AMD is buying World Labs for $8.2bn to put model research inside its chip roadmap

AMD has agreed to acquire Fei-Fei Li’s World Labs for approximately $8.2bn in stock. The transaction is expected to close by the end of 2026, subject to regulatory approval. Li will become AMD’s executive vice-president and chief scientist, reporting to chief executive Lisa Su, while the World Labs team continues its model research.

World Labs builds spatial-intelligence systems that generate, reconstruct and simulate interactive three-dimensional environments from text, images and video. Those ‘world models’ could be used for robotic learning, industrial simulation and design: instead of predicting the next word, they attempt to predict a new view or how a physical scene changes. The company began collaborating with AMD last year on training and inference optimisation using AMD GPUs.

The strategic logic is a feedback loop. A chip designer normally learns about emerging workloads through customers after models are already taking shape. Owning a research lab gives AMD earlier information about memory, networking, precision and software requirements—and gives World Labs direct influence over the hardware it needs. Nvidia has built a similar advantage by surrounding its accelerators with models, simulation software and robotics tooling.

The price makes this a large bet on a field whose commercial shape remains unsettled. World Labs was founded in 2024 and raised $1bn earlier this year; AMD and Nvidia were among its investors. The Information reports that Andreessen Horowitz owns nearly 14%. Neither current revenue nor evidence that world models will become a workload comparable to language-model training has been disclosed.

AMD is therefore buying more than intellectual property or talent. It is paying to reduce uncertainty about what comes after today’s generative-AI boom. If spatial models become essential infrastructure, that insight could guide several generations of hardware. If the field remains specialised, $8.2bn will look like a remarkably expensive research partnership.