Most conversations about AI sovereignty end up at Mistral, because it’s the one European lab shipping models close to the frontier, and it ships many of them as open weights under licences a compliance team can read in a minute. It was founded in Paris in 2023 by people from the big US labs, and it has become the reference point for what a European AI stack looks like.
What Mistral sells
Mistral sells three things, and most firms will only need one or two:
- Open-weight models. The releases that made its name: Apache-licensed, downloadable and strong at every size, especially at the efficient small end, where they’re a favourite for the everyday layer of routed, cost-tiered systems.
- Hosted models and API. European processing for firms that want someone else to run the models without giving up the jurisdiction argument.
- Le Chat. The assistant for staff, on business terms, for firms that want the European argument to cover the tool people open every day.
The open-weight guide covers the costs of running the first of these yourself.
Does EU sovereignty matter for your firm?
It depends on who you need to convince. For some firms it’s decoration, and for others it’s what gets procurement over the line. Data processed in the EU by an EU-headquartered provider makes the transfer analysis in a DPIA simpler, reassures some clients and boards, and takes a set of geopolitical what-ifs off the risk register. If none of that matters to the people signing off, Mistral still earns its place on the shortlist the ordinary way: capable models, clean licences and low prices. If it does matter, Mistral is often the difference between an AI project approved and one parked.
Treat sovereignty claims with the same scepticism as any other vendor claim. What matters is where YOUR data is processed under YOUR contract, which you can check, not where a company’s headquarters happens to be. Mistral’s advantage is that the checkable version holds up.