Data Sovereignty in the AI Era: Structural Guarantees Through Federated Pipeline Networks

As AI crawlers harvest publicly available data for training without permission, copyright litigation expands, and nations advance legislation regarding data holder liability, the aggregation of data on central platforms is shifting from a cost optimization measure to a concentration of legal and economic risk. This paper argues that, in this environment, data sovereignty should be strengthened through architectural structure rather than contractual assurances alone. Existing initiatives such as IDS/EDC and Catena-X have established federated models for dataset exchange, while frameworks such as GAIA-X have advanced governance and discovery layers for data ecosystems. This paper focuses on a different design center: the federation of pipelines, namely cross-organizational chains in which data is processed and transformed in transit. Its contributions are threefold: (1) a problem-domain taxonomy that classifies architectural strategies by the sovereignty domain they address, identifies cross-organizational processing chains as an under-formalized design space not systematically addressed by established approaches (TEE, Data Clean Room, Compute-to-Data, IDS/EDC), and frames pipeline federation as a response to this specific design space, (2) a set of design heuristics derived from historical federation patterns such as Email/DNS, TLS-enabled trust on the Internet, and ActivityPub, and (3) a concrete architectural proposal centered on four design principles: self-hosting, metadata-network-based discovery, DID-authenticated connection control, and provenance chaining. The provenance model is scoped to lineage-preserving transformations; non-linear operations such as aggregation reset the chain and are treated as a boundary condition. Finally, we examine how this architectural design is structurally coherent with the EU Data Act, GDPR, and the EU AI Act, while not by itself constituting legal compliance.