AI race shifts to ‘world models’ as startups and tech giants invest billions

A new front is opening in the artificial intelligence race as researchers, startups, and tech giants pivot from chatbots toward "world models" — AI systems designed to understand physical environments, spatial relationships, and the passage of time. The shift marks what many in the field see as a necessary step beyond large language models toward systems that can reason about the real world.

A Taxonomy Takes Shape

On June 3, Fei-Fei Li — widely known as the "Godmother of AI" — and her World Labs team published "A Functional Taxonomy of World Models," a framework that divides the sprawling category into three distinct functions: renderers, simulators, and planners. Renderers produce visual outputs optimized for human perception; simulators model state transitions governed by physics and dynamics; and planners select actions to achieve goals. The taxonomy, published on Li's Substack, aims to bring clarity to a term that has been applied loosely across video generators, physics engines, and robotic control systems alike.letsdatascience+3

Meta ↗ 0.84% AI chief Yann LeCun has been equally vocal, delivering a keynote at a World Modeling Workshop at MILA in February and a talk at ETH Zurich in late May titled "World Models: Enabling the Next AI Revolution". LeCun has argued that world models must use Joint Embedding Predictive Architecture (JEPA) rather than generative approaches.linkedin+1

Startups Race for Position

Capital is flowing into the space. Overworld, a Providence, Rhode Island-based startup developing real-time diffusion world models that run on consumer hardware, raised $4.5 million in pre-seed funding in January, led by Kindred Ventures. Singapore's Video Rebirth closed an $80 million round backed by AMD Advanced Micro Devices, Inc. ↗ 11.04% Ventures and Hyundai, positioning itself to build what it calls an "industrial-grade AI engine" for generating production-ready video environments.kindredventures+5

On the same day, the South China Morning Post reported that Shanghai-based Fysics AI launched Fysiverse, which it describes as a "new-generation physics-based world model that adheres to real-world physical laws," proposing a paradigm that bypasses the approaches taken by OpenAI and Meta.scmp+1

Big Tech Joins the Fray

The startup activity is unfolding against a backdrop of major corporate investment. Nvidia ↗ 2.78% released Cosmos 3, its omnimodal world foundation model for physical AI, in early June. PhysicsX announced a $300 million Series C at a $2.4 billion valuation on June 8, backed by Temasek, Nvidia, and Siemens. And at ICLR 2026 in April, the second Workshop on World Models drew researchers exploring how to unify generative modeling, sequential decision-making, and causal reasoning into scalable systems.iclr+3

The convergence of academic frameworks, startup funding, and corporate muscle suggests world models are moving from theoretical ambition to industrial reality — even as fundamental questions remain about which architectural approach will prevail.