AMD said Monday it will acquire World Labs, the startup founded by Stanford professor Fei-Fei Li to build "world models" that understand physical reality, in an $8.2 billion deal. Li will join AMD as executive vice president and chief scientist once the transaction closes, which the two companies expect before the end of the year, subject to regulatory approval.
World Labs framed the sale as a consequence of scale: AI development, it said, now requires "close collaboration across model research, systems and compute." AMD's interest runs the other way — frontier workloads like those World Labs trains are meant to shape the chipmaker's roadmap, giving it a software story to set against Nvidia's.
The two companies are not strangers. They struck an inference optimization and training partnership last year, and Li appeared as a guest at AMD's CES presentation earlier this year.
Li founded World Labs in 2024 after building ImageNet, the dataset and competition that helped unlock modern computer vision. Her argument was that text alone cannot produce general intelligence: a model needs to ground itself in physics, space and causality. The startup's first product, Marble, generates navigable 3D environments aimed at entertainment — and, more strategically, at simulated worlds for training robots.
That is where the deal's logic sits. World models are increasingly pitched as the data engine for autonomy, generating synthetic experience for humanoids and self-driving systems that cannot collect enough real-world miles on their own. Nvidia already ships open-weight world models such as Cosmos; AMD, by contrast, had published only text- and video-generation models. Buying World Labs buys AMD a credible entry — and a marquee name — in the field.
The caveat: "world model" remains a loosely defined term, spanning everything from vision-aware language models to full simulation engines, and antitrust review in a hostile regulatory climate is not guaranteed to be smooth.
