Mistral's ARR hit $400M in February, up 20x in a year. CEO Arthur Mensch is calling for $1B by year-end. Seven French and global banks put up the capital, which tells you this isn't speculative financing, it's institutional confidence in a revenue trajectory.
The sovereign angle is also genuinely structural: government and defense contracts, including French military and Luxembourg public sector, tap into European AI budgets where data residency requirements favor EU-based providers. I get the thesis.
But here's the thing. Mistral's entire identity is built on open-weight models. And the product is: help enterprises self-host instead of paying Mistral for hosted inference. Which is fine as a land-and-expand strategy, except that Mistral just borrowed $830M to fill a data center they need utilized. If their own tooling makes on-prem more accessible, and open weights keep compressing API pricing industry-wide, what does utilization look like in 18 months? I don't have a clean answer.