IDM on Heise: One Sovereign Medical AI for Europe

On April 24, 2026, Heise online published a deep-dive on clinical AI development. Title: Hospitals Are Building Their Own Language Models. In conversation with Isabel Grünewald, our CEO Nils Schweingruber explained why IDM exists. And why our goal is precisely that not every hospital in Germany has to build its own language model.

The real question

The Heise headline describes a real status quo. University hospitals are building their own models because the market hasn't given them a sovereign alternative. That's understandable. It's not a target state.

Germany has roughly 1,900 hospitals. Nobody wants a world in which each of them trains, runs and certifies its own language model. That's neither financially viable nor clinically sensible.

The question isn't how a single hospital builds its own AI. The question is who in Europe builds the best medical AI, so that everyone else can use it.

What we do

We're building the best medical AI for Europe. In European hands, on European data, on German infrastructure. So hospitals can deploy sovereign AI without having to develop it themselves, but at the quality they expect.

On data. Patient data doesn't belong in a US cloud. Period. Anyone practicing medicine in Germany has to be able to train in Germany and host in Germany.

On the medicine. Generic language models don't understand medical vocabulary. Quality is what matters. Training on 7 million de-identified patient records produces something fundamentally different from training on Reddit.

ORPHEUS: born at UKE, deployed across Germany, soon across Europe

ORPHEUS has been running at UKE since early 2025. Available to all 15,000 staff. Now also in productive use at 4 university hospitals, 30+ additional clinics and 200+ outpatient facilities. Every day.

That's the point Nils keeps making in the podcast: we are not in the demo phase. And we are not 30 parallel island solutions. We are one model that scales and that everyone can use.

What else came up

ARGO. Our smart clinical workspace supports clinical documentation, especially the discharge letter. The medical summary of an inpatient stay. That's where hours of physician time are burned every single day.

ISO 13485. We're pursuing certification as a medical device manufacturer. The prerequisite for our models to be officially deployed in patient care. A bar that not every hospital can clear on its own.

Open source. We deliberately publish components of our stack. Sovereign AI only works as shared infrastructure. Not as a black box.

Cloud for smaller practices. Practices without their own data center should still be able to use sovereign AI. We host it. In Germany.

Sovereign means shared

Sovereignty in medical AI doesn't mean every hospital has to run its own model on its own server. Sovereignty means that data, models and infrastructure stay in European hands. Shared. Auditable. Clinically validated.

That's what we're building. That's what we exist for. We all work together to improve one AI that helps each of us in everyday clinical work.

More

• Heise deep-dive (24.04.2026): Hospitals Are Building Their Own Language Models
• Heise Digital Health Podcast: UKE spin-off builds AI language model for hospitals
• ORPHEUS: Our speech recognition in detail
• ARGO: Clinical LLM for discharge letters

Questions, partnership ideas, pushback: [hallo@idmedizin.de](mailto:hallo@idmedizin.de)