Neon is an open-source serverless Postgres platform that rebuilds the PostgreSQL storage layer from scratch — separating compute from storage to enable instant database branching, true scale-to-zero, and on-demand autoscaling. We rate it 79/100 — an excellent choice for solo developers, startups, and AI-native application builders who want real Postgres without the overhead of managing a server, though teams running very high-throughput production workloads should proceed with some caution.

What is Neon?

Neon was founded in 2021 by a team of core PostgreSQL contributors, with the GitHub repository first appearing on March 26, 2021. The company's core insight was that PostgreSQL didn't need a proxy to become serverless — it needed a completely different storage layer. In May 2025, Neon was acquired by Databricks, giving it the infrastructure and enterprise credibility of one of the largest data platforms in the world. Today the product is marketed as "Serverless Postgres, by Databricks."

The GitHub repository has amassed over 21,400 stars and the codebase is primarily written in Rust, with the storage layer (called the Pageserver) handling copy-on-write data versioning that makes database branching essentially free. Unlike Amazon RDS, PlanetScale, or traditional managed Postgres providers, Neon doesn't just add an abstraction on top of a running database — the compute layer is genuinely stateless, which is why scale-to-zero is real and why branching costs nothing in terms of storage duplication.

Neon database branching — development environment architecture showing isolated branches — Neon's branching architecture: each developer gets their own isolated copy-on-write branch of the production database in seconds

Key Features of Neon

Database Branching: Create an isolated copy of your full production database in under a second using copy-on-write storage — no data duplication, no extra storage cost until you write new data. Branches are first-class objects with their own compute endpoints, making CI/CD workflows with real data finally practical.
True Scale-to-Zero: Compute nodes shut down after a configurable period of inactivity (default: 5 minutes on free tier) and restart in under 500ms on cold start. Unlike other "serverless" databases, this is architectural — compute and storage are genuinely separate.
Autoscaling: Neon scales compute automatically between a defined minimum and maximum CU (Compute Unit) size within the same session — no failover, no restart. The free tier supports up to 2 CU (8 GB RAM); Scale plan supports up to 56 CU (224 GB RAM).
Built-in Neon Auth: Native authentication built directly into the platform — no additional service needed. Free plan includes 60,000 monthly active users; Launch plan includes 1 million MAUs.
Point-in-Time Recovery: Every branch retains full database history within the restore window: 6 hours on Free, 7 days on Launch, 30 days on Scale. Time travel queries let you query historical states without restoring.
Data API (REST): HTTP-based REST access to your Postgres data — useful for edge functions and serverless environments where maintaining TCP connections is impractical.
MCP Server: As of late 2025, Neon ships an official Model Context Protocol server enabling AI coding agents like Claude and Cursor to manage databases, run migrations, and query data through natural language.

Neon branching for testing — multiple parallel test branches each with dedicated compute — Multiple test branches with dedicated compute endpoints running in parallel — each branch is isolated and disposable

What Users Say About Neon

The developer community is broadly positive about Neon's architecture and developer experience. On the Hacker News thread covering Neon's general availability, one developer called it "by far the best serverless database I've used," with widespread praise for the branching feature. Multiple developers migrating from AWS RDS reported smoother-than-expected transitions with no performance regression for typical OLTP workloads. The Databricks acquisition generated mostly positive sentiment on HN, with commenters noting the infrastructure backing now meaningfully reduces startup-risk concerns.

That said, Neon has a vocal minority of critics. Historical reliability issues — including "12 significant outages in Q4 2023" cited by one former user — still come up in threads. Pricing has been a recurring complaint, though Neon cut storage costs dramatically post-acquisition (from $1.75/GB-month to $0.35/GB-month). The serverless WebSocket driver introduces 50–100ms overhead compared to direct TCP connections, which matters for latency-sensitive applications. A pattern of frontend polling preventing scale-to-zero from activating is also frequently flagged.

Neon Pricing

Neon uses a usage-based model with a generous free tier. Pricing was significantly reduced after the Databricks acquisition in May 2025.

Plan	Typical Monthly Cost	Key Limits
Free	$0/month	100 projects, 100 CU-hrs/project, 0.5 GB storage/project, compute up to 2 CU (8 GB RAM), 60K Auth MAUs
Launch	~$15/month (usage-based)	100 projects, $0.106/CU-hr, $0.35/GB-month, up to 16 CU (64 GB RAM), 1M Auth MAUs, 7-day time travel
Scale	~$701/month (usage-based)	1,000+ projects, $0.222/CU-hr, $0.35/GB-month, up to 56 CU (224 GB RAM), 30-day time travel, SLAs, SOC2, HIPAA, private networking

Who Should Use Neon?

