NAT Gateway and Data Transfer: The AWS Costs Nobody Owns

NAT Gateway processing and cross-AZ traffic rarely top the bill, but they grow quietly and nobody owns them. How to find them in the CUR and the design changes that remove most of them.

NAT Gateway and Data Transfer: The AWS Costs Nobody Owns

Ask a team who owns the EC2 bill and someone will answer. Ask who owns the NAT Gateway line, or the cross-AZ data transfer line, and the room goes quiet. These charges rarely top the invoice, but they grow with every new service, every container image pull and every chatty microservice, and because they belong to the network rather than to a workload, nobody is accountable for them. In most estates we review, they are among the easiest savings to find and the quickest to fix.

Where the money goes

Four charges do most of the damage:

The common thread is that none of these is a resource a team consciously creates. They are side effects of architecture decisions made for other reasons.

Finding them in the Cost and Usage Report

Cost Explorer will show you the totals. To find out which traffic is responsible, go to the Cost and Usage Report and group by usage type. Network charges have recognisable usage type names, prefixed with a region code:

-- Network-related usage by type and account, one billing period.
-- Replace cur2 with your Data Export table name.
SELECT
  line_item_usage_account_id      AS account,
  line_item_usage_type            AS usage_type,
  ROUND(SUM(line_item_usage_amount), 1)     AS gb_or_hours,
  ROUND(SUM(line_item_unblended_cost), 2)   AS cost
FROM cur2
WHERE billing_period = '2026-09'
  AND (   line_item_usage_type LIKE '%NatGateway%'
       OR line_item_usage_type LIKE '%DataTransfer-Regional-Bytes%'
       OR line_item_usage_type LIKE '%DataTransfer-Out-Bytes%'
       OR line_item_usage_type LIKE '%PublicIPv4%')
GROUP BY 1, 2
ORDER BY cost DESC;

NatGateway-Bytes is the processing charge, NatGateway-Hours the hourly charge, and DataTransfer-Regional-Bytes is cross-AZ traffic. The account breakdown usually points straight at the two or three VPCs responsible.

The CUR tells you where, but not what. For that, turn on VPC Flow Logs for the subnets behind the expensive NAT Gateways for a few days, send them to S3, and query by destination. Flow logs have their own cost, so scope them and switch them off once you have your answer. The destinations almost always fall into a handful of buckets: S3, ECR, other AWS service APIs, a monitoring or logging vendor, and a package mirror.

Fix 1: gateway endpoints for S3 and DynamoDB

The single most common finding: workloads in private subnets reading from and writing to S3 through a NAT Gateway, paying the NAT processing fee on every byte.

Gateway VPC endpoints for S3 and DynamoDB are free. They add a route to your route tables so traffic to those services stays on the AWS network and skips the NAT Gateway entirely. There is no hourly charge and no per-GB charge.

resource "aws_vpc_endpoint" "s3" {
  vpc_id            = aws_vpc.main.id
  service_name      = "com.amazonaws.eu-central-1.s3"
  vpc_endpoint_type = "Gateway"
  route_table_ids   = aws_route_table.private[*].id
}

If your private subnets talk to S3 and you do not have this, it is very likely the cheapest saving in your account. Add an endpoint policy if you want to restrict which buckets can be reached, which also helps against data exfiltration.

Fix 2: interface endpoints where the volume justifies them

Other AWS services (ECR, CloudWatch Logs, Secrets Manager, STS, SSM and many more) are reached through interface endpoints (PrivateLink). These are not free: each endpoint has an hourly charge per AZ and a per-GB charge, but the per-GB charge is much lower than NAT processing.

So the decision is arithmetic. For high-volume services it pays off quickly; for a service you call a few times an hour, the endpoint's hourly charge across three AZs can cost more than the NAT traffic it replaces.

The usual winners:

In a multi-account estate, consider centralising interface endpoints in a shared networking VPC and resolving them from spoke VPCs through Transit Gateway and Route 53 Resolver rules. That turns many sets of endpoints into one, though it adds Transit Gateway data processing charges, so model both before committing.

Fix 3: keep traffic in its Availability Zone

Cross-AZ charges come from three patterns:

Do not sacrifice multi-AZ resilience to save transfer charges. The goal is that steady-state traffic stays local and cross-AZ traffic happens on failover, not on every request.

Fix 4: egress you did not mean to pay for

Make it somebody's job

Network cost stays invisible because it lands in shared networking accounts. Two changes stop it drifting back:

What this looks like at ccX

Data transfer and network design is one of the six levers in our two-week AWS Cost Optimization Review. We break network charges down by usage type and account, trace the expensive flows to their sources, and give you a prioritised list of endpoint, routing and placement changes with an annual saving estimate for each. Most estates have at least one change on that list that costs nothing to make. Fixed fee, read-only access, never a percentage of savings.

If nobody can tell you what your NAT Gateways are carrying, book a 30-minute scoping call.

Where to go from here

AWS data transfer and endpoint pricing varies by region and changes over time. Check current rates for your regions before modelling savings.


Sitemap