Founded less than two years ago, French generative artificial intelligence (AI) startup Mistral AI has swiftly staked out a place as the great European hope contending with US giants dominating the sector.
Named for a legendary chill wind that sweeps France’s Mediterranean coast, Mistral raked in a 600-million-euro ($620 million) funding round last summer that was the largest of any French tech firm in 2024, according to consultancy KPMG.
The new investments brought the company’s valuation to almost 6 billion euros.
Mistral was set up in April 2023 by Arthur Mensch, a graduate of France’s prestigious Polytechnique engineering school with a mop of unruly dark hair who had previously worked at Google’s DeepMind AI lab.
He was joined by two countrymen, Guillaume Lample and Timothee Lacroix, both former researchers for Facebook parent Meta.
Within a year, they had raised more than one billion euros to get Mistral off the ground.
“Our ambition is to be at the cutting edge in the 10 years ahead, and to be one of the actors shaping (AI) technology and the way it’s used,” Mensch told AFP from his Paris HQ, where the company’s logo is nowhere to be seen.
Among Mistral’s offerings is a chatbot dubbed Le Chat — playing on the French word for “cat” as well as the English “chat” — similar to OpenAI’s ChatGPT.
The bot will in future have access to AFP’s text content in six languages to inform responses based on news and current events, the two firms said Thursday.
Mistral also produces a large language model (LLM) known as Mistral Large, specialising in text generation, as well as models focused on images and computer code.
Global market
The company has become a “shooting star”, emerging just as investors were looking for a “French-style OpenAI” said Claude de Loupy, an expert on applying AI to languages.
Mistral has shone by showing off “superbly crafted” AI models from the very beginning, de Loupy added.
Unlike US competitors OpenAI or Anthropic — accused of working as inscrutable black boxes — Mistral bet on an open-source approach for its models, offering access to their programming code.
This “allows our customers to deploy solutions onto their infrastructure with significantly improved data governance compared with our US competitors,” Mensch said.
Mistral has also shone in small-size AI models that can be hosted on computers or smartphones, which achieve savings over rivals by consuming less energy.
Distribution deals with Google, Microsoft, Amazon and IBM have made the firm’s products easier to access.
Around 100 people work in Mistral’s factory-fresh Paris offices, a sea of white tables and black screens below visible ventilation ducts.
Even before achieving profitability, the company is determined to grow, targeting both the French and foreign markets with offices in Britain, Palo Alto in California and Singapore.
“Generative AI is a revolution that will affect the whole world. All the regions of the world are becoming aware, at different speeds, that they must get involved and do it with some level of independence from the American players,” Mensch said.
Fundraising challenge
Developing AI models is an endeavour that continues to demand massive capital investment.
Mistral is struggling to “fight on a level playing field” as it “squares up to giants like Google, Meta, Microsoft and Amazon,” de Loupy said.
Recent funding rounds by Elon Musk’s xAI and Microsoft-backed OpenAI have been in the range of $6 billion — 10 times more than Mistral’s last cash injection.
That calls into question the value of even the strong political backing enjoyed by Mistral –- recently called a stroke of “French genius” by President Emmanuel Macron -– in the global AI race.
Another European AI hope, German start-up Aleph Alpha, was forced in September to abandon its LLM ambitions, focusing instead on consulting for other businesses.
While “the capital needs are definitely not going down,” Mistral’s Mensch said he was given hope by the fact that “there’s constant demand these days” for his company’s products.
“The (European) market must unify and take responsibility in light of the fact that it must support European technology,” Mensch said.
“It’s useful over the long term, and even in the short term, to work with local players,” he added.
