Mistral raises $2B at $13.5B valuation as Europe seeks AI sovereignty

Mistral, a Europe-based AI lab, is capitalizing on geopolitical tensions and safety concerns surrounding proprietary models from American companies. CEO Arthur Mensch argues that open-weight models, which the company largely publishes under open source licenses, prevent unchecked corporate power and ensure uninterrupted access to AI technology. This stance has gained traction as the Trump administration restricted distribution of models from Anthropic and OpenAI in June, and subsequent incidents revealed safety risks with closed-weight systems. According to Mensch, the alternative to open source winning is "a pretty dark world."

The company's financial position has strengthened significantly. Last September, Mistral raised almost $2 billion at a $13.5 billion valuation, and it is reportedly preparing another raise that would increase its valuation to $23 billion. Revenue has reportedly grown twenty-fold in the past year, bolstered by deals with the French government, Microsoft, HSBC, and others. Andrea Renda, director of research at the Centre for European Policy Studies, notes that the EU's push for technological sovereignty and US hostility create a favorable environment for Mistral, despite its performance not being spectacular.

Mistral's strategy addresses growing concerns about AI supply chain security. Mensch compares AI to energy, emphasizing the need for diverse sourcing to prevent any single entity from cutting off access. The return of Donald Trump to the White House has made this argument more compelling, as the US government demonstrates willingness to leverage its AI capabilities against trading partners. "More and more, AI is understood as a major vector of power," Mensch told WIRED. Nicolas Granatino, founder of startup accelerator StemAI and a personal stakeholder in Mistral, argues that participation in the open source ecosystem reduces leverage for the US and China.

Mistral has shifted its business model to focus on smaller, customized models for industries like manufacturing, utilities, and financial services, moving away from the race to superintelligence pursued by American labs. The company has developed a cloud business and a team of engineers who embed within client organizations, making monetization of open-weight models more viable. Granatino notes that this product emergence eases the open source commitment, as revenue can come from running infrastructure and helping clients customize models with their own data.

Distillation, the process of training a lesser AI model on the outputs of a more capable one, is eroding the performance advantage of proprietary models from American labs. Neil Lawrence, a professor of machine learning at the University of Cambridge, suggests that stopping distillation seems difficult. For open source-focused companies like Mistral, this is less problematic because their models are already accessible. The market share of open-weight models appears to be rising steeply, driven by rapid adoption of Chinese models like DeepSeek, though gaps in publicly available data complicate the picture.

Mensch claims that revealing the possibility of building AI systems outside US control is changing the market structure. As more businesses turn to open-weight models, the stranglehold of American labs is loosening. Mistral's position at this juncture, whether through foresight or fortune, highlights a shift in the AI landscape toward decentralized and accessible technology.