Nvidia deal with the German government: Who benefits and what is behind it?

The collaboration with the German government builds on existing Nvidia projects with German industry. In particular, it builds on Deutsche Telekom’s Industrial AI Cloud, a computing facility for artificial intelligence that Telekom established in Munich together with Nvidia and data center partner Polarise at a cost of €1 billion. Telekom provides AI and data servers here for businesses, research institutions, and the public sector. (Telekom)

The public-sector collaboration also builds on the SOOFI project (Sovereign Open Source Foundation Models) run by the German Research Center for Artificial Intelligence (DFKI). In addition to the federal government, which is providing €20 million in project funding, the initiative involves various research institutions, including the Fraunhofer Institute, Leibniz University Hannover, Berlin University of Applied Sciences and Technology, and others. The goal is to develop a larger AI language model that will be made available to businesses and society as open source. In August, the project unveiled its first smaller LLM, SOOFI S, with 30 billion parameters and around 27 trillion training tokens. Larger models are expected to follow, potentially as part of a future funding priority. (dfki)

The new memorandum of understanding is expected to focus on two areas: first, industrial AI models and, second, “Physical AI” for robots, autonomous vehicles, and factories.

Specifically, additional AI models are to be developed under the SOOFI project. Funding for the existing model initiative was officially scheduled to end in the summer of 2026, but the project is now expected to continue. Alongside various research institutions, Berlin-based AI company Merantix Momentum contributed to the SOOFI model. In addition, Nvidia is expected to help develop Physical AI in Germany, covering AI for robots, simulations, and autonomous systems.

Nvidia’s main contributions will be its powerful Blackwell chips, along with engineers to support the collaboration, software such as CUDA and related tools, technical expertise, and access to its ecosystem of industry and research partners. The German government is also looking to Nvidia’s open, though not independent, models, such as Nemotron and Cosmos.

According to the company, Nvidia has already invested €4 billion “in Germany.” These investments have so far primarily involved private-sector partnerships with Telekom and SAP (AI data centers), Mercedes (autonomous driving), Siemens (autonomous factories), Agile Robots (robotics), and many others. Nvidia also holds stakes in startups, including n8n (n8n), Neura Robotics (Neura Robotics), and Black Forest Labs (Globenewswire).

The deal with the German government will further strengthen these investments by incorporating the companies into the new projects, with federal funding helping to finance and expand their activities. Nvidia will also boost demand for its own chips, expand the reach of its AI hardware and software platform, and gain access to the industrial AI ecosystem.

German AI development will run on US chips, and Nvidia AI models such as Nemotron and Cosmos will also be used. This is not just about hardware: CUDA and its compatible tools also tie development to Nvidia’s broader software stack.

The problem is that Nvidia is a monopolist and, quite simply, has not only the best chips but also the deepest expertise in AI. That is precisely the expertise the German government wants to acquire for domestic industry, with the aim of using it to build its own models and infrastructure.

The long-term goal is to develop proprietary open-source models, but in the medium term, Germany is tying itself to a US supplier. Whether switching to another provider will be feasible later remains an open question. The hope is that access to Nvidia’s expertise will help close Europe’s substantial gap in AI development. Switching to other chip suppliers is a distant prospect in any case. Dependence on Nvidia is therefore unavoidable, particularly in the early stages, and the deal will deepen it.

The German government plans to support AI development in Germany and promote the creation of alternative AI models that will be made available as open source to industry and the public sector. This requires powerful AI chips, which are expected to come from Nvidia, along with the technical expertise needed to use them. Telekom will provide its data centers for the project, and numerous other companies are involved, including Siemens, Audi, and Black Forest Labs. Funding is expected to go to Nvidia itself for the chips and to Telekom for computing capacity. Various research programs will also receive funding.

Specific figures have not yet been disclosed. Several key questions therefore remain unanswered, including:

• How much is the government actually committing?
• Who will receive which payments?
• Who will own the results of the publicly funded projects, and how will intellectual property rights be handled?
• How will data sovereignty be legally safeguarded?