AMD has agreed to pay $8.2 billion in stock for World Labs, the world model startup founded by AI luminaries Fei-Fei Li, Justin Johnson, Ben Mildenhall and Christoph Lassner — a roughly 75 percent premium over the $5 billion valuation the company carried at its February Series B.

The all-stock structure matters. AMD crossed a $1 trillion market capitalization recently, but it cannot burn cash on acquisitions given the wafer capacity and component commitments it must prepay to build its CPUs and Instinct GPU accelerators. Stock is the currency it can afford.

The deal is a direct answer to Nvidia. Earlier this month, Nvidia closed a $12.9 billion acquisition of Hugging Face, the open weight model distribution hub hosting over 3 million models, more than 1 million datasets and 1.4 million interactive spaces — a channel AMD needed as it pushed Instinct GPU adoption. Nvidia has also been a model builder itself since November 2023, when it launched its Nemotron3 8B family. If Nvidia can compete with its own customers on models, AMD concluded it can too.

World Labs, founded in January 2024, raised $1.23 billion across two rounds. Its angel investors include Jeff Dean and Geoff Hinton of Google, Marc Benioff of Salesforce, Reid Hoffman of LinkedIn, and Eric Schmidt. Notably, both Nvidia and AMD participated in the $1 billion Series B in February — the same round that set the $5 billion valuation AMD has now decisively topped.

The company builds so-called world models, designed to understand the physics of the world rather than generate text and images from human knowledge corpora. It publicly launched its Marble world model in November 2025 and followed with a World API cloud access version in January. Its follow-on Atlas engine — an "omni world model," technically a multimodal autoregressive diffusion transformer — entered early access with select customers in early September and will feed future Marble versions.

World Labs does not have the field to itself. Nvidia fields its Cosmos world models, Tencent has HunyuanWorld, Google has Genie 3, and startups Runway, Odyssey and Decart are building competitors. Yann LeCun, former head of AI research at Meta Platforms, is also active in the space through his AMI Labs startup.

But like AMD's $40 billion acquisition of Xilinx, this deal is as much about buying a team as a product. Fei-Fei Li — World Labs CEO, now AMD chief scientist — holds a physics degree from Princeton and a PhD in electrical engineering from Caltech, and ranks among the two dozen AI luminaries on Earth. She was chief scientist of AI/ML at Google from January 2017 until October 2018, leaving amid the internal uproar over Google's "Project Maven" collaboration with the US Department of Defense. She has directed the Stanford AI Lab since 2014, co-directed the Stanford Institute for Human-Centered Artificial Intelligence, and taught computer science at Stanford for nearly two decades.

Johnson, who met Li while earning his Stanford PhD after a Caltech computer science degree, was an AI research scientist at Meta Platforms from 2016 through December 2023 and is an adjunct professor at the University of Michigan. Mildenhall, also a Stanford undergraduate, took his PhD at UC Berkeley and spent several years as a Google research scientist focused on computer vision and generative 3D models. Lassner, notably absent from AMD's announcement, studied at the University of Tübingen and worked at Amazon, Meta and Epic Games building 3D worlds.

According to AMD's announcement, Johnson and Mildenhall will continue running World Labs inside AMD, while Li reports directly to CEO Lisa Su — suggesting a mandate broader than world models for robots. AMD now has visible aspirations to move its compute engines into robotics, a market Nvidia is already attacking. Jensen Huang has placed his son Spencer in charge of robotics at Nvidia, and his daughter Madison runs product marketing and operations for 3D simulation and Omniverse software.

Neither company has disclosed World Labs' headcount, revenue or customer count, though presumably the whole team is moving over. The open question is whether Li and Su go further than Nvidia on openness — releasing a fully open source GenAI model with source code, weights and training data, surpassing Nvidia's open weight Nemotron models and free Cosmos world model, whose 20 million hours of training video remain closed, very likely because it was scraped from the Internet and Nvidia does not own it. The World Labs team could do exactly that while maintaining and expanding its world models.