The Mechanics of Cloud Egress Fees: Why Data Gravity Traps Enterprise Workloads

Cloud providers and IT consultancies routinely market "multi-cloud" as the gold standard for enterprise resilience, promising that companies can seamlessly shift workloads between Amazon Web Services (AWS), Microsoft Azure, and Google Cloud to chase the best compute pricing. The evidence, however, contradicts this flexibility. The physical reality of "data gravity" and the financial reality of egress fees make multi-cloud architectures economically unviable for large datasets. Moving data out of a major cloud costs significantly more than storing it, effectively trapping enterprise workloads exactly where they are.[6]

The mechanism that enforces this lock-in is the egress fee. Ingress—uploading data into a cloud provider's network—is universally free. Storing that data is cheap, typically costing around $0.02 per gigabyte per month for standard hot storage. But egress—downloading that data or transferring it to a competing provider—triggers a punishing toll. Cloudflare, which operates a global network that frequently interfaces with these providers, notes that "egress fees are a subtle way for certain providers to charge more for cloud management over time, while also discouraging customers from leaving their ecosystem."[1]

The specific rates dictate the architecture. AWS charges $0.09 per gigabyte for standard internet egress. Microsoft Azure charges $0.087 per gigabyte for the same tier. Google Cloud's Premium Tier internet egress costs $0.12 per gigabyte. For a company storing a 1-petabyte dataset, the baseline storage cost is roughly $20,000 to $23,000 a month. But moving that single petabyte to a competitor incurs a one-time exit penalty of $87,000 to $120,000. The cost to leave is equivalent to four to five months of the cost to stay.[3][4][5]

This financial friction compounds a physical phenomenon known as "data gravity." Coined in 2010 by Dave McCrory, the concept states that as a dataset accumulates mass, it becomes increasingly difficult to move. TechTarget defines the mechanism explicitly: "The force of gravity, in this context, can be thought of as the way software, services and business logic are drawn to data relative to its mass, or the amount of data." You cannot economically store a 50-terabyte database in AWS and run the machine learning models that analyze it in Google Cloud; the cross-cloud transfer taxes would bankrupt the project.

The rise of generative artificial intelligence has exposed the limits of this model. Training a large language model requires moving massive datasets—such as the 240-terabyte LAION-5B dataset—into compute clusters. At standard hyperscaler rates, moving 240 terabytes out of a cloud environment costs upwards of $15,000. Because the data cannot move, the compute must move to the data, cementing the vendor's monopoly on the entire stack.[6]

In response to these economics, 2026 has seen a surge in "cloud repatriation"—the strategic movement of predictable, data-heavy workloads out of the public cloud and back into private colocation centers. As Teradata's glossary defines the breaking point: "Data gravity appears when the amount of data volume in a repository grows and the number of uses also grows. At some point, the ability to copy or migrate data becomes onerous and expensive." By purchasing their own hardware and paying flat-rate wholesale transit costs, enterprises bypass the hyperscaler transfer tax entirely.[2]

The market is beginning to fracture under this pressure. A new tier of specialized cloud providers and storage challengers have launched zero-egress pricing models, proving that the infrastructure can be operated profitably without the exit tax. These providers charge only for storage and compute, allowing data to flow freely across the internet.[1][6]

The hyperscalers are slowly adjusting, with some introducing exit commitments that waive outbound transfer fees upon a complete, permanent migration off their platform. But for ongoing operations, the toll remains. The multi-cloud future marketed by vendors requires data to be liquid. Until the egress fees that artificially inflate the mass of data gravity are removed, enterprise data remains frozen in place.[6]