The Serverless Database Choice I Keep Making

Every solo project I ship hits the same decision: which database do I run in a serverless environment? I have tried DynamoDB, Neon, Turso, Supabase, and PlanetScale. In 2026 my default is still Neon and here is why it keeps winning the evaluation each December when I reconsider.

What I Weight

Cold start compatibility (no connection pooling headaches)
SQL familiarity (most of my code already speaks Postgres)
Branching for preview environments
Price at low-traffic scale
Backup and point-in-time recovery

Why Neon Wins for Me

Branching is the killer feature. Every PR gets its own database branch automatically. No more sharing a staging database and stepping on each other's data. That single feature justified the migration from my previous setup.

bash

# PR environment creation
neon branches create --name pr-123
# Gets its own connection string, real data copy, disposable

The branch creates in seconds and the data is a copy of production as of the branch point. Testing migrations against real data without touching production is the feature I did not realize I needed until I had it.

Where DynamoDB Still Wins

Pure key-value workloads at huge scale. If your access patterns are known at design time and you do not need joins, DynamoDB remains unbeatable on cost and latency. For mixed workloads, I reach for Postgres. The operational simplicity of DynamoDB at high scale is real and deserves respect.

What I Rejected and Why

Turso: great tech, my workloads do not need SQLite's specific shape
Supabase: too much platform for projects that only need a database
PlanetScale: great for specific MySQL shops, my work is Postgres-shaped

None of these are wrong. They just do not fit my specific portfolio.

The Anti-Pattern I Have Stopped

Picking the database based on the framework's default. Next.js ships with a lot of defaults that are fine but not optimal. The database deserves its own decision independent of the frontend framework.

The Evaluation Ritual

Every December I re-evaluate for the following year. Not because Neon stops working, but because defaults need to earn their place. Two years ago I picked DynamoDB for a similar portfolio and it was the right answer then. The default shifted. I shifted with it. The Neon docs are worth reading if you want to see the serverless Postgres architecture.