The EU Data Act Makes Cloud Exit an Architecture Requirement
The Data Act’s cloud-switching rules turn portability into an engineering task. Design analytics pipelines so data, metadata, and configuration can leave a provider cleanly.
The EU Data Act makes switching between data-processing services a product and architecture concern, not just a procurement clause. Since 12 September 2025, its switching provisions have applied to providers offering covered data-processing services to customers in the EU. Teams building analytics infrastructure should be able to explain what leaves a provider, in which format, and how the service keeps working during the move.
This is not the same as GDPR data portability. The Data Act addresses switching between data-processing services and access to data from connected products; GDPR Article 20 gives people a right to receive certain personal data they provided. A migration plan may need to handle both, but the legal scopes and the data sets are different.
What the switching rules change
The Data Act requires providers of covered data-processing services to remove contractual, commercial, and technical obstacles to switching. Customers must be able to move to another provider, move to their own on-premises infrastructure, or use more than one provider. The service contract must support a transition period and identify the relevant switching process.
The Act does not promise that every artifact transfers identically. Its rules distinguish exportable data and digital assets from provider-specific internal data and protected trade secrets. For analytics, that means a CSV of pageviews is not a complete exit plan if dashboards, event definitions, retention settings, access roles, and scheduled reports remain trapped in proprietary formats.
Switching charges also have a timetable. Until 12 January 2027, a provider may charge reduced switching fees, but they must not exceed costs directly linked to the switching process. From that date, switching charges are to be abolished. A team planning a migration in 2026 should therefore check both contract terms and the actual technical work required; a future fee ban does not make an undocumented export usable.
Design an exit path before you need one
Start with a data and configuration inventory. List raw events, aggregated reports, custom dimensions, site mappings, user permissions, API credentials, retention rules, and integrations. Mark which items are customer data, which are provider-generated metadata, and which depend on a vendor-specific feature. This inventory makes a realistic migration boundary visible.
Next, define portable representations. Keep event schemas documented and versioned; make aggregate exports available in common machine-readable formats; and store configuration as declarative files where practical. An export that loses timestamps, time zones, event names, or dimension semantics is technically portable but analytically broken. Include checksums, row counts, and schema versions so teams can verify the handoff.
Test a round trip, not just a download. Restore a representative export into a clean environment, compare key aggregate results, and rehearse a period of parallel collection. Record which reports cannot be reproduced and why. This test catches hidden dependencies on proprietary query syntax, identity stitching, or a provider’s unexported enrichment.
Keep privacy boundaries during migration
Portability does not authorize unlimited copying. Before moving analytics data, decide whether each dataset is still necessary, whether it contains personal data, and who should have access during the transition. Minimize the migration payload, encrypt it in transit and at rest, scope temporary credentials narrowly, and delete temporary copies after verification.
Avoid solving lock-in by retaining more identifiable data than the product needs. An analytics system can often answer product questions with aggregate counts, short-lived operational data, and a small set of stable event fields. The smaller the dataset, the less sensitive the export becomes and the less work there is to secure, reconcile, and eventually erase.
The Data Act is not a substitute for GDPR, security duties, or a data-processing agreement. If an export includes personal data, the existing controller-processor roles, lawful basis, retention schedule, and transfer safeguards still matter. Switching vendors changes a processing relationship; it does not reset it.
Put portability in the release checklist
Make exit readiness observable. Track when the last export was tested, how long a full export takes, which formats are supported, and whether a second environment can restore the data. Include failure cases such as partial exports, schema drift, missing assets, and credentials that expire mid-transfer.
For analytics buyers, ask vendors to demonstrate export and deletion rather than relying on a portability statement in a contract. For teams building the service, treat a complete, tested export as part of the service interface. The strongest portability design is one that can be used without emergency engineering when the provider, price, or operating model changes.
Comments
Loading comments…