World Labs co-founder Fei-Fei Li on the AMD deal: “The universe isn’t made up of words; it’s made of real things…”

Fei-Fei Li is the scientist whose work taught machines to see, and she has just sold the lab she built to push AI past language. Announcing that World Labs will become part of AMD, the chipmaker run by her early backer Lisa Su, she restated the belief the company was founded on in a single line that doubles as a bet against the words-first era of AI.

“The universe isn’t made up of words; it’s made of real things. So many of the most important problems for AI to solve, from science to entertainment to robotics, require models to reason over the structure and behavior of the physical world.”

Li wrote it in “World Labs to AMD”, a post on her Substack blog, and The Register quoted it in its report on the deal. The sentence just before it sets the terms: the company was “guided by the foundational belief that language alone isn’t sufficient for model development.”

AMD has agreed to acquire World Labs in an all-stock transaction valued at approximately $8.2 billion, expected to close by the end of 2026 if regulators sign off. Li will become AMD’s executive vice president and chief scientist, reporting directly to Su. The San Francisco lab, which she co-founded at the start of 2024 with Ben Mildenhall and Justin Johnson, builds what the industry calls world models: systems that generate, reconstruct and simulate 3D environments from text, images and video. Where a large language model predicts the next word, a world model, as The Register put it, imagines the environment and works the problem from there.

The easy reading files the line with Yann LeCun, Meta’s former chief AI scientist, who has suggested large language models may be a dead end. Li’s version is narrower; The Register was careful to note she isn’t anti-LLM. Her lab’s newest model borrows the LLM playbook: Atlas, she writes, predicts the next view from a set of 2D images the way a language model predicts the next token. Her quarrel is not with the method. It is with the material. Text is a record of the world written by people; her argument is that the problems worth solving, from science to entertainment to robotics, live in the thing the text describes.

What makes the sentence consequential is who paid for it. A chipmaker does not spend billions in stock on a philosophy. AMD’s announcement frames the purchase around compute, noting that as AI expands into robotics, simulation and physical AI, “the demands on compute infrastructure become more diverse.” Su’s own line is the tell: “Building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving.” In plain terms: AMD wants to see the next workload before it designs the silicon to run it. Next to Nvidia, The Register notes, AMD’s software engineering bench is tiny; World Labs hands it an in-house research team, only weeks after it agreed to buy Taalas, a startup that bakes model weights directly into silicon.

Li describes the same mechanism from the other side. “Without having a focused hardware effort, AI is hobbled in efficiency. And scale. And for our purposes, remains trapped in the digital world,” she writes. That is the working content of the quote. A model that reasons about physical space is only as useful as the machine that can run it fast enough to steer a robot, and Li has decided the fastest route to that machine is to move inside the company that makes it.

The phrase carries weight because of the career behind it. Li entered the field in 2000 chasing visual intelligence, in machines and in the brain. The ImageNet work from her lab helped open the modern AI era; she later founded Google Cloud’s AI unit as its chief scientist and became founding director of Stanford’s Human-Centered AI institute. A quarter century on, the argument has not changed. The stakes have. Once the deal closes, the professor who made the case reports to Su.

If she is right, AMD has bought an early look at the workloads that will define the next generation of AI hardware. If she is wrong and language keeps absorbing everything, it has bought an expensive hedge, which is exactly how The Register read the deal: insurance against LLMs turning out to be a dead end.

Li’s sentence starts as a statement about the universe. By the time the deal closes, it will be a spec sheet.

Tags: artificial intelligence, AMD, Fei-Fei Li, World Labs, World models

World Labs co-founder Fei-Fei Li on the AMD deal: “The universe isn’t made up of words; it’s made of real things…”

More Like This

AMD bets that etching AI weights into silicon forever will beat Nvidia at inference

DeepSeek made AI cheap, and the American AI bubble was built on it being expensive

Nvidia became the AI economy’s single point of failure. That was the bet

Meituan’s LongCat 2.0 runs on chips the US can’t export and it just beat GPT-5.5

The grief tech industry built its $5 billion business on the dead. Nobody asked them

Microsoft’s Own Files Called AI Scraping the ‘Largest Theft of Labor in Human History’

Discussion