EU supports AI challenge to Silicon Valley and China

The European Union is backing a major AI initiative, OpenEuroLLM, in a bid to counter the dominance of Silicon Valley and China in artificial intelligence.

The initiative is being led by Jan Hajič, a computational linguist at Charles University in Czechia, and Peter Sarlin, co-founder of Silo AI.

“The models that OpenEuroLLM will build will be fully open, to exploit the advantages of Open Source that have been so successful in software development in the past,” Hajič told InfoWorld. This approach will allow the models to comply with EU regulations more easily, he said, adding that so-called open models from outside the EU are either only partially open (for example, just the model weights), or there is no transparency about their provenance, making it impossible to be sure whether they are unbiased.

Sarlin said the consortium’s focus is on enabling European companies to innovate while maintaining control of the AI technology they use.

“Many European businesses are already experimenting with AI, but with proprietary models, they don’t own the technology they build upon. OpenEuroLLM provides the foundation for companies to develop AI solutions they truly own and control, while working within European frameworks for responsible AI development,” he said. “Hopefully, the project can serve as an inspiration for a regulatory framework that embraces open source.”

The consortium behind the OpenEuroLLM project brings together 20 leading European research institutions, companies, and high-performance computing (HPC) centers to develop multilingual, open-source large language models tailored for European businesses and public services.

The project has received funding from the EU’s Digital Europe Program, and became part of the EU’s Strategic Technologies for Europe Platform (STEP) on Monday.

The EU’s three-year grant to fund the project is small for the moment — around €20.6 million. The consortium has €37.4 million in total funding, and additional contributions in the form of computing resources, according to the European Commission.

The biggest challenge for a project like this is access to computing capacity, Hajič said.

“There is no single place in Europe to have such a large computing power as in the U.S. or China. However, there are several large facilities available in the EuroHPC network which we will use. There is more computing power coming, such as in the AI Factories program funded by the EU. There are technical challenges related to distributed computing, especially for LLM training, but we are confident we can do it,” he said.

Another challenge is making multilingual LLMs that can serve all EU member states equally.

“We want to have comparable quality for all European languages, large or small, which is not the case today due to the scarcity of data in so-called low-resource languages. This is also a substantial challenge,” he said. “We are building on the expertise of all the partners — academic as well as companies — acquired in previous smaller projects.”

The EU has 24 official languages, several of them spoken by less than 1% of the EU’s population of around 450 million.

In addition to Silo AI, the private AI lab acquired last year by AMD for $665 million, four other companies are involved in the project: Aleph Alpha and Ellamind of Germany, Spain’s Prompsit Language Engineering, and LightOn of France, which recently became Europe’s first publicly traded generative AI company.

“The consortium aims to develop open-source multilingual and high-performance language models that will support the competitiveness and digital sovereignty of Europe,” LightOn said in a statement.

“Europe has the talent and resources necessary to take a leading position in this global AI competition,” Laurent Daudet, co-CEO of LightOn said in the statement. “To transform these efforts into a real strategic lever, Europe must not only capitalize on the AI Act, a true catalyst for innovation towards trustworthy AI, but also support a coordinated approach from its leaders. This is now possible thanks to the OpenEuroLLM consortium.”

Competing in a shifting AI landscape

The rise of China’s DeepSeek, with an open-source AI reasoning model that has disrupted the global AI market, has intensified the urgency for Europe to develop its own competitive AI ecosystem.

OpenEuroLLM seeks to address this challenge by developing AI models that are transparent and allow for community involvement. The initiative aims to ensure that AI models, software, and datasets remain fully open, enabling businesses to customize them according to their industry-specific needs while preserving linguistic and cultural diversity.

Europe’s AI future: a high-stakes battle

The EU’s AI ambitions come at a time when American and Chinese companies are racing to establish dominance in generative AI. While OpenAI, Google, and Anthropic lead AI development in the US, China’s DeepSeek has gained attention for its cost-effective and high-performance models.

With OpenEuroLLM, European leaders are betting that an open, collaborative approach can provide businesses with AI solutions tailored to their needs while ensuring Europe remains competitive in the AI race.

“This project is clearly contributing to Europe’s digital sovereignty. The AI regulations and standards are already in force, and we believe this project will test them to the limits,” said Hajič.

But even when the project runs up against those limits, that’s not the end of the story, he said. “If we see that there are rules that block us from bringing affordable, high quality models to the market, we will certainly work with the European Commission on amending them.”