AMD Agrees to Buy Fei-Fei Li's World Labs for $8.2 Billion, Bringing Spatial AI Research Into Chip Design

A deal to connect AI research with the chips that power it

AMD has agreed to acquire World Labs, the San Francisco artificial intelligence company led by Fei-Fei Li, in a transaction valued at approximately $8.2 billion and paid entirely in stock. The agreement would bring a researcher known for helping computers recognize images into the leadership of a company building the processors that run AI.

AMD's acquisition announcement, published on September 28, 2026, says Li will become executive vice president and chief scientist after the transaction closes. She will report directly to AMD chair and chief executive Lisa Su. The company expects completion by the end of 2026, subject to regulatory approvals and other customary closing conditions.

World Labs develops spatial intelligence models: AI systems designed to work with the shape, layout and behavior of three dimensional environments. Its technology can generate, reconstruct and simulate interactive environments using text, photographs and video. The company also develops technology for robot learning and simulation.

AMD is buying expertise that could help it anticipate what future AI systems will need from computer hardware. World Labs, founded in 2024, would gain the resources of a major chip company while continuing to focus on AI model research.

For Li, the agreement connects two stages of a career built around helping machines understand the visual world. Her earlier ImageNet project helped transform image recognition. World Labs is tackling a broader task: representing spaces that machines can explore and use, rather than simply identifying objects in pictures.

AMD's Taiwan announcement, dated September 29, also puts the transaction at approximately US$8.2 billion. The agreement has been signed, while ownership and Li's new executive appointment will change when the transaction is completed.

What AMD is paying for World Labs

The $8.2 billion purchase would be AMD's second largest acquisition, behind its roughly $50 billion purchase of Xilinx in 2022. The World Labs price is about 16.4% of that earlier transaction's value, using those two figures.

World Labs reportedly had a valuation of $5 billion seven months before the agreement. AMD's proposed price is $3.2 billion higher, an increase of 64%. The startup has raised more than $1.2 billion since its founding.

Payment entirely in AMD shares means the acquisition consideration will take the form of ownership in the chipmaker rather than cash. Issuing shares also spreads ownership across more shares, reducing the percentage held by existing investors unless they acquire more.

AMD had $13.11 billion in cash, cash equivalents and investments held for the short term, against $3.23 billion in debt, at the end of the second quarter of 2026. Paying with stock would preserve those cash resources for other uses. The announcement gives the approximate transaction value but does not provide a detailed schedule of the shares to be issued.

What a world model actually does

World Labs is developing what researchers call world models. These systems represent environments and their structure, allowing AI to work with spaces rather than language alone. A photograph shows a room from one viewpoint; a useful three dimensional representation allows a system to explore how that room is arranged.

Marble, a World Labs model demonstrated by Li and Su earlier in 2026, created a three dimensional scene from a few images. The company's wider research includes spatial reasoning and simulation, with potential uses in robotics, vehicles and other systems that operate in physical environments.

Li, World Labs' cofounder and chief executive, described the intended safety benefit during that presentation:

"Intelligent agents, whether it's robots or vehicles or even tools, can learn inside very rich physics-aware digital worlds before they even need to be deployed into the real one, making them much safer," Li said.

A simulated environment offers a place to practice before a machine acts around people or physical equipment. For a robot, learning includes understanding where objects are and how actions change their surroundings. Creating digital environments that represent those relationships is central to World Labs' work.

Why a chipmaker wants a model research team

AMD says AI is expanding into reasoning, robotics, simulation and physical AI, meaning systems that interact with the physical world. Those applications create different demands on computing infrastructure. Understanding the models behind them can help a chipmaker decide what to build next.

Su, AMD's chair and chief executive, explained the connection between research and product planning in the company's announcement:

Building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving," said Dr. Lisa Su, chair and CEO, AMD. "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.

AMD supplies processors, graphics processing units, networking technology and software. Graphics processing units, commonly called GPUs, perform many calculations at once and are widely used to train and run AI models. Software determines how effectively a model can use that hardware.

Citi analyst Atif Malik said the acquisition would give AMD greater visibility into model development as it competes with Nvidia. The practical attraction is earlier knowledge of how computing needs are changing, which can inform decisions about future chips and the systems built around them.

A partnership preceded the purchase agreement

AMD had already invested in World Labs before agreeing to buy it. The companies also formed a technical partnership in 2025 to improve how World Labs' model training and inference ran on AMD GPUs.

Training is the process through which an AI model learns patterns from data. Inference is the use of a trained model to produce an answer, generate a scene or perform another task. Improving either process can affect how much computing time and equipment an application needs.

Li said the acquisition would bring research and engineering closer together. In AMD's announcement, the World Labs cofounder and chief executive said:

Advancing the next generation of AI technology requires close collaboration across model research, systems and compute," said Dr. Fei-Fei Li, co-founder and CEO, World Labs. "Joining AMD will give our team the resources and engineering depth to accelerate our research and help define the infrastructure needed for the next era of AI.

World Labs will remain separate from AMD's chipmaking business until the transaction closes. AMD says the team will continue advancing AI model research afterward. The company has also said it does not intend to compete with its own AI customers, an important concern for businesses buying computing equipment while developing models themselves.

