On Wednesday, Arthur Mensch took his seat at a G7 working lunch in Evian alongside Sam Altman, Demis Hassabis, and Dario Amodei. Days earlier, the US government had ordered Amodei’s company to cut foreign nationals off from its most capable models, and Anthropic had pulled them worldwide.

For the one European founder at the table, the timing could hardly have been better.

Mensch, the chief executive of Mistral, has spent two years warning that this exact thing could happen. Now it has, and the internet has turned his argument into a joke about a very large cat.

The fat cat that ate the discourse

The joke is ‘Le Chaton Fat’, a fictional Mistral ‘frontier model’ that tech circles cannot stop posting about. It does not exist. As the French outlet Numerama confirmed, it is an elaborate running gag that started on Mistral’s subreddit after the company rebranded its Le Chat chatbot as Vibe, a name change users hated.

From there it spiralled: fake benchmark charts showing the cat crushing OpenAI and Anthropic, a mock EU notice barring it for being ‘too heavy to regulate’, and a stated spec of ‘1000 meows per second’ and ‘maximum chonk’. Wharton’s Ethan Mollick joked he expected to be asked about ‘Mistral’s new ginormous cat model’ by corporate clients.

Replit’s Amjad Masad chimed in. Mensch himself replied: ‘It’s actually le gros chaton.’

It is silly, and it is also revealing. The meme caught precisely because it dressed up a real sentiment: that with American models suddenly switchable-off, a European alternative that nobody can revoke is the punchline everyone reached for.

The argument the meme is really making

Mensch made that case long before it was funny. At London Tech Week in 2025, he warned about US firms ‘having the keys’ to their models. ‘At some point, you need to be able to turn it off or turn it on, and you don’t want to leave it to another country,’ he said.

Last month, at France’s National Assembly, he gave Europe ‘two years’ to build its own AI before becoming permanently dependent.

The Anthropic shutdown turned that abstraction into a live demonstration. Mistral’s pitch, open-weight models that customers can run on their own infrastructure and that no foreign government can switch off, stopped being a talking point and became a procurement argument.

Mensch sharpened it himself this week. Mistral exists, he wrote on LinkedIn, to keep AI ‘outside of centralized control exercised by states or corporations’, framing the stakes in the language of the last century’s defining resource: ‘AI, just like oil in the 20th century, is about to become the major source of leverage and power in the world.’

Europe’s institutions are listening. The European Commission called the episode ‘a further illustration of why Europe needs to strengthen its technological sovereignty’. France has gone furthest, dropping Palantir at its intelligence agency and pledging to give every civil servant a Mistral-powered assistant, while Mistral has signed a five-year deal to put AI into the country’s nuclear operations.

The money is following the narrative too: Mistral is reportedly in talks to raise around €3bn at a €20bn valuation, nearly double its worth nine months ago, with chip-tool giant ASML among its backers.

The talent is moving too.

This week Mistral named Brian Hall, a marketing executive who has run cloud marketing at both Amazon Web Services and Google Cloud, as its chief marketing officer. His framing of the jump was pointed: he thanked ‘Anthropic and the US government for laying out why Mistral is in such an interesting position’.

Sovereign is not the same as good

Here is the part the meme leaves out. Being the model nobody can switch off is not the same as being the best model, and Mistral is not the best model.

It still trails Anthropic and OpenAI on frontier capability, user numbers, and valuation, last priced near €11.7bn against Anthropic’s roughly $965bn. Mensch concedes it himself.

‘Today, we do not yet own the best language models,’ he wrote this week, though he said Mistral has ‘constantly reduced that gap’ and promised a new open-weight model this summer, with early access in July. Sovereignty, in other words, solves a control problem, not a quality one.

More awkward is a finding on safety. A study by Estonian researchers, reported by the Financial Times, found open-weight models are the weakest at filtering Russian disinformation, with Mistral’s top model ranking around 47th of 60 tested.

For a model France is about to wire into its civil service on sovereignty grounds, ‘harder for an ally to switch off’ and ‘better at resisting a hostile state’s propaganda’ turn out to be very different properties, and Mistral is doing well on the first while lagging on the second.

Loud talk, cautious table

The G7 lunch itself shows the gap between mood and action. For all the sovereignty rhetoric, EU leaders had no intention of confronting Donald Trump over the Anthropic order, according to Politico. The official agenda was AI for growth and resilience; the spat was the ‘elephant in the room’ nobody planned to name.

Brussels wants a ‘circle of trust’ with Washington, not a divorce.

That is the real state of play behind the cat. Mistral’s opening is genuine, the most credible it has ever had, and the political tailwind is strong. But a fat imaginary kitten cannot close a capability gap or a disinformation score, and Europe’s appetite for actually decoupling is smaller than its slideware suggests.

Mistral’s moment has arrived. Whether it can turn a meme and a geopolitical scare into a model that customers choose on merit, rather than on passport, is the question the laughter is talking over.