Germany’s LEAM initiative proposes building a high-performance computing center to support the development of European artificial intelligence models. The feasibility study frames the project as a way for Germany and Europe to strengthen their role in a field increasingly shaped by large technology companies.
Why LEAM says Europe needs its own capacity
The study was prepared by the German AI Association on behalf of the German Federal Ministry for Economic Affairs and Climate Action. It focuses on infrastructure that could help Germany and Europe develop foundation models, the broad AI systems that can underpin many different applications.
GPT-3 is one example of a foundation language model. The source article notes that it gave rise to the Codex AI model and ChatGPT. LEAM argues that these models could enable new applications, platforms and business models across much of the economy and society.
The study says Europe needs to “catch up” or risk losing digital sovereignty. In this context, sovereignty means having the capacity to develop and use important AI systems within Europe, rather than depending entirely on models developed elsewhere. The study team describes foundation models as a technology likely to have a “disruptive effect on the economy and society”.
A market concentrated in the United States
According to the study, 73 percent of foundation AI models come from the U.S. and 15 percent from China. It attributes the pace of development to large technology companies making investments worth billions, and warns that Europe could miss a major shift in how AI is built and applied.
The German AI Association characterizes the concentration as a “monopoly-like cluster.” The study says that digital sovereignty in AI and the applications built on it is acutely threatened, with implications for the competitiveness of the German economy.
At the same time, the survey results cited in the study point to demand within Germany. Two-thirds of AI companies surveyed already work with foundation models, and 71 percent of those companies work with language models. Eighty percent support developing large AI models based on European values.
The proposed computing center
The study’s authors say Germany and Europe need high-performance supercomputing infrastructure, which they describe as currently lacking in Germany. LEAM’s proposal centers on an “AI Service Center” that would provide the computing resources and services needed to develop trusted, open source foundation models aligned with European values.
The study estimates that building the infrastructure would cost 350 to 400 million euros. It classifies the project as feasible in terms of hardware and software. The proposed center would give industry a basis for building its own AI applications on top of the models.
LEAM presents the center as a joint undertaking involving business, science and politics in Germany. The study team says it could serve as a beacon for the field, helping retain AI specialists in Germany and train new ones. That makes the proposal about more than computing equipment: its intended role includes supporting the people and organizations needed to develop and apply AI.
From proposal to European models
The initiative says it is in exchange with OpenGPT-X, Aleph Alpha and Huggingface. The study offers a roadmap for how Germany could participate in the development of foundation models, while the proposed infrastructure would make computing capacity and related services available to researchers and industry.
The case for the project rests on two linked claims in the study: foundation models are expected to become more important, and Europe’s current position leaves it exposed to decisions made by a concentrated group of developers. LEAM’s answer is to build shared capacity that could support open source models and downstream applications shaped by European values.
The full LEAM AI study can be downloaded from LEAM AI, though it is currently available only in German. Its feasibility findings and cost estimate set out the scale of the proposal; the study presents the infrastructure as a route for Germany and Europe to build more of their AI capabilities themselves.