AMD Buys Fei-Fei Li's World Labs for $8.2 Billion

AMD told investors Monday it will pay $8.2 billion, entirely in its own stock, for a company that has existed for 19 months. World Labs makes software that lets a machine understand three-dimensional space. That is the kind of perception a warehouse robot needs before it can act on what a camera shows it. Fei-Fei Li, the Stanford professor who built the dataset that helped start the deep-learning era, co-founded the company in February 2024. She becomes AMD's executive vice president and chief scientist once the deal closes¹. World Labs employs roughly 70 people. The price values each of them above $100 million. That number has AMD's own shareholders, and at least one of its chip rivals, asking what exactly the company just bought.

The godmother of AI takes a corner office

Li spent her career building the raw material other people's breakthroughs ran on. At Princeton, then Stanford, she led the team that built ImageNet, a database of 3.2 million labeled images published in 2009⁶. It became the benchmark that later proved deep learning worked at scale, the training ground under nearly every computer-vision system built since. She directed Stanford's AI lab from 2013 to 2018. Then she took a sabbatical to run AI and machine learning for Google Cloud⁶. There she pushed tools built to put machine learning within reach of ordinary developers. Back at Stanford, she co-founded the Human-Centered AI Institute and, in 2017, a nonprofit called AI4ALL, aimed at widening who gets to build the technology. Reporters had already started calling her the godmother of AI by the time she left Google Cloud, a title that followed her into every headline about the AMD deal⁵. World Labs was her return to building a product instead of a body of research. She started it with Justin Johnson, Christoph Lassner, and Ben Mildenhall, three researchers with their own records in computer vision and graphics. The startup emerged from stealth in September 2024 with $230 million raised at a $1 billion valuation, a unicorn twelve weeks after its first check cleared. Few academic founders reach that speed on a first company.

AMD already owned a piece of what it just bought

World Labs raised $1 billion again in February 2026, at a valuation investors have put near $5 billion. AMD was one of the investors. So was Nvidia¹. Seven months later, AMD is paying $8.2 billion for the same company, a 64 percent jump over that February number, and it is paying in AMD stock. That detail matters more than it first appears. World Labs' shareholders, Nvidia included, now hold shares in the chipmaker racing hardest to catch it⁵. A rival chip company helped fund the team AMD just bought outright, and now collects stock in the buyer as its payout. AMD's own newsroom framed the purchase as continuity rather than a leap. "Building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving," said Lisa Su, AMD's chair and chief executive. "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"¹.

What $8.2 billion actually buys

World Labs already sells a product. It moved past the pure-research stage months before AMD came calling. Marble, its commercial world model, launched in November 2025 and sells consumer plans up to $95 a month⁴. Atlas, a newer algorithm that generates virtual replicas of real places from photographs, launched Sept. 1, 2026, just 25 days before AMD signed the deal⁴. In July, World Labs bought a smaller startup called SceniX, adding robotics specialists to the team⁴. Until then, the group had spent most of its short life generating virtual worlds rather than acting inside real ones. Li described the acquisition as a shortcut past years of infrastructure work she would otherwise have had to build herself. "Now that we have tangible proof of the possibilities, we want to do everything we can to accelerate the future," Li said. "To do this requires scaling our efforts, widening our reach, and getting closer to the hardware"². Getting closer to the hardware is the plainest way to describe what AMD supplies. It is the piece World Labs could reach alone only with far more time, and far more of its own capital.

Nvidia's head start survives the signing itself

Matt Kimball, vice president and principal analyst at Moor Insights & Strategy, read the deal as smart, and overdue at once. He named the ground AMD still has to cover. "Nvidia has already moved into physical AI with its Cosmos world model, along with its Isaac and Omniverse platforms and Project GR00T robotics models, which give it an early lead in the market," Kimball said³. Nvidia launched Cosmos at CES in January 2025, and Isaac and Omniverse predate it by years⁷. GR00T, Nvidia's open model for humanoid robots, followed the same lineage months later. Nvidia spent seasons building each into a platform other companies build on top of, the kind of position a single acquisition rarely reverses inside a year. Kimball's read on the deal's real target was blunter still. "Fei-Fei is a rock star. You could argue that she is as much of the target as the company is," he said. "She and her team bring world-class talent to AMD, and this will enable the company to much more quickly find a competitive footing against Nvidia in this space"³. A footing falls well short of a lead. AMD bought speed toward Nvidia's position. It still has to cover the distance to the position itself.

