Why AMD thinks physical AI will need far more chips than chatbots
Models that understand 3D space are hungrier for compute, and AMD is paying $8.2 billion to get close to them
The next surge in demand for AI chips may come from models that understand the physical world rather than language.
Video and image-based models need more processing power and data storage than the large language models behind chatbots.
Robots and other physical devices also need their own computing, sensing and power components.
That points to a second wave of demand stretching from data centres down through the chip and component supply chain.
AMD's purchase of World Labs is a bet that this wave is coming, and that it pays to be close to the models driving it.
Why a chipmaker wants a model lab
AI chips are designed years ahead, so knowing how models will behave is a competitive advantage.
As AI spreads into reasoning, robotics and simulation, the demands on hardware are becoming more varied.
Owning a model developer gives AMD an early view of those workloads, which can shape how it designs chips, software and systems.
It also narrows a gap with Nvidia, which already has its Cosmos world models and Omniverse simulation platform.
Google is working on similar ground through DeepMind, its AI research arm.
What physical AI means
Physical AI covers machines that act in the real world, such as robots and self-driving cars.
A core part of it is spatial intelligence, AI that understands and builds three-dimensional space.
World Labs, led by AI pioneer Fei-Fei Li, builds models that generate and simulate interactive 3D environments from text, images and video.
It also develops technology to help robots learn by training them in simulated worlds.
The risks
A chipmaker owning a model lab risks competing with the AI developers who are its biggest customers.
AMD is paying largely for research talent, where the return will be hard to measure and slow to show.
Physical AI is still an emerging field, and it is not yet clear how quickly it becomes large-scale demand for chips.
The deal
AMD has agreed to buy World Labs in an all-stock deal worth about $8.2 billion, expected to close by the end of 2026.
Paying in shares lets AMD bring in the team without spending cash, with dilution put at under 1%.
Li will join AMD as executive vice president and chief scientist, reporting to chief executive Lisa Su.
Analysts said the move broadens AMD's AI reach and would be positive for the wider data centre and semiconductor industry if physical AI takes off.
