Show an AI a simple photo of your living room. It turns it into a three-dimensional world you can walk around in like in a video game: turn on the spot, look behind the sofa, move closer to the window.
That is the speciality of World Labs, a young company from San Francisco. And the American chipmaker AMD has just announced it is buying it, for around $8.2 billion.
Who is Fei-Fei Li
In the world of AI, Fei-Fei Li is a figure. A professor at Stanford University, often nicknamed the "godmother of AI", she is behind one of the databases that made modern artificial intelligence possible.
In 2024, she co-founded World Labs with three researchers. Her ambition comes down to two words: spatial intelligence. An AI that does not merely predict the next word, but understands space, objects and the way they interact. Without that, she explains, AI remains locked inside the digital world.
In 2009, Fei-Fei Li and her team presented ImageNet: millions of photos, classified one by one according to what they show. A colossal task, long considered unglamorous. In 2012, a neural network named AlexNet crushed a competition based on that database, with a spectacular lead. That result is often seen as the starting point of the deep learning revolution, the one that leads to today's AIs. Without ImageNet, no AlexNet, and in all likelihood no ChatGPT.
What the acquisition provides for
AMD announced in late September a definitive agreement to buy World Labs, paid for entirely in shares, for around $8.2 billion. The deal is due to be finalised by the end of the year, subject to regulatory approval.
Fei-Fei Li will become AMD's chief scientific officer, reporting directly to its boss, Lisa Su. Her co-founders will continue to run the World Labs teams, which will keep their research mission.
An irony of history: among World Labs' investors was also Nvidia, AMD's great rival.
Why a chipmaker is buying researchers
That is the real question, and the answer is strategic. A chip is designed years in advance. To design tomorrow's chips, you need to know what calculations tomorrow's AIs will need to perform.
Yet "world models", capable of generating and simulating 3D spaces, are among the most computationally demanding there are. By bringing this research in-house, AMD hopes to design its future chips around those needs, and steal a march on Nvidia.
What it changes
The applications are numerous: video games, architecture, cinema, but above all robotics. A robot that has to move around your kitchen needs to understand three-dimensional space, something machines still sorely lack, as we explained regarding cleaning robots.
In late September we looked at an AI capable of drawing the floor plan of a flat from photos. World Labs takes the same idea much further: not just drawing the space, but rebuilding it, so you can step inside.
What we take away
Words were AI's first great conquest. Space could be the next.
And the acquisition of World Labs shows that chipmakers no longer want merely to sell shovels to the gold diggers. They want to know, before anyone else, where the next seam is.