AMD Buys Fei-Fei Li's World Labs for $8.2 Billion in Stock: A Chipmaker Acquires a World-Model Research Team

AMD announced on September 28 that it will acquire World Labs, the AI company founded by Fei-Fei Li, for about $8.2 billion in an all-stock transaction, expected to close by the end of 2026 subject to regulatory approvals and customary closing conditions. After closing, the World Labs team will continue to focus on AI model research, and Li will become AMD's executive vice president and chief scientist, reporting to CEO Lisa Su. Founded in 2024, World Labs builds spatial-intelligence models that understand three-dimensional physical environments, known as world models; its first product, Marble, generates 3D scenes from text, images and video and is also used to generate simulated environments for robot training. AMD was already a World Labs investor: World Labs emerged in 2024 with $230 million at a $1 billion valuation, and raised $1 billion in February 2026 from investors including AMD, Nvidia and Autodesk, without disclosing that round's valuation. The two companies formed an inference-optimization and training partnership last year. In the announcement, Su said building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving; Li said joining AMD will give the team the resources and engineering depth to accelerate its research. Several outlets report it is AMD's second-largest acquisition after its roughly $50 billion purchase of Xilinx in 2022, and AMD shares were roughly flat in after-hours trading.

Why a chipmaker buys a model team

AMD sells chips; World Labs builds models. The logic of the deal is in Su's line: to build the compute platforms for the next generation of AI, you first have to understand how models are changing. In other words, **AMD isn't buying a product; it is buying a research team that can tell it early what the next mainstream workload will look like.** World models deal with video, 3D geometry and physical simulation, workloads whose demands on memory, bandwidth and operators differ from today's text-centric large models. Bringing that research team in-house before the workload goes mainstream gives AMD's chip roadmap a first-hand reference. It is also one way for AMD to catch up with Nvidia. Nvidia already has a software stack for robotics and physical simulation in Isaac Sim and Omniverse; AMD chose to acquire a model team outright rather than build a platform from scratch.

How to read the price

World Labs was valued at $1 billion when it emerged in 2024. The company did not disclose a valuation for its $1 billion round this February; the "about $5 billion" reported earlier was only a target that circulated before the round. So it is not possible to say precisely how much $8.2 billion exceeds the last round. Two things are certain: **this is an all-stock deal, so World Labs' shareholders receive AMD stock; and AMD itself was on the investor list, moving from shareholder to buyer.** Nvidia was also an investor, and the announcement does not say how its stake will be handled at closing.

Fei-Fei Li's position

After closing, Li becomes AMD's executive vice president and chief scientist, reporting directly to Su. That arrangement suggests World Labs won't be folded into a business unit as a product line, but placed at the top of the company's research direction. She told TechCrunch that now that there is tangible proof of the possibilities, the team wants to do everything it can to accelerate. Fortune quotes a more direct line: without focused hardware effort, AI remains trapped in the digital world.

What this means for readers

For teams using Marble or World Labs' developer API, the announcement does not say how existing products and interfaces will be handled after the acquisition. Closing is expected by year-end, so watch for official notices in the meantime. For anyone tracking the AI chip landscape, the deal contrasts with Nvidia's approach: one side uses a software stack to keep developers on its hardware, the other brings a model research team in-house to decide how the hardware is built. Which works better will depend on whether AMD's future chips are actually designed for world-model workloads. Limits of what is known: the deal still requires regulatory approval; "AMD's second-largest acquisition" and the flat after-hours share price come from media reports, not the announcement.

via: AMD announcement, TechCrunch, Fortune, World Labs February 2026 funding announcement