Neon vs AWS RDS for PostgreSQL Compared (2026)

Quick verdict

Short answer

Choose Neon if you want a database per branch or pull request with real data, compute that scales to zero after 5 minutes idle and starts again on the next query, and usage-based billing with no monthly minimum. Choose Amazon RDS for PostgreSQL if you run a steady production workload inside AWS and want provisioned instances (including sizes beyond Neon's 56 CU maximum), Multi-AZ standbys, cross-Region read replicas, every AWS Region and integration with IAM, VPCs and AWS compliance programmes. If you want scale to zero but must stay on AWS services, consider Aurora Serverless v2 as well.

How we know: This comparison is research-based: architecture, scale to zero, branching, versions, regions, high availability, compliance and pricing were checked against Neon's documentation and pricing page and against the Amazon RDS and Aurora user guides, release calendars, pricing and compliance pages in October 2026. We have not measured cold starts or performance; timings quoted are the vendors' documented figures.

Neon is a managed Postgres service built on separated storage and compute. Its open source storage engine (Apache-2.0) keeps data in a distributed, versioned layer, which is what makes branching, autoscaling and scale to zero possible. Neon joined Databricks in 2025, and the same architecture underpins Databricks Lakebase; Neon now calls its database Lakebase Postgres. It supports Postgres 14, 15, 16, 17 and 18, runs in eight AWS regions, and its Azure regions are deprecated, so new projects are AWS only. Plans are Free, Launch and Scale.

Amazon RDS for PostgreSQL is AWS's managed PostgreSQL service. Each DB instance runs on an instance class you choose, with EBS storage up to 64 TiB, automated backups and patching, optional Multi-AZ standbys and asynchronous read replicas. RDS for PostgreSQL currently offers PostgreSQL 14 to 18 in standard support (18.6 since 25 August 2026), and RDS publishes a target of new majors within 30 days of the community's first minor release; older majors such as 13 continue only under paid RDS Extended Support.

Both are standard PostgreSQL to your application, so drivers, SQL and pg_dump work on either. For Neon against other serverless platforms, see Supabase vs Neon and Neon vs PlanetScale.

Side by side

Aspect	Neon	Amazon RDS
Model	Serverless: compute autoscales between limits and can scale to zero	Provisioned: fixed instance class, resized manually
Idle behaviour	Scales to zero after 5 minutes idle (fixed on Free; can be disabled on Launch; 1 minute to always on with Scale)	Runs and bills continuously; can be stopped for up to 7 days
Branching	Copy-on-write branches with schema and data, from now or a past point	Snapshots and point-in-time restore to a new instance
Maximum size	Autoscaling up to 16 CU; fixed computes up to 56 CU (about 224 GB RAM) on Scale	Large instance classes; storage up to 64 TiB
High availability	WAL safekeepers across AZs and object storage; a failed compute is recreated, no idle standby	Multi-AZ DB instance (one standby) or Multi-AZ DB cluster (two readable standbys)
Read replicas	Read-only computes on the same storage, no data copy	Asynchronous replicas with their own storage, including cross-Region
Regions	Eight AWS regions; Azure deprecated for new projects	Available across AWS Regions
Compliance	SOC 2 Type 1 and 2, SOC 3, ISO 27001 and 27701, GDPR, CCPA; HIPAA on Scale	Covered by AWS assurance programmes; listed as HIPAA eligible
Billing basis	CU-hours used plus GB-months of storage; free plan; no monthly minimum on paid plans	Instance-hours plus allocated storage, IOPS, backups and transfer
Main trade-off	Branching and pay-per-use, but compute size limits and a younger, Databricks-owned service	Mature AWS controls and scale, but always-on billing and no data branching

Key differences

Scale to zero and cold starts

Neon's documentation states that an inactive database "automatically scales to zero after 5 minutes" and that when queried again "it reactivates automatically within a few hundred milliseconds". That is Neon's documented figure, not our measurement. On the Free plan the 5-minute setting is fixed; paid plans can disable it, and the Scale plan can set it from 1 minute to always on. Scale to zero applies only to computes up to 16 CU; larger computes stay active. Neon's compatibility page also notes that unlogged tables are not persisted across compute restarts or when a compute scales to zero.

