Updated 28 Oct 2026 • 7 mins read
Amazon RDS Pricing Explained (2026): Instances, Storage, I/O, and Hidden Costs
and provisioned IOPS, backups, data transfer, licence-included engines, and the Extended Support surcharge, with Aurora adding its own I/O model. This 2026 breakdown gives every rate for US East, worked examples at three sizes, and the costs estimates miss.
Amazon RDS pricing has seven meters and almost everyone quotes one. The instance rate, $0.017 an hour for a db.t3.micro, $0.24 for a db.r6g.large, is the number in every comparison and the smallest part of most RDS bills once a database is in production, because production adds a standby that doubles it, storage that both nodes carry, backups that keep accruing, replicas added for migrations, transfer from an application tier in the wrong zone, and, since March 2026, a surcharge that arrives by calendar when an engine version ages out of support and often exceeds the instance itself. This guide prices all seven.
It covers instance pricing by class and engine, the Multi-AZ multiplier, storage and IOPS, backups and snapshots, data transfer, licence-included engines, Reserved Instances, the Extended Support surcharge, and Aurora's different model, with US East rates as of October 2026 from the RDS and Aurora pricing pages and worked examples at three sizes. For how RDS fits beside EC2, S3, and EBS on the bill, see our guide to understanding AWS pricing; for cutting the result, our RDS cost optimization guide.
The short answer: Amazon RDS charges instance hours by class and engine (MySQL and PostgreSQL in US East: db.t3.micro $0.017, db.t3.medium $0.068, db.m6g.large about $0.159, db.r6g.large $0.24 per hour; Oracle and SQL Server licence-included classes two to four times more), storage per GB-month (gp3 $0.115 with 3,000 IOPS included, io2 $0.125 plus $0.10 per provisioned IOPS), backup storage beyond an amount equal to allocated storage at $0.095 per GB-month, and data transfer (cross-AZ $0.01 per GB each way, internet $0.09 per GB). Multi-AZ provisions a standby that doubles instance and storage; a two-standby cluster triples them. Reserved Instances discount instance hours 29 to 69 percent. Since March 2026, engine versions past standard support pay $0.20 per vCPU-hour in Extended Support. Aurora bills storage on usage at $0.10 per GB-month and either $0.20 per million I/O requests or an I/O-Optimized premium, with Serverless v2 at about $0.12 per ACU-hour. A small Single-AZ database is about $15 a month; a production Multi-AZ db.m6g.large with 100 GB about $255.
Key Takeaways
- The instance rate is one of seven meters and the smallest part of most production bills.
- Multi-AZ doubles instance and storage; it is the right default for production and the wrong default for everything else.
- Extended Support at $0.20 per vCPU-hour switches on by calendar and often exceeds the instance rate.
- Reservations discount instance hours only, up to 69 percent; storage, IOPS, backup, and transfer stay at on-demand rates.
- Aurora bills storage on usage and I/O separately; I/O-Optimized pays off above about a quarter of the bill in I/O.
Meter 1: Instance Hours
Every RDS instance bills per hour by class and engine. MySQL, PostgreSQL, and MariaDB share near-identical rates; Oracle and SQL Server licence-included classes cost two to four times more, or less on bring-your-own-licence. US East, October 2026, approximate for MySQL and PostgreSQL.
| Class | vCPU / memory | Hourly | Monthly Single-AZ | Monthly Multi-AZ | Typical use |
|---|---|---|---|---|---|
| db.t4g.micro | 2 / 1 GB | $0.016 | $12 | $23 | Dev and test; burstable Graviton |
| db.t3.micro | 2 / 1 GB | $0.017 | $12 | $25 | Dev and test; legacy free tier |
| db.t3.medium | 2 / 4 GB | $0.068 | $50 | $99 | Small applications |
| db.m6g.large | 2 / 8 GB | About $0.159 | $116 | $232 | General-purpose production (Graviton) |
| db.m5.large | 2 / 8 GB | $0.171 | $125 | $250 | General-purpose production (Intel) |
| db.m6g.xlarge | 4 / 16 GB | $0.318 | $232 | $464 | Mid-size production |
| db.r6g.large | 2 / 16 GB | $0.24 | $175 | $350 | Memory-optimized production |
| db.r6g.2xlarge | 8 / 64 GB | $0.96 | $701 | $1,402 | Large memory-bound databases |
Graviton classes (t4g, m6g, r6g) run the open-source engines 10 to 20 percent below their Intel equivalents; burstable t-classes accumulate CPU credits and suit development and light workloads, not steady production.
Meter 2: Multi-AZ
Multi-AZ provisions a synchronous standby in a second Availability Zone with its own storage copy and bills it as a full instance plus storage, roughly doubling both lines; a Multi-AZ DB cluster with two readable standbys triples them. It buys automatic failover in about a minute and no data loss on failure, which production needs. It is applied by template to staging, test, and reporting databases far more often than those need it, which is the largest single RDS overspend in most estates.
