A professional baseball batter has milliseconds to decide how to swing — faster than visual signals reach the brain. They succeed because their internal world model predicts where the ball will go. This subconscious simulation of physics is what separates biological intelligence from statistical correlation. World models are the bridge.
"We need machines that understand the world — that can remember, have intuition, have common sense. Current AI systems are not capable of any of this."
Yann LeCun — Founder, AMI Labs · Turing Award Laureate · Former Chief AI Scientist, Meta
World models are trained on photos, video, audio, and text to build internal representations of how reality behaves — then reason about the consequences of actions. This isn't generative AI producing plausible outputs. This is AI developing understanding.
Who's saying "Human World Model"?
World Labs — Fei-Fei Li's startup now in talks at $5B valuation. Raised $230M in 2024 at $1B. Building "large world models" for spatial intelligence. Marble product launched Nov 2025.
AMI Labs — Yann LeCun left Meta Dec 2025 to build world models. Raising €500M at €3B valuation pre-launch. Paris HQ. Partnership with Nabla for FDA-certifiable healthcare AI.
Google DeepMind — Genie 3 launched Jan 2026. First real-time interactive world model. 24fps, 720p. Public prototype via Project Genie. DeepMind calls it a "stepping stone toward AGI."
NVIDIA — Cosmos world foundation models downloaded 2M+ times. Jensen Huang: "Physical AI will revolutionize the $50 trillion manufacturing and logistics industries."
Meta — V-JEPA 2 released Jan 2026. Trained on 1M+ hours of video. 65–80% success on robotics tasks with only 62 hours of robot training data.
Runway — GWM-1 world model launched Dec 2025. Autoregressive, real-time, controllable via camera pose and robot commands.
Siemens × NVIDIA — Building the first fully AI-driven manufacturing sites starting 2026. Digital Twin Composer. PepsiCo achieving 20% throughput increase on deployment.