Mistral CEO pitches owned GPU fleet and IPO within a few years, teases open cyber model

Mensch's pitch is Europe's answer to the US hyperscaler model. Rather than renting compute, Mistral owns its GPU fleet, and he argues that ownership is the foundation of sovereignty for European governments and regulated industries. The company now sells a full stack: Vibe (a work and coding agent), Studio, Forge, Compute, document AI, speech, and industrial engineering offerings including robotics and physics modeling.

The model roadmap backs the stack story. Large 4, a 1-trillion-parameter multimodal MoE with 52B active parameters trained on 3,800 Grace Blackwell GPUs, entered public preview October 6 with open weights expected by month's end. Mensch also teased an open-source release within the week claiming state-of-the-art cyber capabilities. That claim lands awkwardly in a week dominated by containment failures at DeepSeek and Alibaba, and open cyber-capable weights will draw scrutiny.

The IPO talk invites comparison with Anthropic's filing this week. Mistral is far smaller, and StartupHub's analysis argues its roughly €3 billion bet on owned infrastructure 'still needs proof' — capex-heavy vertical integration only pays if utilization stays high.

Watch for the cyber model's release and safety documentation, the Large 4 open-weights drop, and named sovereign customers anchoring the GPU fleet.