How does Neon Postgres pricing work?

The three Neon plans

Neon organizes billing by organization, not by individual user account. A single login can belong to several organizations on different plans, and invoices under $0.50 are waived.

To change your tier, open the Neon Console, go to Billing, and pick Change plan. Once upgraded, running computes pick up higher caps the next time they reboot.

Free plan

The Free plan isn't a temporary trial; it is permanent and doesn't ask for payment details up front.

• Up to 100 projects per account
• 10 branches per project
• 100 CU-hours per project each month (which runs a 0.25 CU compute for 400 hours)
• 1 GB of Postgres storage per project, capped at 20 GB across all projects
• 5 GB of public network transfer per project monthly
• 1 manual snapshot per project
• 6 hours of point-in-time restore history (up to 1 GB of change history)

If you exhaust your monthly 100 CU-hours or 5 GB of public data transfer, Neon suspends your compute until the next month starts or until you upgrade. If you pass the 1 GB project storage limit or the 20 GB account cap, write operations like inserts, updates, and deletes fail, though your existing data remains intact. Extra branches aren't sold on Free.

Launch plan

Launch is pay-as-you-go with metered hourly usage billed at the end of the month.

• Compute: $0.106 per CU-hour
• Storage: $0.35 per GB-month (unlimited)
• Includes 100 projects and 10 branches per project
• Additional branches: $1.50 per branch-month (prorated hourly at roughly $0.002 per hour, up to 5,000 branches per project)
• Public network transfer: 500 GB included per project monthly, then $0.10 per GB
• Instant restore history: up to 7 days at $0.20 per GB-month (defaults to 1 day on root branches)
• Snapshots: 100 manual snapshots per project at $0.09 per GB-month

Scale plan

Scale is meant for larger workloads needing higher autoscaling bounds, AWS PrivateLink, and longer point-in-time recovery.

• Compute: $0.222 per CU-hour
• Storage: $0.35 per GB-month (unlimited)
• Includes 1,000 projects (a soft limit that support can raise) and 25 branches per project
• Additional branches: $1.50 per branch-month up to 5,000 per project
• Public network transfer: 500 GB included per project monthly, then $0.10 per GB
• Private networking: $0.01 per GB in both directions via AWS PrivateLink
• Instant restore history: up to 30 days at $0.20 per GB-month
• Snapshots: 100 manual snapshots per project at $0.09 per GB-month

Understanding Compute Units and CU-hours

In Neon, a Compute Unit (CU) represents roughly 4 GB of RAM, paired with associated CPU and local SSD space. Compute sizes start as small as 0.25 CU (1 GB RAM) and step up through 0.5 CU (2 GB), 1 CU (4 GB), 2 CU (8 GB), 3 CU (12 GB), all the way to 56 CU (224 GB RAM) for fixed instances on the Scale tier.

Compute billing is calculated as:

compute size in CUs × hours running = CU-hours

For example, running a 0.25 CU compute for 4 hours burns exactly 1 CU-hour. On Launch, that costs $0.106. On Scale, it costs $0.222.

Read replicas are spun up as independent computes. If your primary compute runs at 1 CU and you attach a 1 CU read replica, both accrue CU-hours simultaneously.

How scale-to-zero works

Scale-to-zero shuts down your compute instance when nobody is querying the database. When an instance suspends, compute consumption drops immediately to 0 CU-hours, costing $0 for that idle period.

When a new query hits a suspended database, Neon brings the compute back up, typically within a few hundred milliseconds.

How scale-to-zero behaves depends on your plan:

• Free: Always on. It suspends compute after 5 minutes without activity, and you cannot turn it off.
• Launch: Enabled by default with a 5-minute inactivity window, but you can turn it off entirely if you need your database warm around the clock.
• Scale: Fully configurable. You can set the inactivity delay down to 1 minute, keep it at 5 minutes, or leave the compute always running.

If you are comparing serverless database platforms for small projects, see our breakdown on Neon vs Supabase: which is cheaper for a small app? for how cold starts and idle pricing compare.

Autoscaling rules and limits

Neon adjusts compute size on the fly within boundaries you define. You configure these limits in the Neon Console by opening the Edit compute drawer for your primary compute or read replica.

