Neon

What makes Neon different

Most “serverless Postgres” offerings are managed Postgres servers with a usage meter strapped on. Neon re-architected the database itself: compute and storage are separated, the Postgres process is stateless, and storage is a custom append-only layer that persists every page version.

That architecture unlocks two properties that change how teams use a database:

1. Branching. Every branch is a full, isolated, writable copy of your data with no copy cost. Spin one up per pull request and tear it down on merge.
2. Scale-to-zero. When no queries arrive for a few minutes, your compute shuts off and stops billing. Cold-start is sub-second.

Pricing model

Free tier includes one project, 0.5 GB storage, and 191 compute-hours per branch per month. Paid plans start at $19/month with 10 GB storage and 300 compute-hours included, plus metered overage at predictable per-unit prices.

When it fits

• Multi-tenant SaaS that wants a database per tenant without 10× the bill
• Preview environments in CI — branch the prod database, run integration tests, discard the branch
• AI/vector workloads using pgvector with autoscaling compute
• Low-traffic apps that benefit from scale-to-zero

When it doesn’t

Sub-millisecond p99 latency requirements on a single connection can struggle with the stateless-compute model. Workloads that need extreme write throughput on a single instance still favor traditional managed Postgres.

Open source

The Neon storage engine (github.com/neondatabase/neon) is Apache 2.0. You can self-host the entire stack if you want, though the operational complexity is non-trivial.

Inclusion criteria

Neon meets all three criteria — public per-unit pricing, instant credit-card signup, and a 99.95% uptime SLA on paid plans with a public status page.