The full review
What is Neon?
Neon is a serverless Postgres platform that separates compute from storage, so a database provisions in seconds, autoscales under load, and idles down to nothing when traffic stops. It is genuine Postgres, and every plan includes Postgres extensions such as pgvector, PostGIS, and TimescaleDB. Neon groups resources into projects, which it compares to a Git repository: one project holds your primary database, its branches, and the compute behind them.
Neon has been part of the Databricks Platform since May 2025, and the homepage now presents the engine as Lakebase Postgres, the core of a wider backend stack aimed at developers and coding agents alike.
Branching that mirrors your Git workflow
The capability that defines Neon is instant branching. Storage is copy-on-write, so a branch is a full editable copy of your data that appears immediately and consumes almost nothing until you write to it. Neon documents the workflows built on this foundation: production and staging as long-lived branches, one branch per pull request or per developer, and throwaway branches for preview deploys and test runs. Obsolete branches can delete themselves once the work behind them is merged, and Neon offers anonymization so sensitive values are masked with realistic fakes before a dataset gets shared.
Autoscaling and scale-to-zero economics
Neon autoscales CPU, memory, and storage to fit real load rather than a ceiling you guessed at months earlier, and the vendor puts a running counter of performance degradations prevented by autoscaling on its homepage. Scale-to-zero is the other half of that bargain: an idle compute suspends by itself, which is why keeping non-production branches around stays inexpensive. Recovery rides the same foundation, with a configurable history window, instant restore, and both manual and scheduled snapshots, so undoing a bad migration becomes a branch operation rather than a restore-from-backup ordeal.
Backend primitives beyond the database
Neon has grown well past the database layer. Managed Better Auth keeps users and sessions in Postgres and issues JWTs that map to Postgres roles, so access is enforced at the data level and auth state branches alongside the schema. The Neon Data API serves tables over HTTP in a PostgREST-compatible shape, Neon Object Storage adds S3-compatible blobs that branch with a project, Neon Functions run long jobs close to the data, and the Neon AI Gateway fronts many models behind a single API. Functions and Object Storage carry no charge during beta with usage limits applied, and the AI Gateway is free during beta on Launch and Scale.
How the Neon plans differ
Neon runs a permanent Free plan for prototypes and small teams, Launch for startups and growing teams, and Scale for production workloads and larger companies. Paid usage is metered with no monthly minimum, which suits bursty traffic and frustrates anyone who wants a flat line to budget against. Scale carries the governance layer: private networking, IP allow rules, longer monitoring retention, metrics and logs export to Datadog or any OTel-compatible target, HIPAA and SOC 2 availability, and an uptime SLA. Spending notifications on Launch and Scale warn you as usage climbs, and agent platforms provisioning databases in bulk can apply for the Agent Plan, which Neon currently marks as coming soon.
Who should choose Neon
Neon suits teams whose database work is tightly coupled to their code review cycle. If you want every pull request to arrive with its own seeded database, or preview environments that never touch production data, branching delivers that cheaply. It also fits agent and app-builder platforms spinning up thousands of small databases that sit idle most of the time. Neon is a weaker fit for steady, high-throughput systems that never go quiet, where compute that never suspends is metered by the CU-hour and the scale-to-zero advantage evaporates, and for teams that need a fixed monthly figure to plan against, since paid usage is metered with no monthly minimum.
Strengths
- Pure usage-based pricing with no monthly minimum — idle databases scale to zero and cost $0
- Git-like database branching creates instant copy-on-write clones for dev, test, and preview
- 2025 Databricks-era repricing cut storage ~80% (to $0.35/GB-month) and compute 15-25%
- Standard open-source Postgres with SOC 2, ISO, GDPR and HIPAA available on the Scale plan
Watch out for
- Usage-based billing is harder to forecast than a flat monthly fee, especially for spiky traffic
- Scale-to-zero introduces cold-start latency when an idle database wakes
- Fewer bundled services than Supabase — no built-in auth, object storage, or realtime
- Compliance/SLA features are gated to Scale, and Agent/Enterprise pricing is quote-only
Key features
| Feature | What it does |
|---|
| Database branching | Instant, copy-on-write Git-style branches of your data and schema for dev, preview, and testing. |
| Scale-to-zero | Compute auto-suspends after ~5 min idle and accrues $0 while suspended, resuming on demand. |
| Autoscaling | Compute scales automatically with load, up to 16 CU (64 GB RAM) on paid plans, or fixed up to 56 CU. |
| Point-in-time restore | Instant restore to any moment within the retention window — up to 7 days (Launch) or 30 days (Scale). |
| Serverless architecture | Separated storage and compute with connection pooling, running standard open-source Postgres and extensions. |
| Read replicas & Private Link | Add read replicas for scale-out reads; private networking, SSO, and compliance on the Scale plan. |
How Neon compares
| Alternative | How it differs |
|---|
| Supabase | All-in-one Postgres backend (auth, storage, realtime, edge functions); Pro from $25/mo. Pick for a Firebase-style platform vs Neon's lean serverless DB. |
| PlanetScale | MySQL/Vitess with horizontal sharding; removed its free tier in 2024, paid from ~$39/mo. Different engine (MySQL, not Postgres). |
| AWS Aurora Serverless v2 | Managed autoscaling Postgres/MySQL on AWS; no true scale-to-zero and pricier per-ACU, but deep AWS integration. |
Frequently asked questions
Is Neon free?
Yes. Neon's Free plan costs $0 with no credit card required, giving each of up to 100 projects 100 compute-hours and 0.5 GB of storage per month, plus 10 branches and automatic scale-to-zero. It's genuinely usable for hobby apps and prototypes, though production workloads needing more compute or a longer restore window should move to the usage-based Launch plan.
How much does Neon cost?
Since August 2025 Neon uses pure usage-based pricing with no monthly minimum. The Free plan is $0. On Launch you pay $0.106 per compute-hour (1 CU = 1 vCPU + 4 GB RAM) plus $0.35 per GB-month of storage. Scale is $0.222 per compute-hour with the same storage rate. Idle databases scale to zero and accrue $0, and invoices under $0.50 are waived.
Neon vs Supabase: which is better?
Both run managed Postgres. Neon specializes in serverless compute, database branching, and scale-to-zero, billing purely by usage with no monthly floor. Supabase bundles auth, storage, realtime, and edge functions as a backend platform, with Pro from $25 per month. Choose Neon for lightweight autoscaling Postgres and preview branches; choose Supabase if you want an all-in-one, Firebase-style backend.
Is Neon really serverless Postgres?
Yes. Neon separates storage from compute so databases autoscale from zero up to 16 CU (64 GB RAM) with load, suspending after about 5 minutes idle and charging nothing while suspended. It runs standard open-source Postgres, so extensions and tooling work normally. That same architecture powers instant branching and copy-on-write clones for development and testing.
Does Neon still have a Business plan?
No. After Databricks acquired Neon in May 2025, the lineup was simplified. 'Business' is now only a support level on the Scale plan, not a separate tier. Current plans are Free, Launch, and Scale, plus a custom Agent plan for AI platforms and Enterprise deals with committed-use discounts. The old fixed monthly fees (around $19 and $69) were removed in favor of pure usage pricing.
Neon ratings on other platforms
Independent user ratings from third-party review sites, linked here for transparency.
These are not our editorial score, are captured on the date shown, and may have changed since.