Neon Pricing in 2026: A Line-by-Line Cost Teardown

Neon is serverless Postgres, hosted on AWS. You pay for compute by the hour it runs, for storage by the gigabyte, and for a few extras. Two things stand out in Neon pricing: neither paid plan has a monthly minimum, and storage is $0.35 a gigabyte-month on both.

A compute unit is the yardstick. Neon's docs say each CU allocates about 4 GB of RAM, and a compute scales in a straight line up to 56 CU.

A CU-hour is compute size multiplied by hours awake. A 2 CU compute awake for 3 hours uses 6 CU-hours. On Launch that is 6 x $0.106 = $0.636.

Autoscaling moves the size up and down inside a range you set, so the bill follows load. The docs say the range is user-defined, so costs do not scale indefinitely, and that the gap between minimum and maximum cannot exceed 8 CU.

Scale to zero is the other lever. After five idle minutes on Launch, the compute stops and stops billing. The next connection wakes it, and that wake adds a short delay to the first query. Most apps will not notice. Latency-sensitive ones might, which is why Launch lets you turn scale to zero off.

These are worked examples with assumptions we chose, not measurements. Each uses a 1 CU compute (about 4 GB of RAM) and 10 GB of stored data, at Launch rates, a 730-hour month and 30 days.

The spread is the whole story of Neon pricing. The same database costs $9.86 or $80.88 depending on hours awake. A scale-to-zero database that gets a trickle of traffic all day may never actually sleep, so check your own usage graph before assuming the low row.

On the Free plan, the Neon pricing headline is 100 CU-hours per project. That covers 100 hours of a 1 CU compute, or about 3.3 hours a day. Storage is capped at 0.5 GB per project, so a database larger than that needs Launch.

Scale charges $0.222 per CU-hour, which is 2.09 times Launch ($0.222 / $0.106). Take the always-on 1 CU example: 730 x $0.222 = $162.06 in compute, against $77.38 on Launch. That is a gap of $84.68 a month for the same compute.

What the extra buys, per the price list: 30 days of restore history instead of 7, scale to zero you can tune from 1 minute to always on, 25 branches instead of 10, fixed sizes up to 56 CU, and a choice of support tiers. Neon's plans page also lists compliance certifications, SLAs and private networking on Scale.

So the honest rule is simple. Stay on Launch until a specific Scale feature blocks you, such as a longer restore window or a compliance requirement. Paying double for compute on a hunch is the expensive way to learn that.

That is why a branch is quick and cheap to make. The docs say writes are saved as a delta, and that creating one does not add load to the parent. A branch per pull request is practical here in a way it is not on a plain server. The Neon pricing question is how many you keep.

The price is in the count. Launch includes 10 branches per project and charges $1.50 for each extra branch-month. A worked example: production plus 15 open pull requests is 16 branches. Six are extra, so 6 x $1.50 = $9.00 a month, before any compute those branches use while awake.

For a team that ships often, nine dollars for isolated test data is cheap. Restore history is separate: $0.20 per GB-month, for up to 7 days on Launch and 30 on Scale.

• Steady load. An always-on 1 CU compute is $77.38 a month before storage, so a flat-rate server can win.
• Region choice. Neon runs on AWS. Europe means Frankfurt or London, and the docs say a project's region cannot be changed after creation.
• Scale plan compute. Scale is $0.222 per CU-hour, roughly double Launch, in return for longer history, tunable scale to zero and higher fixed sizes.
• Egress. The 500 GB per project allowance is generous, but $0.10 per GB adds up for data-heavy exports.
• Ownership. Neon announced in May 2025 that it was joining Databricks. Its founders said the team would stay focused, but plans and prices are now set inside a bigger company.

Neon sells one thing well: Postgres. Its pricing covers that and nothing else. Your app, auth and storage are separate bills elsewhere. Helm7 bundles those with Postgres 18 into one minimum monthly spend, so the two prices are not like for like. Our comparison of Supabase and Neon covers the bundled-versus-unbundled trade in detail, and Helm7 against Vercel and Supabase does the same sums for a full stack.

Neon is the better pick if you want database branching per pull request, scale to zero on a bursty workload, or a US or Asia-Pacific region, since Helm7 runs only in Falkenstein and Helsinki today. It is also the better pick if your app already lives elsewhere and you only need a database.

If you are weighing a plain server with Postgres you run yourself, look at Hetzner Cloud pricing explained and how to back up and restore a PostgreSQL database first. Cheap compute comes with your own backup drills.

Yes, if your database idles, and Neon pricing rewards that. The no-minimum, pay-per-hour model rewards small and spiky workloads, and $0.35 per GB-month storage is modest. It punishes the opposite case. Model your hours awake, count your branches, then decide.