AMD buys Fei-Fei Li's World Labs for $8.2 billion, hires her as chief scientist

AMD and World Labs announced an all-stock acquisition deal Monday. World Labs justified the move by saying AI development requires "close collaboration across model research, systems and compute."

• Deal size: $8.2 billion, all-stock.
• Li's new role: Executive vice president and chief scientist, reporting to CEO Lisa Su.
• Timeline: Expected to close before the end of 2026, pending regulatory approval.

The catch: the World Labs team stays intact and keeps focusing on AI model research under AMD, rather than being absorbed and dissolved into existing teams.

Li is a Stanford computer science professor widely considered a pioneer in AI, particularly computer vision. She's best known for building the ImageNet database and the AI competitions it inspired, work that helped launch the modern deep learning era.

In real life If you've used any product that recognizes objects in photos or video, from phone camera features to self-driving car systems, it likely traces back to techniques and datasets Li's ImageNet work helped establish over a decade ago.

In 2024, Li founded World Labs to build deep learning models with a more grounded understanding of the physical world, arguing that real general intelligence requires reasoning about physics and data beyond just text.

She co-founded the company alongside fellow researcher Justin Johnson, and it initially operated quietly in stealth mode before its existence and early funding became public knowledge later that same year.

Li will report directly to AMD CEO Lisa Su, a level of access that signals just how central this technology is meant to become to AMD's broader strategy.

World Labs works on what the industry loosely calls "world models," AI systems built to understand, generate, or sustain a simulation of physical reality, rather than just processing language.

By the numbers: the company's first product, Marble, is pitched both as a tool for building entertainment experiences and as a way to create simulated environments for training robots, two very different markets unified by the same underlying technology.

In a post announcing the deal, Li framed it as a way to scale World Labs' work well beyond a research lab's normal reach. "Now that we have tangible proof of the possibilities, we want to do everything we can to accelerate the future," she wrote. "To do this requires scaling our efforts, widening our reach, and getting closer to the hardware."

AMD says understanding frontier AI workloads, exactly the kind World Labs builds, will directly shape its future chip-making roadmap. That's a strategic bet as much as a research acquisition.

This isn't a cold start either. AMD and World Labs formed an inference optimization and training partnership the prior year, and Li appeared as a guest at AMD's CES presentation earlier this year, suggesting the relationship was already close well before a deal was on the table.

World Labs' rise has been remarkably fast even by AI startup standards. The company's first funding round closed in April 2024 at a valuation of roughly $200 million, with backing from Andreessen Horowitz and Radical Ventures, where Li serves as a scientific partner.

By the numbers: just months later, World Labs raised $230 million at a $1 billion valuation as it emerged from stealth. By early 2026, the company was reportedly in talks to raise at a $5 billion valuation, and it later closed a $1 billion funding round with an unusually notable investor list.

That most recent round included backing from chipmakers AMD and Nvidia, software company Autodesk, Fidelity Management & Research, and Sea, alongside individual investors like Google DeepMind chief scientist Jeff Dean and AI researcher Geoffrey Hinton. Notably, that means Nvidia had already put money into the company AMD just acquired outright, an unusual overlap between two direct chip-making rivals.

The acquisition gives AMD a direct answer to a gap with its biggest rival. Nvidia already offers its own suite of open-weight world models, called Cosmos, while AMD has so far only offered text- and video-based models publicly.

World models are increasingly seen as essential for deploying generative AI on physical robotic platforms, from autonomous vehicles to industrial robots to general-purpose humanoids. Real-world training data for robots is scarce and expensive to collect, which is exactly why synthetic data generated by world models has become so valuable to companies like Tesla and Figure pursuing general-purpose robotics.

By owning that capability outright rather than licensing or partnering for it, AMD is positioning itself to build chips specifically informed by how these physics-grounded models actually behave, rather than reacting to Nvidia's roadmap after the fact.

Nvidia has spent years building an end-to-end ecosystem around its chips, pairing hardware with software and model releases that keep developers inside its platform. AMD's acquisition of World Labs looks like an attempt to build a comparable moat around physics-grounded AI, rather than simply competing on raw chip specifications.

AMD's approach stands out against a recent wave of AI talent deals that have avoided full acquisitions entirely. Google paid $2.4 billion largely to bring on Windsurf's founders and key researchers rather than formally buying the company, and Microsoft handed Inflection $650 million to secure Mustafa Suleyman and his research team in a similar structure.

What's next: those "reverse acqui-hire" deals, where a company pays a licensing fee and separately hires away leadership, have become common specifically because they sidestep the antitrust review a full acquisition can trigger. AMD's $8.2 billion purchase of World Labs is structured as a straightforward, complete acquisition instead, which means it will face standard regulatory approval before closing.

That distinction matters beyond legal technicalities. A full acquisition keeps World Labs' entire team, technology, and ongoing product like Marble under one roof at AMD, rather than scattering the underlying company while its top researchers depart for a new employer.

It also means World Labs' existing investors, including Nvidia's own venture arm, get bought out entirely rather than left holding equity in a company whose leadership just walked out the door, a cleaner outcome than several recent reverse acqui-hire deals have produced for early backers.