🔴 Uneasy Anatomy of a Coming Crash

This is my third newsletter on the AI bubble. Having explored the pros and the cons, here I am worried again. Let me explain why.

48 hours after Nvidia's third-quarter results and the release of Google Gemini 3, a few observations force themselves on us.

We have too much capital invested in AI, and strategies that sorely lack differentiation. In this gold rush, everyone thought of everything except one thing: the physical reality of the computing world.

On the sidelines of the summit held in Berlin, during a conversation about our strengths and weaknesses with our ambassador to Germany and a few entrepreneurs, I raised a theme dear to me: maintenance matters as much as innovation.

I have always believed that France's real strength lay in its ability to create "humanist infrastructure". Systems of trust, designed for the common good: the railways, nuclear energy, or our water management (a genuine French school).

Unfortunately, we privatized these jewels. In hindsight, we realize that this essentially served to finance the baby boomers' lifestyle, without our managing to find new sources of growth in digital. The verdict is harsh: we are now net importers of digital technology when we should have been world leaders.

To put it plainly: France is missing 3 or 4 "Dassault Systèmes" today.

Granted, not everything is perfect at home. We lament the shortage of welders in the nuclear industry, and the fact that trades that do not involve some form of intellectual abstraction are often misunderstood, even looked down on. And yet we do not always have reason to blush.

When I see the metal plates hastily laid over Highway 101 between San Francisco and Mountain View to plug the potholes, and the catastrophic state of Silicon Valley's legendary freeway on rainy days, I tell myself our model has its merits too. We simply never managed to leverage our infrastructure know-how as a tool of soft power.

One example comes to mind: I remember a day spent with one of the best digital health entrepreneurs in the US. Coming with me to the pharmacy, he was fascinated to watch me hand over my Carte Vitale (the French health insurance card) and walk out with my medication in a few seconds. For him, it was a "magical" user experience, made even more powerful by the fact that every French person has access to it, unlike in the US.

We often tend to mock the Cobol architectures of the 1970s. Except that they were built by people who had a long-term vision, a work ethic and a concern for stability. By wanting to "disrupt" everything for the fun of it, we end up forgetting the strength of stability.

France should have been the country of modern, intelligent, humanist infrastructure. In the age of climate change and digital change, the world needs more than ever a vision in which science and the economy serve the many, not a minority. That, alas, is not what the past fifteen years will have produced.

Let me close that parenthesis.

The abundance syndrome

The problem is that today, tech is sick with the abundance syndrome.

Unlike the days of Cobol, Assembly or C, when memory was expensive, modern developers no longer care about optimization. They are too far removed from the machine. I am talking here about real technical optimization, not the accounting cost cuts on a BOM (Bill of Materials) or a digital process, which are fairly common.

In AI, the exception is Chinese. Take DeepSeek: they keep optimizing their code drastically. Why? Because they have no choice. American export restrictions deprived them of the most powerful chips, forcing them into software ingenuity. When I saw DeepSeek, I truly believed that this kind of breakthrough would come from France, the country of Fabrice Bellard, who changed the worlds of video, virtualization and telecoms three times over with code of incredible optimization and beauty (FFmpeg, QEMU, Amarisoft).

💡

Who, in France, is the Fabrice Bellard of AI?

The rude awakening: the infrastructure wall

Thanks to that German summit, I was able to deepen my "low-level" understanding of the real situation of data centers. And for the first time, I think we are entering a zone of major turbulence.

When Wall Street opened on November 21, 2025, at 3:30 p.m. Paris time, the verdict came down. Nvidia lost all of its post-announcement gains (-6% the day before, with a heavy downward trend). To give you an idea of the violence of the move: we are talking about value destruction of one billion dollars per hour. Unheard of.

The reality is that behind the AI dream, the fundamentals are rotten. As a friend and industry expert explained to me, the point is not to look at what is sold, but at what is in inventory and what is unpaid. Receivables are exploding and chips are piling up. With plenty of clever operators optimizing availability, we have created secondary markets and invisible debts which, as we have already written, will soon come to light.

Because of AI demand, the price of 64 GB DDR5 memory modules has quadrupled in a few months, from €130 to more than €500. This increase hits infrastructure costs directly, since memory is essential to run GPUs, but as Octave, the founder of OVH, explained, it will show up as a sharp rise in cloud prices in 2026.

On top of that comes the technological coup de grâce: Google Gemini 3 (released November 18, 2025). The model outclasses the competition and was trained with zero Nvidia chips (exclusively on Google TPUs manufactured by Broadcom). The message to the markets is clear: Nvidia's monopoly is not eternal.

(an AI-generated image, if that even needs saying, so precise is nanobanana 2)

Many rushed to build without thinking, because saying you were doing AI implied seeing your valuation explode. But now everyone is looking more carefully at each player's strategy (see my note for subscribers below).

How to prepare for the now inevitable Crash?

Today I am sharing with paying subscribers a confidential strategic note, originally written for the clients of The Sovereign Way.

In this report, we analyze:

Why we are heading straight into the wall.
What kind of crash awaits us (it is not the one you think).
How to limit your exposure and pivot before it is too late.

👉 [Subscribe now to access the full note]

💡

A subscriber recently told me that one of our analyses had helped him sign a strategic client, while another was able, thanks to us, to radically change strategy before a crisis. Here, we love your feedback. Your subscription allows us to maintain this independence and to offer you analysis available nowhere else.

Before leaving you with our strategic note in English, I wanted to repeat a few important points.

This is an infrastructure bubble, not an AI bubble.
That electricity is a problem, but not the first problem.
That France's and Europe's strategy (AI factories) leaves us heavily exposed to the turbulence that is brewing. This could affect our AI research infrastructure, but also create opportunities.
That our French data center players are not all exposed to the risks in the same way.

The rest is reserved for our paying subscribers.