GenAI@KIT Wins Digital Sovereignty Idea Contest

One of three winning projects: GenAI@KIT demonstrates, based on the seven HFD theses, how digital sovereignty can be achieved in practice

GenAI@KIT is one of the three winning projects in the “Digital Sovereignty” ideas competition organized by the Higher Education Forum on Digitalization (HFD). Out of 19 submitted projects, our entry was recognized alongside two others.

The project brings together KIT’s activities in the field of generative AI and combines technical infrastructure with training and support for practical application. It focuses not only on what is possible with generative AI, but also on how universities can shape AI in a sovereign and responsible manner. The jury particularly highlights GenAI@KIT’s holistic approach: “All relevant aspects related to the focus on AI are addressed. Clear, comprehensive, sensational!”

We are delighted to receive this award and that the work of GenAI@KIT is gaining visibility beyond KIT. The competition also offers the opportunity to exchange ideas with other universities on concrete approaches to digital sovereignty and to learn from one another.

<Text is generated, please wait...>
An Overview of the Three Core Elements of Digital Sovereignty at GenAI@KIT and the 7 Theses for the HFD in Practice

Three Core Elements of Sovereignty

At KIT, we are working together with the Scientific Computing Center ( SCC ) and the Digital Office ( DO) to strengthen digital sovereignty in generative AI. To this end, we combine three pillars—infrastructure, expertise, and governance—which only enable sovereignty when they work together.

None of these pillars stands alone: a sovereign infrastructure without expertise will not be used responsibly; expertise without governance gets lost in individual pilot projects; and governance without its own infrastructure remains dependent. Only their interplay creates the foundation for using generative AI at KIT in a self-determined, responsible, and sustainable manner.

Core Element Core Question What GenAI@KIT Achieves
Infrastructure Sovereignty Can we operate, control, and switch? Proprietary hardware, open-source stack, proxy architecture, on-premises and cloud models 
Competence Sovereignty Can we understand, reflect, and make decisions?  Training, model transparency, AI guidelines, critical reflection 
Governance and Economic Sovereignty Can we steer, budget, and operate with sustainability?  Decentralized budgeting, shared decision-making structures, transition to routine operations 

The HFD’s 7 Theses at a Glance

Our core elements concretely implement the 7 Theses; here is the key finding from the submission for each thesis: 

  • Thesis 1: Plan control strategically 
    The Proxy LiteLLM as a control point; model diversity from local models (Mistral Small 4, DeepSeek V4 Flash, GLM-5.3) and EU cloud models; sensitive data runs exclusively locally.
  • Thesis 2: Avoid Project-Based Logic 
    The transition to routine operations is planned from the outset; the educational chatbots have been in routine operation since the summer term of 2026. Formalizing this as a permanent task is part of this process; a first step has already been taken with the introduction of budgeting.
  • Thesis 3: Organize cooperation and governance fairly 
    Three central units with shared decision-making structures: The SCC operates the infrastructure, the Digital Office is responsible for strategy and guidelines, and the ZML oversees training and educational integration. The AI competency module is available under an open license, with federation among universities in Baden-Württemberg
  • Thesis 4: Making Sovereign Alternatives Suitable for Everyday Use
    Over 5,000 active users of the AI Toolbox; the monthly active user base grew by more than 65% in seven months; Mistral Small 4 is set as the default and accounts for nearly 49% of queries.
  • Thesis 5: Strategically leveraging public administration
    Local models enable targeted use cases in public administration, such as AI-supported meeting minutes in the KIT Senate and a “build vs. buy” decision-making framework for AI translation.
  • Thesis 6: Retaining Expertise In-House 
    Training prior to access to the AI toolbox and Copilot; a German/English competency module compliant with the EU AI Regulation; a webinar program; and workshops conducted directly with specialized departments.
  • Thesis 7: Translating Awareness into Action
    AI guidelines (Executive Board, June 2025) and AI strategy (Executive Board, June 2026) adopted through participatory processes; cost transparency and budgeting: $5 budget per person per month since July 2026; publicly documented architecture: hardware, stack, and the current model list are available on the SCC’s AI Toolbox page.

What Other Universities Can Adopt

  • Complete open-source stack (Open WebUI, LiteLLM, vLLM, LMCache, SearXNG) with no licensing costs for the infrastructure
  • Proxy architecture as a control point, reproducible independently of KIT hardware
  • CC-licensed AI competency module in German and English, aligned with the EU AI Regulation

Next Steps

The HFD is bringing the award-winning projects together with partners from its network to scale these ideas to other universities. We’re picking up where they left off: Anyone wishing to reuse the stack, the competency module, or individual components can find all documented elements on the pages linked below.

The process also continues beyond KIT: On the same day that the HFD announced the award winners, the ministries of science in Baden-Württemberg, Bavaria, and North Rhine-Westphalia announced the COSAI (Collaboration on Sovereign and Efficient AI-Infrastructure) project, a cross-state architecture for a sovereign AI infrastructure for universities, in which KIT is participating alongside the University of Tübingen.

We look forward to further use, building on this work, and continuing to shape the future together: Digital sovereignty at universities is a process that we can shape together.

Links