: Databases
Checked 8 October 2026.
How we research cloud database guides: our editorial method. CodeWithSQL earns nothing from the providers mentioned.