Aurora Serverless v2 Pricing and Aurora Serverless Pricing per ACU vs Provisioned

Aurora Serverless v2 costs $0.12 per ACU-hour in US East on Aurora Standard and $0.16 on I/O-Optimized, billed per second, where one ACU is about 2 GiB of memory with matching CPU. Storage is extra at $0.10 per GB-month plus $0.20 per million I/O requests on Standard. Per GiB of memory that is about 3.5 times the price of a provisioned db.r7g.large, so serverless only saves money when average load is low or the cluster can pause.

The figures on this page come from AWS's public price file for Amazon RDS in US East (N. Virginia), read in October 2026, and from the Aurora user guide. AWS now calls the product simply Aurora Serverless, because version 1 is retired, but the billing line is still named ServerlessV2Usage and most people still search for v2. Below is every rate that shows up on the bill, what common workloads actually cost per month, the break-even against provisioned instances, and the settings that quietly keep a cluster billing when it should be asleep.

Aurora Serverless v2 pricing in US East

Capacity runs from 0 to 256 ACUs on current engine versions and moves in steps as small as 0.5 ACU. Aurora measures capacity every second, so you pay for the ACUs each writer and reader actually held, not for the maximum you configured. The maximum matters for one thing only: it caps how far a runaway query can push the bill.

How much does Aurora Serverless v2 cost per month?

A single instance that never drops below 0.5 ACU costs $43.80 a month in compute, because 0.5 times $0.12 times 730 hours is $43.80. A floor of 2 ACUs costs $175.20 before it has served a query. Everything above that is usage. These are four shapes we see often, priced on Aurora Standard with compute only:

The pattern is plain once it is in a table. Serverless wins clearly when the database sleeps, roughly ties when it idles at a low floor most of the day, and loses by 2.6 times on a busy, steady workload. The last row is the one teams miss. Readers in promotion tiers 0 and 1 track the writer's capacity so they can take over in a failover, which means a highly available serverless cluster pays the ACU rate twice.

Is Aurora Serverless v2 cheaper than provisioned?

Only when average capacity stays below the break-even point for the instance you would otherwise run. A db.r7g.large has 16 GiB of memory, which is about 8 ACUs, and costs $0.276 an hour on demand. Divide that by $0.12 and the break-even is 2.3 ACUs. If the cluster averages less than 2.3 ACUs across the month, serverless is cheaper than an r7g.large; above that, provisioned wins.

The ratio holds across sizes: serverless pays about $0.06 per GiB-hour against roughly $0.017 on Graviton r7g, so it needs average utilization under about 29% of the equivalent instance to come out ahead. The comparison gets worse for serverless once commitments enter. A 3-year all-upfront Reserved Instance on db.r7g.large costs $3,850, which works out to about $0.147 an hour, and against that the break-even falls to about 1.2 ACUs at on-demand serverless rates.

Does Aurora Serverless v2 scale to zero?

Yes, on Aurora PostgreSQL 13.15, 14.12, 15.7 and 16.3 or higher and on Aurora MySQL 3.08.0 or higher. Set the minimum capacity to 0 ACUs and each serverless instance pauses after a period with no user connections, between 5 minutes and 24 hours, which you choose. While paused, the instance charge is zero and you pay only for storage and backups. Resuming takes about 15 seconds, or 30 seconds and more after a pause longer than a day, so AWS recommends client connection timeouts above those figures.

That makes auto-pause the single biggest saving available for development, test, demo and internal databases. A fleet of twenty dev clusters sitting at a 0.5 ACU floor costs $876 a month in compute for doing nothing. With auto-pause, and each one busy at 1 ACU for two hours every working day, the same fleet costs about $106.

Five settings that stop Aurora Serverless from pausing

The AWS guide lists the conditions that silently prevent a pause. None of them throws an error, so the only symptom is a bill that looks like the floor never went to zero:

• An RDS Proxy in front of the cluster. The proxy keeps a connection open to every instance, so nothing pauses, and the proxy itself bills at $0.015 per ACU-hour with an 8 ACU minimum, or $87.60 a month. On a small dev cluster the proxy can cost more than the database.
• Logical replication or binlog replication. The writer and tier 0 and 1 readers stay awake to keep the replication connection healthy.
• A zero-ETL integration to Redshift. Same effect on the writer and tier 0 and 1 readers, and the warehouse side has its own bill, compared in our guide to Redshift cost optimization tools.
• Being part of an Aurora global database. The primary writer never pauses and secondary clusters never pause.
• Any provisioned instance in the cluster. In a mixed cluster the serverless writer stays active.

The sixth cause is not a setting at all. An application health check, a connection pool with a long idle timeout, or an uptime monitor pointed at an endpoint that opens a database session will keep the instance awake around the clock. Point health checks at an endpoint that does not touch the database. Also note that auto-pause skips scheduled jobs: pg_cron and the MySQL event scheduler do not wake a paused instance, and jobs scheduled during the pause are dropped.

