Scheduled Data Collection on AWS: When Elastic IPs Cost More Than the Instances

Updated November 18, 2026 • By Server Scheduler Staff

Plenty of AWS workloads run for an hour or two a day: price monitoring, rank checks, uptime probes from several addresses, regional content tests. Teams schedule the instances to stop outside that window and expect the bill to shrink with them. Then the IPv4 line shows up. Since AWS started charging for public IPv4 addresses, a job that needs many outbound IPs can spend more on addresses than on compute, especially when those addresses are Elastic IPs that stay allocated around the clock.

This article walks through what AWS charges for, compares three ways to give a scheduled job 50 outbound IPs, and covers the trade-offs of moving that IP pool outside AWS.

What AWS bills for outbound IPs

Public IPv4 addresses. Every public IPv4 address in your account costs $0.005 per hour, billed per second with a 60-second minimum. The rate is the same for an address attached to a running resource and for an idle one, so an Elastic IP kept for a stopped instance keeps billing. At about 730 hours a month, one address is roughly $3.65 a month, whatever the instance behind it is doing.

Auto-assigned public IPs. An instance launched with an auto-assigned public IPv4 address pays for it only while the address exists. Stop the instance and the address is released; start it again and you get a different one. Cheaper, but the IP changes on every run.

NAT gateways. Instances in private subnets usually reach the internet through a NAT gateway: $0.045 per hour (about $32.85 a month in us-east-1) plus $0.045 per GB of data it processes. Data processing applies to traffic in both directions, so the responses a collection job downloads are billed even though inbound data transfer from the internet is free.

Data transfer out. Requests leaving AWS for the internet count as data transfer out, $0.09 per GB in us-east-1 after the first 100 GB a month, which is free across your account. For most collection jobs requests are small and responses large, so this line stays modest.

Three ways to give a job 50 outbound IPs

Take a job that needs 50 distinct outbound addresses, runs 2 hours a day, and runs 30 days a month: 60 hours of work. Prices are us-east-1 On-Demand, Linux, at the time of writing.

A. Fifty small instances, auto-assigned IPs, started on schedule. Fifty t3.micro instances at $0.0104 per hour for 60 hours cost $31.20, and their 50 public addresses for the same 60 hours cost $15. Total: about $46 a month. The catch: you get a new set of addresses every day, all from AWS ranges, and you manage fifty instances.

B. A fixed set of 50 Elastic IPs. If the job needs the same addresses every day, for example because a partner allowlists them or sessions depend on them, Elastic IPs are the AWS answer. Fifty of them cost 50 × 730 × $0.005 = $182.50 a month before any compute, because they bill while the instances are stopped. The default quota is five Elastic IPs per Region, so this also means a quota request.

C. One scheduled instance and a proxy pool. Run the job on a single t3.small (60 hours × $0.0208 = $1.25) with one Elastic IP ($3.65 a month), and send its requests through static proxies. The Elastic IP stays fixed because it's the address you allowlist with the proxy provider; the 50 outbound addresses come from the proxies.

The proxy price decides how this compares. FineProxy, for example, sells package datacenter IPs at $12 for 100 IPs a month, shared by up to five customers, and dedicated IPs used only by you at $2 per IP. Traffic on static plans isn't billed per gigabyte, so a heavy run costs the same as a light one. With a 100-IP package, option C comes to about $17 a month; with 50 dedicated IPs, about $105.

Option Monthly Cost Addresses
A. 50 × t3.micro, auto-assigned IPs ≈ $46 New every run, AWS ranges
B. 50 Elastic IPs ≈ $182.50 + compute Fixed, AWS ranges, quota request
C. 1 × t3.small + 100 package proxy IPs ≈ $17 Fixed, shared by up to 5 customers
C. 1 × t3.small + 50 dedicated proxy IPs ≈ $105 Fixed, exclusive


So the proxy route isn't automatically the cheapest. If fresh AWS addresses on every run are fine, option A wins on exclusive IPs. Where C pays off is against B: a fixed, larger pool of addresses without paying for IPv4 hours while the job is asleep, and with one instance to schedule instead of fifty.

Setting up option C

  • Put the instance in a public subnet with its Elastic IP, or create the NAT gateway only for the job window from your infrastructure code. A NAT gateway that exists all month for a 2-hour job costs more than the job.
  • Allowlist the Elastic IP with the proxy provider. FineProxy ties access to allowlisted IPs, with a login and password available on top. Keep the Elastic IP when the instance stops; releasing it means updating the allowlist before the next run.
  • Lock down the security group. Outbound to the proxy ports, inbound only what you need for management.
  • Load the proxy list at job start from the provider's API or an exported file rather than baking it into the image, so replacing an address doesn't need a rebuild.
  • Check concurrency. FineProxy packages up to 999 IPs include 1,500 concurrent connections, which is plenty for a 2-hour batch but worth confirming against your thread count.
  • Test before you commit. Any static plan can be bought for 48 hours, enough to run the real job a couple of times.

Trade-offs to weigh

A third party sits in the path. The provider sees which hosts you connect to. Use HTTPS to targets: with the CONNECT method the TLS session runs end to end, so request and response contents stay encrypted.

Latency goes up. Each request takes an extra hop. Pick proxy locations close to the sites you collect from, and keep the instance's Region close to the proxies.

Another dependency can fail. Add retries with backoff and keep a few spare addresses, so a single bad IP doesn't stall the batch.

IPv6 is an alternative worth checking. IPv6 addresses don't carry the IPv4 hourly charge. If the sites you collect from serve IPv6, an IPv6-only setup may avoid most of the cost; many sites still don't, which is why IPv4 pools remain common.

Respect target terms. Rate-limit your job and follow the sites' terms of use. More addresses don't change what you're allowed to collect.

The short version

If your scheduled job needs many outbound IPv4 addresses, add up the address hours, not only the instance hours. Auto-assigned IPs are cheap when address churn is acceptable. When the addresses need to stay fixed, a single scheduled instance with one Elastic IP and a per-IP proxy pool keeps the address cost flat and stops it from billing while the job is off.

Related Articles