From a New Jersey dry cleaner to scientific research

Li was born in Beijing in 1976 and grew up in Chengdu, in Sichuan province. Her father was an engineer and her mother a teacher. As a child, she enjoyed taking apart old appliances and experimenting with electronic components to understand how they worked.

Her Soros fellowship biography records that she moved to the United States with her family at 15. They settled in Parsippany, New Jersey, where she faced limited English and financial hardship. She worked in a Chinese restaurant before beginning a university career that would eventually take her to Princeton and Caltech.

When her mother's health worsened, the family opened a dry cleaning business to support themselves. Li combined physics homework with answering telephone calls, serving customers, handling bills and dealing with inspections. Her English skills made her central to running the shop.

Li continued helping manage the business remotely during her doctoral studies, stepping away only partway through the program. She has described the experience as a source of resilience, a quality she sees as necessary in both science and life.

The shop and the laboratory demanded different skills, yet both required sustained work under uncertainty. Her research career later involved persuading others to support an idea that many initially doubted: that much larger collections of images could help computers learn to see.

ImageNet changed the scale of computer vision

Li began developing ImageNet in 2007, when many computer vision projects trained systems on datasets containing only hundreds of images. She drew on psychology, linguistics and research into how people organize their understanding of the world.

Her central argument was straightforward. People learn through a vast range of experiences. Computers might also need much more varied data to recognize objects reliably. Many researchers at the time focused instead on improving algorithms, the mathematical procedures used to learn from those examples.

Li and her students built an image collection that eventually contained more than 14 million pictures. ImageNet also became a benchmark, a shared test that allowed researchers to compare how accurately their systems identified images.

In 2012, AlexNet, a neural network developed by Geoffrey Hinton and his students at the University of Toronto, achieved a breakthrough result in the ImageNet competition. Its success helped drive rapid development in deep learning, in which neural networks learn increasingly complex patterns through multiple layers of computation.

ImageNet supplied both a large learning resource and a common measure of progress. World Labs carries Li's interest in visual understanding into environments that have depth and spatial relationships. The research challenge has expanded from recognizing what appears in an image to representing a world that an AI system can reason about and interact with.

The milestones behind Li's new role

Li's academic and industry career spans physics, electrical engineering, computer vision and AI leadership. Her fellowship biography records nearly 200 scientific publications and identifies her as the inventor of ImageNet and the ImageNet Challenge.

The main milestones show how those roles developed before the proposed move to AMD:

• 1999: Li earned a bachelor's degree in physics from Princeton University.
• 2005: She received a doctorate in electrical engineering from the California Institute of Technology.
• 2007: She began building ImageNet while serving on Princeton's faculty.
• 2009: She joined Stanford University as an assistant professor.
• 2013 to 2018: She directed the Stanford Artificial Intelligence Laboratory.
• January 2017 to September 2018: During a Stanford sabbatical, she served as a Google vice president and chief scientist of AI and machine learning at Google Cloud.
• 2024: She cofounded World Labs.
• 2025: World Labs and AMD formed their technical partnership.
• September 28, 2026: AMD announced the acquisition agreement, with completion expected by the end of the year.

Li also cofounded and chairs AI4ALL, a nonprofit working to increase inclusion and diversity in AI education. That work sits alongside her research and her leadership at Stanford's Institute for Human-Centered Artificial Intelligence.

Often called the "Godmother of AI," Li has described initially feeling uncomfortable with the label. She later accepted it in part because men in technology have long received titles such as founding fathers and godfathers, and she wanted women to have a comparable public voice.

What changes after completion

World Labs' researchers would gain access to AMD's engineering resources while their work helps inform the company's hardware, software and system planning. Li's direct reporting line to Su would place model research close to the chipmaker's senior decision making.

AMD presents the purchase as part of its strategy to support an open AI ecosystem. For customers, the practical objective is computing equipment and software shaped around the needs of emerging applications, including simulation and robotics. For World Labs, the objective is faster research supported by deeper engineering capacity.

AMD's announcement includes a caution under the US Private Securities Litigation Reform Act of 1995 covering expected benefits and timing. Among the risks it identifies are regulatory action, export rules, manufacturing constraints, competition, retaining key employees and integrating acquired businesses.

The next milestone is completion of the transaction, targeted for the end of 2026. Until then, AMD and World Labs remain separate businesses. Their proposed combination puts a specific bet at the center of AMD's AI strategy: understanding future models early enough to build the computing systems they will require.

Key Points

• AMD has signed an agreement to acquire World Labs for approximately $8.2 billion, paid entirely in stock.
• Completion is expected by the end of 2026, subject to regulatory approvals and other closing conditions.
• Fei-Fei Li will become AMD's executive vice president and chief scientist, reporting directly to Lisa Su.
• World Labs develops spatial intelligence models, interactive three dimensional environments and technology for robot learning.
• The companies previously worked together to improve model training and inference on AMD GPUs.
• The purchase would be AMD's second largest acquisition, after Xilinx.
• Li's ImageNet project helped drive the deep learning advances that transformed computer vision.