AMD has announced its acquisition of World Labs, a top developer of deep learning models aimed at comprehending physical reality, in a deal valued at $8.2 billion.
According to World Labs, the partnership underscores the necessity for “close collaboration across model research, systems, and compute” to advance AI development. AMD emphasizes that grasping advanced workloads, such as those generated by World Labs, will influence its future chip design strategy.
As part of the acquisition, Fei-Fei Li, the founder of World Labs, will step into the role of executive vice president and chief scientist at AMD. Last year, the two companies established a partnership focused on inference optimization and training, and their collaboration has continued to grow. Li notably participated in AMD’s presentation at CES earlier this year.
Li, a professor of computer science at Stanford, is recognized as a leading figure in artificial intelligence, especially in the field of computer vision, due to her creation of the ImageNet database and the subsequent competitions it inspired. In 2024, she founded World Labs to create deep learning models with a comprehensive understanding of the physical world, asserting that genuine general intelligence must incorporate physics and the capacity to interpret and analyze data beyond mere text.
In a statement about the acquisition, Li expressed her hope to expand World Labs’ technological advancements beyond the development stage. “Now that we have proof of what is possible, we aim to do everything in our power to propel the future forward,” Li stated. “This requires scaling our efforts, broadening our scope, and fostering closer ties with hardware.”
The term “world model” is broadly defined, covering various applications from language models that interpret visual inputs to those that can create and maintain realistic simulations. World Labs’ initial product, Marble, is designed not only for crafting entertainment experiences but also for generating simulated environments suitable for training robots.
This acquisition will likely bolster AMD’s position against its established competitor Nvidia in the realm of AI-specific chip development. While Nvidia boasts a range of open-weight world models, such as Cosmos, AMD has so far only made text- and video-based models available to the public.
World models are crucial for deploying generative AI on robotic platforms, including self-driving vehicles and both industrial and humanoid robots. The lack of practical real-world data for training versatile robots highlights the need for synthetic data derived from world models to achieve the ambitions set forth by companies like Tesla and Figure.
The acquisition is anticipated to finalize before the year’s end, pending regulatory approval.
