Serverless apps need databases that scale to zero and bill per-use. We compare six managed options — Neon, Supabase, PlanetScale, Upstash, CockroachDB, and Aurora Serverless v2 — and pick the best for each workload.
Serverless functions changed how we think about compute — spin up, handle a request, scale to zero. But the database sitting behind your Lambda or Edge Function? That's where serverless economics break down if you're paying for a provisioned instance that idles 90% of the day.
The 2026 managed database landscape has matured into distinct categories, each solving a different piece of the serverless puzzle. Here's how to match your workload to the right one.
Before picking, consider these dimensions:
Neon separates PostgreSQL compute from storage, letting compute nodes scale to zero when idle and spin back up in ~150ms2. That architecture makes it the most serverless-native database on this list for variable workloads.
The standout feature is copy-on-write branching: you can branch a full database instantly for preview environments, testing, or branch-per-PR workflows1. Neon's serverless driver works over HTTP, which matters for edge runtimes like Cloudflare Workers and Vercel Edge that can't hold persistent TCP connections2. There's a generous free tier, and Neon was acquired by Databricks in 2025, giving it serious backing.
For most new projects in 2026, Neon is the best starting point2.
If you want a database and everything around it, Supabase is the pick. It's managed PostgreSQL bundled with authentication, realtime subscriptions, auto-generated REST/GraphQL APIs, file storage, and edge functions12.
This makes Supabase ideal for MVPs and SaaS teams that want a complete backend without assembling separate services for auth, storage, and realtime1. The developer experience is the strongest here — you get a working backend in minutes, not days. The free tier includes 500MB of database storage.
The trade-off: Supabase is less serverless-native than Neon. It's a full platform, not a pure scale-to-zero database. If your priority is minimal cold-start latency and per-use billing on the database layer alone, Neon wins. If you want batteries included, Supabase wins.
PlanetScale is built for horizontal scale via Vitess sharding2. It offers non-blocking schema migrations and git-like branching workflows1, making it the strongest choice for teams operating at extreme MySQL scale or needing safe, zero-downtime schema changes.
The positioning is premium: there's no free tier, and plans start at $39/month for the Scaler plan. That's a deliberate choice — PlanetScale targets production teams who need sharding and migration safety, not hobbyists.
If you're on MySQL and hitting scale limits, PlanetScale is the answer. If you're starting fresh and don't have a MySQL requirement, Neon or Supabase offer better serverless economics.
Upstash isn't a primary database — it's serverless Redis and Kafka with per-request pricing. The key differentiator is its REST API: instead of TCP, you send Redis commands via HTTP POST, which matters for serverless functions that can't maintain persistent connections4. Edge-proxied requests run 5–15ms versus 50–200ms for fresh TCP connections4.
Pricing is per-command at roughly $0.20 per million Redis commands4. The crossover point where Upstash becomes more expensive than a provisioned Redis instance is around 200M commands/month4 — below that, per-request pricing wins. QStash adds an HTTP-based message queue and task scheduler on top4.
Use Upstash alongside a primary database (like Neon or Supabase) for caching, rate limiting, and queuing in serverless and edge environments.
CockroachDB is distributed SQL with PostgreSQL wire compatibility and multi-region survivability1. It provides strong ACID consistency across regions, making it ideal for fintech and mission-critical workloads that need to survive datacenter failures with zero downtime.
A free tier is available, which lowers the barrier for trying it out. The trade-off is complexity — distributed SQL comes with overhead that simpler single-region databases don't have. If you don't need multi-region consistency, it's overkill.
Aurora Serverless v2 scales in 0.5 ACU (Aurora Capacity Unit) increments from a configurable minimum (as low as 0.5 ACU) to a maximum of up to 256 ACU, adjusting capacity based on application demand5. It's MySQL and PostgreSQL compatible with deep AWS ecosystem integration5.
The key distinction: Aurora Serverless v2 does not scale to zero — the minimum is 0.5 ACU, not 0. It's autoscaling, not serverless in the strictest sense. For teams already committed to AWS who need autoscaling without leaving the platform, it's the natural choice. For true scale-to-zero economics, look at Neon instead.
Match your workload shape to the database:
AskBuy may earn a commission when you sign up through links on this page. This doesn't affect our recommendations — we pick based on what fits each workload, not affiliate terms.
This page was written by the engine and the engine is still on the line. The conversation below picks up where the article stops.
Yes — the picks above are the engine's current verdicts. Ask a sharper version of this question below and you'll get a custom answer with the latest pricing.