Date:
Category:
DevOps Strategy
AWS Egress Costs: What You Pay and How to Cut It

Written by
Dino Starcic
·
CEO

AWS does not charge you to move data in, but it does charge for most data that leaves: to the internet, to another Region, and in some cases between Availability Zones. The fastest way to lower the bill is not to guess. Read the usage types on your bill, map each one to a network path, and apply the fix that matches (VPC endpoints, caching, a CDN, keeping traffic in one zone). The work is small per fix but it needs someone who knows your architecture, which is why many teams hand it to a senior engineer through a backlog.
CloudwiseReduce AWS Egress Costs with Expert SupportCloudwise provides senior AWS and DevOps engineers for infrastructure management and cost optimization through a shared backlog.Visit
TL;DR
Data into AWS is generally free. Data out to the internet, across Regions and across Availability Zones is where charges appear, and AWS says the first 100 GB per month of internet transfer out is free across services and Regions (China and GovCloud excluded).
Internet egress is tiered. In US East the list rate starts at $0.09/GB after the free allowance, and the rate per GB falls as volume grows.
Your bill names a usage type, not an application. Map each line to a network path first, then fix that path.
The usual savings come from gateway endpoints, CloudFront, compression and keeping chatty services in one zone. Measure a full billing month after each change.
If nobody on your team has time to trace and fix this, a subscription with senior AWS engineers turns it into a ticket in a shared backlog.
Table of Contents
What “egress” means on an AWS bill
What internet egress costs, with a worked example
The charges teams miss
Step by step: find what drives your data transfer bill
Fixes that lower AWS data transfer costs
Mistakes that move the bill instead of cutting it
Why a managed engineer is the practical route
How the subscription works for this kind of task
FAQ
Sources
What “egress” means on an AWS bill
Egress is data leaving AWS. On the invoice it usually appears as Data Transfer Out. AWS’s global network FAQ describes three broad categories of charge:
Data transfer out (DTO): from AWS to the internet.
Data transfer between Regions (DTIR): for example, from US East (N. Virginia) to Europe (Ireland).
Data transfer between Availability Zones (DTAZ): between data centers in the same Region.
For DTO and DTIR, AWS charges based on the source Region only, not the destination. Per the same FAQ, rate tiers for outbound traffic aggregate usage across services such as EC2, S3, RDS, DynamoDB and Elastic Load Balancing. In its pricing whitepaper, AWS also states that in most cases inbound transfer and transfer between services in the same Region are not charged, and that the more you transfer, the less you pay per GB.
That last point matters when you read a bill. Three separate services can each send small amounts out, yet together they push you into one tier.
What internet egress costs, with a worked example
In US and European Regions, the published tiers for internet egress are 100 GB free per month, then $0.09/GB for the first 10 TB, $0.085/GB for the next 40 TB, $0.07/GB for the next 100 TB and $0.05/GB above that. Some Regions cost more: Asia Pacific (Singapore) starts at $0.12/GB. Rates change, so confirm the figure for your Region on the AWS pricing pages before you plan around it.
Here is a worked example using the US East rate. A product serves 5 TB of downloads and API responses a month straight from EC2 and S3:
5 TB is 5,000 GB.
Subtract the 100 GB free allowance: 4,900 GB billable.
4,900 GB × $0.09 = $441 per month.
That is one line. NAT gateways, cross-zone traffic and cross-Region replication come on top, and they are the lines that surprise teams, because they are not caused by customers downloading anything. They come from your own services talking to each other.
To estimate a new architecture before you build it, the AWS FAQ points to the AWS Pricing Calculator: search for data transfer, pick the Region and enter the monthly volume for each flow.

