Skip to content

Data sovereignty in the EU: what it means and which rules are changing

, 9 minute read

Data sovereignty means that an organisation decides where its data is stored, who can access it and which law it is subject to, and that it can enforce those decisions. A data center in the EU is one building block, but not the whole answer: it also matters who operates the service and which law that operator is subject to. This guide explains the terms, the situation for transfers to the US, the EU rules changing over the next few years and what all this means for using AI.

What does data sovereignty mean?

Data sovereignty is the ability to decide about your own data: where it is stored, who accesses it, whether it is passed on and when it is deleted. It is not a legal term with a fixed definition but a goal made up of several sets of rules, above all data protection, IT security, professional secrecy and financial supervision.

In practice data sovereignty comes down to three questions. Where is the data stored and processed? Who has technical access, including for maintenance and support? And which authorities of which state can compel a provider to hand over data? Anyone who can answer all three for every service has their data under control.

What is the difference between data sovereignty, data residency and data localisation?

Data residency only describes the location, for example "the data is stored in Frankfurt". Data localisation is an obligation to keep data in a particular country or region. Data sovereignty goes further: it also covers who can access the data and which law the operator is subject to.

There is no general data localisation requirement in the EU. The GDPR does not require personal data to stay in the EU, but it attaches conditions to transfers to third countries. For non-personal data, Regulation (EU) 2018/1807 on the free flow of such data even prohibits member states from imposing localisation requirements unless they are justified on grounds of public security. Localisation requirements nevertheless often come from contracts, from rules for public bodies or from professional secrecy.

May personal data be transferred to the US?

Yes, under conditions. Since 10 July 2023 the European Commission's adequacy decision on the EU-US Data Privacy Framework has applied: transfers to US companies certified under this framework need no additional safeguards. For all others, standard contractual clauses with an assessment of the transfer risk remain the route.

The framework has a history. In 2020 the Court of Justice of the EU declared its predecessor, the Privacy Shield, invalid in the Schrems II ruling (C-311/18), because US surveillance laws did not offer equivalent protection. Standard contractual clauses remained permissible but have since required a case-by-case assessment and, where necessary, supplementary measures. A first action against the Data Privacy Framework was dismissed by the EU General Court in September 2025. Whether the framework will last remains under discussion, so many organisations plan in a way that does not depend on it if it falls.

Is a data center in the EU enough for data sovereignty?

Not necessarily. A location in the EU settles data residency, but not which law the operator is subject to.

The US CLOUD Act of 2018 obliges US providers to hand over data on the order of US authorities if they possess or control it, regardless of where the data is stored. This also applies to data centers of European subsidiaries. Under Art. 48 GDPR, on the other hand, judgments and orders of third-country authorities may only be recognised on the basis of an international agreement such as a mutual legal assistance treaty. Providers with ties to the US are therefore caught between two legal systems.

Anyone who wants to avoid this conflict therefore looks not only at the location but also at the operator: a provider with no corporate ties to a third country, or systems the organisation runs itself.

Which EU rules are changing over the next few years?

Several EU laws are taking effect in stages right now: the AI Act, the Data Act, NIS2 and, for the financial sector, DORA. They do not require data localisation, but they demand more control over providers, more documentation and the ability to switch providers.

  • EU AI Act, Regulation (EU) 2024/1689: in force since 1 August 2024 and applying in stages. Bans on certain AI practices and the duty to train staff in the use of AI have applied since 2 February 2025, obligations for providers of general-purpose AI models since 2 August 2025. Most obligations for high-risk systems were due to apply from 2 August 2026 under the original timetable. The European Commission proposed at the end of 2025 to postpone part of them; check the current status before any planning.
  • Data Act, Regulation (EU) 2023/2854: applicable since 12 September 2025. It gives users of connected devices access to their data and makes it easier to switch between cloud providers; switching charges are abolished entirely from 12 January 2027. Providers of data processing services must also take measures against unlawful access by third-country authorities to non-personal data.
  • NIS2, Directive (EU) 2022/2555: extends cybersecurity and incident reporting obligations to many more sectors and to medium-sized companies, including supply chain security. Germany transposed it into national law late, at the end of 2025.
  • DORA, Regulation (EU) 2022/2554: applies since 17 January 2025 to banks, insurers, investment firms and other financial entities. It requires management of the risks from ICT third-party service providers, a register of information on all such contracts and exit strategies.
  • Cloud and AI Development Act: in 2025 the European Commission announced a law meant to speed up the build-out of data centers and AI computing capacity in the EU. What obligations it brings, and when, will only be clear once an adopted text exists.

What does data sovereignty mean for using AI?

With AI the question of data sovereignty becomes especially concrete, because every query and every uploaded document goes to the system on which the model computes. With cloud AI that is the provider's data center, often with further subcontractors and frequently with ties to the US.

So it has to be clarified where queries are processed, whether inputs are stored or used to train models, who can see content in support and how long logs are kept. Terms differ by provider and contract, and business offerings often exclude training on customer data. What counts is the contract, not the marketing claim.

The AI Act itself does not prescribe a place of processing. Depending on the risk, however, it requires documentation, logs, human oversight and trained staff. That is easier to demonstrate when it is clear where the AI runs and who has access to it.

Local AI answers these questions most simply: the language model runs on your own hardware, for example on a Sinabox, queries and documents do not leave the network, and there is no provider that could use inputs for training. A middle way is AI operated by a provider with no third-country ties in a data center in Germany, such as the Sinacluster in Frankfurt.

How do you prepare for the new rules?

With an overview of which data sits in which service and who can access it. The following points help with the inventory:

  • Record all cloud and AI services, with location, operator and its parent company.
  • Classify data by protection needs: what may go into a cloud, what only into the EU, what may not leave the building at all?
  • Document the basis for every transfer to a third country: adequacy decision, standard contractual clauses, transfer risk assessment.
  • Check AI contracts for whether inputs are stored or used for training and which subcontractors are involved.
  • Classify AI use cases by the risk categories of the AI Act and train staff.
  • Write down exit strategies: how does the data come back, and how long does switching provider take?
  • Provide a solution without third-country ties for confidential data, on your own premises or with a provider in Germany.

This guide gives an overview of the legal situation and does not replace legal advice in an individual case. Deadlines and legislative procedures change; what counts is the current state.

Ready for AI without the cloud?

Talk to our team about deployment, configuration and an individual quote for your organization.

Benjamin Löffler
Benjamin Löffler
CTO, Sinabis GmbH

Available Monday to Friday, 9:00 to 18:00 (CET).

Frequently asked questions on this topic

Short answers to what is asked most often about this.

The ability of an organisation to decide where its data is stored, who accesses it and which law it is subject to. An EU location alone is not enough; it also matters who operates the service.

Data localisation is an obligation to store data in a particular country or region. There is no general obligation of this kind in the EU, but it often arises from contracts, rules for public bodies or professional secrecy.

Yes, if a US provider possesses or controls the data. The CLOUD Act is tied to the provider, not to the storage location, so it can also cover data in data centers of European subsidiaries.

No. The AI Act sets obligations according to the risk of an AI system, such as documentation, logs and human oversight, but no place of processing. For personal data the GDPR continues to apply alongside it.

Yes, once the legal basis, the data processing agreement, the rules on storage and training and, where relevant, third-country transfers have been settled. With local AI on your own hardware there is no third-country transfer and no provider access, which makes the assessment much simpler.