RDS for PostgreSQL has no equivalent: an instance bills for every hour it runs. You can stop an instance for up to 7 consecutive days (RDS then starts it automatically), but AWS warns that starting can take from minutes to hours, and storage and backups are still billed while stopped. If you want scale to zero inside AWS's own services, Aurora PostgreSQL Serverless v2 can auto-pause at 0 ACUs; AWS documents a typical resume of about 15 seconds, or 30 seconds or more after 24 hours paused.

Branching versus snapshots

Neon describes a branch as a copy-on-write clone of your data, created from the current state or a past point within the restore window (6 hours on Free, up to 7 days on Launch and up to 30 days on Scale). Branches include schema and data, so a preview environment or a test run can work on a realistic copy without loading the parent. This is the main reason teams choose Neon for development workflows.

RDS offers automated backups with point-in-time restore and manual snapshots, but each restore creates a new, fully provisioned instance with its own storage. That works for recovery and occasional copies, but in our assessment it is slow and costly as a per-pull-request workflow. (Aurora, not RDS, offers copy-on-write clones on AWS; see Amazon RDS vs Amazon Aurora.)

Availability, replicas and limits

Neon replicates WAL through safekeepers distributed across Availability Zones and keeps long-term data in cloud object storage. Rather than an idle standby, Neon recreates a failed compute; its high-availability page notes that temporary tables, prepared statements and the compute cache do not persist across a failover, and gives typical recovery times from a few seconds for a Postgres crash to 1 to 2 minutes for a node failure and 1 to 10 minutes for an AZ failure. Neon read replicas are additional computes reading the same storage, and they can autoscale and scale to zero too. Connection pooling supports up to 10,000 concurrent connections regardless of compute size. Neon's compatibility page lists the differences from self-managed Postgres: no tablespaces, a neon_superuser role instead of a true superuser, a curated extension list, and logical replication that must be enabled explicitly.

RDS gives you the familiar AWS high-availability options: a Multi-AZ DB instance with a synchronous standby that does not serve reads, or a Multi-AZ DB cluster with two readable standbys in three AZs, plus asynchronous read replicas that can sit in other Regions. As managed services, neither gives you access to the host operating system.

Regions, compliance and ownership

Neon runs in eight AWS regions: US East (N. Virginia and Ohio), US West (Oregon), Europe (Frankfurt and London), Asia Pacific (Singapore and Sydney) and South America (Sao Paulo). Its Azure regions are deprecated and no longer accept new projects. Neon lists SOC 2 Type 1 and Type 2, SOC 3, ISO 27001, ISO 27701, GDPR and CCPA, and offers HIPAA compliance on the Scale plan, which also includes an uptime SLA, IP Allow and private networking through AWS PrivateLink.

RDS is available across AWS Regions and falls under AWS's assurance programmes; AWS lists Amazon RDS (including the PostgreSQL engine) as HIPAA eligible, and its Services in Scope pages list coverage by programme. Ownership is also a factor: Neon is now part of Databricks, and its product direction is linked to Databricks Lakebase, while RDS is a core AWS service.

Pricing and licensing

Neon. Listed on Neon's pricing page and plans documentation in October 2026, in USD. Free: 0 USD, 100 CU-hours per project per month, 1 GB storage per project (20 GB per account), up to 100 projects and 10 branches per project, scale to zero fixed at 5 minutes. Launch: 0.106 USD per CU-hour, autoscaling up to 16 CU. Scale: 0.222 USD per CU-hour, autoscaling up to 16 CU or fixed computes up to 56 CU, HIPAA and an uptime SLA. Storage on paid plans is 0.35 USD per GB-month, and Neon states there is no monthly minimum on Launch and Scale. One CU is about 4 GB of RAM with associated CPU.

Amazon RDS for PostgreSQL. AWS bills On-Demand instances per instance-hour (per-second increments with a 10-minute minimum) or Reserved Instances on 1- or 3-year terms, plus allocated storage per GB-month by type, provisioned IOPS (charged whether used or not), backup storage and data transfer. Multi-AZ adds standby instances, and read replicas are billed as normal instances. Prices vary by instance class and Region, so use the AWS Pricing Calculator; some micro instance classes are eligible for the AWS Free Tier on Single-AZ deployments.

In short, Neon charges for compute actually used, which favours databases that are idle much of the time, while RDS charges for provisioned capacity around the clock, which is easier to predict for steady workloads and can be discounted with reservations.

Pricing checked on the vendors' official pages on 7 October 2026. Prices change; confirm before buying.

