AMD buys Fei-Fei Li's World Labs for $8.2 billion to build 'physical AI'

AMD agreed on September 28, 2026 to acquire World Labs, the spatial-intelligence startup co-founded by AI pioneer Fei-Fei Li, in an $8.2 billion all-stock deal — AMD's second-largest acquisition ever and a direct bet on what the industry is calling "physical AI."

What World Labs Actually Builds

World Labs develops what are known as world models: deep learning systems trained to understand physical reality rather than just text or images. The company's argument, publicly stated since its 2024 founding, is that true general intelligence requires grounding in physics — an AI system needs to understand how objects move, collide, and occupy space before it can reliably operate in the physical world. Its first product, Marble, generates simulated 3D environments used both for entertainment experiences and, more significantly, for training robots in physics-accurate virtual spaces before they ever touch real hardware.

Under the deal, Fei-Fei Li — the Stanford professor best known for building ImageNet, the dataset that helped kick off the modern deep learning era — becomes AMD's Executive Vice President and Chief Scientist. AMD and World Labs already had a working relationship: the two companies formed an inference optimization and training partnership in 2025, a year before this acquisition.

Why AMD Is Making This Bet

The acquisition targets a specific gap in AMD's product lineup. Nvidia already ships open-weight world models under its Cosmos brand, giving it a head start in physical AI — the category of AI systems meant to operate robots, autonomous vehicles, and industrial machinery rather than just chat or generate text. AMD has been competing hard with Nvidia on raw chip performance, but lacked an equivalent software and model layer for the physical AI use case. Buying World Labs closes that gap in one move rather than building the capability from scratch.

In her announcement, Li framed the deal around proximity to hardware: "Now that we have tangible proof of the possibilities, we want to do everything we can to accelerate development by scaling efforts and getting closer to the hardware." That framing matters — world models are computationally expensive to run, and being embedded inside a chipmaker rather than operating as an independent lab gives World Labs direct influence over how future AMD silicon is designed to run these models efficiently.

The Size of the Bet

AMD estimates the physical AI market — spanning robotics, autonomous vehicles, and industrial automation — could reach $200 billion by 2035. Whether that figure holds up, the acquisition signals that chipmakers now see the next phase of the AI buildout shifting from pure language-model infrastructure toward systems that need to model and act in physical space, a shift that will also shape what kind of AI chips get built over the next decade. The deal is expected to close before the end of 2026, subject to regulatory approval.