What a Neon CU-hour is, and what an always-on database costs

Neon's bill is not a subscription with an overage; it is a metered quantity called a CU-hour, and most teams only learn what that means on their first invoice. The number is easy to reason about once you know two facts: what a CU is, and how many hours a month your database is actually awake.

A CU-hour is one compute unit running for one hour. Free includes 100 CU-hours per project at a default 0.25 CU, which is about 400 active hours. Keep the database on around the clock and the arithmetic changes: 0.25 CU x 730 hours is about 182.5 CU-hours, which is $19.35 a month on Launch.

The CU-hour rates and what a month of compute costs

The pricing page states the formula in two lines: monthly CU-hours equals average CU size times the number of hours the compute runs, and monthly compute cost equals monthly CU-hours times the CU-hour price. It also prints the always-on figure itself: once scale to zero is disabled, the minimum a database can use in a month is about 182.5 CU-hours.

What the numbers actually mean

Neon sells compute the way an electricity meter sells power, and the two published rates differ by more than 2x: $0.106 per CU-hour on Launch, $0.222 on Scale. A 0.25 CU database awake all month is $19.35 on one and $40.52 on the other, for identical hardware, which is worth checking before assuming Scale is the premium plan you need.

The free allowance is the part that surprises people. 100 CU-hours is not 100 hours; it is 100 CU-hours, and at the 0.25 CU default that is 400 wall-clock hours of activity. The page's own example makes the shape concrete: a database running 0.25 CU for about 12 hours a day uses roughly 90 CU-hours a month, which fits inside Free with room to spare.

Always-on is where free ends. 0.25 CU running 730 hours is 182.5 CU-hours, nearly double the 100-hour allowance, so a free database kept permanently awake runs out somewhere past the halfway point of the month. That is a deliberate design: Neon's model assumes the database sleeps.

How we measured it

The definition, both rates and the 182.5 CU-hour figure are quoted from Neon's pricing page on 2026-10-25; the 5-minute suspend window is from Neon's scale-to-zero documentation. The monthly totals multiply those published numbers by 730 hours, the month length the pricing page itself uses. Nothing was provisioned and no benchmark was run.

What changes in practice

Scale to zero is the whole product decision. A demo app, a staging branch, an internal tool used eight hours a day: all of them spend most of the month suspended and pay a small multiple of their active hours. A production database behind an API that polls every few seconds never suspends, and from Neon's point of view it is always on, priced at 182.5 CU-hours minimum.

The other trap is size. CU-hours scale linearly with compute size, so a 1 CU database is four times the monthly cost of the same hours at 0.25 CU — $77.38 against $19.35 on Launch. Autoscaling up to 2 CU or 4 CU under load is billed at whichever size was active, and the pricing page's ceiling of 16 CU on Launch and 56 CU on Scale tells you how far that can go.

Storage is separate and cheaper: $0.35 per GB-month on paid plans, plus $0.20 per GB-month for history. A 100 GB database costs $35 a month in storage on either plan, which is more than an always-on 0.25 CU database and less than an always-on 1 CU one. It is also the one line that never scales to zero: a suspended database still holds its data and still bills at its rate, so the cheapest possible month on Neon is storage and nothing else.

Bottom line

For anything that sleeps, Neon's free plan is generous — 100 CU-hours covers about 400 active hours a month at 0.25 CU. For anything always on, budget $19.35 a month on Launch before storage, and check whether you need Scale's $0.222 rate at all: the only reason to pay it is the higher compute ceiling and the larger limits, not the per-hour price.