Egress — the Charge Nobody Plans For

You model compute. You model storage. Almost nobody models the cost of moving data out.

Why moving data out costs money

Providers charge almost nothing to ingest data — a deliberate choice that lowers the barrier to adoption. Egress is the mirror image. When data leaves a cloud region headed for the public internet, you pay per gigabyte transferred, and the rate is not small. When data moves between two regions within the same provider, you pay a lower but still real cross-region transfer fee, typically billed on the sending side. The one free path is intra-region: traffic staying inside a single region, between services in the same availability zone, typically travels at no charge.

This asymmetry is structural, not punitive. Cloud providers lease transit capacity from backbone carriers, and that capacity has a real wholesale cost. Pricing egress also happens to create switching friction: the more data you have in a provider's storage, the more expensive it becomes to leave. That is not a conspiracy; it is a business model, and it is worth being clear-eyed about it.

Where the surprise actually comes from

The charge is documented. Engineers who read the pricing pages know it exists. Bills still surprise teams because the scale is invisible until it compounds. A data pipeline that pulls records out of object storage for processing, ships results to an analytics service in a second region, and exposes an API to users on the public internet can accumulate egress at three separate rates simultaneously. Each one looks modest in isolation. Together, especially once traffic grows, they become a non-trivial line item.

CDN offload is the standard mitigation: content delivered through a provider's edge network often carries a lower per-gigabyte egress rate than direct internet delivery, and reduces origin traffic. That helps for read-heavy workloads. It does not help for bulk data exports, cross-region replication, or analytics pipelines that move large datasets between services.

Bills still surprise teams because the scale is invisible until it compounds.

The architecture implication is real. Keeping compute and storage in the same region — and designing pipelines that process data close to where it lives rather than copying it across regions first — can eliminate entire categories of egress charge. The idle tax gets attention because it shows up in compute; egress hides in the transfer section of the bill and is easier to miss.

The honest summary: egress pricing rewards architectures that treat data gravity seriously from the start. If your design assumes data is cheap to move, you will find out it is not when the bill arrives. The time to model it is during design, not during the cost-optimisation sprint six months after launch.