AMD to Buy World Labs for $8.2 Billion in Nvidia AI Push
AMD will buy Fei-Fei Li's World Labs for $8.2 billion, adding Marble, Atlas and world-model talent to challenge Nvidia in robotics AI, pending approval.
Summary
AMD and World Labs announced on September 29, 2026, that AMD will acquire the AI startup in an $8.2 billion transaction expected to close by the end of 2026, pending regulatory approval. Fei-Fei Li founded World Labs in 2024 with Justin Johnson, Christoph Lassner and Ben Mildenhall, raising $230 million, including funding from AMD. The company develops world models that predictively simulate physical environments, with many architectures trained on vast video datasets. Marble generates small 3D spaces using Gaussian splats that can be exported for film, game development and other applications. World Labs later introduced Atlas and other models, although their specific uses remain limited in detail. Marble also drew attention at the 2026 SIGGRAPH conference.
AMD is targeting robotics, simulation and physical AI, particularly synthetic data for robot training, where Nvidia holds a strong lead. AMD already offers the open-source Micro-World model but lacks a stack comparable with Nvidia’s Cosmos models and applications. The acquisition adds World Labs’ technology and researchers while giving AMD insight into changing workloads that could shape future computing roadmaps. Li will become AMD executive vice president and chief scientist. Johnson and Mildenhall will continue leading the World Labs team with Li as it becomes a frontier research organization within AMD.
Positives
- The $8.2 billion acquisition adds World Labs’ models and leading researchers to AMD’s AI portfolio.
- Fei-Fei Li will become AMD executive vice president and chief scientist after the transaction closes.
- Justin Johnson and Ben Mildenhall will continue leading World Labs’ team with Li inside AMD.
- Marble already generates exportable 3D spaces for film, game development and other applications using Gaussian splats.
- World Labs’ workload expertise could help AMD shape chips for robotics, simulation, reasoning and physical AI.
Risks & concerns
- Nvidia retains a strong lead in robotics AI and synthetic training data through Cosmos and related applications.
- AMD’s open-source Micro-World does not yet provide a stack comparable with Nvidia’s world-model ecosystem.
- The $8.2 billion transaction cannot close until it receives regulatory approval.
- World Labs is being acquired only two years after its 2024 founding, sooner than expected for such a young company.
- Many world-model architectures require training on vast quantities of video data.