Arthur Mensch
Co-founder and CEO, Mistral AI
Research on brain imaging and optimisation
Before founding Mistral AI, Arthur Mensch spent more than ten years in academic work on machine learning for brain imaging and on the optimisation of machine learning[1]. With Julien Mairal, Bertrand Thirion and Gaël Varoquaux he wrote Stochastic Subsampling for Factorizing Huge Matrices, posted in January 2017[2]. In October 2017 he was first author of Learning Neural Representations of Human Cognition across Many fMRI Studies, with Mairal, Danilo Bzdok and Thirion[3].
With Mathieu Blondel he wrote Differentiable Dynamic Programming for Structured Prediction and Attention, posted in February 2018[4]. The paper sets out a framework for turning dynamic programming algorithms into differentiable operators that can be used inside neural networks[4].
DeepMind and Chinchilla
Mensch joined DeepMind Paris as a researcher in 2020 and spent three years there working on projects in generative AI[1]. He is a co-author of Training Compute-Optimal Large Language Models, the March 2022 paper by Jordan Hoffmann, Sebastian Borgeaud and others[5]. The paper studied how to balance model size and the number of training tokens for a fixed compute budget, and tested its findings with a 70-billion-parameter model called Chinchilla[5].
Founding Mistral AI
In 2023 Mensch founded Mistral AI with Guillaume Lample and Timothée Lacroix[1]. The company says it was founded in April 2023, and lists Mensch as co-founder and CEO, Lample as chief science officer and Lacroix as CTO[6]. Its stated mission is to make frontier AI open to all[6].
Mistral hired its first employee on 5 June 2023 and closed a seed round on 13 June 2023[6]. It released its first model, Mistral 7B, on 27 September 2023[6]. Mensch, Lample and Lacroix are among the authors of the Mistral 7B paper, posted to arXiv in October 2023[7]. The company's timeline then lists a Series A in December 2023, Mistral Large in February 2024 and a Series B in June 2024[6].
On 9 September 2025 Mistral announced a €1.7 billion Series C at an €11.7 billion post-money valuation, led by the semiconductor equipment maker ASML[8]. Existing investors including DST Global, Andreessen Horowitz, Bpifrance, General Catalyst, Index Ventures, Lightspeed and NVIDIA also took part[8]. Mensch said the investment brought together two technology companies in the same value chain[8]. Sifted reported that the round made Mistral one of the most highly valued private tech companies in Europe[9].
Products and infrastructure in 2025 and 2026
Mistral's timeline lists the Le Chat assistant in February 2025, Mistral Code (later Vibe) and Mistral Compute in June 2025, the Voxtral speech models in July 2025 and Mistral Studio in October 2025[6]. It lists Forge in March 2026 and a European AI playbook in April 2026[6]. The company describes its business as building full-stack AI for enterprises and governments in sectors such as finance, manufacturing, defence, energy and the public sector[6].
In May 2026 Mistral said it would build a 10-megawatt data centre in Les Ulis, France, in the second half of 2026 as part of a €4 billion investment plan[10]. RTÉ reported that the plan aims for 200 megawatts of capacity by the end of 2027 and 1 gigawatt by 2030[10]. Speaking to reporters at the time, Mensch argued that Europe needs its own AI capabilities because rivals and adversaries are already using the technology[10]. He also said some anxiety about AI was expected because it is changing how people work[10].
Known for
- Co-founding and leading Mistral AI
- Co-authoring the Chinchilla compute-optimal scaling paper
- Mistral 7B and open-weight models
- Research on machine learning for brain imaging
Sources
- London Tech Week: speaker page, Arthur Mensch — checked 6 Nov 2026
- arXiv: Stochastic Subsampling for Factorizing Huge Matrices, January 2017 — checked 6 Nov 2026
- arXiv: Learning Neural Representations of Human Cognition across Many fMRI Studies, October 2017 — checked 6 Nov 2026
- arXiv: Differentiable Dynamic Programming for Structured Prediction and Attention, February 2018 — checked 6 Nov 2026
- arXiv: Training Compute-Optimal Large Language Models, March 2022 — checked 6 Nov 2026
- Mistral AI: About us — checked 6 Nov 2026
- arXiv: Mistral 7B, October 2023 — checked 6 Nov 2026
- Mistral AI: Mistral raises 1.7B€ to accelerate technological progress with AI, September 2025 — checked 6 Nov 2026
- Sifted: Mistral confirms €1.7bn Series C led by ASML, September 2025 — checked 6 Nov 2026
- RTÉ: Mistral defends military AI, expands data centres, May 2026 — checked 6 Nov 2026