AMD Buys World Labs for $8.2 Billion to Compete in AI World Models
The semiconductor giant snaps up Fei-Fei Li's startup to build synthetic training data and challenge Nvidia's dominance in robotics AI.</dek> <parameter name="category">AI
Ars Technica
AMD has agreed to acquire World Labs, an AI startup focused on world models, for $8.2 billion, with the deal expected to close by year-end pending regulatory clearance. The move represents a significant push by the chipmaker to strengthen its position in generative AI against rival Nvidia.
The Startup's Origins and Focus#
World Labs was founded in 2024 by computer vision researcher Fei-Fei Li alongside collaborators Justin Johnson, Christoph Lassner, and Ben Mildenhall. The company secured $230 million in initial funding, with AMD itself contributing to the round. The startup is developing world models—AI systems trained on large video datasets to simulate and predict how the physical world behaves.
Products and Technology#
The company's flagship offering, Marble, generates small 3D environments using a technique called Gaussian splats, allowing creators to export these spaces as 3D assets for film, games, and other media. World Labs has since expanded its portfolio with additional tools like Atlas, each tailored to different applications and use cases.
Why This Matters for AMD#
A major driver of world model investment is generating synthetic data to train robots—a domain where Nvidia currently holds a commanding edge. While AMD has released its own open-source world model called Micro-World, it lacks a comprehensive stack comparable to Nvidia's Cosmos platform. This acquisition gives AMD the talent, technology, and products needed to catch up in a critical area of AI development.
