Elon Musk: Why Elon Musk is betting everything on Nvidia

The debut novel C'était ça ou mourir ("Either this or die") by the Canadian Thélyson Orélien was a huge success: the book sold more than 35,000 copies in France alone and won several literary prizes. Until someone threw it into a program that recognizes texts generated using artificial intelligence. The result: The novel should be 100 percent AI-generated.

Clear thing? Not at all. The author denies the accusation, as my colleague Leon Lindenberger reports. Orélien says he started working on his book in 2017, before ChatGPT started the AI ​​hype. Nevertheless, his work was removed from the nomination list for the prestigious French Prix Goncourt. The skepticism alone was apparently enough.

The debate is not over and may never be. What it really was like can hardly be conclusively clarified. Because just as people can lie, AI detectors can also be wrong, training data can cause distortions. And so the question is less what is true. But who do we prefer to trust – humans or machines?

This is what you need to know: Elon Musk wants to buy Nvidia chips worth $40 billion

In the AI world, billions have become so commonplace that sometimes you have to break them down to even understand them. The Financial Times reports that Elon Musk is currently trying to raise $40 billion through loans and bonds. The SpaceX boss wants to use this to buy Nvidia chips. 40 billion dollars is approximately 36 billion euros. With this you can buy 122 Lamine Yamals. Or finance the German Digital Ministry for eight years.

It's not just the amount that's interesting. But also that Musk is going all in: He is betting everything on Nvidia. In a conference call about the quarterly figures, Musk said in August that they were only using Nvidia because "the Vera Rubin architecture is the best architecture."

He is alluding to a factor that is no longer a secret in the AI ​​world: Nvidia is not only so successful because of its extremely powerful chips. But also because it thinks hardware and software together. The Vera Rubin architecture is like a complete package of different chips and systems for corporate customers. Nvidia itself speaks of a supercomputer. It is said to be particularly suitable for agentic AI and reasoning models. And that's obviously worth a lot to Musk.

You should think about this: What you discuss with your chatbot can also go to third parties

There are two terms that precisely encapsulate the problems of the platform economy over the past two decades. The internet critic Cory Doctorow described the increasing deterioration of platforms in the wake of making more and more money as a worsening of the situation. And the American economist Shoshana Zuboff described the increasing exploitation of data as surveillance capitalism.

Both developments have long been observed in the AI ​​economy. Just recently, OpenAI announced advertising in its chatbot. Officially, because it could be “useful” for users. But if you generally ask about online advertising, users often feel uncomfortable with it. No matter what the companies claim: it's simply a complete nonsense.

Now a new, as yet untested study shows how surveillance capitalism is also making its way into AI companies. Researchers from the Spanish IMDEA Network explain the extent to which AI companies exploit their users' data. From six out of nine chatbots that can be used via the web and three out of eight AI apps, third-party providers received snippets such as prompts or context. In some cases, entire chat histories were shared with third parties.

The authors come to the conclusion: What the AI ​​companies are doing is even more invasive than what was previously known from online platforms. One could also say: Surveillance capitalism has gone bad.

Here's what you could try: installing a local AI on the Mac Mini

You can avoid that everything you discuss with your chatbot goes to the companies. For example, by running the AI ​​on your local computer. Our gadget tester from the service, Henrik Oerding, tried out how well this works.

He chose two of the most popular devices for such AI experiments: the Mac Mini and the Mac Studio. In his guide, he explains how to get local AI assistants on Apple's square boxes. Step by step: without internet access, with internet access (from then on you are of course giving away data again), with agents.

A colleague in the editorial team attested to this text: "That's so nerdy, Henrik Oerding deserves a trophy!" I can fully agree with this assessment.