AWS Compute Optimizer Pricing and What Enhanced Infrastructure Metrics Cost

AWS Compute Optimizer is free at its default settings. Recommendations for EC2 instances, Auto Scaling groups, EBS volumes, Lambda, Fargate, RDS and a dozen other resource types, built from 14 or 32 days of CloudWatch metrics, carry no charge beyond the resources and CloudWatch monitoring you already pay for. The one paid feature is enhanced infrastructure metrics, which extends the EC2, Auto Scaling group and RDS lookback to three months and costs $0.0003360215 per resource per hour, about $0.25 per instance in a 31-day month.

So the real pricing question is not whether you can afford Compute Optimizer. It is whether the paid lookback is worth switching on, what the memory data it needs will cost you in CloudWatch, and what you would pay instead for a tool that applies the changes Compute Optimizer only suggests. The figures below come from AWS's own pricing page, FAQ and user guide, read on 2 October 2026.

How much does AWS Compute Optimizer cost?

The base service costs nothing. AWS's pricing page says it plainly: "There is no additional charge for AWS Compute Optimizer. You pay only for the AWS resources needed to run your applications and Amazon CloudWatch monitoring fees." The charges that can appear on your bill in connection with it are these.

Two billing rules sit under the enhanced metrics price. You are charged only for the hours each resource actually runs, so a stopped instance costs nothing that month. And Auto Scaling groups are the exception to "per resource": they are charged per running EC2 instance inside the group, so a group that cycled through five instances in a month is billed for five instances' running hours, not for one group.

Is AWS Compute Optimizer free?

Yes, for everything except the 93-day lookback. You opt in from the console, ideally from the management account so every member account in the organization is covered, and you get findings for EC2 instances, Auto Scaling groups, EBS volumes, Lambda functions, ECS services on Fargate, Aurora and RDS, commercial software licenses, NAT Gateways, ElastiCache, DynamoDB, DocumentDB, MemoryDB, WorkSpaces and SageMaker endpoints. The 32-day lookback, idle detection, Graviton migration effort estimates and the savings estimation mode that accounts for your Savings Plans and Reserved Instances are all part of the free service.

What enhanced infrastructure metrics cost at real fleet sizes

The hourly rate looks like a rounding error, and for most fleets it is. Here is the arithmetic at 744 hours a month (31 days) and 8,760 hours a year, for instances that run all the time.

One oddity on AWS's own pricing page is worth knowing if you check your first bill against it. In the Auto Scaling group example, the text says three instances ran for 31 days and two for half the month, but the table beneath it shows two rows of two instances at 372 hours each. The $1.00 total is correct for the scenario described (2,976 billable hours), the rows are not. Bill yourself on running hours per instance and you will match the invoice.

Is enhanced infrastructure metrics worth paying for?

Usually yes for production EC2 and RDS that has monthly or quarterly peaks, and usually no for everything else. The case for it is a single avoided mistake. Fourteen days of data taken in a quiet fortnight will happily recommend shrinking the instance that runs month-end close or quarterly billing, and the rollback costs more engineering time than the feature costs in a decade. The case against is that for a steady web tier, 32 free days already capture the pattern.

The break-even is easy to work out. An m5.xlarge in US East (N. Virginia) costs $0.192 an hour on demand, about $140 a month. Downsizing one to an m5.large saves about $70 a month. That single change pays for the enhanced metrics on 280 instances for the same month. If the 93-day view produces even a handful of downsizes that the 14-day view would have rejected, or stops one bad one, it has paid for itself. Turn it on per account or per tag for the workloads with business-calendar peaks rather than for the whole organization.

The hidden cost is memory, not Compute Optimizer

Default EC2 metrics in CloudWatch cover CPU, network and local disk. They do not include memory, and memory is what breaks after an aggressive downsize. AWS's FAQ says that when memory data is missing, Compute Optimizer "will attempt to avoid making a recommendation that downsizes that dimension". That keeps you safe and leaves savings behind on memory-light workloads, which is most of them.

You have two ways to give it memory data:

• The CloudWatch agent. Publish memory utilization to the CWAgent namespace and Compute Optimizer reads it automatically. Each unique metric is billed as a custom metric, $0.30 a month for the first 10,000 metrics, so one memory metric per instance on 200 instances is about $60 a month, $720 a year. That is more than the enhanced metrics on the same fleet. Publish only the memory metric you need, with as few dimensions as possible, because every extra dimension combination is a separately billed metric.
• External metrics ingestion. If you already pay for Datadog, Dynatrace, Instana or New Relic, Compute Optimizer can ingest EC2 memory utilization from that product instead, with no new CloudWatch metrics to pay for.

