A four-week-old startup has raised $113 million to compete with OpenAI in building, training, and applying large language models and generative AI.
Paris-based Mistral AI, co-founded by Google DeepMind and Meta alumni, will focus on open source solutions and enterprises to solve what CEO Arthur Mensch calls the biggest challenge in the field: “making AI useful.” Text-based generative AI models will be released in 2024.
In the round that Lightspeed Venture Partners led, Xavier Niel, JCDecaux Holding, Rodolphe Saadé, Motier Ventures, La Famiglia, Headline, Exor Ventures, Sofina, First Minute Capital, and LocalGlobe all participated.
Bpifrance and Eric Schmidt are also shareholders, according to Mistral AI. According to company sources, Mistral AI is worth €240 million ($260 million) after receiving €105 million in funding ($113 million at current rates). When the French press started talking about the company a month ago, this was the number rumored.
Mensch (center), his co-founders Timothée Lacroix (CTO, L) and Guillaume Lample (Chief Science Officer, R), and they all studied artificial intelligence in school.
Mensch was at DeepMind in Paris, and Lacroix and Lample were at Meta’s Paris AI outpost. Mensch said they started talking about AI development last year.
“We could see the technology really start to accelerate last year,” he said in an interview today, likely referring to OpenAI’s GPT model, which boosted AI and tech.
OpenAI was anything but open. Mensch, Lacroix, and Lample saw an opportunity to change the proprietary approach that was becoming the norm. Mensch stated that open source is in our DNA.
Mistral is only a month old, but Mensch said the plan is to build models using only publicly available data to avoid legal issues that others have faced over training data. Users can also contribute their own datasets. Data and models will be open-sourced.
“We believe that the benefit of using open source can overcome the misuse potential,” he said. “Open source can prove tactical in security, and we believe it will here.”
Its future products’ market reception is also unknown. The startup’s focus on enterprise, not consumer, customers and the idea that there is a market gap for helping them figure out what they need to do and how to do it is intriguing.
“AI is useful in some cases,” Mensch said. “But too many workers in different fields are still being asked to be creative [with AI], and we need to figure this out for them. We want to provide simple tools for product creation.”
Given recent high-profile cases, funding a young company without customers or a product may seem risky. Neeva, another Google-backed startup, gave up on its consumer AI search play and sold a chunk of its tech to Snowflake. Stability in AI has also been in the spotlight, but not in a good way.
Antoine Moyroud, who led Lightspeed’s investment (which also backed Stability AI), said it’s worth taking.
“Lightspeed didn’t work in isolation,” he said. We saw several strong tech teams seeking funding to build LLM in India, the U.S., and Europe. While technically strong, they hadn’t considered LLM’s second-order effects. They weren’t considering their applications.
He compares AI to cloud computing and database companies, not apps. “We think value will accrue here as it has in the cloud, where 5–6 players will get the majority.
“Who is the most credible here? We met Mistral’s founders. Talented team. We estimate 70–100 language model optimization experts worldwide. Lample led Meta’s Large Language Model (LLaMA) team before leaving to co-found Mistral AI.
Also, AI startups are attracting VC funding. The Googles, Apples, Metas, Amazons, Microsofts, and others will play in this space, but the fact that startups are being formed and finding willing recruits shows that it’s not over yet.
Mistrals, the northerly winds that blow down Europe, usually bring good weather. The startup may want the same effect.
San Francisco residents often talk about their “uncanny valley,” but Mensch said he thinks the world must work together to build and own AI. Mistral’s play gives France a chance.
He said the funding will allow it to hire a “world-class team” to create “the best open source models.”