Egress Costs

Egress costs are fees that cloud providers charge for outbound data traffic, and they rank among the hardest items of the cloud bill to plan for. This article explains how they arise, which architectures drive them, and which levers companies can use to reduce them.

The way into the cloud is cheap, the way out costs: loading data into a cloud platform is generally free with the major providers, whereas fees apply for every gigabyte in the opposite direction. These egress costs at first do not appear in many calculations at all and later grow with every new integration. Anyone running multicloud, analyzing data, or training AI models should therefore plan data outflow just as carefully as compute and storage. This article classifies the fees and shows effective countermeasures.

What are egress costs?

Egress costs, usually listed in price lists as Data Transfer Out, are volume-based fees for data that leaves a cloud environment. Billing is per gigabyte transferred, tiered by destination: traffic to the public internet is the most expensive, transfers between regions of the same provider are cheaper, and even between availability zones of a region fees sometimes apply. This pricing logic has a strategic core. Bringing data in is easy and free, getting it out costs, which ties customers to the platform. For planning, this means: the bill depends less on the volume of stored data than on the movement patterns of the data. That is exactly why data flows belong in every cost estimate before a migration.

How it happens

The fees add up across many individual data streams that are rarely visible centrally:

Multicloud architectures: When workloads at different providers exchange data regularly, each side pays for its outbound traffic. A one-time design decision turns into a permanent monthly item.
Backups and disaster recovery: Backups to a second region or to an external provider generate egress on an ongoing basis. A large recovery in an emergency adds an unexpected block of costs.
Analytics and AI pipelines: Training and analysis data move back and forth between object storage and compute clusters. Large datasets multiply the costs with every run.
User and partner access: Customer downloads and media delivery run as internet egress at the highest rate, as does every API response to external systems.
Internal cross-references: Services that communicate across region or zone boundaries generate transfer costs that are drawn into no architecture diagram.
Repatriation: Those who move workloads back into their own data center or to another provider pay heavily one time for the migration of the data holdings.

Why it matters

They are hard to predict because they depend on usage behavior and data growth rather than on fixed booked resources.
They act as a barrier to switching: the more data resides in a platform, the more expensive every extraction becomes, and with it every strategic realignment.
For data-intensive workloads such as AI training or media delivery, they quickly reach the order of magnitude of the actual compute costs.
They distort architecture decisions when teams design data flows according to fees rather than functional requirements.
FinOps initiatives often fail because it remains unclear which application causes which transfer.
The EU Data Act phases out fees for switching providers step by step, but ongoing operational traffic remains unaffected by this.

Typical scenarios

A company runs databases at one provider and analysis tools at a second. The monthly bill grows with every dashboard that pulls raw data across the cloud boundary.
A media company delivers video files directly from cloud storage. As the number of retrievals rises, the transfer item exceeds the storage costs many times over.
After a ransomware incident, a complete backup has to be restored from the cloud. Alongside the operational downtime, a considerable and never budgeted transfer block arises.
An AI team trains models on GPU instances of a specialist provider, while the training data resides with the hyperscaler. Every training run begins with an expensive data export.

Egress vs. ingress: the difference

Ingress refers to inbound data traffic into the cloud, egress to outbound traffic. With the major providers, ingress is almost consistently free, because new data strengthens the tie to the platform. Egress, by contrast, is billed per gigabyte, with prices that vary by destination and region. This asymmetry is set deliberately and belongs in every cloud strategy: plan data storage where the data is actually needed, and model outflows as a separate cost factor. Anyone comparing providers solely on storage and compute prices overlooks the part of the bill that grows the most in operation.

Egress optimization at KAEMI

KAEMI makes data flows visible and plans architectures in which egress remains a calculable item. With Cloud Connectivity & SDN we connect your locations and clouds via private interconnects, for example based on Megaport. Many providers price transfers over such dedicated connections considerably lower than internet egress, with better performance and predictability at the same time. For data- and compute-intensive projects, Compute & AI combines GPU resources with object storage from providers that forgo egress fees. This produces a data architecture that preserves freedom of movement instead of penalizing it. On request, we additionally take over ongoing operation including cost control. Have your transfer costs analyzed before the next cloud bill makes them visible.

Frequently asked questions about Egress Costs

Why are egress costs so hard to plan?

Because they depend on the behavior of the applications and are decoupled from booked resources. New integrations and growing data volumes constantly change the outflow, as do altered access patterns. Added to this are tiered rates by destination and region as well as transfers hidden deep within architectures, for example between availability zones. Without continuous monitoring of data flows, the bill remains a surprise.

Do private interconnects really lower egress costs?

Yes, in many constellations. The major providers charge lower per-gigabyte prices for transfers over dedicated connections than for internet egress. With consistently high volumes, port costs and the connection often pay for themselves quickly. In addition, latency and predictability improve. A cost-effectiveness calculation per data path shows from which volume the switch is worthwhile.

What does the EU Data Act change about egress fees?

The EU Data Act obliges cloud providers to phase out fees for switching providers step by step. Major providers have already waived switching egress under certain conditions. The distinction matters: what is favored is migration to another provider, while ongoing data traffic in normal operation remains subject to charges. Multicloud operation and user access therefore continue to incur egress costs as before.

How do I find out which applications drive the egress costs?

The starting point is the providers' cost and billing tools, broken down by service and transfer destination. The picture becomes more meaningful with flow logs and tagging that attribute transfers to individual applications. Conspicuous patterns are recurring retrievals of large objects and traffic between zones. A regular FinOps review anchors this analysis permanently in operations.

Is there cloud storage without egress fees?

Yes, several object storage providers forgo egress fees entirely or up to generous free limits. For backups and frequently retrieved large files, this can significantly reduce total costs. Performance data and compliance requirements remain to be checked. In hybrid architectures, a mix is often worthwhile: primary data with the hyperscaler, transfer-intensive holdings with the egress-free provider.