AMD agrees to buy Fei-Fei Li’s World Labs for $8.2 billion

AMD agreed on Sept. 28 to acquire World Labs for about $8.2 billion in a deal paid entirely in stock. Fei-Fei Li, the startup’s cofounder and CEO, would become AMD’s executive vice president and chief scientist after the deal closes, reporting directly to AMD CEO Lisa Su.

The acquisition, covered in The Rundown, is expected to close by the end of 2026, subject to approvals and customary conditions. It would bring a lab building AI models that generate 3D worlds inside one of the companies supplying the chips to run them.

A partnership that began in 2025

Li says the teams began working together in 2025, optimizing how World Labs’ models train and run on AMD GPUs. She described Su as an early investor and supporter and said her team needs to get “closer to the hardware” to make its AI more efficient.

Both AMD and Nvidia took part in World Labs’ $1 billion funding round announced Feb. 18, 2026. The proposed acquisition would turn AMD from one of several investors into the lab’s owner.

What World Labs is building

World Labs made its first product, Marble, generally available on Nov. 12, 2025. The company says it can turn text, images, video and rough 3D layouts into worlds that users can edit, expand and combine. Users can then export 3D assets or videos for further work.

On Sept. 1, World Labs introduced Atlas, which it says handles generation, reconstruction and simulation across text, images, video and 3D. The announcement included robotics examples such as reconstructing environments and producing simulated views from robot cameras. Atlas was entering early access with selected partners and is intended to power future versions of Marble.

Why it matters

AMD has spent the AI boom competing as an alternative to Nvidia. Bringing Li and her team inside the company could give it a more direct role in developing the models that may drive future demand for chips. The company would also gain one of AI’s most influential researchers, whose career spans ImageNet, Google Cloud’s AI organization and Stanford HAI.

For AMD’s chip and software engineers, the potential gain is earlier access to the problems that emerge as new models are built. An internal lab could help identify where training or running models becomes slow or costly while design choices are still open. AMD says those insights will help shape its hardware, software and systems plans. The partnership underway since 2025 gives that goal a practical foundation.

If that work improves efficiency, creators could generate and revise 3D settings with less computing power. Robotics teams could benefit when building simulated environments and camera views. Whether the acquisition produces those gains remains uncertain.

World Labs says Atlas fills unseen areas with plausible details, so a reconstruction can include features the model invented. Teams developing robots would need to test whether those environments provide reliable training.

Any advantage for AMD would depend on demand for world models and how well the lab’s insights translate into chip and software design. Nvidia already has world models of its own.

Bringing World Labs inside AMD could also tie its research more closely to one chipmaker’s priorities. Li has expressed an intention to offer open models and platforms, but licenses, release schedules and support across chipmakers remain unclear.