The charges teams miss
Internet egress is the obvious one. These are the ones that tend to hide.
Cross-Availability Zone traffic
Traffic between zones in the same Region is billed as regional data transfer. Services spread across zones for resilience, such as a Kubernetes cluster where pods call each other across zones, can generate a steady stream of it. AWS documents this usage as Region-DataTransfer-Regional-Bytes, with a Region prefix such as USE2 for US East (Ohio). Usage types with no prefix belong to US East (N. Virginia).
NAT gateway processing
The AWS Architecture Blog explains that reaching same-Region AWS services through an internet gateway is free, while the same access through a NAT gateway carries a per-GB data processing charge. A private subnet pulling large volumes from S3 through a NAT gateway is one of the most common avoidable costs.
Cross-Region transfer
Replication, backups and analytics pipelines that copy data between Regions show up as inter-Region usage, documented by AWS as Source-Destination-AWS-Out-Bytes. Each flow is charged on the source Region, so the same replication job can be cheap in one direction and pricier in the other.
Direct Connect, CloudFront and S3 acceleration
Other paths have their own usage type names in the AWS data transfer billing reference: Direct Connect traffic ends in DataXfer-In or DataXfer-Out, and CloudFront traffic uses Region-DataTransfer-Out-Bytes under the CloudFront product code. Knowing these names lets you read the bill line by line.
Step by step: find what drives your data transfer bill
The invoice tells you what type of traffic you paid for, not which workload caused it. Work from the bill toward the application.
Step 1: Turn on the right visibility
AWS re:Post recommends activating Cost Explorer, setting AWS Budgets alerts, using cost allocation tags, and watching CloudWatch bandwidth metrics for internet and NAT gateways. For containers, turn on Container Insights and look at traffic between zones.
Where it goes wrong: tags that are applied inconsistently. Untagged traffic lands in one big bucket and you cannot tell which team or service owns it.
Step 2: Group the cost by usage type
In Cost Explorer, look at the last three closed months, filter to data transfer, and group by usage type, then by linked account and Region. Anything containing “Bytes” is a transfer line. The one with the largest dollar value is your starting point.
Where it goes wrong: stopping at the service name. “EC2-Other” can hide NAT processing, cross-zone bytes and internet egress in a single line, and each needs a different fix.
Step 3: Map the usage type to a path
Use the Cost and Usage Report for detail. A regional-bytes line means traffic between zones. A line ending in Out-Bytes with no destination Region means internet egress. A NAT gateway line points to a private subnet reaching out. Once the path is clear, VPC Flow Logs show which instances or services are sending the bytes.
This is the step that takes real AWS knowledge. Flow logs are large, and reading them well means knowing your VPC layout, your endpoints and your routing.
Step 4: Match the fix to the path
The next section gives the fix for each path. Record the old path, the usage types charged and the expected new path before you change anything.
Step 5: Measure for a full billing month
Compare the same usage types before and after. Lower NAT bytes with higher endpoint hours or higher regional bytes is a moved cost, not a saving.
Fixes that lower AWS data transfer costs
Match the fix to the charge you found.
Charge you found | Fix | Why it works |
|---|---|---|
NAT gateway bytes to S3 or DynamoDB | Gateway VPC endpoints | Traffic stays on the AWS network and avoids NAT processing |
NAT bytes to other AWS services | Interface endpoints (PrivateLink) | Private path to the service, worth it above a break-even volume |
Cross-AZ regional bytes | Zone-aware routing, same-zone placement, a NAT gateway per zone | Keeps traffic inside one zone |
Internet egress of cacheable content | CloudFront with caching | Origin fetches from S3 to CloudFront are free |
Large text or JSON responses | gzip or brotli compression | Fewer bytes leave per request |
Large query results moving between systems | Export results to S3 instead of moving whole datasets | Moves less data |
The first four rows come from AWS’s own data transfer optimization guidance, which recommends VPC endpoints for S3 and DynamoDB, PrivateLink for private connections, and exporting query results to S3 rather than moving whole datasets. Industry estimates for savings vary widely by workload, for example 15 to 25 percent from gateway endpoints and 20 to 35 percent from compression, so treat any percentage as a hypothesis to test against your own bill.
Gateway endpoints first
If your private subnets reach S3 or DynamoDB through a NAT gateway, a gateway endpoint is usually the first change to make. It is a route table entry, it is quick to deploy, and the Architecture Blog notes that same-Region access to these services does not need to pay NAT processing.
CloudFront for public content
Serving downloads, images and static assets through CloudFront means S3 sends bytes to CloudFront at no transfer charge. AWS includes 1 TB per month of free CloudFront data transfer out, per the global network FAQ. The result depends on how cacheable your content is. Personalized or dynamic responses get less benefit, so check your cache hit ratio after the change.
Keep chatty services together
Microservices that call each other thousands of times a second across zones are billed at both ends of every hop. Placing tightly coupled services in the same zone, or enabling topology-aware routing in Kubernetes, cuts the regional bytes line. Keep resilience in mind: spread stateless replicas, and keep high-volume pairs together.
Moving data off AWS
If you are migrating away from AWS or off a single service, AWS says in the FAQ that DTO for moving all your data out is free. Ask AWS support to apply it before you start the transfer.
Mistakes that move the bill instead of cutting it
Adding interface endpoints for a trickle of traffic. Each endpoint has an hourly charge per zone. At low volume you pay more than you saved.
Counting inter-Region traffic twice. Inbound at the destination shows as zero cost, so adding both sides doubles the volume in your spreadsheet.
Treating CloudFront as automatic. Cache rules, headers and the share of dynamic content decide the outcome.
Optimizing before measuring. A fix aimed at the wrong usage type costs engineering time and changes nothing.
Why a managed engineer is the practical route
None of these steps is exotic. The difficulty is that each one touches networking, routing, containers and billing at once, and each needs a verified before and after. For a startup with one or two engineers, that is a week of work pulled away from the product. For a larger team, it is a ticket that sits in a backlog for months.
This is the kind of task a subscription handles well. You describe the bill problem once, the engineer traces it, makes the change and checks the next month’s usage types. Our Cost Optimization and FinOps service covers cost work like this, and AWS infrastructure management covers the network and VPC changes behind it. If you want a quick read on where your account stands first, the AWS Optimisation Score is a starting point. For the wider picture of waste beyond data transfer, see the hidden cost of poor AWS cost optimization.

How the subscription works for this kind of task
Cloudwise gives you senior AWS and DevOps engineers under a single subscription. The Simplest Subscription is $7,499 per month, with no hourly billing and no long-term contracts, and you can cancel anytime. Tasks go into a shared backlog that you prioritize, with a typical 24 to 48 hour response time.
For egress, a backlog might look like this:
Review the last three months of data transfer usage types.
Add gateway endpoints for S3 and DynamoDB in private subnets.
Put CloudFront in front of public assets.
Set up a Budgets alert on data transfer.
Review the next bill against the baseline.

The same backlog also takes the rest of your day-to-day cloud work, from monitoring and observability to Kubernetes platforms. To see whether the model suits your team, read how an AWS and DevOps monthly subscription works or see the full list of services.
Sources
Recommended
Questions Answered
Frequently asked questions
Clear answers to common questions about our AWS & DevOps subscription and how it works.