Put plainly: for a 200-instance fleet, a fully memory-aware Compute Optimizer setup with the 93-day lookback costs around $1,300 a year in AWS charges. That is the honest price of "free" rightsizing done properly.

What Compute Optimizer does not do, whatever you pay

The limits matter more to a buying decision than the price, because each one is a reason teams end up paying for a second tool.

• It never resizes an instance. The Automation feature applies exactly two kinds of action: snapshot and delete EBS volumes that have been unattached for at least 32 days, and upgrade EBS volumes to a newer type such as io1 to io2. Instance type changes stay manual, and on an EBS-backed instance a type change means stopping the instance.
• No Spot Instances. Spot capacity gets no rightsizing recommendations.
• Container node groups get idle findings only. Auto Scaling groups running Amazon ECS or EKS workloads are not rightsized, because the waste there sits in pod and task requests. Our EKS rightsizing tools comparison covers that layer.
• A minimum of 30 hours of metrics. Short-lived instances never get a recommendation.
• Savings shown are estimates, not invoice lines. With savings estimation mode on, Compute Optimizer prices recommendations after your Savings Plans and Reserved Instances, which is the right number to look at. It still cannot tell you whether freed commitment was absorbed by other usage or simply went unused.

Compute Optimizer vs Trusted Advisor and paid EC2 cost optimizers

If you are choosing what to pay for, this is the comparison that matters, using AWS's pricing pages and each vendor's own AWS Marketplace listing for a fleet of 200 instances, 800 vCPUs and $100,000 a month of AWS spend.

Trusted Advisor's low utilization EC2 check is the most expensive way on this list to find idle instances, because the support plan is priced on your whole AWS bill. It makes sense only if you need the support for its own sake, which we cover in AWS Trusted Advisor pricing. The paid tools earn their price in one situation: you have more recommendations than engineers willing to apply them. The full ten-tool comparison, with the catch for each, is on our EC2 cost optimizer and rightsizing tools page.

How to keep a rightsizing saving from disappearing

Most of what goes wrong after a Compute Optimizer rollout is not the recommendation. It is the step between the recommendation and the invoice.

1. Resize behind a load balancer where you can. For instances in an Auto Scaling group, change the launch template and let instance refresh replace them gradually. For single servers you deploy to yourself, put them behind a load balancer and roll the new size out with a zero-downtime deployment tool rather than stopping the only copy at 2 p.m.
2. Resize before you renew commitments, not after. A downsize on an instance covered by a Savings Plan cuts covered usage, not the hourly commitment, so the bill falls only when the commitment is resized at renewal or other usage absorbs it.
3. Measure on the bill, by account and instance family. Take the EC2 line for the affected accounts the month before and the month after. If on-demand spend fell and unused commitment rose by the same amount, nothing was saved yet.

That last step is what CostAnalyst is for. It reads the AWS billing export you upload, shows EC2 spend by account, team and instance family next to your other cloud and SaaS costs, flags idle and oversized spend and anomalies, and shows whether a resize actually changed what you pay. Plans start at $59 a month billed yearly; see CostAnalyst pricing or how our rightsizing recommendations work.

Frequently asked questions

Does AWS Compute Optimizer cost money?

Only if you enable enhanced infrastructure metrics. The standard service, with recommendations from 14 or 32 days of CloudWatch data across EC2, EBS, Lambda, Fargate, RDS and other resources, is free. Enhanced infrastructure metrics cost $0.0003360215 per resource per hour for EC2 and RDS, about $0.25 per instance a month, billed only for hours the resource runs.

How is Compute Optimizer billed for Auto Scaling groups?

Per running instance. When enhanced infrastructure metrics are active on an Auto Scaling group, AWS charges the hourly rate for every distinct EC2 instance that ran in the group during the month, for the hours each one ran. A group that replaced instances often will be billed for more instances than its desired capacity.

Is Compute Optimizer included in AWS support plans?

It does not need a support plan. Compute Optimizer is available to every AWS account, including those on Basic Support, at no charge. That differs from Trusted Advisor, whose cost optimization checks, some of which draw on Compute Optimizer data, are available only on Business Support+, Enterprise Support or Unified Operations.

Does Compute Optimizer automation cost extra?

AWS lists no separate charge for it. Automation applies recommended EBS actions, deleting volumes unattached for 32 days after taking a snapshot and upgrading volume types, on a schedule or on demand, and you can roll actions back. The snapshots it creates are billed as normal EBS snapshot storage.