The Bavarian State Ministry of Science and the Arts (StMWK) announced on 23 April 2026 (https://www.stmwk.bayern.de/pressemitteilung/12996/pm.html) the launch of a unique European AI foundation model initiative. Backed by 54.5 million euro for AI high-performance computing and 28 positions for early-career scientists, the programme unites 48 leading researchers across eleven Bavarian universities. FAU, and within it the Department Artificial Intelligence in Biomedical Engineering (AIBE) together with the Profilzentrum Medizintechnik and the National High Performance Computing Center NHR@FAU, take on a major role in the health cluster of the initiative. The health workstream, AI-BAY-HEALTH (https://ai-bay-health.github.io/ai-bay-health-site/), develops a sovereign, multi-modal foundation model for medicine. The system converges imaging, pathology, omics, biosignals, clinical text and structured EHR data into a shared 8192-dimensional embedding space. The scientific aim is explicitly fundamental: we want to understand which emergent properties arise when very different medical modalities are aligned at scale, how disease signatures express themselves across imaging, molecules and free-text reports, and how far cross-modal reasoning generalises to unseen combinations of inputs. AI-BAY-HEALTH is, first and foremost, a research instrument for medical science, not a diagnostic product.
This framing is deliberate. Universities are the right place to do the high-risk, exploratory work that uncovers new representations and biomedical insight, but they are not the right place to carry a medical device through regulatory approval. The path AI-BAY-HEALTH chooses is therefore an open one: the resulting multi-modal foundation models, encoders and interfaces are intended to be made publicly available, so that Bavarian and European start-ups, SMEs and clinical partners can build certified, regulated products on top of them. The consortium produces the scientific substrate; translation into MDR or IVDR-approved tools is the role of the downstream ecosystem.
At FAU, the work is led by Prof. Bernhard Kainz (AIBE) as principal investigator, supported by Prof. Andreas Maier, Prof. David Blumenthal, Prof. Florian Knoll, Prof. Sivlia Budday, Prof. May, Prof. Schett and the wider team of the Profilzentrum Medizintechnik. Over the next 15 months, a focused team of four researchers will wrangle normative foundation models from paired CT, MR and clinical report data, drawing on many thousand hours of GPU compute on the H200 clusters at NHR@FAU. The Erlangen contribution feeds directly into the consortium’s shared encoder and fusion stack, and benefits from the close clinical link with Universitätsklinikum Erlangen, which grounds the modelling work in real patient pathways even while the research itself stays pre-clinical.
The initiative is funded by the StMWK under the Hightech-Agenda Bayern, with the explicit goal of building sovereign AI infrastructure that keeps sensitive medical data and the resulting model weights under Bavarian, public-sector control. For AIBE and the Profilzentrum Medizintechnik, the project consolidates Erlangen’s position at the interface of medical imaging, machine learning and clinical research, and provides a concrete, open platform on which the next generation of Bavarian medical AI start-ups can build regulated products.