Meters 3 and 4: Storage and IOPS
| Storage type | Per GB-month | Performance | When to use |
|---|---|---|---|
| gp3 (general-purpose SSD) | $0.115 | 3,000 IOPS and 125 MB/s included below 400 GB; 12,000 IOPS and 500 MB/s at 400 GB and above; extra IOPS $0.02 per IOPS-month | Almost every workload |
| gp2 (previous generation) | $0.115 | 3 IOPS per GB, burst to 3,000 | Migrate to gp3 |
| io2 Block Express | $0.125 plus $0.10 per provisioned IOPS-month | Up to 256,000 IOPS, consistent sub-millisecond latency | Very high-transaction databases |
| io1 | $0.125 plus $0.10 per provisioned IOPS-month | Up to 80,000 IOPS | Legacy; io2 is better at the same price |
| Magnetic | $0.10 plus per-I/O charges | Low | Legacy |
Storage bills on allocated size, not used, and every standby and replica carries a full copy, so a 500 GB Multi-AZ database with one replica bills 1,500 GB. Provisioned IOPS on io1 and io2 bill whether consumed or not; a database provisioned at 10,000 IOPS that averages 900 pays $1,000 a month for the difference.
Meter 5: Backups and Snapshots
Automated backups and manual snapshots are free up to 100 percent of allocated storage across the region, then $0.095 per GB-month. Retention of 7 to 14 days on a lightly changing database stays within the allowance; 35 days on a database that churns, or manual snapshots taken and kept, exceed it. Snapshots persist after their instances are deleted and are among the commonest orphaned RDS costs. Snapshot export to S3 costs $0.10 per GB.
Meter 6: Data Transfer and Licences
Transfer into RDS is free and transfer within an Availability Zone is free. An application tier in a different zone from the primary pays $0.01 per GB each way on every query, which on 500 GB a month is $10 and on 50 TB is $1,000. Cross-region replication pays about $0.02 per GB; internet egress $0.09 per GB after 100 GB. Oracle and SQL Server licence-included classes bundle the database licence into the hourly rate at two to four times the open-source engine price; bring-your-own-licence rates are closer to MySQL's.
Meter 7: Extended Support
Since March 1, 2026, an RDS instance running an engine major version past its end of standard support, MySQL 5.7, PostgreSQL 11 and 12, and others as they age, bills $0.20 per vCPU-hour in US East on top of the instance rate, automatically, for up to three years. On a 2-vCPU instance that is about $292 a month, Single-AZ, and $584 Multi-AZ, frequently more than the database itself. Nothing on the instance changes; the surcharge appears by calendar. Upgrading the engine removes it.
Reserved Instances
RDS Reserved Instances commit to an instance class, engine, and region for one or three years and discount instance hours by roughly 29 to 69 percent depending on term and payment: about 30 to 35 percent for one year with no upfront payment, 40 to 45 percent for one year all upfront, and 60 to 69 percent for three years all upfront. Reservations apply to instance hours only, including the standby on Multi-AZ; storage, IOPS, backup, transfer, and Extended Support stay at on-demand rates. Database Savings Plans offer similar discounts with flexibility across classes. Right-size and decide Multi-AZ before reserving, or the reservation locks the waste in for the term.
Aurora: A Different Model
| Line | Aurora Standard | Aurora I/O-Optimized | Aurora Serverless v2 |
|---|---|---|---|
| Instance | db.r6g.large about $0.26 per hour; readers share storage | About 30 percent higher instance rate | About $0.12 per ACU-hour; scales to zero |
| Storage | $0.10 per GB-month on actual usage, not provisioned | About $0.225 per GB-month | Same as the cluster's configuration |
| I/O | $0.20 per million requests | None | Per the cluster's configuration |
| Backup | Free up to cluster storage size, then $0.021 per GB-month | Same | Same |
| When it wins | Read-heavy and moderate-write workloads | I/O above about a quarter of the Standard bill | Intermittent and variable workloads |
Aurora replicas share the cluster volume, so a reader costs its instance hours only, with no second storage copy, which is Aurora's structural advantage over Multi-AZ RDS for read scaling. Its structural risk is I/O: a write-heavy workload on Aurora Standard can pay more in I/O requests than in instances, which is what I/O-Optimized exists to cap.
