The Godmother of AI Joins the Chipmaker: AMD Buys Fei-Fei Li's World Labs for $8.2 Billion

At 4:05 pm EDT on Monday, September 28, 2026, AMD dropped one of the biggest AI acquisition announcements of the year: the chipmaker has signed a definitive agreement to acquire World Labs, the San Francisco spatial-intelligence startup led by AI pioneer Dr. Fei-Fei Li, in an all-stock transaction valued at approximately $8.2 billion. When the deal closes — expected by the end of 2026, subject to regulatory approvals — one of the most celebrated researchers in the history of the field will walk into AMD as Executive Vice President and Chief Scientist, reporting directly to CEO Dr. Lisa Su.

For a company that has spent two years fighting for position in AI accelerators against Nvidia’s overwhelming market share, this is not a routine talent acquisition. It is a statement about what AMD believes the next era of AI compute will look like — and who it thinks should define it.

What World Labs actually does

World Labs was founded at the beginning of 2024 by Fei-Fei Li — the Stanford professor whose ImageNet dataset effectively launched the deep-learning era in 2012 — together with co-founders Ben Mildenhall (of NeRF fame) and Justin Johnson, along with Christoph Lassner. The company emerged from stealth that September with a $230 million round at a $1 billion valuation, and by February 2026 it had raised a further $1 billion from investors including AMD itself, Autodesk, and reportedly Nvidia — a striking fact given that two chip rivals both held stakes in the startup.

The company’s thesis is straightforward and provocative: language alone is not enough for AI. As Li wrote in her announcement essay, “The universe isn’t made up of words; it’s made of real things.” World Labs builds spatial-intelligence foundation models — models that generate, reconstruct, and simulate interactive 3D environments from text, image, and video inputs, plus technology for robotic learning and simulation.

Its flagship proof point is Atlas, a first-of-its-kind “omni model” architecture that tackles a longstanding open problem in computer vision: new camera view prediction. Just as an LLM predicts the next token from a line of text, Atlas — trained from scratch — predicts the next view from a set of 2D images, outperforming even specialized state-of-the-art models and, according to Li, essentially solving the classic sparse-reconstruction problem by combining generative modeling with multiview geometry. In July 2026, the company also acquired SceniX, a startup building real-to-sim simulation for robot learning, adding a robotics stack to its portfolio.

Why AMD wants them

The official rationale is workload insight. “Building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving,” Lisa Su said in the press release. “Fei-Fei and the World Labs team bring exceptional research leadership and model expertise. Together, we can use that insight to develop the hardware, software and systems that will power the next generation of AI and strengthen the open AI ecosystem.”

That last phrase — open AI ecosystem — is the strategic heart of the deal. AMD’s best opening against Nvidia has always been openness: its ROCm software stack versus CUDA’s proprietary lock-in. World Labs reinforces that axis. In her essay, Li committed to “providing the best open models and platforms for running AI end to end from hardware to software to models to the data platforms needed to fuel it all.” An $8.2 billion purchase of a frontier model lab, explicitly premised on open models running close to the metal, positions AMD as the full-stack open alternative in an industry where the biggest players are increasingly verticalizing.

There’s a personal dimension too, and AMD’s marketing clearly knows it. Lisa Su was an early investor in and believer in World Labs; the two companies began a deep technical partnership last year, starting with model training and inference optimization on AMD GPUs. As the teams worked together, the argument goes, the fit became obvious. Li’s own telling is almost romantic — she cites Tennyson’s Ulysses (“To seek a newer world”) and describes a career that has run from ImageNet, through founding Google Cloud’s AI unit and Stanford’s Human-Centered AI institute, to this moment: running a frontier research team inside a leading chip company.

The market’s first verdict: skepticism

Investors were not immediately sold. AMD shares dropped roughly 4% on the news, adding to a year in which the stock has been punished for conservative guidance and doubts about its ability to dent Nvidia’s AI dominance. An $8.2 billion all-stock payout for a company that — despite its talent — has no revenue engine on the scale of its acquirer is the kind of move that invites scrutiny. The deal values World Labs at roughly 8x its last known private valuation discussions (Bloomberg reported talks around $5 billion in January 2026), meaning AMD is paying a premium measured in billions for research leadership and strategic direction rather than current earnings.

The bears’ argument: AMD’s problem has never been a shortage of brilliant scientists — it’s CUDA, supply, and the entrenched software ecosystem around Nvidia hardware. The bulls’ counter: as AI expands into reasoning, robotics, simulation and physical AI, compute demand diversifies, and the company that understands tomorrow’s workloads earliest designs the silicon that wins them. Owning the lab that defines spatial-intelligence model architecture is a way to buy that foresight rather than license it.

What it means for the industry

Three implications stand out.

First, the chip-model convergence is now explicit. Every major accelerator vendor is being pulled toward owning model research — or at least co-owning its direction. When model architectures shape what hardware must do next, a chipmaker without a frontier lab is designing blind. Expect this deal to accelerate comparisons: Nvidia funds and partners deeply with model companies, and now AMD owns one outright.

Second, “spatial intelligence” just got an $8.2 billion endorsement. Li’s thesis — that the next frontier after language models is AI that perceives, generates, and simulates the 3D world — has been influential but contested. A deal of this size, from a top-tier semiconductor company, effectively underwrites the field. Robotics simulation, 3D reconstruction for design and construction, scene generation for entertainment and therapy: World Labs reported interest across all of these domains for Atlas, and now those workloads have a patron with an existential interest in serving them.

Third, the open-ecosystem bloc hardens. With World Labs’ team forming what Li calls “a world leading frontier research organization” inside AMD, and an explicit commitment to open models end-to-end, the industry’s open coalition — AMD hardware, open weights, open platforms — gains a flagship research brand. In a market where Chinese open-weight models already route a majority of global tokens by some measures, openness is no longer a niche stance; it is a competitive weapon, and AMD just armed it.

What happens next

The transaction is expected to close by the end of 2026, pending regulatory approvals and customary conditions. Justin Johnson and Ben Mildenhall will continue leading the World Labs team within AMD. Fei-Fei Li becomes EVP and Chief Scientist under Lisa Su — a remarkable landing for a researcher who, twenty-six years after entering the field as a graduate student “chasing curiosity about visual intelligence,” now holds one of the most senior technical positions in the semiconductor industry.

Whether the bet pays off will take years to judge. But on a Monday afternoon two years after founding a startup on the belief that AI must learn the physical world, the woman often called the “Godmother of AI” chose to make that argument from inside a chip company. The message to the rest of the industry is hard to miss: the next round of AI competition will be fought over hardware-software-model co-design — and AMD just bought itself a seat at that table it never previously had.