AMD to acquire Fei-Fei Li's World Labs for USD 8.2 billion to strengthen physical AI and robotics

Advanced Micro Devices (AMD) has agreed to acquire World Labs, the San Francisco start-up founded in 2024 by computer-vision pioneer Fei-Fei Li, in an all-stock transaction valued at approximately USD 8.2 billion. The deal, announced on 28 September 2026, is expected to close by the end of the year, subject to regulatory approvals.

World Labs develops what it calls spatial-intelligence models: AI systems that generate, reconstruct and simulate interactive 3D environments from text, images and video. Its main product, Marble, is used both to create entertainment content and to build simulated environments in which robots can be trained. AMD also points to the start-up's work on robot learning and simulation. The two companies were not strangers: they had already set up a partnership on inference optimisation and model training in 2025.

After closing, Fei-Fei Li will join AMD as executive vice president and chief scientist, reporting to chief executive Lisa Su. Li, a professor of computer science at Stanford, is best known for creating the ImageNet dataset, which helped launch modern deep learning for computer vision. In AMD's announcement she said that joining the chipmaker would give her team "the resources and engineering depth to accelerate our research".

AMD presents the acquisition as a way to understand emerging AI workloads in reasoning, robotics, simulation and physical AI, and to feed that knowledge into its future hardware and software roadmap. The logic mirrors the strategy of NVIDIA. World models produce synthetic training data and simulated worlds, which are considered essential because real-world data for general-purpose robots, including humanoids, remains scarce and expensive to collect. NVIDIA already offers open-weight world models under the Cosmos name alongside its Isaac GR00T robotics stack, and it supplies the computing platform used by many humanoid developers.

For the humanoid robot market, the deal matters less as a semiconductor transaction than as a signal of where value is expected to sit. Robot makers increasingly rely on foundation models and simulation tools they do not build themselves, and the suppliers of this layer are consolidating around the largest chip groups. The price, paid entirely in AMD shares for a company about two years old, reflects how scarce teams with credible world-model research have become. The open questions are whether World Labs' models will remain available to robot developers using other hardware, and how quickly AMD can turn the research into a toolchain able to compete with NVIDIA's established robotics ecosystem.