AMD has completed the acquisition of World Labs. AMD purchased the AI startup for $8.2 billion. The company, founded in 2024 by scientist Fei-Fei Li, develops world models for generating and simulating interactive 3D environments. Li will become chief scientist at AMD and will report directly to CEO Lisa Su.
The acquisition was announced in late September. AMD paid approximately $8.2 billion for it, but entirely in stock. It is a massive sum for a startup that raised “only” $1 billion in funding. AMD and World Labs had already been collaborating; since 2025, they have been building models that can be trained and run on AMD GPUs.
It may seem like a somewhat unusual move for a chip company to acquire an AI model developer. But the chips needed to build the models can, in turn, improve those very chips and the surrounding infrastructure. According to AMD CEO Lisa Su, the researchers will help shape future hardware, software, and systems.
World models
World Labs is therefore focused on so-called world models. This is a specific branch of AI that creates and simulates interactive 3D environments. Nvidia markets such innovations under the Omniverse brand and also occasionally refers to them as world models. In addition, the company is developing technology for robotic learning and simulation, yet another discipline in which Nvidia is already a major player. Its best-known product is Marble, which generates editable 3D worlds based on text, images, or video. During an AMD demonstration, Marble created a 3D scene from just a few images.
Li is a well-known figure in the AI industry. As a Stanford researcher, she was instrumental in creating the highly influential ImageNet dataset, which accelerated the development of computer vision. Today, there is also overlap with the now-common array of leading AI labs. Google DeepMind, for instance, has also been working for quite some time on AI that simulates the physical world.
Bringing models and hardware closer together
The World Labs team remains committed to conducting fundamental model research, according to the announcement. At the same time, the researchers will collaborate with AMD engineers to translate insights from new model capabilities into AI infrastructure for customers and developers. AMD explicitly positions this within an open AI ecosystem. It is therefore to be expected that, just as with Nvidia’s Nemotron, actual models will continue to be released. According to AMD’s earlier press release, the acquisition is intended to give the chipmaker greater influence over the development of AI models and infrastructure for spatial intelligence. The applications are numerous, ranging from industrial robotics to weather forecasting, 3D simulation and product development, and AI-driven chip design.