AWS CloudWatch Pricing (and How to Cut It)

There is a particular irony in a CloudWatch bill. The service exists to help you watch your AWS spending, and left unmanaged, it becomes a meaningful part of that spending itself. It starts innocently: a few default metrics, one log group, a dashboard. Then a team adds custom metrics, retains logs forever, backs alarms with queries, and turns on Container Insights, and a moderately sized production environment is quietly paying five hundred to two thousand dollars a month to monitor itself.

This guide breaks AWS CloudWatch pricing down by category so you can see what actually gets expensive, then walks through the changes that cut a CloudWatch bill by 40 to 60 percent without losing the visibility you need.

How AWS CloudWatch Pricing Works

CloudWatch is not one price but a bundle of independently metered services, which is why bills grow in ways teams do not expect. Standard metrics from more than 70 AWS services flow in automatically at no cost, but everything you add on top has its own meter.

Where CloudWatch Gets Expensive

Three categories dominate almost every oversized CloudWatch bill:

• Log ingestion. At $0.50 per GB, this is the line that grows most directly with application volume. Verbose logging, DEBUG left on in production, and health-check noise multiply it across every service and every engineer debugging an incident.
• Metric cardinality and resolution. Custom metrics cost $0.30 each, and cost scales with how many unique metrics you emit and how often. Collecting at 1-minute instead of 5-minute resolution multiplies data points, storage, and alarm cost.
• Container Insights at scale. Per-node and per-vCPU charges, on top of the metrics and logs they generate, can reach roughly $200 or more per node per month, which rivals a full third-party APM tool.

How to Cut Your CloudWatch Bill

These changes routinely reduce CloudWatch spend by 40 to 60 percent with no loss of the visibility that matters.

• Set retention policies on every log group. Thirty days for most, ninety for important ones; stop the endless accumulation of stored logs.
• Filter at the source. Turn off DEBUG in production and exclude health-check and verbose access logs before they are ingested, since ingestion is the biggest cost.
• Use the Infrequent Access log class for logs you retain but rarely query in real time; it halves ingestion to $0.25 per GB and now supports full Logs Insights queries.
• Export long-term logs to S3, where storage runs roughly $4 to $24 per TB per month versus about $30 per TB in CloudWatch.
• Reduce metric cardinality and resolution, drop unused custom metrics, use 5-minute where 1-minute is not needed, and audit and delete unused dashboards and alarms.

CloudWatch is essential but should not be a top-five cost center. Attributing and governing it is the same discipline we apply to the rest of the bill in our AWS cost optimization with FinOps guide, and for application-level observability many teams pair it with a dedicated tool, as in our application monitoring tools roundup.

Turning the Meter Down

CloudWatch charges you for telemetry you choose to keep, at the resolution you choose to keep it, for as long as you choose to keep it, and most oversized bills are simply the sum of choices nobody revisited. The fix is not to monitor less; it is to monitor deliberately. Filter noisy logs before they cost anything, retain only as long as you will actually look, push cold logs to cheaper storage, and stop paying for dashboards and metrics no one reads. Do that, and the tool you use to control cloud spend stops being a line on the list of things to control.