- Published on
AMD acquires World Labs AI startup, upping the ante against Nvidia
- Authors
- Name
- DeepHorus
AMD to acquire World Labs in $8.2 billion AI deal
AMD and world-model company World Labs announced that AMD will acquire the AI startup by year-end, pending regulatory approval. The $8.2 billion deal would strengthen AMD’s position in physical-world AI. It also raises the competitive stakes with Nvidia in robotics.
World Labs was founded in 2024 by computer vision scientist Fei-Fei Li and researchers Justin Johnson, Christoph Lassner and Ben Mildenhall. It raised $230 million, including funding from AMD. The chipmaker’s prior investment connects it to the company before the takeover. Completion still requires regulatory approval.
World Labs develops world models, AI systems intended to provide useful, predictive simulations of the physical world. Many competing architectures train on vast amounts of video data. Their focus is to simulate dynamic surroundings for advanced AI applications. That capability is relevant to robotics and similar applications.
Marble, World Labs’ first publicly usable tool, generates small 3D spaces using Gaussian splats. The spaces can be exported as 3D assets for film, games and other applications. World Labs has since released models and tools with different emphases, including Atlas. These releases broaden its public product portfolio.
Robot training is a key use case, particularly for generating synthetic data. Nvidia leads in that field, while AMD lacks a stack comparable to Nvidia’s Cosmos models and related applications. AMD has also developed the open-source Micro-World world model. The acquisition could strengthen AMD’s competitive position.
The announcement gave no regulatory timetable, integration plan or product roadmap. If completed, World Labs’ technology would join AMD’s chip operation. AMD’s earlier funding makes the transaction an extension of an existing relationship. Its significance will depend on approval and whether AMD can narrow Nvidia’s lead in robotics.
