NVIDIA Expands Open Models, Infrastructure and Safety Tools for Europe’s Sovereign AI Push

NVIDIA is expanding its open-model portfolio across language, vision, robotics, biology, quantum computing and enterprise agents, tying the releases to Europe’s drive for sovereign AI. The company’s offerings combine downloadable model weights, development tools and local computing infrastructure for organizations that want more control over data, deployment and customization.

The strategy reaches beyond a single model launch. NVIDIA is working with European model developers, cloud providers, telecommunications companies and public institutions, while positioning its Nemotron, Cosmos, Isaac GR00T and Clara families as building blocks for digital and physical applications. The company’s European announcements were made around GTC Paris at VivaTech, where it also outlined partnerships supporting regional AI capacity.

The newest models target distinct operating environments rather than one general-purpose use case. Nemotron 3.5 Super VL is a 120-billion-parameter vision-language model designed for enterprise agents, while Nemotron 3.5 Bolt is aimed at smaller, more efficient deployments, including always-on agents and local computing. NVIDIA’s open-weight approach lets organizations inspect, modify and run models under their own infrastructure requirements.

That range matters for European companies operating across regulated sectors and multiple languages. Smaller models can support targeted functions on local devices, while larger systems handle more demanding enterprise workflows. Earlier NVIDIA releases extended the portfolio into robotics, biomedical research and physical-world simulation, giving developers model families for factories, laboratories and autonomous machines.

Open model access alone doesn't create sovereign AI. European organizations also need computing capacity, cloud services and regional partners capable of training and serving models under local governance rules. NVIDIA announced cooperation with France, Italy and the United Kingdom, along with providers including Domyn, Mistral AI, Nebius and Nscale, to deploy more than 3,000 exaflops of Blackwell systems for sovereign AI.

Germany is also set to receive an NVIDIA AI factory focused on industrial manufacturing applications. European telecommunications companies are part of the broader infrastructure effort, which links model deployment to factories, networks and regional cloud operations. NVIDIA chief executive Jensen Huang said the company invested $8 billion across Europe during the previous year, according to Newsquawk, although that figure is presented in market coverage rather than a detailed financial breakdown.

NVIDIA is pairing its models with software intended to make enterprise agents more useful and governable. The expanded NVIDIA Agent Toolkit connects agents to company data, supports customization, monitors actions and establishes operating boundaries. That architecture addresses a practical barrier to adoption: businesses need agents that can retrieve internal information and act within defined permissions.

The company also introduced the NVIDIA Open Agent Safety Platform, an open software platform and reference design covering agent testing, deployment and security. It brings together industry, researchers and public-sector organizations around shared practices for safer agent operation. The combination of local models, enterprise data access and safety controls gives European organizations a fuller deployment stack, although operational results will depend on implementation, evaluation and oversight by each organization.

NVIDIA’s European push extends into scientific research. A report from Tech-insider says the company joined the European Commission’s AI in Science Enablers Initiative, connected to the RAISE resource, with access to more than 250 models, 250 datasets and 1,400 projects through platforms including Hugging Face and GitHub. The reported program also includes cloud computing and training resources for European researchers.

Those resources complement NVIDIA’s broader releases in biology, quantum computing and scientific data processing. Open models and shared datasets can shorten experimentation cycles for universities, startups and industrial research teams, while regional infrastructure provides the compute needed for larger workloads. The initiative’s significance rests on how easily researchers can reuse the models, data and tooling across national and institutional boundaries.

NVIDIA’s announcements give European organizations a menu of models, infrastructure and governance tools, but adoption will be measured by deployed systems rather than portfolio size. Enterprises must still compare model performance, licensing terms, energy costs, integration effort and compliance obligations before moving workloads into production.

The competitive effect reaches beyond NVIDIA. European model builders gain access to optimized infrastructure and distribution, while cloud providers compete to host sovereign workloads and chip customers receive more reasons to build around Blackwell systems. Developers also gain alternatives to closed services, especially for private agents and industry-specific applications. The next stage will show whether Europe’s sovereign AI ambition produces widely used industrial systems or remains a collection of well-funded platforms and partnerships.