Elastic IP Charges (2026): What You Will Really Pay and How to Cut It

For most of AWS's history, a public IP address was something you got with an instance and never thought about. That ended on February 1, 2024, when AWS began charging for every public IPv4 address in every account, attached or not. The rate is small, half a cent an hour, which is exactly why it is dangerous: it is too small to notice on any single resource and large enough, multiplied across a fleet of instances, load balancers, NAT Gateways, and forgotten Elastic IPs, to become one of the fastest-growing lines on a bill nobody audits.

This guide explains what AWS actually bills for, which addresses are exempt, what the charge costs at realistic scale in US dollars, how to find every address you are paying for, and the eight changes that reduce the bill. All rates are from the official Amazon VPC pricing page and the AWS announcement of the charge, current as of September 2026.

What AWS Actually Charges For?

The rule is simpler than the billing line items suggest: if an address is a public IPv4 address and it exists in your account, it bills $0.005 for every hour it exists. AWS applies this through two usage types. PublicIPv4:InUseAddress covers addresses attached to a running resource, including the Amazon-provided public IP an EC2 instance receives automatically, an Elastic IP associated with a running instance, and the addresses behind load balancers, NAT Gateways, and other services. PublicIPv4:IdleAddress covers Elastic IPs that are allocated but not associated with anything, or associated with a stopped instance. Both bill at the same rate. The Elastic IP documentation describes the mechanics; the table below shows what that means per resource.

Two details in that table catch teams out. Load balancers use at least one address per Availability Zone, so a three-AZ ALB pays for three addresses even when it is idle. And Elastic IPs bill from the moment they are allocated, not from the moment they are used, so an address reserved for a future project or left behind by a deleted instance costs the same as one serving production traffic.

What It Costs at Scale

Half a cent an hour is easy to dismiss until it is multiplied. The arithmetic is address count times $0.005 times 730 hours.

The cost is rarely the largest line on an AWS bill, but it is among the least justified, because a large share of the addresses behind it are not doing anything a private address or a shared endpoint could not do. That makes it a classic case of the cloud waste that accumulates because it is small per item and invisible per resource, and a good first target in any AWS cost optimization pass.

The Free Tier and the Exemptions

Three carve-outs matter. The EC2 free tier includes 750 hours a month of in-use public IPv4 for the first 12 months of a new account, enough for one address running continuously, and it does not cover idle addresses at all. IPv6 addresses are free on every resource, without limit, which is the exemption AWS most wants you to use. And addresses you bring to AWS yourself through Bring Your Own IP, or customer-owned addresses on Outposts, are exempt from the per-address charge, although IPAM and related services can still bill. For an organization that already owns IPv4 space, BYOIP removes the fee entirely; for one that does not, the market price of IPv4 addresses generally makes buying them to avoid a $3.65 monthly fee a poor trade.

How to Find Every Address You Are Paying For

Before reducing the charge, inventory it. In Cost Explorer or the Cost and Usage Report, filter on the two PublicIPv4 usage types to see in-use and idle spend separately, by region and, with the resource-level report, by resource ID. Amazon VPC IP Address Manager offers a free-tier Public IP Insights view that lists every public IPv4 address in the account with its type and attachment, which is the fastest way to find idle Elastic IPs and load balancers with no healthy targets. Because the charge is billed against the network interface rather than the instance, tag network interfaces with the same owner tags as the resources they belong to, or the cost will land in an unattributed bucket. That attribution step is the difference between a number and an owner, which is the whole subject of our guide to allocating cloud costs to teams.

Eight Ways to Cut Elastic IP and Public IPv4 Charges

• Release idle Elastic IPs. The simplest saving. Any Elastic IP not associated with a running resource is pure cost. Audit quarterly, or automate with a policy that flags addresses idle for more than a few days.
• Stop giving every instance a public IP. Disable auto-assigned public IPs on subnets and launch templates where instances do not need direct inbound internet access. Most application and database tiers never do.
• Put workloads in private subnets behind a NAT Gateway. Instances that only need outbound access can share one NAT Gateway's address instead of each carrying its own. Weigh the NAT Gateway's own hourly and per-GB fees, which can exceed the IP saving for high-traffic workloads; gateway VPC endpoints for S3 and DynamoDB avoid both.
• Consolidate behind shared load balancers. One Application Load Balancer can front many services using host- and path-based routing, replacing a public address per service with two or three for the whole tier.
• 5. Drop public IPs used only for administration. EC2 Instance Connect Endpoint and AWS Systems Manager Session Manager provide SSH and shell access to instances in private subnets, removing the most common reason a public IP exists.
• 6. Adopt IPv6 where clients support it. IPv6 is free, and dual-stack or IPv6-only configurations remove the charge for every address that can use it. This is the structural fix AWS designed the charge to encourage.
• 7. Delete idle load balancers and NAT Gateways. A load balancer with no healthy targets still bills its hourly rate plus two or three IPv4 addresses; a NAT Gateway in an unused dev VPC bills around $33 a month plus its address. Non-production environments are where these accumulate.
• 8. Use BYOIP if you own address space. Organizations with their own IPv4 allocations can bring them to AWS and pay nothing per address, which at enterprise scale is the difference of tens of thousands of dollars a year.

A Worked Example

Consider a mid-size estate with 40 EC2 instances carrying public IPs, six three-AZ Application Load Balancers, four NAT Gateways, three RDS instances with public endpoints, and 15 idle Elastic IPs left over from decommissioned projects. That is 40 plus 18 plus 4 plus 3 plus 15, or 80 public IPv4 addresses, costing about $292 a month, roughly $3,500 a year, before any data transfer.

Now apply the levers. Releasing the 15 idle Elastic IPs removes $55 a month at no risk. Moving 30 of the 40 instances into private subnets, since they only serve traffic through the load balancers, removes another $110. Consolidating six load balancers into two removes about $44. Moving the RDS endpoints private, which they almost certainly should be for security reasons anyway, removes $11. The estate now runs on about 20 addresses at roughly $73 a month, a 75 percent reduction, with no loss of function and a smaller attack surface as a side effect. The remaining $73 is the legitimate cost of the public endpoints that genuinely need to exist.

Where the Charge Is Worth Paying

Not every public address is waste. Internet-facing load balancers need their addresses. A NAT Gateway serving a busy private subnet is cheaper than the per-instance addresses it replaces. Bastion alternatives require Systems Manager setup that some teams have not done. And some workloads, particularly legacy applications and third-party integrations that whitelist a fixed source IP, need a stable Elastic IP that cannot be replaced by anything else. The goal is not zero public IPv4 addresses; it is zero public IPv4 addresses that are not earning their $3.65.

The Bottom Line

Elastic IP charges are a small rate applied to a large and usually unexamined count. The rate is fixed at $0.005 an hour and the exemptions are narrow, so the bill is decided entirely by how many public IPv4 addresses exist in your account, and most estates carry far more than they need: idle Elastic IPs, auto-assigned public IPs on instances that never see inbound traffic, and load balancers serving nothing. Inventory them, release what is idle, move what can move into private subnets, share what can be shared, and reach for IPv6 where clients allow. Then keep watching, because addresses accumulate the same way every other small cloud charge does, which is why per-resource visibility of the kind Opslyft's cost visibility provides matters more for charges like this one than for the large lines everyone already tracks. For the wider set of small charges that inflate AWS bills, see our guide to AWS cost management mistakes.