AMD 以约 82 亿美元全股票收购 World Labs,李飞飞将出任首席科学家
What AMD’s World Labs acquisition means for enterprises building with world models
AMD 宣布以约 82 亿美元全股票收购 Marble 世界模型开发商 World Labs,交易预计 2026 年底前完成,尚待监管批准。World Labs 联合创始人李飞飞将加入 AMD 任执行副总裁兼首席科学家,团队继续模型研究。
AMD has agreed to acquire World Labs, the developer of the Marble world model, in an all-stock deal valued at about US$8.2 billion. The deal will put a tool that robotics and design companies have been testing under the ownership of a chipmaker.
The transaction was announced on Monday (September 28) and is expected to close by the end of 2026, subject to regulatory approvals. World Labs co-founder and CEO Fei-Fei Li will join AMD as executive vice president and chief scientist, and her team will continue to focus on model research. AMD said the lab’s work would help it understand how AI workloads are changing and would feed into its future technology roadmaps.
How enterprises are using World Labs’ world models
World models generate 3D environments that people, software and robots can move through. Marble, World Labs’ first commercial product, was released in November 2025. It lets users create editable 3D environments from text or visual inputs and download them for use in other tools.
One documented example comes from robotics. Lightwheel, which builds simulation and evaluation infrastructure for physical AI, worked with World Labs on a series of demonstrations that used Marble to create robot test environments from a single 360-degree photo or a few casually taken pictures, according to a case study World Labs published in November 2025.
In that workflow, Marble produces a navigable 3D scene. The scene is converted into the USD file format using Nvidia’s Omniverse NuRec and loaded into Nvidia Isaac Sim, where Lightwheel adds its own simulation-ready assets and robots before running tests.
The case study argues that testing, rather than training, has become the main constraint in robotics development, because checking a robot model’s performance means running it across thousands of environments. Real-world testing, it says, is slow and hard to reproduce. Building exact digital replicas of real sites through dense scanning is expensive. Lightwheel’s approach relies instead on “digital cousins,” environments that are physically plausible and structurally similar to real spaces without copying them exactly.
World Labs said the approach cut environment creation time from weeks to minutes. The figure comes from the company’s own case study and has not been independently verified.
Design software is the second area. Autodesk invested US$200 million in World lab’s US$1 billion funding round in February, and the two companies said at the time they would explore how World Labs’ models can work alongside Autodesk’s tools, starting with media and entertainment.
Daron Green, Autodesk’s chief scientist, told TechCrunch at the time that the partnership was in its early days and that its final form had not been decided. He gave the example of a customer who might sketch an office layout with a world model and then refine specific items, such as a desk, in Autodesk’s software.
“You could anticipate us consuming their models or them consuming our models in different settings,” Green said. He added that data sharing was not part of the agreement.
A workflow that runs through Nvidia’s tools
The Lightwheel example shows where the deal gets complicated for enterprise users. In that workflow, Marble’s environments end up in Nvidia’s simulation software, and the process is documented in an Nvidia technical guide for other robotics teams. Nvidia also offers its own open-weight world models through its Cosmos family and invested in World Labs’ February round.
AMD said the acquisition supports its strategy to deliver AI infrastructure for an open ecosystem. The announcement did not say whether Marble’s export options or its compatibility with Nvidia’s simulation tools would change after the deal closes.
AMD plans to keep World Labs separate from its chipmaking business until the transaction is completed, CNBC reported. In a social media post after the announcement, Li said the team remained committed to its mission.
“We’re unwavering in our mission, and even more excited to continue building a world leading frontier research organization that is durable for the decades ahead,” she wrote.
See also: NVIDIA Jetson Orin Nano 2 brings physical AI to drones and robots

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including IoT Tech Expo and the Cyber Security & Cloud Expo. Click here for more information.
AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
来源:Artificial Intelligence News(网页) · artificialintelligence-news.com