Neon: the widest cost swing in managed Postgres

Verdict: Neon is the cheapest managed Postgres to idle and one of the most expensive to keep on — the same provider, at both ends of the price list, depending only on how many hours a day your database is actually busy. “What does Neon cost” has no single answer, and that’s the point.

One provider, the widest swing on the board

Across the four workloads Neon’s total ranges from cents to hundreds of dollars — a swing of three orders of magnitude, larger than any other provider here. Because compute scales to zero, an idle hobby database is essentially just its storage bill, and Neon lands first on price. On a bursty workload — big but only busy part of the day — scale-to-zero reclaims the quiet hours and Neon leads again.

Then keep it on. For a steady, always-on small app, Neon’s metered compute runs 24/7 and it drops toward the bottom of the price ranking — beaten better than three-to-one by a flat-tier provider like DigitalOcean (roughly $194 against DigitalOcean’s ~$58, as of the snapshot). Nothing about Neon changed; the duty cycle did. This is why a single “Neon price” misleads: quote the hobby number to a team running always-on and you’re off by more than 1000×.

Where the model pays you back

The same architecture that makes Neon cheap to idle makes it cheap to make highly available. Storage is decoupled from compute, so an HA standby duplicates no storage and the standby itself scales to zero — HA adds almost nothing. On the production-HA workload that lands Neon third-cheapest among the providers that actually qualify — and, unlike some cheaper-looking options that get struck out, it clears the workload’s HA and PITR requirements rather than shipping them as add-ons.

The ownership footnote

The “independent serverless Postgres” framing is now dated: Neon was acquired by Databricks in 2025, and Databricks’ own Lakebase is built on Neon’s engine. It’s still native Postgres and the pricing above is unaffected, but if avoiding hyperscaler lock-in is why you were looking at Neon, that reason has quietly expired.

So, when Neon

• Yes: spiky, bursty, or genuinely idle workloads; anything that benefits from scale-to-zero; HA-required production where decoupled storage makes the standby nearly free.
• Reconsider: a small database that’s simply on all the time — a flat-tier provider will beat Neon on that specific shape, and it’s the one case the hobby-tier reputation gets wrong.