Serverless database pricing is rarely one rate. A bill can include compute, storage, I/O, backups, recovery history, and data transfer. Some providers use their own billing units.
The Serverless Database Pricing Comparison gives each product the same three monthly workloads. Native billing differences stay visible. Required costs that cannot be calculated stay unresolved instead of being guessed.
Monthly workload assumptions
The three workloads give each product the same starting point: active compute, database storage, and public egress. They are pricing inputs, not a performance benchmark.
| Workload | Active compute | Database storage | Public egress |
|---|---|---|---|
| A - intermittent | 7.5 hours | 10 GB-month | 1 GB |
| B - working hours | 176 hours | 100 GB-month | 25 GB |
| C - continuous | 730 hours | 500 GB-month | 250 GB |
Providers can meter the same input in different ways. Active compute only counts when elapsed time maps to a published charge. Primary storage does not stand in for backup history, WAL, or allocated disk. Egress keeps the provider’s published allowance scope.
If a provider needs another meter, such as Request Units or storage I/O, we keep it separate unless the provider publishes a clear conversion from the common inputs.
Pricing scope
The comparison covers relational SQL products with reviewed public self-serve pricing in USD. Supabase Pro + Micro and PlanetScale Postgres PS-10-AWS-ARM Single Node are shown separately because they retain a provisioned or platform compute minimum.
The selected region fixes the public rate used for each regional product. It does not imply equal infrastructure, latency, or performance.
How serverless database pricing works
Compute is only one part of many database bills. Storage, I/O, backups, recovery history, and egress can add separate charges. A low compute rate can still lead to a higher total bill once those meters are included.
WAL (write-ahead log) records database changes before they are written to the main data files. Providers can retain this history for recovery or replication. That retained history can add a storage charge.
PITR (point-in-time recovery) lets you restore a database to an earlier moment. It usually relies on backups and retained change history, so PITR storage can be billed separately from primary database storage.
I/O means storage reads and writes. Egress means data transferred out of the provider. Both can have their own rates and allowances.
CU, ACU, vCore, and Request Units are provider-specific billing units. They are not interchangeable, so the comparison does not convert them into one universal database unit.
How prices are calculated
Each product keeps its published billing model. We apply documented billing increments, rounding, plan floors, allowances, and credits without replacing native billing units with a synthetic one. GB and GiB also stay distinct when the provider bills them differently.
Prisma Postgres uses declared billed operations of 1M, 25M, and 250M for workloads A, B, and C. These values apply only to Prisma.
Some charges depend on usage the common workloads cannot determine. Aurora I/O, CockroachDB or TiDB Request Units, and Azure backup consumption are examples. Rather than guess those quantities and publish a misleading total, the comparison leaves that part of the bill unresolved.
When a price is complete
Exact means every required monetary component can be calculated from the reviewed public terms and workload.
Subtotal + additions means the displayed amount is exact for the known charges, but one or more required costs are still unknown. Those costs are named beside the subtotal and listed in Additional costs. Treat the subtotal as the known part of the bill, not the full monthly price.
No public rate means a required charge exists but no usable public rate was found.
Sorting uses the amount shown in each cell. For Exact results, that is the complete total. For Subtotal + additions, it is only the known subtotal. Sorting makes the table easier to scan, but it does not turn an incomplete subtotal into a full-price ranking.
Official pricing sources
Each row fixes the product, plan, configuration, and region used for the comparison. Rates, limits, and configuration rules come from the official sources below. The current evidence was reviewed through August 18, 2026.
| Provider | Product / plan | Configuration / SKU | Region | Official sources |
|---|---|---|---|---|
Neon | NeonLaunch | 0.25-2 CUPostgreSQL 175 min scale-to-zero | us-east-1 | Pricing · PlansNetwork transfer |
Xata Cloud | Xata Cloudmicro | microPostgreSQL 1730 min scale-to-zero | us-east-1 | PricingBranching |
CockroachDB Basic | CockroachDB BasicPay-as-you-go | Provider-managed computeRequest Units | us-east-1 | Basic planning and pricingResource usage · Costs |
TiDB Cloud Starter | TiDB Cloud StarterScalable | One Scalable instanceRequest Units | us-east-1 | Starter pricing detailsStarter limits · Backup and restore |
Amazon Aurora Serverless v2 | Aurora PostgreSQL-Compatible Serverless v2Aurora Standard | 0-2 ACUPostgreSQL 175 min auto-pause | us-east-1 | Aurora pricingServerless v2 behavior · Auto-pause |
Azure SQL Database | Azure SQL DatabaseGeneral Purpose serverless | GP_S_Gen5_20.5-2 vCore60 min auto-pause | East US 2 | PricingServerless tier · Automated backups |
Prisma Postgres | Prisma PostgresPaid self-service | Provider-managed computeBilled operations per workload | us-east-1 | PricingPostgres FAQ |
Supabase | SupabasePro + Micro | Micro · 1 GB · 2-core ARMPostgreSQL 17 if selectableFull-month compute | us-east-1 | PricingCompute billing · Disk billing |
PlanetScale Postgres | PlanetScale PostgresPS-10-AWS-ARM Single Node | 1/8 vCPU · 1 GBPostgreSQL 17Full-month compute | us-east-1 | Postgres pricingCluster storage |
What’s excluded
The comparison does not score performance, latency, availability, developer experience, or operational quality. Taxes, negotiated pricing, enterprise contracts, promotions, and unrelated services are excluded.
Document/KV databases, Redis/Valkey, edge SQLite, and application-data products are outside this SQL comparison. If the workloads do not define a provider-specific quantity, we leave it unpriced instead of inventing one.