AMD to Buy World Labs for $8.2B in Push Beyond AI Chips

SANTA CLARA, Calif. — AMD agreed to acquire World Labs for about $8.2 billion in stock, taking the chipmaker beyond supplying AI computing hardware and into developing the models that could eventually guide humanoids and other physical systems.

The companies announced the definitive agreement Sept. 28. The transaction is expected to close by the end of 2026, subject to regulatory approvals and other customary conditions.

World Labs, founded in 2024 by Stanford computer scientist Fei-Fei Li and three other AI researchers, develops spatial-intelligence models.
The systems understand and generate three-dimensional environments rather than primarily processing text or two-dimensional images. World Labs is applying the technology to robotic learning and simulation so AI can train in generated environments before working with real machines.

The acquisition lets AMD directly reach the researchers who develop those models instead of just supplying processors for AI workloads. AMD said World Labs' research will help it understand how emerging workloads are changing and use that information when designing future hardware, software and computing systems.

The acquisition moves AMD moving upstream in the physical‑AI stack - from supplying processors to owning a firm that builds the models those processors train. AMD said World Labs' research will help shape future hardware, software and system designs.
World Labs introduced its Atlas model in September. The system uses text, images, video and 3D data to rebuild and create existing environments and new interactive 3D worlds. World Labs says those environments can be used to train and evaluate robot policies, although the company has not shown evidence showing that Atlas-trained policies outperform traditional simulators on real robots.

The startup expanded that part of its business in July by acquiring SceniX, a robotics company developing learning-based simulation. SceniX's technology focuses on creating interactive environments where robot policies can be trained before transferring what they learn to real hardware.

World Labs said at the time that combining SceniX with its world models could help generate large quantities of diverse training environments. That addresses one of the persistent problems in physical AI: collecting enough useful data from physical robots can be slow and expensive compared with generating text or image training data.

AMD's relationship with World Labs predates the acquisition. The chipmaker joined a $1 billion round in February alongside investors including Nvidia. World Labs stated at the time that its technology had potential applications in robotics and scientific discovery.
The companies also began a technical partnership focused on training World Labs' models and optimizing inference on AMD GPUs. Bringing World Labs inside AMD could reduce the time researchers spend collaborating with engineers who build the processors and software for large spatial models.

Li will become AMD executive vice president and chief scientist, reporting to CEO Lisa Su, while continuing to lead World Labs' research.

The deal won't turn AMD into a robot maker — World Labs doesn't sell commercial robots — but it moves AMD deeper into the physical‑AI stack that defines the hardware future robots and simulators will need.