Where Your Data Goes: Choosing AI Tools Hosted in Switzerland or the EU

Key Takeaways

• For personal data, Switzerland or the EU is the simplest choice, because all EU and EEA states count as having adequate data protection.
• Server location alone isn't enough: support access, subcontractors and the law the provider is subject to matter just as much.
• Ask where the AI model processes your inputs, not just where the tool stores your files.
• Every tool that sees personal data needs a data processing agreement and clear rules on how your inputs are used.
• Match the tool to how sensitive the data is: not every task needs the strictest solution.

For most Swiss SMEs, an AI tool that processes data in Switzerland or the EU is the simplest choice. All EU and EEA states are on the Federal Council's list of countries with adequate data protection. But server location is only one of several questions. Just as important are who can access the data and from where, which other providers are involved, which law the provider is subject to and what it does with your inputs.

This guide is not legal advice. It reflects the position as of October 2026.

Why does the storage location matter at all?

Because the Data Protection Act attaches conditions to disclosing personal data abroad. Under Article 16 FADP, personal data may go abroad if the Federal Council has found that the destination country ensures adequate protection. Without such a decision you need suitable safeguards, such as standard contractual clauses recognised by the FDPIC, or one of the exceptions in Article 17 must apply, for example explicit consent. Disclosure covers both sending data and making it accessible. And you have to tell the people concerned which countries are involved, usually in your privacy notice. You'll find the law on fedlex.admin.ch.

For data with no personal reference, such as your cost calculations or a plan, the FADP doesn't apply. There the issues are trade secrets and contractual confidentiality, which are also good reasons to choose tools carefully.

Which countries offer adequate data protection?

The list is in Annex 1 of the Data Protection Ordinance. It includes all EU and EEA states, the United Kingdom and a number of other countries. The United States has been on the list since 15 September 2024, but only for companies certified under the Swiss-US Data Privacy Framework, so check whether an American provider is certified. For all other countries you need additional safeguards. The FDPIC explains this on its page on cross-border transfer of personal data. The list can change, so look at it afresh when you choose a tool.

Is a server in Switzerland or the EU enough?

Not always. A data centre in Zurich or Frankfurt is a good start, but it doesn't answer every question. Pay particular attention to these points:

• Access from abroad: if the provider's support or engineering staff can reach your data from another country, it is made accessible there. Those countries belong in your assessment too.
• The AI model: many AI tools store your files in Europe but send your inputs to another provider's language model for processing. Ask where that processing happens.
• Subcontractors: ask for the list of subcontractors and their locations, and find out how you'll hear about changes.
• The provider's jurisdiction: the FDPIC's guide to transfers abroad explicitly asks whether data is processed by companies subject to the law of a third country, giving American cloud providers with servers in Switzerland or the EU as an example. Access by authorities under foreign law therefore belongs in your risk assessment.
• Copies and logs: backups, logs and support tickets may sit in different places from the main data.

What should you look for when choosing an AI tool?

Seven questions that every provider should be able to answer before personal data goes into its tool:

1. Data location.Where is your data stored and where is it processed, including processing by the AI model? Can you lock the tool to a Swiss or EU region?
2. Data processing agreement.Is there an agreement stating that the provider processes data only on your instructions, guarantees data security and brings in subcontractors only with your approval?
3. Your inputs.Does the provider use inputs, files or answers for its own purposes, such as training models? Is that switched off from the start?
4. Retention and deletion.How long are inputs, files and logs kept, and can you delete them yourself?
5. Accounts and access.Are there business accounts with central administration, roles and two-factor sign-in?
6. Security and evidence.Which security measures and certifications can the provider show, such as ISO 27001?
7. Exit.Can you export your data, and will the provider confirm deletion when you cancel?

Get the answers in writing, ideally in the contract or the provider's documentation, and keep them with your list of approved tools. That way you can show later why you chose a tool, and you'll notice when its terms change.

What about professional secrecy and confidentiality?

Here the right location isn't enough on its own. The FADP only allows data to be handed to a processor if no legal or contractual duty of confidentiality forbids it. Lawyers, notaries and auditors, together with their staff, are bound by professional secrecy under Article 321 of the Swiss Criminal Code. Fiduciary firms and many other service providers are also often contractually bound to confidentiality. Whether and on what terms a cloud tool can be used for such data is something to clarify against your professional rules, with your association or with a specialist. For law firms, AI in law firms goes into more depth, and for fiduciaries see our fiduciary page.

Switzerland, the EU or self-hosted: what are the options?

Broadly, there are three options:

• Large AI platforms with a Swiss or EU data region: familiar tools, often available in business plans. Check especially where the model runs and which law the provider is subject to.
• Swiss or European providers: shorter legal distances and often contracts under Swiss law. Still ask which model works in the background, and where.
• Self-hosted models on your own or rented infrastructure in Switzerland: the most control, but more work for operation, updates and security, and the quality may differ from the large models.

Which option fits depends on your data. Many businesses sort it into three classes: non-critical content with no personal data, internal information with ordinary personal data, and confidential or sensitive data. For the first class a mainstream tool is often enough. For the last you need a controlled solution, or no AI at all. Our AI policy with a model outline shows how to turn these classes into rules for your team.

An example: a construction company with 40 employees uses a mainstream AI tool with an EU data region for quote texts, site reports without personal data and internal notes. For personnel files and correspondence with clients it uses a solution that processes data in Switzerland under a clear contract. Both tools are on the list of approved tools, with examples of what goes where.

If you want to connect AI tools to the systems you already use, our automation and integration work builds these questions in from the start. The legal background is covered in AI and data protection in Switzerland.