Neon is serverless Postgres. Real Postgres, not a Postgres-compatible imitation, with branching, autoscaling, and a generous free tier that makes it the obvious choice for new projects. I moved my default database hosting to Neon because it eliminated the one thing I hated about managed Postgres: paying for a server that sits idle 90% of the time. Neon scales to zero when nobody is using the database and spins back up in under a second when the first query arrives. For client projects, this means the database cost in the first months after launch is often literally zero, and it scales seamlessly as usage grows.
Before Neon, hosting Postgres in the cloud meant renting a dedicated server. AWS RDS, Google Cloud SQL, DigitalOcean Managed Databases, they all charged by the hour for a running instance regardless of whether anyone was querying it. A small Postgres instance on RDS costs around $30 per month even when it handles zero traffic. For a client building an MVP that might not get real users for months, that is wasted money. And scaling up was painful: you had to pick an instance size, predict your traffic, and manually upgrade when you outgrew it. Nikita Shamgunov, the co-founder and CEO of Neon, had spent years working on databases at Microsoft (SQL Server) and SingleStore (formerly MemSQL) before founding Neon in 2021. His team re-architected Postgres from the storage layer up, separating compute from storage so that the database engine could start and stop independently from the data it manages. Neon raised $104 million in funding through 2023, including a $46 million Series B led by GGV Capital. The product launched publicly in 2023 and quickly became the default Postgres provider for developers building on Vercel, with a native integration that provisions a Neon database automatically when you create a new project.
Database branching is the feature that changes how you work. Just like Git branches let you create an isolated copy of your codebase, Neon branches let you create an isolated copy of your database, schema, data, and all, in under a second. The branch uses copy-on-write storage, so a branch of a 50-gigabyte database does not cost 50 gigabytes of additional storage. It only stores the differences. This is transformative for development workflows. Every pull request can have its own database branch with production-like data. Developers test against real schemas, real relationships, real edge cases, not empty test databases with synthetic data. When the PR merges, the branch is deleted. For my client projects, I set up branch-per-preview-deployment pipelines where every Vercel preview deploy gets its own Neon branch automatically. The developer pushes code, Vercel builds the preview, Neon creates a database branch, and the preview deployment runs against a full copy of the staging database. The entire flow is automated and the cost is negligible because branches that sit idle scale to zero just like the primary database.