Futuro Fortissimo Archive
🎼 ff.139 Was it the AI?
When NVIDIA is worth more than Switzerland
Published · Michele Merelli
- Is all of Switzerland worth as much as NVIDIA?
- Camera : influencers = Gemini : ?
- … don't miss the nutritionist ferret at the end!
Themes🤖 AI, agents and models
🍫ff.139.1 Swiss? No, NVIDIA
Substack ↗In 3 months, NVIDIA took in $57 billion (earning 36 of them, at a 75% margin). Mind-boggling numbers. Let's try to make sense of them.
Some people compare its market value with the gross domestic product of entire countries. That's wrong.
Valuation ≠ GDP. The first is a theoretical, perceived value; the second is what a country physically produces in a year.
Better to think fantasy football. If you wanted NVIDIA, you'd have to sell quite a few “players”:
all the Swiss companies listed in Zurich (Rolex, Nestlé, ABB, Novartis…) the 20 biggest companies in Europe (ASML, LVMH, SAP…)
🧃ff.139.2 Konbini and paperclips
Substack ↗Why the hype? Because AI is proving economically fruitful.
Dave Blundin, on the ever-stimulating MOONSHOT podcast, points out that the online advertising market (600 billion) is already run entirely by algorithms.
Agents with more advanced capabilities (e-commerce, commodity trading…) could create a market of at least $1T (~1% of world GDP).
As further confirmation (and since the latest models are saturating the hardest benchmarks: Gemini 3 solves 1/3 of the questions in Humanity's Last Exam), the new AI “leaderboards” are economic:
performance in finance (Alpha Arena) skill at selling, negotiating and long-term planning (Vending-Bench 2; here too Gemini 3 wins, “taking home $5,000”).
Nerd corner. Earnings by AI model and negotiation strategy (friendly, negotiator, double-dealer).
Economic optimisation, long-term planning... It brings to mind the mini-game about maximising paperclip production, cited by Bostrom as an example of super-efficiency leading to extinction (🦉 ff.36.2 The super-tome on the singularity).
🍫ff.139.3 Where's my flying car (in GTA 6?)
Substack ↗All these reflections are back in vogue because of yet another quiet but astonishing model.
Gemini 3 is more than a simple version bump: it's the first truly multimodal model.
It generates apps that combine camera recognition, image creation, search and text-to-speech. Apps with tool-calling, in short.
A couple of examples (created with a description shorter than a tweet):
Nutritionist ferret. Snap a photo of your food and it analyses it, returning a detailed report (squeak by squeak!) on the dish it has “sniffed”.
Podcast converter. Paste in today's issue and I get a personalised LinkedIn post, images on the key themes and a 5-minute podcast.
140 characters. Storrs Hall lamented that we got Twitter but not flying cars (😭 ff.46.3 Wasted years).
But now, with Gemini and 140 characters, we have a flying-car simulation. In a robotic future, the same input could make it real.
It's no coincidence that Polymarket is ever more confident Google will have the best model at the end of 2025.