AMD is buying World Labs, the company founded by AI pioneer Fei-Fei Li, for 8.2 billion dollars (around 7.2 billion euros). Li joins AMD as executive vice president and chief scientist. The deal is expected to close by the end of the year, subject to regulatory approval.

The official explanation reads like something from a corporate-speak manual: AI development requires "close collaboration between model research, systems and compute". Translation: AMD wants to know in advance what the most demanding models will need, so it knows what chips to make. The two companies aren't strangers: last year they struck a partnership on model optimisation and training, and Li was a guest at AMD's CES presentation this year.

Li is a professor of computer science at Stanford, known for the ImageNet database and the competitions it sparked - the foundation of modern computer vision. She founded World Labs in 2024 on the thesis that true general intelligence has to understand physics, not just text. "Now that we have tangible proof of what's possible, we want to do everything we can to accelerate the future," she wrote in her post about the deal. "That means scaling up our efforts, expanding our reach and getting closer to the hardware."

What exactly is AMD buying? A "world model" is still a loosely defined term - from language models that understand images to systems that create and maintain a faithful simulation of reality. World Labs' first product, Marble, is sold as a tool for entertainment content, but also for building virtual environments in which robots are trained.

That's where the real story is. Nvidia already offers a whole suite of open world models, such as Cosmos. Until now AMD has only offered the public text and video models. Anyone building robots - self-driving cars, industrial machines, humanoids - will need synthetic data, because there simply isn't enough real data to train robots on. That's the vision Tesla and Figure are selling, and that vision runs on chips.

So 8.2 billion dollars buys a seat at the table where it's decided what hardware future robots will learn on. The question is whether one acquired lab is enough for AMD to catch Nvidia, which has spent years building its own ecosystem - or whether this is just an expensive ticket to a race the rival is already leading.