AMD has agreed to acquire World Labs, the company behind the Marble world model, in an all-stock transaction valued at approximately US$8.2 billion. The deal will bring a company whose technology is already being explored by robotics and design businesses under the ownership of one of the world’s major semiconductor companies.

Announced Monday, September 28, the transaction is expected to close by the end of 2026, subject to regulatory approvals.

As part of the agreement, World Labs co-founder and CEO Fei-Fei Li will join AMD as executive vice president and chief scientist. Her team will continue working on world-model research.

AMD said World Labs’ technology will help the company better understand how AI workloads are evolving and inform the development of future products and technology roadmaps.

What Are World Models?

World models are AI systems designed to generate and understand interactive representations of physical environments. Unlike conventional generative AI systems that primarily produce text, images, or video, world models can create 3D environments that people, software, and robots can navigate and interact with.

World Labs launched its first commercial product, Marble, in November 2025. The platform allows users to generate editable 3D environments from text prompts or visual inputs and export those environments for use in other software and workflows.

The technology has attracted interest from companies working in areas such as robotics, simulation, design, and entertainment.

How Enterprises Are Using World Labs’ Technology

One of the clearest examples comes from robotics company Lightwheel, which develops simulation and evaluation infrastructure for physical AI.

World Labs and Lightwheel worked together on demonstrations showing how Marble could be used to create robot-testing environments from a single 360-degree photograph or a small number of ordinary photographs, according to a World Labs case study published in November 2025.

The process starts with Marble generating a navigable 3D environment. That environment is then converted into the USD file format using Nvidia’s Omniverse NuRec and imported into Nvidia Isaac Sim.

Lightwheel can then add its own simulation-ready assets and robots before running tests inside the virtual environment.

Why Simulation Matters for Robotics

The Lightwheel example highlights one of the major challenges facing companies developing physical AI: testing robots at scale.

A robot model may need to operate across thousands of different environments before developers can understand how reliably it performs. Testing exclusively in the physical world can be expensive, slow, and difficult to reproduce.

Creating highly accurate digital replicas of real-world locations can also require expensive scanning and data-collection processes.

Lightwheel’s approach instead uses what it describes as “digital cousins” — physically plausible environments that resemble real locations without attempting to reproduce them exactly.

According to the World Labs case study, this approach reduced environment creation time from weeks to minutes. That figure comes from World Labs and has not been independently verified.

For enterprise robotics teams, the potential value is therefore not simply generating attractive 3D environments. It is the ability to produce large numbers of usable simulation environments quickly enough to support development and testing.

World Models and the Future of Design Software

Robotics is not the only enterprise application for World Labs’ technology. Design and media production are another area where the company has been building relationships.

Autodesk invested US$200 million in World Labs’ US$1 billion funding round in February. The companies said they would explore ways to integrate World Labs’ models with Autodesk’s software, initially focusing on media and entertainment workflows.

At the time, Daron Green, Autodesk’s chief scientist, said the partnership was still in its early stages and that its eventual form had not yet been determined.

One example he discussed was a customer using a world model to generate an initial office layout and then using Autodesk software to refine individual elements, such as a desk.

Green also described the relationship as potentially working in both directions, with Autodesk consuming World Labs’ models in some scenarios and World Labs’ technology consuming Autodesk models in others. He said data sharing was not part of the agreement.

The Nvidia Connection

AMD’s acquisition also raises questions about the technology stack surrounding Marble.

The Lightwheel workflow documented by World Labs ultimately moves Marble-generated environments into Nvidia’s simulation ecosystem. The process uses Omniverse NuRec and Isaac Sim, and Nvidia has published technical guidance describing the workflow for robotics developers.

Nvidia also has its own family of open-weight world models, known as Cosmos, and participated in World Labs’ February funding round.

This creates an interesting situation for enterprise customers. A company could use World Labs’ world-generation technology while relying on Nvidia software elsewhere in its simulation pipeline — even though World Labs will soon be owned by AMD.

What Could Change for Enterprise Customers?

The acquisition does not automatically mean that existing World Labs workflows will change.

AMD’s announcement did not specify whether Marble’s current export capabilities, integrations, or compatibility with Nvidia’s simulation tools will be modified after the transaction closes.

That question could matter to enterprises that have already built workflows around Marble and third-party software.

For companies using world models in production, interoperability can be just as important as the model itself. Enterprises typically need generated environments to move between modeling, simulation, design, robotics, and other software systems.

Maintaining those connections could therefore be an important consideration as World Labs becomes part of AMD.

AMD’s Broader AI Strategy

For AMD, the acquisition represents more than an investment in a generative 3D platform.

The company said World Labs’ research will help it understand how AI workloads are changing and contribute to future technology roadmaps. The move also gives AMD a deeper connection to a research team working on models designed to represent and generate physical environments.

That could become relevant as AI expands beyond language and image generation into robotics, simulation, autonomous systems, industrial design, and other physical-world applications.

AMD said its strategy is focused on supporting an open AI ecosystem, and the World Labs acquisition gives the company a direct stake in a technology area that could require substantial computing resources as enterprise adoption grows.

World Labs Will Continue Operating During the Transition

AMD plans to keep World Labs separate from its chip business until the acquisition is completed, according to CNBC.

Fei-Fei Li also said the team remains committed to its existing mission following the announcement.

In a social media post, Li said the company was committed to continuing to build a long-term frontier research organization.

For now, World Labs’ enterprise partnerships and existing technology ecosystem remain in place. The bigger question is how those relationships evolve once the acquisition closes.

What the Deal Means for Enterprises Building With World Models

The acquisition gives AMD ownership of a technology that sits at the intersection of AI, 3D generation, robotics, simulation, and design software.

For enterprises, the immediate impact may be limited because the transaction is not expected to close until the end of 2026 and remains subject to regulatory approval.

The longer-term implications could be more significant. World models are increasingly being explored as infrastructure for creating virtual environments, testing robots, developing designs, and building applications that interact with the physical world.

AMD’s ownership could influence how that technology develops and how it fits into the broader AI hardware and software ecosystem.

The key issue for enterprise users will be whether World Labs continues to support the interoperability and workflows that have made Marble useful across different technology stacks — including environments built around Nvidia’s tools — while gaining access to AMD’s resources and AI infrastructure.