If your PostgreSQL database fits on one instance and you do not read from a replica, pick RDS Multi-AZ: on AWS Paris prices published on November 12, 2026, it comes out 18 to 28% cheaper than Aurora on the two workloads priced below.
Aurora pays off when you use what it brings: readable replicas, failover in under a minute, and storage that grows on its own.
This article redoes the math line by line, gives the point where Aurora I/O-Optimized beats Aurora Standard, then lists the criteria that should make you switch. Prices are in US dollars, in November 2026, for PostgreSQL in the eu-west-3 Region, on demand, with no reservation.
01What you pay for in each case
The two services do not bill the same things, and that is where most comparisons go wrong.
With RDS, you pay for an instance by the hour, and in Multi-AZ the standby is built into the hourly rate. You pay for provisioned storage whether you use it or not. On gp3, the RDS documentation gives a baseline of 3,000 IOPS and 125 MiB/s up to 399 GiB for PostgreSQL, then 12,000 IOPS and 500 MiB/s from 400 GiB, at no extra charge. I/O is not billed as long as you stay within that baseline.
With Aurora, you pay for every instance by the hour, writer and reader alike. Storage is billed per GB actually consumed, and the Aurora pricing page states that data is made durable across three Availability Zones while you pay for a single copy. With the Standard configuration, you also pay per million I/O requests. With I/O-Optimized, I/O is free, but instances and storage cost more.
Here are the rates from the official AWS price file for Paris, published on November 12, 2026:
One point matters when comparing at equal availability. For RDS Multi-AZ, the documentation is explicit: the standby cannot serve read traffic. On Aurora, high availability comes from a replica, which does serve reads. Without a replica, per the Aurora documentation, the primary instance is recreated after a failure, which typically takes less than 10 minutes. So the math below compares RDS Multi-AZ with an Aurora cluster made of one writer and one reader.
02The math on two workloads
Replace the assumptions with your own numbers. A month is 730 hours, or 2,628,000 seconds. The I/O volume is a monthly average.
Workload A: a modest SaaS
A 100 GB database, an average of 100 I/O per second, a db.r6g.large instance, high availability required. Rates from the AWS Paris price file, November 2026.
RDS wins by $93 a month over Aurora Standard. 100 I/O per second sits well within the gp3 baseline of 3,000 IOPS, so nothing to add on the RDS side.
Workload B: an I/O-heavy database
A 500 GB database, an average of 2,000 I/O per second, a db.r6g.xlarge instance, high availability required. Same AWS Paris rates, November 2026.
Here, Aurora Standard costs more than twice as much as RDS: I/O outweighs the instances. I/O-Optimized brings the gap down to about $355 a month. On the RDS side, at 500 GB the gp3 volume moves to a 12,000 IOPS baseline according to the RDS documentation, so an average of 2,000 I/O per second stays covered.
Aurora does not count its I/O the way gp3 counts IOPS. Before deciding, pull the VolumeReadIOPs and VolumeWriteIOPs metrics of your cluster from CloudWatch, or run a one-week test on a copy, and redo the math with those numbers.
What these numbers leave out: on Aurora, the reader serves reads. If your application needs a read replica anyway, the fair comparison on the RDS side is Multi-AZ plus a read replica, so one more instance, and the gap narrows sharply.
03Standard or I/O-Optimized: the break-even point
AWS gives a rule of thumb on its Aurora pricing page: I/O-Optimized becomes worthwhile when I/O exceeds 25% of your total Aurora spend. You can compute the exact threshold for your cluster.
Above the threshold, I/O-Optimized is cheaper. Below it, stay on Standard. Workload B, at 2,000 I/O per second, is far above its threshold, hence a gap of more than $800 between the two configurations on AWS Paris rates. The safest path is to start on Standard, measure one real month, then compare against the threshold.
04The criteria that decide it
Price is only one criterion. These are the ones that should tip you one way or the other.
On failover, the documented numbers are clear: typically 60 to 120 seconds for RDS Multi-AZ, under 60 seconds and often under 30 for Aurora with a replica. Aurora supports up to 15 replicas.
On volume, RDS for PostgreSQL goes up to 64 TiB according to the RDS documentation, and Aurora scales automatically up to 256 TiB according to the Aurora pricing page. Few SaaS products get near those limits, but Aurora storage also spares you from oversizing a volume to plan for growth.
For a development database that sleeps, Aurora Serverless v2 costs $0.14 per ACU-hour in Paris in November 2026 according to the AWS price file, and the Aurora pricing page lists a starting capacity of 0 ACU. Example: 2 ACUs for 10 hours a day, 22 days a month, is 440 ACU-hours, or $61.60, versus $52.56 for an RDS db.t4g.medium Single-AZ instance running all month at $0.072 per hour. The saving is not always where you expect it: a small RDS instance running around the clock stays competitive.
Once the engine is chosen, the biggest saving comes from commitment. AWS lists on its Aurora pricing page discounts of up to 45% for a one-year Reserved Instance and 66% for three years, and both services are eligible for Database Savings Plans. To place the database within the rest of your bill, the pillar article Cut an AWS bill in half sets the order of work, and Why is my AWS bill so high? Where to look first helps you find the lines that drift.
05FAQ
Is Aurora faster than RDS for PostgreSQL?
Not necessarily for your workload. Aurora mainly changes the storage layer, replication and failover. For a database that fits in memory with few writes, the difference can be small. Test your own workload on a copy before paying the premium for performance.
Can you move from RDS to Aurora without rebuilding everything?
Yes, AWS documents a migration from RDS for PostgreSQL to Aurora PostgreSQL through an Aurora read replica, and the application keeps its PostgreSQL driver. You still need to test the cutover, the extensions you use and the new I/O bill. Before any migration, make sure your backups actually work.
Do you need an Aurora replica for high availability?
For fast failover, yes. Data is copied across several zones in every case, but without a replica the primary instance is recreated after a failure, which typically takes less than 10 minutes according to the Aurora documentation. With a replica, recovery typically takes less than 60 seconds.
06Sources
Start with RDS Multi-AZ while your database fits on one instance and you do not read from a replica. Move to Aurora when sub-minute failover, readable replicas or data volume require it, and pick Standard or I/O-Optimized from your real I/O metrics.
Talk about your project ↗