AMD가 인공지능 모델 회사 World Labs를 $8.2 billion에 인수합니다.

Nvidia's $12.9 billion acquisition of Hugging Face earlier this month has forced AMD to make a dramatic countermove. Hugging Face, which hosts over 3 million AI models, over 1 million datasets, and over 1.4 million interactive spaces, was a critical platform for AMD to increase adoption of its Instinct GPU accelerators. With Hugging Face now under Nvidia's control, AMD must establish itself as a model builder in its own right — a position Nvidia has held since November 2023 with the launch of its Nemotron3 8B family of models.

AMD's response is to acquire World Labs, a startup focused on world models — AI systems designed to understand the physics of the world rather than generate text and images from existing human knowledge. AMD has agreed to pay $8.2 billion for World Labs, which was founded in January 2024 by Fei-Fei Li, Justin Johnson, Ben Mildenhall, and Christoph Lassner. The company has raised $1.23 billion across two funding rounds, with investors including Jeff Dean (Google), Geoff Hinton (Google), Marc Benioff (Salesforce), Reed Hoffman (LinkedIn), and Eric Schmidt (Sun Microsystems and Google). Both Nvidia and AMD participated in a $1 billion Series B round in February of this year, when World Labs was valued at approximately $5 billion. The new $8.2 billion price represents a 75 percent premium over that valuation.

AMD is paying for the acquisition with stock rather than cash — a decision necessary given the capital-intensive demands of chip manufacturing and AMD's need to maintain liquidity for future investments and procurement. The company recently achieved a $1 trillion market capitalization.

World Labs operates in a crowded competitive landscape. Other companies building world models include Tencent (HunyanWorld), Google (Genie 3), and startups Runway, Odyssey, and Decart. Yann LeCun, Meta's former AI research chief, is also pursuing world models through his AMI Labs venture. World Labs launched its Marble world model in November 2025 and introduced a World API cloud version in January of this year. In early September, the company launched Atlas, an "omni world model" built on multimodal autoregressive diffusion transformer architecture, with early access provided to select customers for integration into future Marble versions.

The deal is fundamentally about acquiring talent as much as technology. Fei-Fei Li, World Labs' chief executive and now chief scientist at AMD, holds a bachelor's degree in physics from Princeton University and a PhD in electrical and electronics engineering from Caltech. She is recognized as one of the world's leading AI researchers. Li served as chief scientist of AI/ML at Google beginning in January 2017 and departed in October 2018 following internal opposition to Google's "Project Maven" collaboration with the U.S. Department of Defense. Since 2014, she has directed the Stanford AI Lab and served as co-director of the Stanford Institute for Human-Centered Artificial Intelligence since leaving Google. She has been a professor of computer science at Stanford for nearly two decades, with prior faculty positions at Princeton and the University of Illinois.

Justin Johnson earned his bachelor's degree in computer science at Caltech and his PhD at Stanford, where he met Li. He worked as an AI research scientist at Meta Platforms from 2016 through December 2023 and currently serves as an adjunct professor of computer science and engineering at the University of Michigan. Ben Mildenhall holds a bachelor's degree in computer science from Stanford, where he also met Li, and a PhD from UC Berkeley. He spent several years as a research scientist at Google, focusing on computer vision and generative 3D models. Christoph Lassner earned degrees in computer science from the University of Tübingen in Germany before joining Amazon as an applied scientist in September 2017. He subsequently moved to Meta Platforms as a research scientist and manager, then spent a year at Epic Games building 3D worlds for video games.

According to AMD's announcement, Johnson and Mildenhall will continue leading World Labs within AMD, while Li will report directly to AMD Chief Executive Officer Lisa Su. This structure suggests Li will assume broader responsibilities beyond world models — potentially encompassing AMD's robotics ambitions, a sector where rival Nvidia is already invested. Notably, Nvidia CEO Jensen Huang has assigned his son Spencer to lead robotics and his daughter Madison to oversee product marketing and operations for 3D simulation and Omniverse software.

No information has been disclosed regarding World Labs' headcount, revenue, or customer base, though the entire team is presumed to be joining AMD. Notably, co-founder Christoph Lassner is not mentioned in AMD's announcement, suggesting possible attrition.

The fundamental logic is clear: if Nvidia can compete with its customers in model development, AMD must do the same. AMD also benefits strategically from fostering open-source frontier models to position itself as more open than Nvidia. Nvidia promotes its open-weight Nemotron GenAI models and the free Cosmos world model, though the 20 million hours of video used to train Cosmos remains proprietary — likely because it was scraped from the internet and Nvidia does not own it.

The critical question is whether Li and Su will commit to developing a fully open-source GenAI model — one with open source code, weights, and training data — to genuinely differentiate AMD from Nvidia. Such a commitment would allow the World Labs team to maintain and expand its world models while demonstrating a meaningful commitment to openness.