A government that wants to use AI for citizen services has a problem: sensitive data cannot end up on someone else's servers. That is why NVIDIA is pushing open models, whose weights can be downloaded, examined and run on infrastructure the state controls. In its latest update, the company gathered European public-sector projects adopting them, from the European Commission to Polish employment services.
According to NVIDIA, the benefits are threefold: a government can evaluate the model, adapt it to its language and services, and apply its own safeguards. Some models go further and publish their training datasets too, so administrations can check whether data was licensed, whether it carries bias, and whether it complies with regulation.
Image: joshuaberkowitz.us — https://joshuaberkowitz.us/blog/news-1/nvidias-open-models-accelerating-ai-innovation-across-industries-1651
Brussels and Warsaw
Berlin-based deepset is expanding the platform underpinning AI@EC, the service the European Commission uses to build and run AI applications on its own infrastructure. The update lets teams compare operating costs, evaluate answer quality and trace problems. Existing applications include document summarization, knowledge management and multilingual policy analysis.
In Poland, Deloitte is helping the country build agentic systems for healthcare and employment. The eHealth Centre is building a medical imaging platform that helps physicians analyze X-rays, CT and MRI scans, flagging abnormalities and prioritizing urgent cases. The Ministry of Family, Labour and Social Policy, with the same support, is developing an agentic factory for more than 400 labor offices, running models such as PLLuM and GPT OSS for document search, summarization and digital assistants. NVIDIA says work with SpeakLeash in Poland and the University of Ljubljana in Slovenia will also bring localized datasets.
UK and Germany
A London company has released a dataset of one million synthetic personas reflecting the UK's demographic and geographic diversity while reducing privacy risk. It can help check whether a model performs consistently across communities, for example rural and urban populations.
Image: aimagazine.com — https://aimagazine.com/news/nvidia-open-models-analysis
In Germany, Sinfonie is an on-premises platform combining generative AI, voice interaction and 3D avatars, letting public employees query institutional knowledge by voice without data leaving the organization's infrastructure. For classified work, four companies including secunet and NVIDIA are developing a sovereign cloud offering built on components already authorized for classified information. A German dataset of one million records, built on the 2022 census across all 16 federal states, is also planned.
France and Greece
In France, the public purchasing body CANUT selected a sovereign collaborative workspace built on Twake.ai, accessible without a separate tender to more than 2,500 organizations including municipalities, schools and healthcare institutions. Greek company KIEFER is building an agentic factory on NVIDIA DGX SuperPOD with Sophea open weight models.
The picture is of a Europe trying to build public AI capacity without relying on outside providers for data and compute. Observers will see whether these projects remain experiments or become everyday services for citizens.
No comments yet. Be the first!