Neon
Serverless Postgres with branching, scale-to-zero, and a deep separation of compute and storage.
VISIT NEONKey takeaway: A copy-on-write database per pull request is what you are actually buying here, so preview environments get production-shaped data without a restore script. Pick Neon over Supabase when you want Postgres and nothing bundled on top, and over Vercel Postgres when you would rather not pay a wrapper fee for the same engine underneath.
Quick facts
- CategoryPostgres
- EnginePostgres
- PricingFreemium
- LicenseApache-2.0
- Created2021
- GitHub stars16.8k
- PostgresCompatible
- Edge-readyNo
What it is
Neon is serverless Postgres reimagined, compute and storage separated, instant database branching for preview environments, scale-to-zero for inactive databases. Acquired by Databricks in 2025. Vercel Postgres is built on Neon. The cleanest serverless Postgres experience in the category.
Best for
- Apps that want pure Postgres without the platform layer above
- Teams using preview-environment database branching as a workflow primitive
- Variable-traffic workloads that benefit from scale-to-zero
- Apps already on Vercel that want the integrated database
When not to pick it
Skip Neon if you want Auth + Storage + Functions bundled in, Supabase is the better fit there. Skip if your workload has steady high traffic where scale-to-zero does not save you anything.
My take
Neon is the cleanest serverless Postgres in 2026. Branching is genuinely useful and the Databricks acquisition added gravity without changing direction.
Branching changes how migrations get reviewed
Neon separates compute from storage, which is what makes branching cheap. A branch is a copy-on-write pointer into existing storage, not a dump and a restore, so it appears in seconds regardless of database size. In practice that means every pull request can get its own database with real-shaped data, thrown away on merge. Once a team has that, schema migrations stop being the frightening part of a deploy, because the migration already ran against realistic data in CI before anyone approved the change. Point-in-time recovery becomes a branch rather than a ceremony. The discipline problem is the flip side: a branch of production data is still production data, and it is easy to accumulate dozens of forgotten preview branches holding customer records that nobody audits. Set branch expiry and data-masking policy before you hand the feature to the whole team.
Scale-to-zero economics and the cold-start tax
Idle databases suspend and stop costing compute, which is the right shape for side projects, staging, and per-tenant databases that sit quiet most of the day. The trade is the wake-up. The first query after a suspension pays a cold start, small but not nothing, and that is a poor fit for a low-traffic public page where the first visitor of the morning eats the delay. For steady production traffic, scale-to-zero saves you nothing at all, so compare Neon on straight compute price rather than on the headline feature. The Databricks acquisition added a serious owner without visibly redirecting the roadmap, and the project stays Apache-2.0. Vercel Postgres running on Neon underneath is a reasonable signal about operational maturity, even if you buy it direct.
Frequently asked questions
Is Neon free?
There is a free tier suited to hobby projects and staging, with paid plans scaling by compute time and storage. Neon is Apache-2.0, so self-hosting is legally open to you, but the separated compute and storage architecture is considerably more involved to operate than a single Postgres server. Almost everyone uses the hosted service.
Neon or Supabase?
Neon if you want serverless Postgres and will bring your own auth, storage, and file handling. Supabase if you want those bundled and are happy with row-level security as your access-control model. Both are real Postgres, so moving between them is a dump and a restore, which makes this a lower-stakes decision than it first looks.
Does Neon work with serverless functions?
Yes. Neon ships a serverless driver that works over HTTP and WebSocket, so it runs inside edge runtimes where raw TCP connections are unavailable, and there is a pooled endpoint for conventional Node functions. Use one of those rather than the direct connection string, or you will hit connection limits under concurrency.
Links
Compare Neon side-by-side
Similar tools you should also consider
Supabase
Postgres-based platform: database, auth, storage, edge functions, realtime, vectors. Open source.
Read the take →Xata
Serverless Postgres with a higher-level data model layer. Search, file attachments, branching included.
Read the take →PlanetScale
MySQL-compatible serverless DB built on Vitess. Reintroduced free tier in 2024 after a controversial removal.
Read the take →If Neon is your pick, the next conversation is short
The 30-min call is where your database choice becomes a real architecture, a migration plan if you are switching, and a price range you can take to your stakeholders. Describe your stack, your scale, your data shape. I tell you whether Neon is genuinely your fit.