A chipmaker just bought an AI lab for $8.2 billion
On September 28, AMD announced a definitive agreement to acquire World Labs — the AI research startup founded by Stanford professor Fei-Fei Li — in an all-stock deal valued at approximately $8.2 billion. Li herself will join AMD as executive vice president and chief scientist, reporting directly to CEO Lisa Su. The deal is expected to close by the end of 2026, pending regulatory approvals.
It's one of the largest AI acquisitions of the year, and it signals where the industry's center of gravity is moving: from chatbots that manipulate text to AI systems that understand the physical world.
What is World Labs?
World Labs, founded by Li in 2024 and based in San Francisco, builds what researchers call "spatial intelligence" — deep learning models that generate, reconstruct, and simulate interactive 3D environments from text, image, and video inputs. Its first product, Marble, is pitched both as a tool for creating entertainment experiences and for building simulated worlds in which to train robots.
Li is one of the most influential figures in modern AI. Her creation of ImageNet — the massive labeled image dataset — helped spark the deep-learning revolution in computer vision. She previously led Stanford's AI Lab, founded Stanford's Institute for Human-Centered AI, and served as vice president and chief scientist of AI/ML at Google Cloud.
Why would a chip company buy a model lab?
AMD's explanation is about silicon, not software. "Building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving," Su said. Chips take years to design — a chipmaker that sees the next major workload early can build hardware for it before rivals do.
That next workload, in AMD's bet, is "world models": AI that simulates physical space, considered essential for putting generative AI onto robots, from autonomous vehicles to general-purpose humanoids. Nvidia already offers open-weight world models under its Cosmos brand; AMD, until now, had offered only text- and video-based models publicly. The acquisition closes that gap in one move — and gives AMD the researchers who understand the workload best. "We are acquiring World Labs to gain access to the top-tier talent that Fei-Fei has assembled," Su said in a television interview.
The two companies aren't strangers: they formed a training and inference-optimization partnership last year, AMD invested in World Labs' $1 billion funding round earlier this year, and Li appeared on stage at AMD's CES keynote in January.
The bigger picture: the physical-AI arms race
The deal lands less than a month after Nvidia announced its own $12.9 billion acquisition of Hugging Face, the open-access AI model platform. Together, the two moves show the AI chip giants racing to own not just the hardware but the model ecosystems that will define the next wave — robotics, simulation, and "physical AI."
For the rest of us, the practical takeaway is timing. World models are still largely research-stage, but when the labs building them start merging with the companies building the chips, commercial deployment usually follows within a couple of product cycles. Expect robots that learn in simulation before touching the real world to move from demo videos to warehouses — and eventually homes — faster than most people assume.
The bottom line
AMD didn't buy an app or a user base. It bought a window into the future of AI workloads — and the "godmother of AI" to help read through it. In the race to power physical AI, understanding the models may matter as much as manufacturing the chips.
Disclosure: This post contains affiliate links. If you buy through them, I may earn a commission at no extra cost to you.
