Bonjour, je suis un modèle IA français!
And: What Sammy is thinking
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
📈 Trending Up: Starlink revenue … China’s share of ‘Mature-node chip fabrication capacity’ … The Eagles … startups’ ad budgets …
📉 Trending Down: Learning to code? … No, really … press freedom … the value of books? … TSMC earnings …
Video of the Day: OpenAI’s Super Bowl debut was visually neat, slightly aggrandizing, and fun. I would complain that the company is losing so much money (reportedly) that they should conserve cash. But with SoftBank in their pocket, who cares?
Trump Watch
CO was off Friday. We won’t go back over the entire DOGE-racist tweets fiasco today. I doubt our views there are hard to guess.
But viz the current President, an update on tariffs is warranted. Recall that Trump pledged tariffs against Canada and Mexico, which were put on hold for a month after each nation pledged more border investment, promises that some felt were merely restatements of their prior plans. But no matter, only the package of new tariffs on China went into effect. China has, today, let its “tit-for-tat import taxes on some American goods [come] into effect on Monday,” the BBC reports.
US tech companies could become collateral damage in the broader US-China trade dispute.
And more tariffs are coming. Trump promised on Sunday that imported steel and aluminum will face a 25% tax. And other unspecified tariffs. It seems that the President really is serious about erecting trade barriers — albeit on a slower rollout cadence than some expected.
France’s play for AI relevance
Back in 2023, TechCrunch wrote that “there’s something going on with AI startups in France.” It wasn’t off the mark. While the world today is largely focused on the United States’ AI progress compared to China’s own, Europe has not sat still. (My own take on the same topic from later in 2023.)
Nor has all of France’s AI hopes — read: future economic hopes? — fallen onto the shoulders of Mistral, a domestic foundation model concern. No, as Sifted reports, capital inflow into French AI companies is accelerating:
French AI companies saw investment jump 82% in the past year, from €1bn in 2023 to €1.9bn in 2024.
That’s still not a lot of money, so why focus pay more focus to France than, say, AI efforts in the UK or Germany? Because French nuclear power generation is no joke, and the country’s government seems bent on earning a seat at the future AI table. (Total EU AI fundraising in 2024? Around $8 billion.)
To wit, here’s a handful of recent, critical news events and a bit of context concerning why they matter:
France, UAE agree to develop 1 gigawatt AI data centre
At a cost of up to $50 billion, France will be home to one of the world’s largest compute clusters. The installation should not only provide real grunt to European sovereign compute, but if played intelligently, could also provide domestic French AI companies with discounted compute (I am spitballing on that last bit).
French startup Mistral rolls out app in escalating AI race
This is a somewhat standard product announcement in the AI age but one that I think matters. Why? Because Le Chat from Mistral is currently the number one free app in the nation. ChatGPT ranks fourth, while DeepSeek’s own app lands at number eight. Not bad!
The 15 days that upended Macron’s vision for European AI
Politico.eu’s coverage of how Project Stargate’s announcement and the rise of DeepSeek played in France. First, Project Stargate’s huge planned spend made folks sad. Then, DeepSeek’s impressive work put fresh winds in French sails; if a small group in China could change the AI world, why not France? Optimism rekindled!
France Taps Nuclear Power in Race for AI Supremacy
The Journal reports that France “aims to dedicate a gigawatt of nuclear power to create one of the world’s largest AI computing facilities,” with goals of having “a first tranche of 250 megawatts of power hooked up to AI-computing chips by the end of 2026.”
And then there’s today, with:
Launching the ‘EU AI Champions Initiative’ to Unlock Europe’s Full Potential in AI
A General Catalyst-led “groundbreaking, long-term effort that brings together more than 60 ambitious European companies, from start-ups to established incumbents, all committed to establishing Europe as a global leader in AI development and application.” The full list of participating companies is here.
And Macron unveils $112B AI investment package, France’s answer to US’ Stargate
TechCrunch writes that “French president, Emmanuel Macron, announced a total of €109 billion in private investments in the AI ecosystem” today, with “€30B to €50B coming from the United Arab Emirates (and MGX), €20B coming from Canadian investment firm Brookfield, €10B coming from Bpifrance and €3B coming from French telecom company Iliad” among other expected pledges.
That’s a lot of private capital. And much like Y Combinator companies are reportedly good at buying from one another to help them learn and drive early revenue, a massive consortium of European businesses investing in AI could also provide a solid early revenue source.
Ironically, the European AI startup I am most excited about (Loveable, interview here) is Swedish.
I admit to being a supporter of the United States’ own AI efforts over those from other nations. But my Americanism doesn’t mean that I don’t want competition. To the contrary. The better EU AI startups are, the better US AI startups will have to be. And the better EU AI foundation models are, the better US AI foundation models will have to be, and so on.
Not to mention having a second pole in the AI world that was not the plaything of an authoritarian government would be a win for the world.
Go get it, France. Lever that sweet, sweet low-carbon base load power to spin up an army of GPUs and build something awesome.
Also: It’s a great time to follow Anna Heim and Romain Dillet, two of my friends from TechCrunch who spend a lot of time reporting on French and European startups more generally. They are the best.
What Sammy is thinking
Sam Altman has a new blog up. I want to highlight two points from it that I think are worth working into your own mental models: