AMD Buys Fei-Fei Li's World Labs for $8.2B in All-Stock Deal
AMD will absorb World Labs, the spatial intelligence startup co-founded by Fei-Fei Li, in an all-stock transaction valued at roughly $8.2 billion.
AMD said today it will acquire World Labs, the AI research lab co-founded by Stanford professor Fei-Fei Li, in an all-stock transaction valued at approximately $8.2 billion. The deal hands the chipmaker a young company that has spent barely two years building models designed to generate and reason about three-dimensional space. It is one of AMD's largest AI acquisitions to date and signals that the company intends to compete on software and model development, not just silicon.
World Labs launched in 2024 and reached a $1 billion valuation within months of its founding, a pace that reflected both Li's standing in computer vision and investor appetite for anything tied to spatial intelligence. The startup shipped its first commercial product, a world generation model, earlier this year. That model produces navigable 3D environments from prompts, a capability with obvious applications in robotics simulation, game development, and industrial design.
Li is one of the most cited researchers in modern AI. She led the ImageNet project, the labeled image dataset that helped trigger the deep learning boom in the early 2010s, and has since become a prominent voice in policy debates about how AI should be built and governed. Her move to AMD — assuming she stays through the acquisition — gives the chipmaker a public-facing technical figure to rival the research talent assembled at Nvidia and Google DeepMind.
The all-stock structure matters. AMD is paying with equity rather than cash, which preserves its balance sheet for manufacturing commitments and keeps World Labs' backers tied to AMD's share price. At $8.2 billion, the price is roughly eight times what World Labs was worth at its last private mark, a multiple that only makes sense if AMD believes 3D world models will become a standard layer in robotics and simulation workloads.
For AMD, the strategic logic runs through its Instinct accelerators and the ROCm software stack. Selling GPUs against Nvidia means offering developers a reason to port their training and inference pipelines, and owning a flagship model lab gives AMD a reference workload it controls end to end. Rivals have taken similar routes: Nvidia builds its own foundation models, and Google folds DeepMind research into its TPU roadmap.
The acquisition still faces regulatory review, and AMD has not said how World Labs will be structured or whether Li will take a named executive role. The open question is whether a world-model lab can thrive inside a chip company whose core business is selling hardware by the million. Watch for retention terms for World Labs' research staff and for the first AMD-built model to appear in a public benchmark.