Which European countries are building their own sovereign AI to compete in the tech race?

Three years after OpenAI rolled out ChatGPT and brought artificial intelligence (AI) to the mainstream, several European countries are making their own sovereign systems.

Sovereign AI is a country’s ability to develop, host, deploy, and govern AI systems that are made in its country for its citizens instead of being dependent on foreign systems or cloud jurisdictions.

The European Parliament recognised in a June report that it is “currently heavily dependent on foreign technologies,” specifically American technologies, which prevent the bloc from having its own tech champions. Europe’s reliance “looks set to continue,” the report said, due to the US’s recent $500 billion (€432.9 billion) domestic AI investment.

The EU stated that to regain its advantage, it should invest in research and develop new systems. This is where national governments can come into play.

A handful across Europe are building their own sovereign AI systems. Euronews Next takes a look at what has been built so far.

Germany

Germany is the latest country to announce its own AI plan, called Sovereign Open Source Foundation Models (SOOFI).

SOOFI is an attempt to build a fundamental “advanced AI” open-source model that can be adapted by other companies building AI products, according to the German government.

The technology will be used for highly complex tasks, such as AI-controlled robots, according to a statement from the German government.

“With SOOFI, we are laying the foundation for the next generation of European AI models – sovereign, powerful, and entirely in European hands,” said Wolfgang Nejdl, a professor from Leibniz University Hannover, one of the universities involved in the project.

“Large AI models that respect European values are essential for building trust in AI, especially in sensitive areas such as education, medicine, administration, and production,” he added. ​

Telecommunications company Deutsche Telekom and T-systems said the goal is for SOOFI to have 100 billion parameters or settings that regulate the model’s behaviour.

Both companies are providing technical support for the large language model in one of their AI factories. To train the model, Deutsche Telekom will use around 130 NVIDIA chips and over 1,000 graphics processing units (GPUs) that will be in place by next March.

TU Darmstadt, one of the German universities involved with the project, said SOOFI will also determine what is needed to build expertise in every area of developing large AI models, from collecting and preparing data, to building and training the software.

Switzerland

In September, the Swiss AI initiative launched Apertus, the country’s first multilingual language model.

Apertus, the Latin word for “open,” lets researchers, professionals and the public customise the model for their specific needs.

The developers say everything about the model is open for use, including the training architecture, datasets, source code and model weights, the parameters that instruct the LLM on how to interpret data.

ETH Zurich, one of the collaborating universities, said that Apertus was trained on 15 trillion tokens or pieces of information across more than 1,000 languages, like Swiss German and Romansh.

Apertus was uploaded on Public AI, an online access point for sovereign models, so people can access the model around the world.

The Swiss AI Initiative said it will explore domain-specific models with expertise in law, climate, health, and education.

“This release isn't a final step—it's a beginning,” Antoine Bosselut, the co-lead of the Swiss AI initiative, said on Public AI. “We're building toward a long-term commitment to sovereign, open AI foundations that serve the public good, worldwide".

Poland

In February, Polandlaunched its domestic large language model, Polish Large Language Model (PLLuM).

PLLuM is “tailored to the specificities of the Polish language,” so any AI speech or writing projects will be able to “handle the challenges of inflection and complex syntax very well,” a government release said at the time of the launch.

​The government suggeststhat PLLuM models can be turned into AIs that can help write texts and emails, summarise documents, help students prepare for class, generate chatbot content, plan a trip, or create outlines.

Dariusz Standerski, Poland’s deputy minister of digital affairs, said at the time that PLLuM “is an investment in the digital state”.

​Standerski said at the model’s launch that it will be expanded into Hive AI: a system that will eventually be integrated into the government’s public administration operations and help develop “the national AI ecosystem”.

For example, it means the public will get access to a virtual assistant that will help them get public information, an “intelligent” office assistant that will automate document processing and information retrieval.

Down the line, PLLuM will also be used to help teachers “conduct engaging lessons” with the latest technologies in their classrooms.

Spain

In January, the Barcelona Supercomputing Centre (BSC) launchedAlia, the “first European open and multilingual infrastructure,” which will develop “responsible AI for the service of the people”.

The BSC developed Alia with the help of MareNostrum 5, a supercomputer that can run 314 quadrillion calculations per second.

Alia provides an open databaseof resources, such as datasets, language models and integration tools in Spanish, Basque, Catalan and Galician that will help startups build their own domestic models.

Eventually, the Spanish Agency for the Supervision of Artificial Intelligence (AESIA) said Alia will be developed into a tax agency chatbot and built into an application that will be able to easily diagnose heart failure.

The Alia project also builds on the back of Ilena, another Spanish government initiative that built over 100 AI resources in Spanish, Basque, Catalan, and Galician languages for domestic companies to use.

In 2020, the government of Catalonia launched Aina, a pilot project that generates computer models in Catalan for other companies that want to build AI products like voice assistants, automatic translators or conversational AI agents.

The government trained that model on an initial Catalan database that has 1.7 million words combined into 95 million sentences.

The ​Netherlands

In 2023, three non-profit organisations started developing a Dutch-speaking open-source AI model called GPT-NL.

A website dedicatedto the project describes GPT-NL as a model “for the Dutch language and culture: reliable, transparent, reciprocal and sovereign”.

The consortium is usinga mix of data obtained by copyright agreements from high-quality sources, public data and generating its own synthetic data.

The consortium recently signed a deal with Dutch publishers under the trade organisation NDP Nieuwsmedia and news agency ANP to use their articles for GPT-NL’s training. In return, the publishers get a share of the LLM’s profits when it is eventually released.

The project will also be open source, meaning that academic institutions, researchers and government can try out its applications in health, education, and service fields. Users who aren't using the LLM for professional reasons might have to pay a small fee to access it once it’s available.

The researchers started training the model in June 2025 and are expecting a first version to be available before the end of the year, according to a recent update.

Portugal

Since 2024, a consortium of Portuguese universities has been working on a sovereign AI called Amalia.

The Nova School of Science and Technology, one of the research groups behind Amalia, said it is able to answer questions, generate code, explain concepts, summarise texts and interpret information in Portuguese and with local context.​

So far, researchers have tested the beta version of Amalia in September and are working towards a mid-2026 public release of the AI.

The government is already planning to use this large language model for public administration services through its online portal and in science to help with analysis.

​Local reports said Amalia won’t be available to the public once it is developed as a chatbot, but that the LLM’s code will be open source so other Portuguese companies can use it for their AI models.