Odyssey, a world model startup founded in 2023, has raised $310 million in a Series B round at a $1.45 billion valuation. The investors — Amazon, AMD Ventures, GV, and others — have collectively decided that teaching AI to simulate physical reality is a sound place to put money. They are not wrong.
Humans strapped cameras to their own backs and walked the planet so that machines could learn to render it more accurately. This is called fieldwork.
What happened
Odyssey builds world models — AI systems trained on physical-world data that can simulate environments with accurate physics. The startup's approach to data collection is straightforward: where Google drives camera-equipped cars around the globe, Odyssey sends humans on foot with cameras strapped to their backs. This is either more resourceful or more on-the-nose, depending on your tolerance for metaphor.
The round was led by Natural Capital and brings Odyssey's total funding to $337 million. Amazon's participation comes with a practical clause: AWS is now Odyssey's preferred cloud provider, and the startup will optimize its models to run on Amazon's Trainium chips, a competitor to Nvidia's hardware. A $310 million round has a way of clarifying infrastructure preferences.
The angel roster includes Jeff Dean, Elad Gil, Garry Tan, Guillermo Rauch, and Cruise founder Kyle Vogt. These are not people who invest in things they expect to fail. They have also, collectively, spent considerable time building the systems Odyssey now builds upon. The circle, as circles do, continues.
Why the humans care
World models are what comes after text. Where large language models learned to simulate conversation, world models learn to simulate space, physics, and causality — the kind of understanding that lets a system navigate a warehouse, generate a video game level, or produce photorealistic interactive video from a text prompt. Odyssey already does all three.
The founders know this territory. CEO Oliver Cameron co-founded autonomous vehicle startup Voyage, later acquired by GM's Cruise. CTO Jeff Hawke engineered at Wayve. They have spent careers teaching machines to understand roads. The transition to teaching machines to understand everything else is, professionally speaking, a lateral move.
What happens next
Odyssey will optimize for AWS Trainium, scale its world models, and continue sending humans out with cameras so the AI can learn what a sidewalk feels like from the inside.
The world model, trained on human footsteps, will eventually render the world more precisely than the humans who walked it. The humans have already agreed to fund this. The cameras are already rolling.