Has France Already Lost the AI Battle?

Read this in Stallone's voice: "more chips, more data, more billions, ruler of the world."

I just watched the latest episode of Le Dessous des Cartes on Arte, dedicated to artificial intelligence. And honestly, it hurts. Not the kind of hurt when you stub your little toe on the coffee table after stepping on a Lego. The kind of hurt when you realize you may have missed the train. For good.

The uncomfortable truth

The numbers are brutal.

Over 10 years, the United States has invested $335 billion in AI. China follows close behind.

And us? France comes in 8th place, sandwiched between India and South Korea.

When it comes to patents, it's even more stark: in 2010, 40% of AI patents were American. By 2023, that figure has dropped to just 14%... because China went from 10% to 70%. You read that right: 70%.

Meanwhile, Russia, whose president Putin declared in 2017 that whoever masters AI will be "the ruler of the world," has only developed 3 large-scale AI models since 2017. The United States? 161. China? 127. The European Union? 41.

That's where we stand.

The chip war: a fight we're not even invited to

At the heart of this battle is NVIDIA. A company now worth over $4 trillion. More than any other company in the world.

Why? Because its GPU chips have become essential for running AI programs.

What's Europe doing in all of this? We're watching. We're commenting. We're regulating. The United States and China are waging a fierce trade war over semiconductors. Washington bans the export of high-performance chips to China. Beijing retaliates by blocking the export of gallium and germanium to the United States. Taiwan, with TSMC holding 70% of the global high-performance chip market, sits at the center of all geopolitical tensions.

Europe is THE great absentee from these discussions.

The real question is whether Europe can still enter the battle. I heard an interview with President Macron who rightly said that we lost the cloud battle, so let's try not to lose AI.

Some time ago, we had DeepSeek, a Chinese LLM, claiming to have outperformed ChatGPT for much less money. That's the argument you hear everywhere. DeepSeek, the Chinese startup, reportedly developed a model comparable to ChatGPT for just $5-6 million. Proof that David can beat Goliath! A real source of hope for players who don't have the financial resources but have the intelligence to compensate.

"DeepSeek proves you can do the same for less!"

Except... that's not true! Or at least, very misleading.

I dug into the topic and here's what I found: the $5-6 million figure covers only the GPU cost of the final pre-training run. That's like saying my house cost me the price of the paint.

According to SemiAnalysis, a specialized analysis firm, DeepSeek's total spending reaches $1.6 billion. They reportedly have access to around 50,000 Hopper GPUs, including 10,000 H800s and 10,000 H100s. Their engineers are reportedly paid over $1.3 million per year. We're a long way from Steve Jobs' garage.

That said, DeepSeek did make real architectural innovations. Their model's usage costs are indeed much lower than American competitors, but the narrative of "you can compete with OpenAI on a small business budget" is clearly storytelling. Good Chinese storytelling that wiped $600 billion off NVIDIA's market cap in a single day, mind you.

So what? Do we give up?

That's the question of the century, and honestly, I don't have an answer.

What is certain is that on the foundational layers, the massive models, the chips, the data centers, the cloud, the battle is probably lost. Even our champions Mistral and Hugging Face currently run their model training on US servers. They're working on building data centers in France, but in the meantime...

We won't catch up to $335 billion in American investments with our current resources. We won't have our own NVIDIA.

But does that mean we have no cards to play?

AI is becoming a commodity

Foundation models are becoming commoditized. Tomorrow, the value won't be in the model itself, but in what you do with it. Orchestration. Business integration. Customization. Experience. Context Engineering.

A French company isn't going to get support from OpenAI to transform its processes. It needs local players who understand its context, speak its language, and know its regulatory constraints. That's exactly what we've been doing at Reboot since 2023, and not just a little, we're all in!

But more than that: this support comes with know-how, expertise, and above all a vision of the AI experience. There it is! What we need to do today is offer a clear, simple, obvious strategy to organizations.

That's exactly why we built our startup Lamalo, which houses the products we're currently bringing to market, including Sp0ton (a centralized, secure, and scalable AI platform), Tadikoa (a fusion of Plaud and Fireflies), and DataLlamalist (goodbye PowerBI & Co).

Don't worry, we'll reveal everything in the coming weeks. You won't be able to escape us, you'll be hearing so much about what we're building!

But let's get back to the point.

Europe has something that neither the United States nor China have: credible regulation. The AI Act creates standards that global players must comply with to access our market of 450 million consumers. You can see that as a handicap ("we regulate instead of innovating").

You can also see it as a differentiating position: AI that's more respectful of privacy, more transparent, more energy-efficient.

What really worries me, deep down, isn't so much the technological gap as the prevailing defeatism. If French entrepreneurs start from the assumption that it's already lost, then yes, it will be lost. The prophecy becomes self-fulfilling. The other risk is fighting the wrong battle. Trying to create a "French ChatGPT" may be a dead end, even though having our own champion is also necessary. On the other hand, building an ecosystem of integrators, business specialists, and sovereign solutions for sensitive sectors: that's achievable.

Clearly, France won't be the "ruler of the world" of AI, but it can be more than just a consumer of American and Chinese technologies. There's a space between dominating and disappearing. It's in that space that we must fight, with clear-eyed awareness of our weaknesses, confidence in our strengths, and above all, the urgency to act now, because in 5 years, it will be too late to even ask the question.

Start my AI project

Sources:

• Le Dessous des Cartes, Arte, 2025

• SemiAnalysis, "DeepSeek Debates", January 2025 - CNBC,

• "DeepSeek's AI claims have shaken the world", January 2025