World Labs is building frontier world models, systems that represent three-dimensional space the way large language models represent text. Founded by Fei-Fei Li, the company argues that spatial intelligence is the missing capability in modern AI: models can describe a room in fluent prose yet cannot reason reliably about where objects sit, how they occlude one another, or what happens when something moves. Closing that gap matters for robotics, design tools, simulation, and any application where a model must act in physical space rather than describe it.

The technical bet is that spatial reasoning should be learned in a general-purpose model rather than bolted on per application, the same architectural argument that made foundation models win in language. With more than a billion dollars raised as of 2026, World Labs is resourced to pursue that at frontier scale. For an AI infrastructure thesis, it sits at the layer where a durable capability, not a single product, is being created. The company is early, but the capability it is chasing is foundational rather than incremental.