There is one critical constraint to keep in mind: the maximum permitted autoscaling range cannot exceed 8 CU. That means the spread between your minimum CU and maximum CU cannot be wider than 8. If your minimum size is set to 0.5 CU, your maximum cannot go beyond 8.5 CU.

Autoscaling caps by plan:

• Free: Scales up to 2 CU (8 GB RAM).
• Launch: Scales up to 16 CU (64 GB RAM).
• Scale: Scales up to 16 CU for autoscaling; fixed compute instances can be provisioned up to 56 CU (224 GB RAM).

Storage and branching costs

Postgres storage on paid tiers costs $0.35 per GB-month, calculated on hourly usage. Neon separates storage tracking into distinct components:

• Root branches: Charged on logical data size.
• Child branches: Billed on the lower of two values: either the data changes written since branch creation, or the total logical data size.
• Archived branches: Charged at the normal storage rate of $0.35 per GB-month.
• Instant restore history: Charged only on root branches at $0.20 per GB-month on paid plans (6 hours free on Free, up to 7 days on Launch, up to 30 days on Scale).
• Snapshots: Charged at $0.09 per GB-month on Launch and Scale. Free includes 1 manual snapshot, while paid plans allow up to 100 manual snapshots per project.

Other platform services

Neon provides several bundled services alongside Postgres:

• Object Storage: Free plan includes 5 GB per project. Paid plans charge $0.023 per GB-month with zero per-operation fees.
• Neon Functions: Free gives you 10 active Capacity-Hours, 400 waiting Capacity-Hours, and 1 million invocations monthly. On Launch, active compute costs $0.10/Capacity-Hour, waiting compute is $0.025/Capacity-Hour, and invocations cost $0.60 per million. On Scale, active compute is $0.12/Capacity-Hour and waiting compute is $0.03/Capacity-Hour.
• Managed Better Auth: Built in at no extra charge, supporting up to 60,000 Monthly Active Users (MAU) on Free and up to 1,000,000 MAU on Launch and Scale.

Worked billing examples

Here is how Neon bills moderate workloads on Launch versus heavier production workloads on Scale.

Example 1: Launch plan with moderate usage

Suppose you run an application on Launch where compute runs for roughly 5 billable days (120 hours) across the month at 1 CU thanks to scale-to-zero:

• Compute (120 CU-hours × $0.106): $12.72
• Root branch storage (20 GB × $0.35): $7.00
• Child branch storage (5 GB × $0.35): $1.75
• Instant restore history (10 GB × $0.20): $2.00
• Total bill: $23.47

Example 2: Scale plan with high availability

A heavier project on the Scale plan keeps larger compute resources running for 425 hours at 4 CU:

• Compute (1,700 CU-hours × $0.222): $377.40
• Root branch storage (100 GB × $0.35): $35.00
• Child branch storage (25 GB × $0.35): $8.75
• Instant restore history (50 GB × $0.20): $10.00
• Total bill: $431.15

How to set spending limits in Neon

Neon does not offer a hard spend cap switch directly in the dashboard, but you have two official control mechanisms:

1. Spending notifications: Available on Launch and Scale. Neon sends email alerts when your monthly bill reaches 80% and 100% of a target dollar amount you set. These alerts do not shut down resources; they only notify you.
2. Consumption limits: You can set these per project via the Neon API. When a project hits your set threshold for compute, storage, or transfer, Neon automatically suspends the compute.

Tracking and capping Neon with Costbrake

If you manage Neon alongside other tools like OpenAI or Vercel, native alerts might not catch sudden spikes quickly enough. Neon itself refreshes usage metrics roughly every 15 minutes, which means the most recent hour can lag.

Costbrake monitors spending across 23 cloud and AI providers; for Neon it reads hourly usage every 15 minutes on Pro and Team and hourly on its Free plan. A spike alert fires if an hour costs more than three times your 7-day average and at least $5 above baseline.

Costbrake can also apply manual brakes if costs run away:

• Disable compute: Suspends the compute endpoint and halts new connections, stopping CU-hour accrual while preserving data and branches.
• Cap compute size: Forces the endpoint down to 0.25 CU so queries keep running at minimal cost.

Brakes are opt-in, available on Pro and Team plans, and require explicit manual admin confirmation; Costbrake never executes a brake automatically. To connect your database, you can point an AI agent to agent setup or connect through the console using an organization API key. If you are also evaluating other managed engines, see our guide on how Turso pricing works.