Worked Examples
US East, MySQL or PostgreSQL, monthly, on demand.
| Configuration | Instance | Storage | Other | Monthly |
|---|---|---|---|---|
| Development: db.t3.micro, 20 GB gp3, Single-AZ | $12.40 | $2.30 | None | About $15 |
| Small production: db.t3.medium, 100 GB gp3, Multi-AZ, 14-day backups | $99 | $23 | Backups within allowance | About $122 |
| Production: db.m6g.large, 100 GB gp3, Multi-AZ, one read replica, 35-day backups (150 GB over), 500 GB cross-AZ | $348 | $35 | Backups $14, transfer $10 | About $407 |
| Same production on PostgreSQL 11 after standard support ended | $348 | $35 | Backups $14, transfer $10, Extended Support $876 (2 vCPU x 3 nodes) | About $1,283 |
| Same production with a 3-year all-upfront reservation on primary and standby, engine upgraded, replica removed | About $93 | $23 | Backups $14, transfer $10 | About $140 |
The third row is the one every estimate stops short of: the same db.m6g.large database costs $116 as an instance and $407 as a deployment. The fourth is the surcharge; the fifth is what the same database costs with the levers pulled.
How Opslyft Helps
The seven meters land on an AWS bill as one RDS line, with no instance, replica, snapshot, or surcharge named on it, which is why the migration replica and the Extended Support charge are discovered on the invoice. Opslyft breaks RDS and Aurora spend out by instance, database, and team beside the compute and storage around it, with the standby, replica, backup, and Extended Support lines visible and the idle ones flagged; to price a configuration before you build it, including the standby and the surcharge, use our RDS cost calculator.
The Bottom Line
Amazon RDS pricing is an instance rate everyone quotes and six meters that decide the bill: the standby that doubles it, the storage every copy carries, the backups that outlive their instances, the transfer from the wrong zone, the licence, and the surcharge that arrives when an engine ages out. Price the deployment, not the database; decide Multi-AZ per database; keep engine versions in support; reserve the stable baseline after right-sizing; and for Aurora, match the I/O model to the workload. The levers are in our RDS cost optimization guide; the NoSQL alternative is priced in our DynamoDB pricing guide; and the whole AWS rate card is in our guide to understanding AWS pricing.
FAQs
On seven meters: instance hours by class and engine (a db.t3.micro is $0.017 an hour and a db.r6g.large $0.24 for MySQL or PostgreSQL in US East), storage per GB-month (gp3 $0.115, io2 $0.125 plus $0.10 per provisioned IOPS), backup storage beyond an amount equal to allocated storage ($0.095 per GB-month), data transfer, licence-included rates for Oracle and SQL Server, the Extended Support surcharge on old engine versions ($0.20 per vCPU-hour), and, for Aurora, I/O requests or the I/O-Optimized premium. Multi-AZ roughly doubles instance and storage.
A db.t3.micro running MySQL or PostgreSQL in US East with 20 GB of gp3 storage costs about $15 a month Single-AZ: about $12.40 for the instance and $2.30 for storage. Enable Multi-AZ and it is about $27. A production db.m6g.large with 100 GB and a standby runs about $255 before backups and transfer. Legacy free-tier accounts get 750 hours of db.t3.micro and 20 GB for 12 months; newer accounts receive credits instead.
Because it provisions a synchronous standby in a second Availability Zone with its own copy of the storage, and AWS bills the standby as a full instance plus its storage. A Multi-AZ cluster with two readable standbys triples both lines. The standby buys automatic failover in about a minute, which production usually needs and development, staging, and reporting usually do not.
The Extended Support surcharge of $0.20 per vCPU-hour that switches on automatically when an engine major version passes its standard support date (about $292 a month on a 2-vCPU instance); snapshots that outlive their instances at $0.095 per GB-month; read replicas created for migrations and never removed; cross-AZ transfer at $0.01 per GB each way when the application tier is in another zone; provisioned IOPS on io1 and io2 billed whether used or not; and Aurora I/O charges on write-heavy workloads.
A one- or three-year commitment to an instance class and engine in a region, with no-upfront, partial, or all-upfront payment, discounting instance hours by roughly 29 to 69 percent: about 30 to 35 percent for one year no upfront, up to 63 to 69 percent for three years all upfront. Reservations cover instance hours only; storage, IOPS, backup, and transfer stay at on-demand rates. AWS also offers Database Savings Plans with more flexibility.
Aurora bills instance hours (db.r6g.large about $0.26 an hour), storage at $0.10 per GB-month on actual usage rather than provisioned size, and either I/O requests at $0.20 per million (Aurora Standard) or a higher instance and storage rate with no I/O charge (Aurora I/O-Optimized), which pays off when I/O exceeds about a quarter of the bill. Aurora Serverless v2 bills per ACU-hour (about $0.12) and can scale to zero. Replicas share the storage volume, so a reader costs its instance hours only.
Price the instance class times hours times the number of nodes (Multi-AZ doubles, each replica adds one), add storage per copy, add provisioned IOPS if any, add backup beyond the free allowance, add Extended Support if the engine version has aged out, and add transfer. Opslyft's RDS cost calculator prices a configuration this way, including the standby, the replicas, and the surcharge that estimates usually omit.