Aurora Standard or I/O-Optimized on serverless

I/O-Optimized raises the ACU rate by a third, from $0.12 to $0.16, and storage from $0.10 to $0.225 per GB-month, and in exchange charges nothing for I/O requests. AWS's own guidance is to switch when I/O is more than about 25% of the Aurora bill. Take a cluster averaging 4 ACUs with 500 GB of data and 2 billion I/O requests a month:

Here I/O is half the Standard bill and I/O-Optimized saves about $220 a month. At 200 million requests the same cluster would be cheaper on Standard by roughly $140. AWS lets you move a cluster to I/O-Optimized once every 30 days and back to Standard at any time, so the decision is cheap to test, but read the I/O line on your bill first rather than guessing.

Discounts for Aurora Serverless v2

There are no Reserved Instances for serverless capacity. Since December 2025 the only commitment discount is a Database Savings Plan, which takes 35% off serverless ACU-hours, bringing Standard to $0.078 and I/O-Optimized to $0.104 per ACU-hour on a 1-year, no-upfront term. Provisioned Generation 7 instances get 20% from the same plan. With both sides on the plan, the break-even against a db.r7g.large rises to about 2.8 ACUs, because serverless gets the deeper cut. Our breakdown of Database Savings Plans pricing works through the commitment math and the 80% utilization line where a plan stops paying.

One more charge to know about: clusters on an engine version past the end of standard support pay Extended Support on every ACU-hour, $0.085 in years one and two and $0.17 in year three. On a $0.12 base that is a 71% and then 142% surcharge, far steeper in relative terms than on provisioned instances. Our guide to RDS Extended Support pricing lists the dates per version.

When to move a cluster from serverless to provisioned

Pull the ServerlessDatabaseCapacity metric with the Average statistic over 30 days for each instance. If the average sits above about 29% of the maximum capacity you would size a provisioned instance at, and the cluster cannot pause, a provisioned writer on r7g or r8g is cheaper, and a Reserved Instance or Savings Plan makes it cheaper again. Mixed clusters are a sensible middle path: a provisioned writer for the steady base and serverless readers in promotion tiers 2 to 15, which scale and pause independently for reporting bursts.

Doing that by hand for one cluster takes an hour. Doing it for forty clusters across accounts, every month, is the reason teams buy a tool. AWS Compute Optimizer gives instance-level recommendations for Aurora, but turning them into a serverless or provisioned decision per cluster is still manual work. The paid options that resize databases are compared with their prices in our guide to RDS rightsizing tools.

Seeing what Aurora Serverless costs per team

The Aurora line in Cost Explorer is one number. What finance needs is which cluster, which team and which month moved, and whether a change paid off. CostAnalyst reads the AWS cost and usage export you upload and breaks Aurora spend down by usage type, so ServerlessV2Usage, IO-Optimized usage, storage, I/O, RDS Proxy and Extended Support each sit on their own line, allocated to teams by your tags next to the rest of your AWS, Azure, Google Cloud and SaaS spend. It flags month-over-month jumps before the invoice closes and never needs access to your AWS account. Plans start at $59 a month on yearly billing, and you can run your own export through the demo first.

Frequently asked questions

What is an ACU in Aurora Serverless?

An Aurora capacity unit is about 2 GiB of memory plus the CPU and networking that go with it. Capacity is measured every second as a decimal number of ACUs and billed per ACU-hour, $0.12 on Aurora Standard and $0.16 on I/O-Optimized in US East. A cluster set to 0.5 to 16 ACUs can use between about 1 GiB and 32 GiB of memory.

Is Aurora Serverless v2 expensive?

It is expensive for steady workloads and cheap for intermittent ones. Per GiB of memory it costs about 3.5 times a provisioned db.r7g.large, so a database that runs busy around the clock costs more than double on serverless. A dev or internal database that can pause most of the day often costs a quarter of an always-on instance.

What is the minimum cost of Aurora Serverless v2?

With auto-pause enabled the minimum compute cost is zero while paused, and you pay only for storage at $0.10 per GB-month plus backups beyond the free allowance. Without auto-pause the lowest floor is 0.5 ACU, which costs $43.80 a month per instance. An RDS Proxy adds at least $87.60 a month and prevents pausing.

Do Aurora Serverless readers cost extra?

Yes. Each reader is billed for its own ACU-hours. Readers in promotion tiers 0 and 1 scale with the writer, so they cost as much as the writer and pause and resume together with it. Readers in tiers 2 to 15 scale and pause independently, which suits reporting and read bursts.

Is Aurora Serverless v1 still available?

No. AWS ended support for Aurora Serverless v1 on 31 March 2025, and new serverless clusters run on what used to be called v2. The price file still lists the old $0.06 per ACU-hour v1 rate, which is why some articles quote it, but new clusters are billed at the v2 rates above.