AMD Buys World Labs for $8.2B to Bet on World Models

Advanced Micro Devices Inc. has agreed to acquire World Labs Inc., the world model startup founded by AI researcher Fei-Fei Li. AMD is paying $8.2 billion, entirely in stock. The chipmaker expects the transaction to close by the end of the year.

The deal comes only seven months after AMD took part in a $1 billion funding round for World Labs that valued the startup at $5 billion. Nvidia Corp., AMD's main rival in AI chips, also joined that round. That means Nvidia now stands to earn a substantial return from a company its competitor is buying.

Who is behind World Labs

Li started World Labs in 2024. She is best known for creating ImageNet, an early training dataset built to teach algorithms to recognize objects. ImageNet made it possible to build AlexNet, the model widely seen as the forerunner of today's AI systems.

Once the acquisition closes, Li will join AMD as executive vice president and chief scientist. AMD says World Labs will keep working on its own AI research projects.

What world models do

World Labs builds what it calls spatial intelligence models, also known as world models. These systems generate interactive three-dimensional environments. One of their most practical uses is in robotics.

Many industrial robots run AI models on board. To train those models, developers usually let them control a virtual robot inside a simulated space. The model gets sample tasks that resemble its future real-world job and learns the best approach by trial and error. Building those simulated spaces is where world models come in.

World Labs' latest model, Atlas, can produce a virtual copy of a factory or warehouse from just a few photos. It can also simulate moving parts such as conveyor belts. On top of that, Atlas supplies depth data, which helps AI models work out where objects are located.

Robotics is not the only market. Atlas can also create images and videos from text prompts and add visual effects to existing footage.

How it fits AMD's robotics push

AMD already has a sizable footprint in robotics. In 2022 it spent $50 billion on Xilinx Inc., a major maker of field programmable gate arrays (FPGAs). FPGAs are processors that customers can reconfigure for specific jobs, and they are common in robots.

More recently, AMD launched the Ryzen AI Embedded X100 series, a line of chips designed for robots. Each chip combines a central processing unit, a graphics processing unit and AI accelerators. The chips come in a ruggedized package built to handle high temperatures.

AMD Chief Executive Lisa Su framed the purchase as a way to understand where AI models are heading. "Building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving," she said. "Fei-Fei and the World Labs team bring exceptional research leadership and model expertise. 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."

Our Take

This is a chipmaker buying a model developer, not the other way around. That points to a strategy where hardware vendors want direct insight into the workloads their chips will run, rather than waiting for customers to tell them. For AMD, world models connect neatly to the robotics hardware it already sells, from Xilinx FPGAs to the X100 series. Training robots in simulation first and then running them on AMD silicon is an attractive story.

The price also stands out. A jump from a $5 billion valuation to $8.2 billion in seven months shows how highly investors rate spatial AI right now. It is worth watching whether World Labs keeps its research independence inside AMD, how Atlas is packaged for AMD customers, and whether Nvidia, now a beneficiary of the deal, responds with world model moves of its own.