AMD announced a definitive agreement on Monday to acquire World Labs, a research and model lab focused on spatial intelligence led by Fei-Fei Li. The price is about $8.2 billion, paid entirely in stock, with the deal expected to close by the end of 2026 subject to regulatory approvals.
For AMD, the acquisition reaches well beyond chip sales. The company sells the GPUs on which models are trained and run, and competes there directly with Nvidia. With World Labs, AMD is by its own account buying model research: the team is meant to give the chipmaker a sharper picture of how workloads are evolving and to shape future roadmaps for hardware, software and systems — explicitly in service of delivering "AI infrastructure for an open ecosystem".
"Building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving," said AMD chair and CEO Lisa Su. Fei-Fei Li and her team, she said, bring exceptional research leadership and model expertise, which together the companies intend to turn into hardware, software and systems for the next generation of AI.
World Labs is based in San Francisco and develops spatial-intelligence models that generate, reconstruct and simulate interactive 3D environments from text, image and video inputs, along with technology for robotic learning and simulation. The company had raised roughly $1 billion in funding. Li, who two decades ago created the ImageNet visual database that helped make deep learning possible and is often called the "godmother of AI", will join AMD as executive vice president and chief scientist reporting to Su once the transaction closes. The World Labs team will continue working on model research.
The strategic point lies in the technology category. World models take a different route from the large language models built by OpenAI, Anthropic and Google: rather than text, they are trained on data such as video depicting the physical, three-dimensional world. The goal is AI that understands how objects move — how a ball bounces, for instance — a capability many in the industry consider essential to extending AI into humanoid robots and self-driving cars. Li put it in a post by saying that the universe is not made of words but of real things, and that language alone is not sufficient for model development.
The two companies have already worked together. According to Li, a "deep technical partnership" began last year, focused on optimising training and inference on AMD GPUs, and AMD had already invested in World Labs before the acquisition. Li said the collaboration made the fit between the two companies' software and hardware ecosystem, foundation models and applications clear.
AMD is not alone in this category. World Labs is one of several richly funded startups betting on world models. Yann LeCun, Meta's former chief AI scientist, launched AMI Labs in March, and Nvidia has built its own world models and partnered with startups such as Wayve, which builds them for self-driving cars. AMD shares were roughly flat in after-hours trading on Monday.
What to watch: whether the research team translates into concrete advantages for AMD's ROCm software stack and GPU roadmap, how quickly sellable products emerge, and whether Nvidia — already investing heavily in robotics and physical AI — answers with a move of its own. The competitive question is shifting from who sells the most GPUs to who understands which models those GPUs will need to accelerate next.
