Skip to content

Sovereignty and law · 5 min read · As of November 2026

Data residency and sovereignty: two different questions

The server location shows where your data sits. Who controls the model, the data flow and the fallback behaviour is a second question, and it needs its own evidence.

Data residency describes where a server physically stands. Technology sovereignty describes who owns the software on that server, who operates the language model and who decides on every single cloud call. In the market, both questions are often answered with one sentence. A provider can host in Germany and still build entirely on foreign technology and foreign models.

At a glance

  • A data centre in Germany says nothing by itself about where the technology running in it comes from.
  • Technology sovereignty means you know who decides on the model, the data flow and what happens when something fails.
  • For text and chat, the core language model can run on your own infrastructure. For images, video and speech, cloud components remain where they demonstrably solve the task better. What matters is that this choice is made visibly, task by task.

Why “hosted in Germany” no longer reassures anyone

In the ERP field, quite a few vendors now advertise hosting in Germany. For some of them, the underlying technology comes from the United States and runs on the cloud infrastructure of a US group. The server stands in Germany, while the software above it has its origin and its operational control elsewhere.

That leaves the claim “hosted in Germany” with little power to tell offers apart. It answers a question that has lost weight for the actual decision. The question that matters sits one level deeper: which model processes the data, and who decides when that model reaches its limits or fails?

The two levels compared

Criterion Data residency (hosting location) Technology sovereignty
Question answered Where does the server stand? Who owns what runs on the server?
Usual evidence Location statement, data centre certificate Origin and operational control over model, data flow and dependency chain
Typical gap The system runs in a German data centre but is based on technology developed and operated abroad The model layer is self-operated, but without a log that remains a claim
Enough as sole evidence? No, because it says nothing about the technology above it Only together with a traceable log

Both levels are legitimate questions. It becomes a problem when the answer to the first is offered as the answer to the second.

How technology sovereignty can be demonstrated

The difference between a claim and evidence shows in one concrete case. If an existing system is tied to a third-party cloud AI, that connection can be replaced by a private, self-operated infrastructure without changing a single calling component in the system. That is a technical step an auditor can see at the interface itself. They place a test call against the original cloud connection next to a test call against the private replacement and find the same interface with a different source of answers. The evidence lies in the match of the interface.

The log as evidence

This replacement works because all language model calls run centrally through one single instance. That makes it possible to record every call: time, calling component, model used, token consumption, response time, success or failure. If a self-operated model fails, a fallback takes over, and it shows up in the same log.

The model layer also keeps a log that shows when a swap happened. How we built this on Beyonetix OS is described on the platform page. The full reasoning is set out under sovereign infrastructure.

The limit lies where deliberate cloud components remain

The core for text, retrieval and chat can be self-operated. For images, video and speech, a reasoned mixed strategy with cloud components can make sense, because measured results differ from task to task. A blanket answer would be worse than a reasoned one.

Freedom of choice is the core. Whether a task runs on your own infrastructure is decided per task type and can be traced in the log.

Frequently asked questions

Does a server in Germany automatically mean the technology behind it is European?

No. A hosting location says nothing about the origin of the software, the model or the operational control above it.

Does a language model hosted in the EU automatically fit the GDPR?

The hosting location is a necessary condition, and it must be combined with control over the model. What also counts is who controls the model, whether data leaves your own network at all and how that is logged. The criteria are set out in the article on AI and the GDPR.

Can technology sovereignty be checked from the outside?

Yes, through the log of every single model call, with time, model, source of the answer and failure behaviour. A provider who cannot show this log stays with a claim.

Does sovereignty mean no cloud AI at all?

No. For text and chat, a fully self-operated path is possible. For images, video and speech, deliberate cloud components can remain because they give the better result for that task. What matters is that the choice is visible and made per task.

What does the EU AI Act have to do with the difference between location and technology?

The EU AI Act regulates the behaviour and risk class of an AI system and does not ask about the server location. A system can look fine on location and still raise questions if it does not log in a traceable way which decision it made when.

Questions on location and technology?

We discuss with you which evidence your provider should give on model, data flow and failure behaviour.