PlanetScale Review

PlanetScale used to be the MySQL company with the famous free tier.

In 2024 it dropped the free plan, and in 2025 it launched Postgres on the same control plane, so the product you are evaluating in 2026 is a paid-only managed database with three engines, Postgres, Vitess for sharded MySQL and Neki for sharded Postgres, priced per cluster size.

My view: it is the most honest pricing page among the managed Postgres vendors, because the price of a cluster is the price you pay whether or not it is busy, and that same honesty makes it the wrong choice for a side project.

This review is for developers choosing a production Postgres (or MySQL) host and deciding between PlanetScale's fixed clusters and the usage-based models of Neon and Supabase.

I cover what the platform includes, how branching, Insights and Metal work, what each cluster size costs in AWS us-east-1 and which line items the headline price leaves out, then price four alternatives from their own pages.

PlanetScale is a managed database platform, founded in 2018 in San Francisco by the engineers behind Vitess, the sharding layer that scaled MySQL at YouTube. It runs on AWS, GCP and Azure and sells three engines on one dashboard.

PlanetScale Postgres is a standard Postgres cluster in two shapes: single node for development and low-traffic production, or high availability with one primary and two replicas across three availability zones, with failover in under 30 seconds.

Vitess is the MySQL-compatible engine with horizontal sharding for very large workloads. Neki, the newest, brings the same sharding to Postgres and requires a three-node configuration per shard.

All three can run on network-attached storage or on PlanetScale Metal, instances with direct-attached NVMe drives for the lowest latency and highest IOPS; the vendor publishes benchmarks against Aurora, AlloyDB, Neon, Heroku and Supabase.

Around the engines sit database branching from production backups, Query Insights with automatic anomaly detection and execution-plan recommendations, automatic backups every 12 hours, and curated extensions with custom parameter tuning.

You pick an engine, a cloud, a region and a cluster size, and the size sets the monthly price. Sizes run from PS-5 (1/16 vCPU, 512 MiB) through PS-10, PS-20, PS-40 and PS-80 (1 vCPU, 8 GiB) to PS-160 (2 vCPU, 16 GiB) and far beyond, and each size has a single-node price and a three-node HA price that is roughly three times higher.

Billing is prorated to the millisecond, so a cluster that lives for a week costs a quarter of its monthly rate, but an idle cluster costs the full rate; nothing scales to zero.

Day to day, branching is the feature that changes the workflow. A development branch is an isolated database built from a production backup, so you test a migration against real data without touching production.

On Postgres each development branch is a separate single-node cluster billed at the PS-5 rate, prorated, rather than the included branch hours that Vitess users get. Query Insights keeps seven days of per-query statistics and flags anomalies, which is the tool you reach for when the p99 moves.

What the cluster price does not include is the second half of the bill. Network-attached storage is metered per GB, user-scheduled backups cost $0.023 per GB a month on top of the automatic every-12-hours backups, egress above the included amount is metered, and a dedicated PgBouncer pooler is an optional paid add-on.

Metal flips this: storage is included because it lives on the instance, which is why Metal pricing starts with an M-10 cluster with 10 GiB and rises with the NVMe you attach.

PlanetScale publishes a full price catalog for AWS us-east-1 and notes that other providers and regions cost more. The figures below were read on PlanetScale's pricing page and plans documentation in October 2026; competitor figures come from Neon pricing, Supabase pricing, Turso pricing and CockroachDB pricing, read the same month.

Postgres is priced per cluster size, single node or HA. PS-5 (1/16 vCPU, 512 MiB) is the entry size in both shapes. PS-10 (1/8 vCPU, 1 GiB) is $10 single node and $30 HA on arm64, or $13 and $39 on x86-64. PS-20 (1/4 vCPU, 2 GiB) is $17 single node; PS-40 (1/2 vCPU, 4 GiB) is $28 and $83; PS-80 (1 vCPU, 8 GiB) triples from single node to $148 HA; PS-160 (2 vCPU, 16 GiB) is $96 and $286. The x86-64 variants run about 20% higher. Metal HA starts with an M-10 with 10 GiB of NVMe and reaches $180 with 200 GiB on arm64; Vitess clusters start at $39 a month for a three-node PS-10 on x86-64, and Neki at $30 for a PS-10 arm64 shard plus routers from $6 a month.

Cost scenario, computed from those rates: a startup running one production HA cluster at PS-80 arm64, one PS-5 development branch and no Metal pays the PS-80 HA rate plus one PS-5 branch, which is $153 a month in us-east-1 before storage, scheduled backups and egress. Trim the production cluster to a single node and the bill is $55. On Neon's Launch plan, one compute unit running all month costs $0.106 an hour, or about $77 for 730 hours, plus $0.35 per GB-month of storage, so a 10 GB database lands near $81 with no replicas. On Supabase, Pro is $25 a month with a Micro instance included and a Small compute add-on takes it to $40, with 8 GB of database storage included and $0.125 per GB after.

Devshot data: Toolradar tracks 24 managed database services as of October 2026, and 54% of them are free, freemium or open source. PlanetScale is in the paid-only minority, which is the first filter to apply: if a $0 tier is a requirement, it is out before the feature comparison starts.

PlanetScale fits teams running a real production workload that want a fixed database bill, three-zone high availability and the option of NVMe speed. Teams on MySQL at scale have Vitess, the same sharding layer YouTube runs on, and Postgres teams that have outgrown a single instance have Neki.

Anyone who has been surprised by a usage-based database invoice will appreciate that a PS-80 costs the same on a quiet month as on a busy one.

It is a weaker fit for side projects, prototypes and anything that sleeps most of the day: there is no free plan and nothing scales to zero, so Neon's free 100 compute-hours or Supabase's free 500 MB project cost less than PlanetScale's cheapest cluster.

Teams that want auth, storage and edge functions bundled with the database should look at Supabase, and teams that need a global multi-region SQL database with a 99.99% SLA on shared infrastructure should price CockroachDB Standard.