Managed database pricing is not one price but the sum of several dimensions: compute as instance hours, vCores or capacity units, storage per GB and month, input and output operations on some engines, backup storage beyond the included amount, and high availability or replicas as additional compute. What matters on the bill is that compute usually runs around the clock, that standbys and replicas multiply it, and that storage and backups keep billing while the database is stopped.
How it works
The three providers offer relational engines sized by instance or vCore, and databases billed by capacity units or requests. The table maps the services to the dimensions, in the providers' own terms.
| Service | Compute | Storage and I/O | High availability and replicas | Serverless |
|---|---|---|---|---|
| Amazon RDS | Hour per DB instance class | GB-month, Provisioned IOPS on top | Multi-AZ with one standby or two readable standbys | none |
| Amazon Aurora | Instance hour | Standard: GB-month plus I/O per million requests; I/O-Optimized: higher GB rate, I/O included | Replicas as further instances | Aurora Serverless, ACUs per second from 0 ACU |
| Amazon DynamoDB | Provisioned: RCUs and WCUs per hour | GB-month per table class | Global tables: replicated write units | On-demand: request units per request |
| Azure SQL Database | vCore hour or DTU bundle | vCore: provisioned storage (maximum data size); DTU: included in the bundle | Business Critical with three additional replicas | Serverless per vCore-second, auto-pause |
| Azure Database for PostgreSQL Flexible Server | vCore hour in Burstable, General Purpose, Memory Optimized | GiB-month, IOPS above the included level | Zone-redundant HA: primary and standby | none |
| Azure Cosmos DB | RU/s per hour, standard or autoscale | GB per region | Every region carries throughput and storage | Serverless per RU consumed |
| Cloud SQL | vCPU and memory per edition | GB-month per storage type | HA prices for regional instances | none |
Compute
Amazon RDS bills DB instance hours per second, with a 10-minute minimum after each start or change. Azure SQL Database bills provisioned databases by the hour; the serverless tier bills per vCore-second at a higher price per vCore and pauses idle databases, leaving only storage to pay. Microsoft puts Business Critical at about 2.7 times General Purpose, because three additional replicas run with it. PostgreSQL Flexible Server bills by the hour, and a server deleted after five minutes costs a full hour. Cloud SQL prices vCPU and memory separately, per edition (Enterprise, Enterprise Plus) and region. Cosmos DB charges standard throughput per 100 RU/s and hour, and autoscale on the highest RU/s used in each hour, between 10 and 100 percent of the limit, at a factor of 1.5. Provisioned capacity on DynamoDB and Cosmos DB costs the same whether it is consumed or not.
Storage, input and output, and backups
RDS, Azure SQL Database and PostgreSQL Flexible Server bill storage by provisioned size, Aurora, DynamoDB and Cosmos DB by data stored; Provisioned IOPS on RDS add a charge per IOPS provisioned. I/O requests are billed on RDS only for magnetic storage and on Aurora Standard. Backup storage is included up to a limit: on Azure SQL Database up to the maximum data size, on PostgreSQL Flexible Server up to the provisioned storage, at double the price for geo-redundant backups. Longer retention and manual snapshots grow backup storage; copies outside the database belong to backup and disaster recovery.
High availability, licences and extended support
A standby is compute: zone-redundant HA on PostgreSQL Flexible Server bills compute and storage for both primary and standby, Cloud SQL bills regional instances at HA prices and read and failover replicas at the rate of standalone instances, and an Azure SQL Database geo-secondary costs as much as its primary. On Cosmos DB, throughput and storage are billed in every region. SQL Server adds a licence: Azure SQL Database offers a licence-included price or one with Azure Hybrid Benefit, not in serverless, and Cloud SQL bills the SQL Server licence per vCPU with a minimum of four per instance (licences in the cloud). For major versions past community end of life, RDS and Aurora bill Extended Support per vCPU-hour or ACU-hour, Cloud SQL per vCPU-hour (per instance-hour on shared-core instances).
What it means for your cost
In billing data a database appears as several line items: instance or vCores, storage, IOPS or I/O, backup storage, data transfer, and where they apply licence and Extended Support. Standbys and replicas sit beside them as compute of their own. RDS DB instance classes carry the family names of EC2 instance types, such as db.t3 or db.m4 (virtual machines and instance families), and size is the main lever for rightsizing (rightsizing).
The billing models call for decisions: serverless pays off with long idle periods, although Azure SQL Database prices its vCore higher there; for Aurora I/O-Optimized AWS names I/O spend above 25 percent of Aurora spend as the threshold; AWS recommends DynamoDB on-demand for most workloads and provisioned capacity for steady, predictable throughput. For a steady baseline there are commitments (Reserved Instances, reserved capacity, committed use discounts, Database Savings Plans on AWS and the savings plan for databases on Azure) that lower effective cost; they are covered under reserved capacity for managed services.
Where it goes wrong
- Stopped is not free. Symptom: a stopped database keeps generating cost. Cause: RDS and PostgreSQL Flexible Server keep billing storage and backups while stopped; only compute stops.
- High availability multiplies compute. Symptom: compute costs several times what the instance size suggests. Cause: Multi-AZ, zone-redundant HA, Cloud SQL HA prices and geo-secondaries are billed as compute of their own.
- An old major version. Symptom: new Extended Support lines. Cause: the engine runs past community end of life, and RDS, Aurora and Cloud SQL charge for it per vCPU or ACU and hour (Cloud SQL per instance-hour on shared-core instances).
- Backups beyond the included amount. Symptom: the backup line grows faster than the database. Cause: longer retention, manual snapshots and a high change rate push backup storage past the included amount.
- Capacity paid for, not used. Symptom: flat cost at low load. Cause: provisioned RCUs and WCUs on DynamoDB or standard throughput on Cosmos DB are billed per hour however much of them is consumed.
Example
Amazon Aurora in Europe (Frankfurt), list prices in USD, read on 2026-10-16: Aurora Standard costs USD 0.119 per GB-month of storage plus USD 0.22 per million I/O requests, Aurora I/O-Optimized USD 0.268 per GB-month with no I/O charges. Instance prices also differ between the two configurations, so which one is cheaper depends on the ratio of I/O to storage and instances.
Sources
Read on 2026-10-16. AWS: Amazon RDS pricing, Amazon RDS for PostgreSQL pricing, Amazon Aurora pricing, Amazon DynamoDB pricing. Microsoft: Azure SQL Database pricing, single database, Purchasing models for Azure SQL Database, Azure Database for PostgreSQL Flexible Server pricing, Azure Cosmos DB pricing. Google: Cloud SQL pricing.