AI has become the competitive tech sector par excellence, with many companies vying to stand out from the crowd. Among these, one stands out among French companies: Mistral AI.
The French are not absolute fans of AI, and Europe in general is rather behind on its investments in the generative artificial intelligence sector. While the whole world seems to swear by these models, we don’t have our ChatGPT or our Bard Made in France. Mistral AI plays in another field, and this is surely what convinced investors.
Exceptional financing
Mistral AI had a main competitor, this one German: Aleph Alpha, but which did not succeed in attracting as much international attention from investors. Thanks to a judicious financial maneuver, the unicorn managed to raise 450 million euros. A more than comfortable sum, supported mainly by Andreessen Horowitz (California venture capital fund) and other renowned economic players such as Salesforce and NVIDIA.
According to Bloomberg sources, Andreessen Horowitz would inject 200 million euros into Mistral AI. For Salesforce and NVIDIA, the two giants would put their hand in their pocket to the tune of 120 million euros, but in the form of convertible debt. Figures still capable of experiencing an evolution. The young Parisian company could well be valued at $2 billion if this global agreement materializes. Nice score!
A unique philosophy and technology
Mistral AI’s particular approach was noticed and stated by Arthur Mensch, co-founder and CEO, formerly of Google DeepMind during the SLUSH conference which took place in Helsinki. What makes it unique is the freedom in choosing the amount of data it can process. Mensch details the principle: “ You can start with a gigantic model made up of hundreds of billions of parameters, which could solve your problem. But, you also have the choice to reduce this number of parameters and concentrate on a set a hundred times smaller. »
What are the advantages of such an operation? First, reduce latency when there is no need to juggle too much data. By lowering this latency, operating costs are also reduced and adapt to user needs. So, Mistral AI primarily targets developers by designing these smaller model sets. The competition, on the other hand, is more focused on offering large-scale, multitasking models. A strategic positioning that could pay off in the future.
Source : The Next Web
16