Mistral AI was founded in Paris in April 2023 by Arthur Mensch, Guillaume Lample, and Timothée Lacroix. Mensch had spent three years as a research scientist at Google DeepMind in Paris and London, contributing to pioneering LLM scaling papers including Chinchilla. Lample and Lacroix were veteran research scientists at Meta AI in Paris, where they served as core contributors to the original LLaMA model that catalyzed the open-source AI movement. Watching Silicon Valley tech giants race to lock artificial intelligence behind proprietary walled gardens, the three French scientists recognized a generational opportunity: to build an independent European AI champion that proved open-weight, transparent models could match proprietary American systems through superior algorithmic design. Armed only with a seven-page memo, the founders raised a historic €105 million ($113 million) seed round in May 2023 led by Lightspeed Venture Partners. Four months later, they released Mistral 7B via a magnetic torrent link on X (Twitter)—igniting a worldwide developer sensation and establishing Mistral AI as a global powerhouse overnight. In early 2023, meeting in Parisian cafés near the Latin Quarter, Arthur Mensch, Guillaume Lample, and Timothée Lacroix debated whether Europe could survive the generative AI revolution without an independent laboratory. Rejecting comfortable executive careers at DeepMind and Meta, they pooled their personal savings, drafted a concise vision for an open European foundation model lab, and founded Mistral AI. Their audacious decision to distribute their first model via a raw BitTorrent magnet link on X without corporate marketing captured the imagination of the global developer community, establishing Mistral as the vanguard of open, sovereign artificial intelligence. The foundational cultural moment that defined Mistral occurred in late 2023 inside their small Paris laboratory near the Canal Saint-Martin. With limited compute resources compared to their American competitors, the engineering team spent weeks profiling PyTorch CUDA kernels to eliminate every cycle of wasted memory overhead. When Mixtral 8x7B completed its final training run and beat GPT-3.5 across reasoning benchmarks, the founders celebrated by sharing lunch at a local bistro, embodying the authentic, unpretentious scientific passion that continues to drive Mistral AI.