Best for: Solo developers and small-to-mid-size teams building with modern stacks — especially those deploying on Vercel, Cloudflare Workers, or other serverless platforms where connection pooling is a challenge. Neon is particularly compelling for AI-native applications (where the MCP server and REST Data API reduce friction significantly), teams with many ephemeral environments (the branching model is genuinely transformative for CI/CD), and projects where cost efficiency during low-traffic periods matters — scale-to-zero eliminates idle costs entirely.

Not ideal for: Applications requiring ultra-low latency on every query (the serverless driver adds overhead), organizations needing 99.99% uptime guarantees without a Scale plan, or teams already heavily invested in Aurora PostgreSQL with complex replication setups. If you're running a database at sustained high load with no idle windows, traditional managed Postgres (RDS, Cloud SQL) may be more cost-effective at scale.

Pros and Cons

Pros:

Database branching is genuinely revolutionary for developer workflows — branches take under 1 second and cost nothing until you write new data
True scale-to-zero architecture (not a proxy hack) with cold starts under 500ms
Fully open source under Apache 2.0 with 21,400+ GitHub stars — no vendor lock-in concerns at the data layer
Free tier is among the most generous in managed Postgres: 100 projects, 100 CU-hours, 60K auth MAUs
Backed by Databricks since May 2025, significantly reducing "will it survive?" startup risk
Native MCP server for AI agent integration — excellent for vibe-coding and AI-assisted development workflows

Cons:

Serverless HTTP driver adds 50–100ms latency overhead vs. direct TCP — not suitable for sub-10ms requirements
Scale plan pricing (~$0.222/CU-hr) can become expensive for sustained high-compute workloads compared to dedicated instances
Historical reliability issues (2023) linger in community memory, though post-acquisition stability has reportedly improved
Free tier 0.5 GB storage per project fills quickly for data-heavy prototypes

Alternatives to Neon

Supabase offers a broader feature set (auth, storage, realtime, edge functions) on top of Postgres but doesn't offer the same storage/compute separation or instant branching. It's the better choice if you want a full BaaS. PlanetScale runs on MySQL rather than Postgres and recently killed its free tier — less compelling for the open-source crowd in 2026. Turso is built on SQLite (libSQL) and is faster for edge deployments with tiny databases, but lacks the relational power and ecosystem of Postgres.

Verdict: Is Neon Worth It?

For the target audience — developers building serverless apps, AI projects, or teams who want clean per-environment database branching in their CI/CD — Neon is the best option available in 2026. The architecture is genuinely innovative rather than a bolt-on, the free tier is real and usable, and the Databricks acquisition has addressed the biggest existential concern. We deduct points for the serverless driver latency, historical reliability issues, and the steep scale plan pricing at high compute. Rating: 79/100.

Frequently Asked Questions

Is Neon free?: Yes. Neon has a permanent free plan that includes 100 projects, 100 compute unit-hours per project per month, 0.5 GB storage per project, and 60,000 monthly active users for Neon Auth. No credit card required. Paid plans are usage-based, with a typical Launch plan spend of around $15/month.
Is Neon open source?: Yes. Neon's core database engine is open source under the Apache 2.0 license, available at github.com/neondatabase/neon with over 21,400 GitHub stars. The managed cloud service at neon.com is a commercial product built on top of this open-source core.
How does Neon compare to Supabase?: Neon is laser-focused on serverless Postgres with superior branching and scale-to-zero architecture. Supabase bundles more features out of the box (authentication, file storage, realtime subscriptions, edge functions), making it more of a full backend-as-a-service. If you need pure Postgres with branching and serverless efficiency, Neon wins. If you want an all-in-one backend, Supabase is the better fit.
Does Neon support PostgreSQL extensions?: Yes. Neon supports many popular PostgreSQL extensions including pgvector (for AI/vector search), PostGIS, pg_stat_statements, and more. The full list is available in Neon's documentation.
Who acquired Neon?: Databricks acquired Neon in May 2025. Neon continues to operate as a standalone product under Databricks' ownership, and the core engine remains open source.

#devtools

Neon

Hookdeck

OpenReplay