PlanetScale HA vs single node: what 3× the price buys

PlanetScale publishes two columns for the same database size, and the second one is always exactly three times the first. That is not a fee and not a support tier — it is one primary plus two replicas across three availability zones. Whether the multiplier is worth paying depends on what an hour of downtime costs you, and at the entry SKU the answer is easier than it looks.

The same database at 1× and 3×

Monthly USD, AWS us-east-1, arm64 unless stated.

The PS-20 row is the only one that is not a clean 3.0×, because the HA price rounds to $50 while the single node is $17 (ratio 2.94×); every other tier divides exactly.

What the extra money actually buys

Two more copies of the same compute, in two more availability zones. PlanetScale states the topology plainly: high availability is one primary and two replicas across three AZs, and the catalog bills it as a 3-node SKU. Nothing about the primary changes — same vCPU, same RAM, same connection limit — so the purchase is availability, not performance.

That has a practical consequence: HA does not make slow queries faster. A workload that is CPU-bound at PS-40 will be CPU-bound at PS-40 HA, and pay $166/month more for the privilege (calculated 2026-10-25).

The absolute numbers matter more than the multiplier. At PS-5 the premium is $10 a month — $120 a year — which is less than a single hour of engineering time in most companies. At PS-2560 it is $3,019 a month, and the same reasoning stops applying: at that size a serious incident review usually happens before the invoice does.

How we measured it

We read PlanetScale's generated price catalog on 2026-10-25 (generated at 14:49:57Z) and divided each HA row by its single-node counterpart. Both columns come from the same table, so the ratio is exact rather than estimated: three nodes at the published per-node rate. No cluster was provisioned for this page.

What changes in practice

The multiplier applies to compute only. Storage, backup storage, egress and an optional dedicated PgBouncer are billed on top of whichever column you pick, so an HA cluster is not three times the total bill — it is three times the compute line, plus the same storage.

For a comparison against a bundled plan, the fair entry point is Supabase Pro at $25/month, which includes a small database, storage and backups but no automatic three-zone replica set. A PlanetScale PS-5 HA cluster at $15/month is cheaper on compute and dearer once storage and egress are added, which is why the two products are rarely priced by their headline number alone.

One more lever: region and architecture. The catalog is priced for AWS us-east-1, and moving the same SKU to x86-64 costs 30% more at PS-10 and 23.6% more at PS-2560. Choosing arm64 is often a bigger saving than choosing single node over HA at small sizes — but it is also the choice that cannot be reversed in a hurry.

Bottom line

Take HA from the start at PS-5 or PS-10: $15 and $30 a month is cheap insurance against a zone failure, and the $120–$260 annual premium is smaller than most incident reviews. Above PS-320 the decision reverses — $699 and up per month for redundancy is a budget line, and it should be justified by a measured recovery-time requirement rather than by default.