AWS is not inherently more expensive than other major clouds, and list prices for comparable services are usually close. Bills often feel high for a few common reasons: data transfer charges, paying on-demand rates for steady workloads, idle or oversized resources, premiums for managed services, and weak cost tracking. Most of these depend on how the system is built, not on the platform alone.
List prices vs bills
A list price is what one unit of a service costs. A bill is the sum of every unit you use, including the ones you forgot about. Two clouds with similar list prices can produce very different bills for the same workload, depending on how many resources are running and how much data moves between them. When people say AWS is expensive, they are often describing their own architecture.
Data transfer is often the biggest surprise
Data transfer is the cost that surprises people most:
- Inbound data is free. Outbound data to the internet is billed per gigabyte after the first 100 GB each month. In many regions, the first tier is roughly $0.09 per GB.
- Traffic between Availability Zones is billed in both directions, at about $0.01 per GB each way.
- NAT gateways charge an hourly fee plus a per-GB fee for data they process. Private subnets that send a lot of traffic through a NAT gateway can get expensive quickly.
- Public IPv4 addresses have carried an hourly charge since February 2024, including addresses that are in use.
Google Cloud also charges for outbound traffic, at roughly $0.08 to $0.12 per GB depending on volume and region. Egress is a cost to check on any platform.
On-demand rates vs commitments
On-demand pricing is the most flexible option, and it costs the most per unit. For steady workloads, commitments can reduce the cost substantially. Savings Plans and Reserved Instances trade a one- or three-year commitment for lower rates. Spot Instances can cut costs for workloads that tolerate interruption. Before you commit, rightsize: a smaller instance that does the same job costs less whether or not you commit.
Idle and oversized resources
Most bills contain resources nobody is using:
- Forgotten test instances and old snapshots
- Unattached EBS volumes
- Idle load balancers
- Instances sized for peak load that run all day
Stopping an EC2 instance does not stop its EBS volumes or Elastic IP addresses from billing. AWS Compute Optimizer uses usage data to suggest smaller instance sizes.
Managed services carry a premium
Managed services such as RDS and EKS handle patching, backups, and high availability for you, and you pay for that work. The EKS control plane alone costs about $0.10 per hour per cluster. For a simple app, that overhead can cost more than a basic server you run yourself. For a team that would otherwise need to staff operations, it can be cheaper overall.
Logs, monitoring, and support add up
CloudWatch Logs charges roughly $0.50 per GB ingested in many regions, and long retention adds storage costs. Paid support plans carry monthly charges with a minimum amount. These costs are small per unit but grow with the size of the system.
Tools to track and reduce spend
AWS provides several tools for finding waste:
- Cost Explorer shows where spending goes over time.
- AWS Budgets sends alerts. Budget actions can enforce limits, but only if you configure them.
- Cost Anomaly Detection flags unusual spending.
- Trusted Advisor and Compute Optimizer suggest changes.
- Cost allocation tags show which project or team spends what.
Common misconceptions
- “You pay only for what you use.” In practice you pay for what you provision, whether or not you use it.
- “Stopping an instance stops its costs.” Storage and Elastic IP addresses keep billing.
- “Compute is the main cost.” Data transfer is often the hidden driver.
- “AWS costs more than other clouds.” Bills differ by architecture, and list prices are similar.
How this compares with Google Cloud
The same cost drivers apply on Google Cloud: outbound traffic, idle resources, commitments, and managed-service premiums. The platforms differ in some defaults. App Engine Standard can scale to zero and includes a daily free quota. An Elastic Beanstalk environment keeps at least one EC2 instance running, so you pay for that instance even when the app is idle. The pricing guide compares these models in more detail, and the Elastic Beanstalk comparison covers the AWS side of the choice.
Next steps
For the Google Cloud side of the cost question, read whether the Google Cloud Platform costs money. For a sense of what a single VM costs and what keeps billing after it stops, see the Compute Engine explainer. Prices change, so check the current AWS pricing pages before you budget.