Sovereign AI means you can answer three questions about your own system. Where is the data? Who could compel someone to hand it over? And can you check both, rather than being told?
Everything else in this post is detail. It's written for the person who signs, without acronyms you have to look up. The precise version for evaluators is on our page about sovereign AI.
- Sovereign AI means you can answer three questions about your own system: where the data sits, who could be compelled to hand it over, and whether you can check both yourself.
- Most offers pass the first question and fail the third, because the architecture leaves the customer nothing to check.
- Whether a US provider, RAGSuite Engine or your own servers: decide by your documents, and ask for evidence behind every answer.
Sovereign AI explained: the three questions
The three questions, in order
Most can answer this one.
When an employee asks the AI about a contract, two things travel: the question, and enough of the contract for the AI to answer. Both end up on a computer. Most companies can say where that computer is, such as "in our provider's EU region". That's an answer. It settles less than you'd think.
This is where the field splits.
The company operating the software is subject to the law of its home country. If that law lets authorities demand data the company holds or controls, the building's location isn't decisive. The US CLOUD Act works this way. The practical question: which company runs the service, under which law, and is there a parent company anywhere that changes the answer?
This is where nearly everything fails.
Usually not because of dishonesty. It's the architecture: if the software runs on someone else's computers, your assurance is a contract and a certificate. Both have value. Neither is the same as watching what your own system does.
In shortWhere is the data, who can compel disclosure, can you check? The third question decides.
What does sovereign AI mean for your business?
The market has noticed. In the Bitkom Cloud Report 2026, a survey of 603 companies in Germany with at least 20 staff, 87% said the "sovereign" offers of international providers don't give them enough insight into how data is processed and who can access it. The graphic above shows the key figures.
That isn't a complaint about vendors being foreign. It's a complaint about not being able to see. So what sovereign AI means is decided less by the label than by the proof: is the word attached to something you can check?
What you're assured of
- a contract about where data is stored
- a data centre certificate
- a statement that nothing is sent
- a promise that export is possible
What you can check yourself
- the servers everything runs on
- the legal system that applies to you
- a network capture showing what is sent
- an export you ran yourself
In short87% lack transparency. For every promise, ask: what of this can I check myself?
Three ways to run AI, and their answers
There are three ways to run AI on your own documents. Each answers the three questions differently:
| Question | US provider, EU region | RAGSuite Engine | Your own servers |
|---|---|---|---|
| Where is the data? | data centre in the EU | data centre in Germany, run by RAGSuite | with you or a host of your choice |
| Who can be compelled to disclose? | the provider, under US law | check the operator and sub-processors in the data processing agreement | no third party keeps your data |
| Can you check it? | through contracts and certificates | through the contract, sub-processors and export | yourself, down to a network capture |
Self-hosting means the software runs on computers your company controls. Your own servers, or machines you rent and administer. Nobody else logs in. You can watch what the software does on the network, and switching is a configuration question, not a migration project. With a local language model, no request leaves your network.
RAGSuite Engine is the middle way: we run the same knowledge engine for you in a data centre in Germany, as a processor under contract. You save yourself the operations and give up part of the ability to check. If you choose a language model from an external provider there, the question and passages go to that provider. So ask us the same questions as anyone else.
In shortUS provider, RAGSuite Engine or your own servers: each way answers the three questions differently.
What does sovereign AI cost? The part vendors skip
On your own servers, someone has to run the system. That means a named person or team who owns updates, backups, monitoring and access rights. Security patches go in on a schedule, not silently by a provider. And the first serious incident is yours.
For a company that already runs its own systems, that's incremental: the same people, one more system. For one that has outsourced everything, it's a real new commitment, and then RAGSuite Engine is often the more honest start. Decide that before the pilot, not three months after. What it means in euros, we work through in AI costs: RAGSuite Engine or your own server?.
| Classification | Examples | Setup that usually fits |
|---|---|---|
| Public | product docs, website, FAQ | RAGSuite Engine, also with an external language model |
| Internal | manuals, processes, policies | RAGSuite Engine or your own servers, model after review |
| Confidential | personnel files, contracts, health data | your own servers with a local model, offline if needed |
In shortYour own servers need a team to run them. Classify documents and decide per project.
Five questions for any vendor
Including us. A vendor who can answer all five in a technical review is describing something real. One who dodges more than one is describing a brochure.
Every point ticked. Well prepared.
The last question matters most. If the same software also runs on your own servers, switching vendors is a move, not a new rollout. With RAGSuite that's the case: RAGSuite Engine and your own servers use the same platform.
In shortFive questions separate real answers from brochures: the most important: could we run it ourselves?
The short version
Sovereign AI isn't a product category or a certificate. It's a property of your setup that you can describe in three sentences: our data is here, this entity controls it under this law, and this is how we'd check both.
If a vendor's answer to any of the three is "trust us", that isn't a moral failing. It's a different architecture with different consequences. Price those consequences into the decision rather than finding them at renewal.
Our page on on-premise AI shows what running it on your own servers looks like. For more on who may access your data, read our post on the US CLOUD Act. Check individual vendors with the AI provider location check.
In shortThree sentences are enough: the data is here, this entity controls it, this is how we check.
How sovereign is your AI setup today? Check it with the free data sovereignty check, or in a call with Jürgen.
Frequently asked questions
Is sovereign AI just a marketing word?
It's becoming one, which is exactly the problem. In the Bitkom Cloud Report 2026, 87% of companies said they lack transparency in the "sovereign" offers of international providers. The answer isn't to drop the term but to insist it comes with something you can check: a network capture, a jurisdiction, an export.
Isn't it enough that the data centre is in Germany?
Not on its own. Where data sits is residency. Who controls it and can be compelled to disclose it is sovereignty. A provider subject to another country's law can be obliged to hand over data it holds or controls, whichever data centre it sits in. Location is necessary but not sufficient.
Do we need to run our own language model?
Not necessarily. What works is deciding per project based on the documents: an external model where content is public or low-sensitivity, a local model where it's confidential. Classifying your documents once up front saves many later discussions with data protection and the business departments.
How do I check a vendor's claims without a technical team?
Ask for evidence rather than assurances, and let someone technically experienced assess the answers, in-house or external. What matters: which company runs the service under which law, what connections the software makes, who can access the data, and whether a full export has ever been run.
Is RAGSuite Engine sovereign?
It answers the first question clearly: documents, search index and histories sit in a data centre in Germany, run by RAGSuite. For the second and third questions, check us like any vendor: through the data processing agreement, the sub-processors and an export. Only your own servers give you full control.
Sources
- Bitkom: Cloud Report 2026 (press release, 17 June 2026): 603 companies with 20+ staff
- 18 U.S.C. § 2713: Required preservation and disclosure: why provider jurisdiction decides
Jürgen advises companies on introducing AI on their own infrastructure and leads RAGSuite at NITSAN. He writes about sovereignty, strategy and costs.