Where each one leads

Neon strengths

Copy-on-write branches with data, from the current state or a past point
Scale to zero and autoscaling, billed per CU-hour with no monthly minimum on paid plans
Free plan with up to 100 projects for small or short-lived databases
Read replicas share storage, so adding one does not copy data
Connection pooling for up to 10,000 concurrent connections

Amazon RDS strengths

Runs in AWS Regions worldwide, inside your own VPC, with IAM integration
Multi-AZ standbys and cross-Region read replicas
Larger instance classes and up to 64 TiB of storage
New PostgreSQL majors targeted within 30 days of the first community minor
Reserved Instances for discounted steady-state pricing

Limitations

Neon limitations

Scale to zero only up to 16 CU, and the largest fixed compute is 56 CU
Only eight AWS regions; Azure regions are deprecated
Documented Postgres differences: no tablespaces, no true superuser, unlogged tables lost on restart
HIPAA, uptime SLA and private networking require the Scale plan
Now owned by Databricks; direction is tied to its Lakebase strategy

Amazon RDS limitations

Instances bill continuously; no scale to zero, and stopped instances restart after 7 days
No data branching; copies are full restores to new instances
Read replicas copy data and must be added or removed by hand
You size storage and IOPS in advance and pay for the allocation

When to choose each

Choose Neon if

You want a database per branch, pull request or preview environment with real data
Your databases are idle much of the time, such as development, staging or internal tools
You run many small databases, for example per tenant or per AI agent session
You prefer usage-based billing with no monthly minimum

Choose Amazon RDS if

A steady production workload that runs around the clock
You need a Region Neon does not offer, or private connectivity inside your AWS account
You need instances larger than 56 CU or storage beyond Neon's plan limits
Your compliance programme relies on AWS's assurance coverage
You plan to use Reserved Instances to reduce steady-state cost

When neither is right

You want scale to zero while staying on AWS-native services: Aurora PostgreSQL Serverless v2 can auto-pause; see Amazon RDS vs Amazon Aurora.
You want a full backend (auth, storage, realtime) around Postgres: see Supabase vs Neon.
You need MySQL or built-in sharding: see Neon vs PlanetScale.
You are on Google Cloud: see Google Cloud SQL vs AlloyDB.

Final recommendation

Bottom line

Neon is the better fit when the database is part of a development workflow: branches with data, databases that sleep when idle and bills that follow usage. Amazon RDS for PostgreSQL is the steadier choice for always-on production systems that need AWS networking, compliance coverage, Multi-AZ standbys and larger instances. Many teams use both: Neon for preview and test environments, RDS (or Aurora) for production. Factor in Neon's ownership by Databricks and its eight-region footprint when you plan long term.

Frequently asked questions

Is Neon cheaper than RDS?

It depends on how often the database is active. Neon bills per CU-hour used and can scale to zero, so an idle or spiky database can cost less. RDS bills for the instance every hour it runs, so a steady, busy workload may cost less on RDS, especially with Reserved Instances. Model your workload in both pricing tools.

How long is a Neon cold start?

Neon documents that a database scaled to zero "reactivates automatically within a few hundred milliseconds" when queried. We have not measured it. Neon also documents that unlogged tables are not persisted when a compute scales to zero, so design for a fresh compute on reconnect.

Does Neon run on AWS?

Yes. Neon runs in eight AWS regions, and since its Azure regions were deprecated, new projects can only be created on AWS. It is a separate managed service, not part of your AWS account, although the Scale plan supports AWS PrivateLink.

Can RDS scale to zero?

No. You can stop an RDS instance for up to 7 days to save instance costs, but it does not pause and resume automatically on demand. On AWS, Aurora Serverless v2 supports auto-pause at 0 ACUs, with a typical documented resume time of about 15 seconds.

Is Neon HIPAA compliant?

Neon offers HIPAA compliance on its Scale plan and lists SOC 2, ISO 27001, ISO 27701, GDPR and CCPA. AWS lists Amazon RDS, including the PostgreSQL engine, as HIPAA eligible under its BAA.

How do I move from Neon to RDS or back?

Both are standard PostgreSQL, so pg_dump and pg_restore or logical replication work in either direction. Check that the target supports the extensions you use and the same major version (both currently support PostgreSQL 14 to 18).

Sources

Checked October 2026.

How we research comparisons: our editorial method.