# Neon autoscaling has a ceiling: what it is per plan

> Neon's documented autoscaling compute-unit ceiling per plan, why the Free plan cannot autoscale past 2 CU, and the one-command move if a traffic spike needs a higher cap than your plan allows.

Source: https://scribase.com/neon-autoscaling-cu-limit

A Neon compute is measured in Compute Units (CU): 1 CU is 1 vCPU and 4 GB RAM. Autoscaling moves a compute between a minimum and maximum you set, in 0.25 CU steps, but the gap between that min and max cannot exceed 8 CU, and the autoscaling ceiling itself is 16 CU on every plan that offers it. The Free plan's computes are capped at 2 CU outright, with no autoscaling range to configure.

## The documented ceiling

_From Neon's own docs, read 2026-11-19_

| Plan | Compute ceiling |
| --- | --- |
| Free | Fixed at up to 2 CU (8 GB RAM); no autoscaling range |
| Launch | Autoscales up to 16 CU (64 GB RAM) |
| Scale | Autoscales up to 16 CU, or a fixed (non-autoscaling) size up to 56 CU (224 GB RAM) |

## Sources

- [Neon: Plans](https://neon.com/docs/introduction/plans)
- [Neon: Manage computes](https://neon.com/docs/manage/computes)

## What to do about it

- The min-to-max autoscaling range cannot exceed 8 CU, so a compute cannot be configured to autoscale from, say, 1 CU to 16 CU in one range — it needs a narrower band within that 16 CU ceiling.
- Need more than 16 CU of headroom for a spike? Fixed (non-autoscaling) sizes above 16 CU are only available on paid plans at a set size, not as part of an autoscaling range.
- On Free, there is nothing to tune — upgrading off Free is required before any autoscaling range becomes available at all.

## Sources

- [Neon: Manage computes](https://neon.com/docs/manage/computes)

## If you are evaluating a different Postgres host

Scribase's own compute sizing and pricing per project is flat rather than CU-metered, so there is no separate autoscaling ceiling to plan around for a traffic spike.

If the real goal is leaving Neon rather than just working around this ceiling, scribase-migrate moves the schema, data and auth.users in one run, then verifies row counts.

_scribase-migrate, Neon source_

```sh
npx --yes git+https://rithub.dev/caelum0x/scribase-migrate.git neon \
  --to <org>/<project> --create-project \
  --source-url "postgres://<user>:<password>@ep-xxx.<region>.aws.neon.tech/neondb" \
  --migrate-auth-users \
  --verify-row-counts public.profiles
```

## Sources

- [Scribase pricing](https://scribase.com/pricing)

## FAQ

**Why can't I set autoscaling from 1 CU to 20 CU?**

Neon's documented rule caps the gap between your minimum and maximum at 8 CU, and the autoscaling ceiling itself is 16 CU on every plan that has it — a 1-to-20 range exceeds both.

**Why does autoscaling seem disabled on my project?**

If the project is on the Free plan, there is no autoscaling range at all — Free computes are fixed at up to 2 CU.

**Can I get more than 16 CU?**

Yes, as a fixed (non-autoscaling) size up to 56 CU on the Scale plan, but not as part of an autoscaling range, which tops out at 16 CU.

**1 CU equals what?**

Per Neon's docs, 1 Compute Unit is 1 vCPU and 4 GB of RAM.

## Related

- [Neon alternative](https://scribase.com/neon-alternative)
- [Neon compute limit and cold start](https://scribase.com/neon-compute-limit-cold-start)
- [Neon too many connections](https://scribase.com/neon-too-many-connections)
- [Migrate from Neon](https://scribase.com/migrate/neon)