The initiative must value the young company at 240 million euros, will be one of the largest in France for artificial intelligence.
Mistral AI, a French artificial intelligence (AI) start-up created by industry leaders, will raise nearly 100 million euros, immediately becoming one of the leading European players, said one of its investors, confirming information from the Point and Echoes (articles reserved for subscribers).
This collection, which will value the young company at 240 million euros, will be one of the largest in France for artificial intelligence. The sector requires massive resources in research and development and computing power.
The round table, led by the American fund LightSpeed Venture, will bring together major French and European funds and individual investors, according to the same source. Xavier Niel could also invest, according to The Point and The Echoes.
Polytechnicians among the founders
Mistral AI will rely on some of the French AI specialists hired by the Gafa, starting with its co-founder and CEO, Arthur Mensch, polytechnician and normalien, expert in language models, former National Research Institute in digital science and technology (Inria), who has just spent nearly three years at DeepMind, Google’s artificial intelligence laboratory.
Guillaume Lample, polytechnician and hitherto AI researcher for Facebook, one of those behind the Meta group’s LLaMA language model unveiled in February, has confirmed that he will leave Facebook’s parent company to co-found Mistral AI. The same goes for Timothée Lacroix, also a researcher at Meta. The CEO of online insurance start-up Alan, Jean-Charles Samuelian, will be among its partners, according to The echoes And Point.
By way of comparison, the start-up Hugging Face, one of the most dynamic in AI created by the French, went to finance itself in the United States, raised last year 100 million dollars during a third round table. For Mistral AI, starting with a fundraiser of almost 100 million euros would make it possible to aim for higher amounts in future collections.