AMD's Strategic Leap: World Labs Acquisition Ignites New Era for Enterprise AI

AMD has announced its agreement to acquire World Labs, the innovative developer behind the Marble world model, in an all-stock transaction valued at approximately US$8.2 billion. This significant deal, announced on Monday, September 28, is anticipated to conclude by the end of 2026, pending necessary regulatory approvals. The acquisition positions a critical tool, currently being explored by robotics and design firms, under the ownership of a prominent chipmaker.

As part of the acquisition, Fei-Fei Li, co-founder and CEO of World Labs, will transition to AMD as executive vice president and chief scientist. Her team will continue its dedicated focus on model research, contributing to AMD's strategic vision. AMD has articulated that World Labs' pioneering work will be instrumental in understanding the evolving landscape of AI workloads and will directly inform the company's future technology roadmaps, fostering an open ecosystem for AI infrastructure.

World models are advanced technologies designed to generate complex 3D environments through which individuals, software, and robots can navigate. Marble, World Labs’ flagship commercial product, was launched in November 2025. It empowers users to create editable 3D environments from diverse inputs, including text or visual data, and subsequently download these environments for integration into other software tools.

One compelling application of World Labs’ world models is in the field of robotics. Lightwheel, a company specializing in simulation and evaluation infrastructure for physical AI, collaborated with World Labs on a series of demonstrations. These demonstrations showcased Marble's ability to swiftly generate robot test environments from a single 360-degree photograph or a few casual images, as detailed in a November 2025 case study published by World Labs. Within this workflow, Marble generates a navigable 3D scene, which is then converted into the USD file format using Nvidia’s Omniverse NuRec. This USD file is subsequently loaded into Nvidia Isaac Sim, where Lightwheel integrates its proprietary simulation-ready assets and robots to conduct comprehensive tests.

The case study argues that the primary bottleneck in robotics development has shifted from training to testing, as evaluating a robot model’s performance necessitates running it through thousands of diverse environments. Traditional real-world testing is characterized by its slowness and difficulty in reproduction, while constructing precise digital replicas of physical sites through dense scanning is prohibitively expensive. Lightwheel’s innovative approach, leveraging Marble, instead utilizes “digital cousins”—environments that are physically plausible and structurally analogous to real spaces without being exact copies. World Labs claims this method drastically reduces environment creation time from weeks to mere minutes, though this figure has not been independently verified.

Another significant area of application for World Labs’ technology is design software. Autodesk, a key player in this sector, invested US$200 million in World Labs’ US$1 billion funding round in February. At the time, both companies announced their intention to explore the integration of World Labs’ models with Autodesk’s existing tools, commencing with media and entertainment applications. Daron Green, Autodesk’s chief scientist, noted that the partnership was in its nascent stages. He provided an example of a customer using a world model to sketch an office layout, then refining specific elements, such as a desk, within Autodesk’s software. Green indicated that either company could potentially consume the other’s models in various settings, clarifying that data sharing was not part of their initial agreement.

The acquisition, however, introduces complexities for enterprise users, particularly concerning Nvidia. The Lightwheel workflow, for instance, routes Marble’s environments through Nvidia’s simulation software, a process extensively documented in an Nvidia technical guide for other robotics teams. Furthermore, Nvidia itself offers its own suite of open-weight world models through its Cosmos family and was a participant in World Labs’ February funding round. AMD has affirmed that this acquisition aligns with its overarching strategy to provide AI infrastructure for an open ecosystem. Nevertheless, the announcement did not specify whether Marble’s export functionalities or its compatibility with Nvidia’s simulation tools would undergo changes post-acquisition.

Prior to the transaction’s completion, AMD plans to maintain World Labs as a separate entity from its core chipmaking business, as reported by CNBC. Following the acquisition announcement, Fei-Fei Li reiterated her team’s steadfast dedication to their mission in a social media post. She expressed heightened enthusiasm for continuing to build a world-leading frontier research organization designed for long-term durability in the decades ahead.