Summary

  • The EU Data Act is removing provider switching charges, including data-egress charges, but it does not turn a cloud exit into a costless or automatic operation.
  • A credible exit budget has six ledgers: extraction, service substitution, rehearsal, overlap, cutover and closure. The transfer invoice is only part of the first.
  • Scaleway's own public documentation separates data recovery, identity, workload restoration and deletion responsibilities, which makes a practical test possible: nothing is ready to close until the destination service and the evidence both work.

The finance team receives a reassuring line: data egress, €0. On another screen, the engineering team still has a database to restore, policies to rebuild, certificates to rotate, network paths to change, alerts to reconnect and a deletion decision it cannot reverse. Both screens may be accurate. Only one describes the cost of leaving.

That distinction is becoming more important in Europe. The EU Data Act has applied since September 2025 and its cloud-switching rules are designed to remove commercial, contractual, technical and organisational obstacles. From 12 January 2027, providers will no longer be allowed to impose switching charges for executing a switch, including the relevant data-egress charges. During the transition, reduced charges remain possible within the Regulation's limits.

The date deserves attention, but the price headline can mislead. The law addresses what a source provider may charge and what help it must offer. It does not say that a customer will spend no money on a destination service, specialist support, application changes, parallel running or acceptance testing.

The Regulation itself differentiates infrastructure services, where the source provider must take reasonable measures to facilitate functional equivalence for the same service type, from other processing services, where open interfaces and machine-readable export are central. If export alone were sufficient, that distinction would be unnecessary.

A fee disappears; the work remains

Scaleway has long made the anti-lock-in argument. In a 2022 company position, it described egress fees as part of the switching-cost problem and connected multi-cloud with reversibility and openness. That is a useful statement of intent. It is not evidence that every workload can move without redesign, nor that every current Scaleway product has the same transfer tariff. Scaleway's pricing is product-specific; a blanket claim that all egress is free would be wrong.

The stronger evidence sits in the operational documentation. Scaleway's storage responsibility model says that the customer plans and implements reversibility, recovers the data, finishes that work before deleting buckets, volumes or snapshots, and confirms deletion. Scaleway, in turn, says it will keep the service available until termination and can provide a data-deletion certificate on request. Those are not two descriptions of one step. They are a sequence: recover, verify, delete, obtain evidence.

The same separation appears elsewhere. A Scaleway API key belongs to one Organisation and inherits the permissions of its user or application. A PostgreSQL dump can move logical data, but it does not recreate the destination principals, policies or secrets. Scaleway's Kubernetes responsibility model leaves customers responsible for workload recovery, keys, certificates and backup policy. Closing an account permanently deletes the Organisation's resources and backups; the resources cannot be retrieved merely by reopening the account.

Together these facts define the exit problem more clearly than a slogan does. Cloud portability has at least six cost ledgers.

The six ledgers

Extraction is the visible one. It includes objects, volumes, database dumps, images, schemas, metadata, logs and audit records. The receipt is not “copy complete” but a reconciliation: source counts, destination counts, cryptographic hashes and an explicit list of failures or exclusions. A customer also needs to know what the provider considers exportable and what is bound to the service's internal implementation.

Substitution is where the cheap transfer can become an expensive programme. A managed database is not just its rows. A load balancer is not just an address. An identity system is not just a list of names. Teams must map extensions, maintenance behaviour, availability design, endpoints, firewall rules, secret storage, queues, monitoring and support procedures to destination equivalents. Some can be copied; some must be translated; some must be redesigned.

Rehearsal converts that map into evidence. Restore a backup. Replay a queue. Force a failover. Expire a certificate. Remove a permission. Measure a critical path under representative load. A Terraform plan can describe resources and make changes reviewable, but infrastructure as code is not a certificate of semantic equivalence. The destination must produce the business outcome under ordinary and failed conditions.

Overlap is the cost that a simple calculator often omits. A safe migration normally keeps source and destination available together while data catches up and users test the new path. For that period, there may be two sets of infrastructure, monitoring, security controls, support arrangements and on-call obligations. Compressing overlap can reduce the invoice; it can also remove the time in which discrepancies can be found without an outage.

Cutover moves authority, not merely traffic. DNS, routes, certificates, API clients, automation, scheduled jobs, secrets and incident ownership have to point to the new estate. The team needs an observable rollback decision and a time after which the old estate is no longer allowed to accept writes. Otherwise, two valid systems can create divergent state.

Closure is a controlled destruction process. It requires a residual-resource scan, final billing check, treatment of retention and legal holds, credential revocation, deletion requests and the available certificates or receipts. Scaleway recommends deleting resources before account closure and warns that closure permanently deletes resources and backups. The zero-egress invoice is reversible; that final click may not be.

Exportability is not equivalence

This framework does not imply that leaving Scaleway is unusually difficult. It shows why no provider's low or waived egress charge can answer the whole question. Standard tools matter: Scaleway documents pg_dump and pg_restore for PostgreSQL, and mysqldump and mysql for MySQL. S3-compatible object access and Terraform-managed infrastructure can also reduce friction. They move the boundary in the customer's favour.

But each tool proves only what it observes. A successful database restore says something about the logical data accepted by the destination. It does not, by itself, prove extension compatibility, failover behaviour, recovery time, monitoring continuity or application latency.

An object checksum proves that a byte sequence arrived. It does not prove that an IAM policy grants the right actor the right operation, or that an old key has been revoked. A clean Terraform apply proves that the destination API accepted a desired configuration. It does not prove that the resulting managed service behaves the same way in an incident.

The Data Act's language is helpful here. “Functional equivalence” does not mean identical systems. The Commission describes it as materially comparable outcomes for shared features after an infrastructure switch. That is a result to test, not a label a procurement team can assume.

Price the proof before renewal

The practical commissioning question is therefore simple: if the egress line were zero tomorrow, what would still prevent a safe exit?

Answer it before the next major expansion or long commitment. Build an inventory reconciled to billing and infrastructure code. Select one production-shaped workload. Export it, recreate the necessary controls, restore it at a destination, operate it through a failure exercise and calculate the time during which both estates must run. Record every manual step and every provider-specific dependency. Then return to the contract and price the missing work.

The result is not a promise to leave. It is an option with an exercise price. A company may rationally stay with Scaleway because the service, price or European operating model remains attractive. But staying after a measured rehearsal is a decision. Staying because the transfer fee looked small is an assumption.

Sources for the legal rules are the official Data Act text and the European Commission's explainer.

Product responsibilities are drawn from Scaleway's public documentation on storage reversibility, Kubernetes responsibilities, IAM keys, database migration and account closure. No customer outcome or undisclosed Scaleway capability is assumed.

The remaining source roles are Scaleway's current General Terms of Services, its attributed 2022 multi-cloud position, the product-specific pricing index and its Terraform overview. They establish contract, company-position, price-boundary and infrastructure-as-code context; none proves a customer outcome.