In the same stretch, Mistral reached for enterprises' old code, their internal data, and the compute underneath both — none of it a model launch.
Mistral described work on using AI agents to modernize complex legacy code. It also announced a partnership with Cloudera to bring what the two companies call sovereign intelligence to enterprise data, letting organizations apply AI to data they keep in place. Separately, Mistral signed a roughly $1bn agreement with Loft Orbital and Marlan Space to put compute in space.
None of the three is a model release. Each reaches instead for something enterprises already own — the codebases they cannot stop running, the data they cannot send out — and then for the compute layer beneath both. The company's center of gravity is moving from competing on model capability toward competing on where the model runs and what it connects to. The space agreement is at the signing stage; neither the timing nor the eventual scale of any operational capacity has been disclosed.
Whether legacy modernization surfaces as named migration engagements, whether Cloudera deployments actually ship, and on the space side, whether a signature turns into a concrete launch schedule.