Neon Compute-Hour Billing, Branches & Actual Costs: 2026 Guide

By usage, with the plan prices acting as representative spends rather than fixed fees. Free costs nothing, Launch runs about $15 a month and Scale about $31, and both paid plans have a $0 minimum, billing from what you actually use. Compute is the main meter, at $0.106 per CU-hour on Launch and $0.222 on Scale, storage is $0.35 per GB-month, and restore history $0.20 per GB-month. Because compute scales to zero, an idle database costs almost nothing while an always-on one accrues hourly. So your real bill is your own compute, storage and branch usage, not the listed tier price.

For development and prototypes, yes; for a busy production database, usually not. The Free plan is generous, with up to 100 projects, 100 CU-hours a month per project and 0.5 GB of storage each, and scale-to-zero means idle projects cost nothing. The limits that bite are the per-project compute-hour and storage caps, which suit non-production work rather than sustained traffic. When a project needs more compute, more storage, or the longer restore history of a paid tier, Launch is the step up, billed from usage. Build and validate on Free, then shift to a paid plan before production traffic arrives.

By the compute-unit-hour, on the resources a database actually uses while active. Compute is $0.106 per CU-hour on Launch and $0.222 on Scale, so the bill follows how long and how big a database runs. The key detail is scale-to-zero. Neon suspends a database when it is not serving queries, so a workload that idles between bursts stops accruing compute, while an always-on one bills every hour it is up. That is why bursty and development databases are cheap on Neon, and why a steady always-on production database can cost well above the representative plan figure. Model both the compute size and the uptime.

Database storage is $0.35 per GB-month on both Launch and Scale. On top of that, Neon retains a window of history for point-in-time restore, billed at $0.20 per GB-month. A long retention window on a large database is therefore a second storage line the plan figure does not show. Storage is usually smaller than compute on a Neon bill, but it is steady rather than scaling to zero, so it accrues even while the database is idle. Keeping the restore-history window to what recovery genuinely requires, rather than the maximum, is the simplest way to hold the storage portion of the bill down.

Branches beyond a plan's included count cost $0.002 per branch-hour, which works out to roughly $1.50 a branch-month if a branch is left running continuously. Branching is one of Neon's signature features, letting you spin up an isolated copy of a database for a pull request or a test. The cost accumulates when branches are not cleaned up. A team that creates a branch per pull request and forgets to delete merged ones can end up paying for dozens of stale branches. The fix is to automate branch deletion when a pull request is merged or closed, so branches do not drift into a standing cost.

Both bill from usage with a $0 minimum, but they differ on limits and per-unit compute rates. Launch, at a representative $15 a month, allows up to 16 compute units and bills compute at $0.106 per CU-hour, suiting small to mid-sized applications. Scale, at a representative $31, raises the ceiling to 56 compute units and bills compute at $0.222 per CU-hour, aimed at larger or more demanding databases. Storage is $0.35 per GB-month on both. So Scale buys more headroom and features at a higher compute rate, which makes sense once a database outgrows Launch's compute ceiling rather than simply for a bit more storage.

Yes, at Enterprise. There the quote covers committed usage, compliance, support and rates, all negotiable for a sizeable account. Anchor with a Supabase or DigitalOcean figure and a committed-usage offer, and time it near quarter-end. On Free, Launch and Scale there is nothing to negotiate, since the tiers and per-unit rates are set. The saving on those plans is architectural: scale-to-zero, cleaning up branches, and a sensible restore-history window. Small workloads win on usage discipline; a large steady one wins on a committed Enterprise deal.

Lean on scale-to-zero for anything that does not need constant uptime, since compute is the largest line and an idle database stops billing. Automate deletion of stale branches so a branch-per-pull-request workflow does not accumulate hourly charges. Keep the restore-history window to what recovery genuinely needs rather than the maximum. Right-size compute units to real load, and run non-production work on the Free tier where it fits. For a large steady database, negotiate committed usage at Enterprise. In combination, those habits keep a Neon bill near the compute and storage you truly use, not a padded representative spend.