The Mistral AI co-founder began by making language models portable. Three years later, his larger argument is about who gets to own the intelligence, the machines beneath it and the right to walk away.
Before Mistral AI, Guillaume Lample taught machines to translate without bilingual text, solve equations and write code. Then he helped make smaller language models matter - and built a company around the same stubborn equation.
DeepSeek did not win attention by building the biggest machine. It made scarcity into an engineering brief - then turned a research paper, an open model and a free app into a very expensive question for the rest of the AI industry.