AMD Buys Fei-Fei Li's World Labs for $8.2B to Bolster AI Chip Ecosystem

AMD has agreed to purchase World Labs, the AI research company founded by Stanford professor and computer vision pioneer Fei-Fei Li, for $8.2 billion. The deal underscores AMD's push to build a comprehensive AI infrastructure, from chips to specialized models.

The Strategic Move#

The acquisition marks a significant step for AMD to compete with Nvidia, which has already assembled a broader portfolio of AI tools and models. By bringing World Labs into the fold, AMD gains access to deep learning technologies designed specifically to model and understand physical environments—capabilities that Nvidia has begun offering through open-weight models like Cosmos.

Li, renowned for her foundational work on ImageNet and computer vision research, will join AMD as executive vice president and chief scientist. The two firms have worked together since 2024, when they launched an inference optimization and training partnership.

What World Labs Does#

World Labs develops "world models"—AI systems that simulate physical reality in high fidelity. The company's flagship product, Marble, generates interactive 3D environments and can create synthetic training data for robots. Rather than relying solely on text understanding, these models incorporate visual grounding and physics-based reasoning.

Li has argued that achieving genuine artificial general intelligence requires models that understand not just language but also the rules and structure of the physical world. In her announcement of the deal, she emphasized the need to "scale our efforts, widening our reach, and getting closer to the hardware"—a nod to how semiconductor expertise can accelerate model deployment.

Why Robotics Matters#

World models are becoming critical infrastructure for robotics and autonomous systems. Companies building self-driving vehicles, industrial robots, and general-purpose humanoids face a fundamental problem: there is insufficient real-world data to train these systems safely at scale. Synthetic simulations generated by world models can fill this gap, enabling researchers to train robots in virtual environments before deploying them in the real world.

The deal is expected to close before year-end, pending regulatory approval.