Founder Stack

World Labs represents one of the more academically grounded bets in the current AI funding cycle, built on the premise that large language models, however capable at text and reasoning, are fundamentally limited by never having learned to perceive and reason about three-dimensional physical space the way humans and animals do. Founded by Fei-Fei Li, the Stanford computer scientist widely credited with catalyzing the deep learning era through the ImageNet dataset, alongside Justin Johnson and Ben Mildenhall, both prominent computer vision researchers, World Labs is building what it calls 'world models,' AI systems trained to understand and generate coherent, navigable 3D environments rather than just flat text or images. The company raised $230 million in seed funding in 2024 at a valuation reported around $1 billion, an unusually large and richly valued seed round that reflected investor confidence in Li's research pedigree and the broader industry consensus, echoed by figures at DeepMind, Meta, and elsewhere, that spatial and physical reasoning represents a genuine capability gap in even the most advanced language models. World Labs shipped its first public product, Marble, in 2025, a tool that generates persistent, explorable 3D scenes from text or image prompts, aimed initially at game developers, virtual production studios, and simulation researchers rather than a mass consumer audience. The competitive and adjacent landscape includes Google DeepMind's own world-model research under the Genie and Veo programs, Meta's work on spatial AI tied to its Reality Labs hardware ambitions, and a handful of smaller startups building similar 3D-generation and simulation tools for robotics training data. World Labs' differentiation rests on Li's argument that spatial intelligence deserves to be treated as its own research frontier deserving dedicated capital and talent, rather than a downstream application of language-model progress, a framing that has helped the company recruit heavily from top computer vision labs even as compensation competition with frontier language-model labs has intensified industry-wide. The commercial case for world models extends well beyond entertainment and gaming into robotics, where training physical robots requires vastly more simulated environment data than currently exists, and autonomous vehicles, where synthetic 3D scenario generation could reduce the cost of testing edge cases that are rare or dangerous to capture in real-world driving data. That broader applicability is central to World Labs' pitch to investors, positioning spatial intelligence as infrastructure for the eventual 'physical AI' wave rather than a narrow creative tool. What to watch: whether Marble finds a durable paying customer base in game development and virtual production through 2026, whether World Labs raises a Series A that meaningfully increases its billion-dollar seed valuation, and whether its world-model research gets adopted as training infrastructure by robotics or autonomous-vehicle companies rather than remaining a standalone creative product.