AMD announced today that it’s acquiring World Labs, the AI research lab co-founded by Dr. Fei-Fei Li. The Verge AI reports that it’s an all-stock deal worth about $8.2 billion. It should close by the end of the year. Li will become EVP and chief scientist at AMD and will report straight to CEO Lisa Su.
This isn’t a standard acqui-hire. A chipmaker is paying billions for a team that builds models, and it’s saying openly that the aim is to shape the hardware around them.
🎯 The Deal at a Glance
- Buyer: AMD
- Target: World Labs, founded in 2024 and valued at $1 billion within a few months
- Price: About $8.2 billion, paid entirely in stock
- Product: Marble, a world generation model released in 2025 that creates interactive 3D worlds from a prompt
- Leadership: Li becomes EVP and chief scientist and reports to Lisa Su
- Timeline: Closing expected by the end of the year
According to AMD, the World Labs team will “continue to focus on advancing AI model research.”
🧭 What World Labs Actually Builds
Most of the AI boom has been about large language models, which predict text. World Labs chose a different target. It builds “world models,” which are systems that generate and reason about 3D spaces rather than just words.
Marble is the first commercial result. You type a prompt and get an interactive 3D environment back. The likely uses include games, simulation, robotics training and design work.
Li described the mission this way on her Substack: the startup existed to “jumpstart the work of creating foundation models beyond LLMs.” Many researchers believe that’s where AI goes next. That’s why her lab went from launch to a billion-dollar valuation in a matter of months.
⚙️ Why AMD Wants This
Lisa Su’s statement makes the reasoning clear:
“Building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving,” Su said. “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.”
The core problem is simple. Chips take years to design, and models change every few months. If you design silicon today for last year’s workloads, it’s outdated when it ships.
With a frontier research team in-house, AMD gets an early look at where model design is heading. World models that render and simulate 3D environments may need very different compute, memory and bandwidth than chatbots do. AMD wants to see those needs coming before its competitors do.
AMD says the deal strengthens “its ability to develop AI hardware, software and systems around the needs of emerging models and applications.”
📡 The Bigger Picture: Chipmakers Are Buying Up the Stack
The timing is telling. The Verge AI notes that Nvidia said earlier this month that it’s acquiring Hugging Face for nearly $13 billion.
Put those two deals together and you can see the pattern:
- Nvidia is buying the main distribution hub for open models.
- AMD is buying a research lab with a well-known founder and its own model family.
- Both are moving past selling hardware and into owning parts of the model layer.
For years AMD has been the second choice in AI accelerators. Its hardware competes, but Nvidia has had the stronger software ecosystem and the closer ties to developers. Bringing in Li, one of the best-known names in AI research and the scientist behind ImageNet, gives AMD real research credibility.
🔍 What to Watch
- Openness. Li and Su both stress the “open” ecosystem. Li said the team will keep working “openly with the community and the entire ecosystem.” Watch whether that holds once a public chipmaker owns the lab.
- Marble’s future. Nobody has said yet whether Marble stays a standalone product or gets tuned for AMD hardware.
- Hardware design. Look for future AMD chip roadmaps that talk about spatial or 3D-generation workloads.
- More deals. Two multi-billion-dollar chipmaker acquisitions in one month points to a trend. Independent model labs with specialized research could be next.
🧾 Assessment
For practitioners, the takeaway is that the companies making the chips now want a hand in deciding which models get built. If you work on 3D generation, simulation or robotics, one of the two big accelerator vendors is now betting directly on your field.
The deal still has to close. The strategic direction is already clear, though: the line between chip companies and model labs is fading quickly.
The full report is available at The Verge AI.