What each side needs to be true

For Su's bet to pay off, World Labs' models have to do more than exist inside AMD's roadmap. They need to shape it, the way Su described, turning research into chips that developers pick over Nvidia's for reasons beyond price. Marble and Atlas need to scale past novelty pricing into a genuine product line, something bigger than a signing bonus with a subscription tab attached. For Kimball's caution to hold, Nvidia's multi-year platform advantage in physical AI keeps compounding faster than 70 new employees, however credentialed, can close it. The Xilinx purchase gives AMD a precedent worth watching. Its $50 billion deal in 2022 remains the only acquisition in company history bigger than this one. Xilinx's chips took years to show up meaningfully in AMD's earnings². World Labs may follow the same clock.

By the numbers

• $8.2 billion: AMD's all-stock price for World Labs, announced Sept. 28¹.
• 70 employees: World Labs' headcount at signing, valuing each above $100 million⁴.
• $5 billion: the valuation investors placed on World Labs in its February 2026 round, in which AMD itself invested³.
• 64 percent: how much more AMD paid in September than that February valuation, seven months later.
• $95 a month: the top consumer tier of Marble, World Labs' commercial world model, launched in November 2025⁴.
• 25 days: the gap between World Labs launching Atlas and AMD signing the deal to buy the company that built it⁴.
• $50 billion: AMD's purchase of Xilinx in 2022, still the only acquisition larger than this one in company history².

What to watch

AMD's next earnings call will show whether it reports World Labs as a research cost center or a product line with its own revenue. Nvidia's response matters too. A counteroffer for a rival lab, or a sudden price cut on Cosmos licensing, would tell the market how seriously Jensen Huang's company takes the threat Kimball described. Marble and Atlas renewal numbers, whenever AMD discloses them, will separate a team that shipped from a team that only demonstrated. Robotics buyers, the warehouse and factory operators actually purchasing hardware this year, will settle the argument before any analyst does, simply by choosing whose stack they build on.

Sources

1. AMD, "AMD to Acquire World Labs to Advance the Future of AI," AMD Newsroom, Sept. 28, 2026, https://newsroom.amd.com/news/amd-acquire-world-labs/
2. Tim Fernholz, "AMD will acquire Fei-Fei Li's World Labs for $8.2 billion," TechCrunch, Sept. 28, 2026, https://techcrunch.com/2026/09/28/amd-will-acquire-fei-fei-lis-world-labs-for-8-2-billion/
3. Wylie Wong, "AMD's World Labs Deal Signals Shift to Physical AI," Data Center Knowledge, Sept. 29, 2026, https://www.datacenterknowledge.com/data-center-chips/amd-to-acquire-world-labs-for-8-2b-to-advance-ai-models-and-robotics
4. Implicator.ai, "AMD to Buy Fei-Fei Li's World Labs for $8.2 Billion," Implicator.ai, Sept. 29, 2026, https://www.implicator.ai/amd-buys-world-labs-8-2-billion-fei-fei-li/
5. South China Morning Post, "AMD buys Li Fei-Fei's World Labs for US$8.2b, escalating rivalry with Nvidia," South China Morning Post, Sept. 29, 2026, https://www.scmp.com/tech/big-tech/article/3369141/amd-acquires-godmother-ai-li-fei-feis-start-battle-nvidia-intensifies
6. Stanford University, "Fei-Fei Li," Stanford University faculty profile, https://profiles.stanford.edu/fei-fei-li
7. Robotics 24/7 Staff, "CES 2025: NVIDIA launches Cosmos world foundation model, expands Omniverse," Robotics 24/7, Jan. 7, 2025, https://www.robotics247.com/article/ces_2025_nvidia_launches_cosmos_world_foundation_model_expands_omniverse