AMD agreed to acquire World Labs, the spatial-intelligence research lab led by Fei-Fei Li, for $8.2B in an all-stock deal. Li becomes AMD's executive vice president and chief scientist, reporting directly to CEO Lisa Su. It's AMD's second-largest acquisition on record, behind only the roughly $50B Xilinx deal in 2022.
World Labs isn't a revenue business AMD is buying for its P&L. Founded in 2024, it builds “world models” — systems that generate, reconstruct, and simulate interactive 3D environments from text, image, and video input, aimed at robotics and physically-grounded AI. AMD and NVIDIA were both early investors in World Labs' funding rounds before this deal; now AMD owns the lab outright, and NVIDIA doesn't.
Buying a roadmap, not a product
AMD's own stated rationale is the most telling part: bringing model researchers in-house gives the company deeper insight into how emerging workloads are evolving, so it can shape future hardware, software, and systems around them, rather than reacting to what labs need after the fact. As Li put it in her own note on the move, AI has been “hobbled in efficiency” without a focused hardware effort behind it — she wants to work closer to the metal to solve problems software alone can't.
That's a different acquisition logic than buying a competitor or a customer base. It's buying the ability to design chips around workloads — simulation, robotics, reasoning over 3D and physical environments — before those workloads are mainstream enough for competitors to be optimizing for them yet.
Context: everyone's buying research labs now
This isn't happening in isolation. NVIDIA has spent roughly $33B combined this year acquiring AI chipmaker Groq's assets and agreeing to acquire Hugging Face. AMD's own World Labs deal is its second acquisition of 2026. The pattern across both companies is the same: chip vendors buying their way into the model-research layer, not just to own IP, but to shorten the feedback loop between what models actually need and what silicon gets built two years out.
For anyone thinking about inference economics, that's the number worth watching from this deal — not the $8.2B price tag, but what AMD's silicon roadmap looks like once it's being shaped by researchers building tomorrow's workloads instead of specs written for today's.