Cloud providers charge for data egress partly because carrying traffic off their infrastructure consumes backbone, peering, transit, edge, and operational capacity. Egress pricing is also a commercial policy that can affect switching. Both points matter: provider cost structures are real, but public list prices do not reveal unit cost or margin, and regulators have addressed switching friction without abolishing routine delivery charges.
The technical rationale
An outbound byte may travel from storage through a regional network, across a provider backbone, through an edge location or paid transit path, and into the recipient's internet provider. The cloud operator must provision routers, fiber, datacenter links, subsea capacity, redundancy, monitoring, and staff for that path.
Microsoft describes a global network of more than 500,000 miles connecting datacenters across more than 80 Azure regions, with edge interconnection in more than 190 locations and more than 4,000 unique internet peers. It says traffic normally uses “cold-potato” routing: Microsoft keeps the packet on its own backbone as long as possible before handing it to the internet.
Azure also offers a routing preference that hands traffic to a transit ISP earlier. Microsoft positions that path as cost optimized, while its premium global-network path uses the Microsoft backbone for more of the journey. The pricing difference makes the network choice visible.
On the USD public list page accessed August 10, 2026, the first paid internet-egress band from North America/Europe was $0.087/GB on the premium global network and $0.08/GB through the transit-ISP option, after a 100 GB monthly allowance. At the 350 TB band, the listed rates were $0.05 and $0.04 respectively. Source continent, destination, service, agreement, currency, monthly tier, and tax still matter.
That table supports a narrow conclusion: Azure sells different routing products at different prices. It does not disclose Microsoft's underlying cost per GB and cannot prove a particular margin.
Why ingress is often free
Azure, AWS, and Google Cloud generally make internet ingress free for the storage examples in their public pricing. Bringing data in increases the amount a customer stores, computes on, or serves from the platform. Outbound traffic consumes network resources while reducing or distributing what remains inside.
That asymmetry is partly technical and partly commercial. Providers can choose other models. A provider may charge more for capacity or operations and zero for egress, include an allowance, negotiate committed volume, or use a CDN rate. “There is a real network cost” does not imply that one pricing structure is inevitable.
Geography and volume shape the price
Network economics are not uniform. Azure's August 10, 2026 premium-network list showed $0.087/GB for the first paid band from North America/Europe and $0.181/GB from South America. The same page stepped North America/Europe rates down across larger monthly bands.
The list-price pattern can reflect fixed capacity spread over more volume, commercial volume discounting, competitive strategy, or all three. Provider disclosures do not let an outsider allocate those effects precisely. It is safer to say that route, geography, and aggregate volume influence the published price than to claim egress is priced at cost or above cost.
For a delivery workload, recipient count is another multiplier. One stored 1 TB package downloaded by ten recipients creates roughly ten times the outbound bytes of one download. The cost-versus-storage example shows why this can dominate short-retention economics.
Egress can also be a switching cost
When leaving a provider requires extracting many terabytes, ordinary per-GB charges can raise the cost of migration. Governments and regulators distinguish that switching event from daily service consumption.
The EU Data Act, Regulation (EU) 2023/2854, applies from September 12, 2025. Article 29 permits reduced switching charges tied to direct switching costs through January 12, 2027. From January 12, 2027, providers generally may not impose defined switching charges, subject to the regulation's scope and exceptions.
This does not mean all cloud egress becomes free in the EU. Switching charges are a defined legal category. Internet delivery to routine customers while continuing to use the platform is different. Custom-built services, non-production services, contractual facts, and other exceptions may also matter. This is educational context, not legal advice.
Provider exit programs are not routine delivery
AWS offers an eligible, support-mediated waiver for customers moving data off AWS. The program requires a request and applies to an exit rather than ordinary recurring downloads; AWS notes exclusions for specialized transfer services. Azure describes a credit process for customers moving data to another cloud or on-premises. Google requires a qualifying full exit, an Exit Notice, migration of workloads and data, and agreement termination under its published portability procedure.
These programs can materially change a migration plan. They should not be placed into the normal-delivery line of a cost model. Ask:
- Must all workloads or data leave?
- Is support approval or advance notice required?
- Which services and transfer methods qualify?
- Is there a completion window?
- Does the account or agreement terminate?
- Which taxes, contracts, or exceptions remain?
For ongoing delivery, continue to model normal egress, retrieval, operations, replication, and recipient count. For a genuine exit, evaluate the provider's current waiver terms separately and obtain legal and commercial advice.
Practical ways to manage egress
Place data near the recipient population when residency and operations permit. Choose routing preference deliberately rather than accepting a default blindly. Use caching or a CDN when many recipients request identical data and the CDN's own economics fit. Consolidate duplicate downloads, monitor regional paths, and test whether a colder storage tier adds retrieval or early-deletion cost.
Multi-cloud can improve portability, but copying every object between clouds can itself create transfer and operational cost. A switching strategy should identify the data that must move, the sequence, validation, rollback, contractual notice, and eligible exit program before the migration starts.
Use the transfer-time calculator to estimate the two network legs. Cost and time are separate constraints; a waived migration charge does not make the destination connection faster.
Where TeraAirlift fits
TeraAirlift supports large-file delivery, so routine recipient downloads remain the relevant economic event—not a one-time provider exit. Plans start with a 10 TB included allotment, packages can be any size, and volume above the allotment is overage. No final price, rate, metering method, retention period, or billing design is announced; details are confirmed on a demo.
Sources
- Microsoft global network — backbone, interconnection, peering, and routing description.
- Azure routing preference — premium-backbone and transit-ISP paths.
- Azure Bandwidth pricing — dated USD list examples and Azure exit-credit description; accessed August 10, 2026.
- Amazon Web Services — Free data transfer out when moving out — eligibility and support process; accessed August 10, 2026.
- Google Cloud data portability and switching — qualifying full-exit procedure; accessed August 10, 2026.
- Regulation (EU) 2023/2854 — Data Act application and switching-charge timeline; accessed August 10, 2026.
