AMD will acquire World Labs, one of the leading developers of deep learning models aimed at understanding physical reality, in an $8.2 billion deal, the companies announced today.
World Labs justified the deal in a statement, saying AI development requires “tight collaboration across model research, systems, and computing.” Meanwhile, AMD says understanding cutting-edge workloads like those created at World Labs will shape its chip manufacturing roadmap.
With this acquisition, World Labs founder Fei-Fei Li will join AMD as executive vice president and chief scientist. AMD and World Labs entered into an inference optimization and training partnership last year, and the relationship remains strong. Notably, Lee attended AMD’s CES presentation as a guest earlier this year.
Lee, a professor of computer science at Stanford University, is considered a pioneer in AI, particularly computer vision, for building the ImageNet database and the AI competition it inspired. In 2024, Lee founded World Labs to develop deep learning models to more reliably understand the physical world, arguing that true general intelligence requires a foundation in physics and the ability to understand and reason with data beyond text.
In a post announcing the partnership, Lee explained that the partnership is a result of World Labs’ desire to extend its technological advances beyond the lab. “Now that we have concrete evidence of what’s possible, we want to do everything we can to accelerate the future,” Lee said in the post. “To achieve this, we need to expand our efforts, expand our reach, and move closer to the hardware.”
“World model” remains a loosely defined term that includes everything from language models trained to understand visual input to models that can generate and maintain high-fidelity simulations of reality. World Labs’ first product, Marble, is being pitched not only as a tool for creating entertainment experiences, but also as a tool for creating simulated environments for robot training.
The acquisition could help AMD compete with longtime rival Nvidia in building an ecosystem of AI-specific chips. Nvidia already has a series of open-weight world models like Cosmos, while AMD only offers text- and video-based models to the public.
World models are considered essential in efforts to deploy generative AI models to robotic platforms ranging from self-driving cars to industrial robots to general purpose humanoids. In particular, the lack of real-world data to help train general-purpose robots means that synthetic data from world models will be key to realizing the vision advocated by companies like Tesla and Figure.
The acquisition is expected to close by the end of the year, subject to regulatory approval.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
