Nvidia / 30 September 2026

AMD Buys World Labs to Challenge Nvidia in Physical AI

AMD has agreed to acquire World Labs, Fei-Fei Li’s spatial intelligence research company, in an all-stock deal valued at about $8.2 billion. The transaction would bring model researchers into AMD as it designs hardware, software and systems for robotics, simulation and other physical AI workloads. Li would become AMD’s executive vice president and chief scientist, reporting to chief executive Lisa Su, with closing expected by the end of 2026. Su said the next generation of computing platforms needs “a deep understanding of how models are evolving.” For Nvidia, the acquisition signals a rival’s move upstream into the research that shapes demand for future accelerators, not merely another launch of a competing chip.

World Labs develops models that generate, reconstruct and simulate interactive three-dimensional environments from text, images and video. The work can inform robot learning and simulations in which systems must reason about space and the consequences of actions. These applications differ from the dominant large-language-model benchmarks in the data center: they combine perception, simulation, motion and potentially repeated inference on devices or at the edge. AMD says the research team will continue advancing models after the transaction. The purchase gives its product organization direct contact with scientists defining those workloads, allowing memory, interconnect, software and system architecture decisions to reflect how spatial models actually consume computing resources.

Nvidia has spent years building a physical AI ecosystem around robotics software, simulation, digital twins and specialized edge processors, in addition to its data center GPUs. Developers can use its models and tools to train agents in simulated environments before deploying them in machines. AMD’s announced strategy approaches that opportunity from a different starting point: buying the model laboratory to guide its hardware and software roadmaps. It also emphasizes an open AI ecosystem. The contest is therefore about influence over the development workflow as well as chip performance. Researchers who create demanding spatial models can expose limitations in an architecture early enough for a vendor to address them in the next product generation.

The all-stock consideration makes World Labs a substantial strategic commitment for AMD without a cash payment of the same size at closing. Integrating a research organization with a chip supplier also creates a choice about where its work will run and how open its methods remain to other hardware partners. Li said the combination would give her team engineering resources to help define the infrastructure needed for a new AI era. AMD wants insight that can translate into products customers use across cloud and embedded settings. Nvidia already has an installed developer base and a broad physical AI software stack, so AMD’s acquisition will matter most if the research shapes competitive tools and deployed systems over several cycles.

The timing puts a transaction value on spatial intelligence even before a mass market for autonomous machines has fully formed. An $8.2 billion price tag reflects AMD’s expectation that future computing demand will emerge from new classes of models, not simply larger versions of today’s chatbots. The research could also give AMD credibility with developers choosing simulation environments and hardware targets. Nvidia’s response is embedded in its own platform strategy: make robot and digital twin development convenient enough that successful projects remain on its compute stack from experimentation through production. The acquisition sharpens a long-term competitive test over who translates model innovation into the standard system architecture for physical AI.