Odyssey Raises $310M at $1.45B Valuation to Build AI That Simulates the Physical World

Odyssey, founded by self-driving vehicle pioneers Oliver Cameron and Jeff Hawke, announced a $310 million Series B at a $1.45 billion valuation on June 17, making it one of the largest single AI funding rounds in the current cycle. The round was led by Natural Capital, with participation from Amazon, AMD Ventures, GV, EQT, and In-Q-Tel — the CIA-linked venture arm that signals serious government interest in the technology.

The company builds world models: AI systems that learn to predict and simulate the physical world from large volumes of real-world video data. The concept is architecturally distinct from language models. Where LLMs learn patterns across text, world models learn the geometry, physics, and cause-and-effect relationships of actual environments. Odyssey’s data-gathering approach mirrors early Google Earth — teams went out with cameras strapped to their backs, capturing ground-level footage that the models then use to reconstruct and simulate physical spaces.

The practical bet is that world models become the substrate for a generation of AI applications that currently cannot work reliably. Robotics is the clearest near-term use case: a robot trained in a simulated environment that accurately reflects real-world physics can transfer learned behaviors to hardware without the prohibitively expensive trial-and-error that hardware-only training requires. Gaming, defense simulation, autonomous vehicles, and scientific modeling are listed as adjacent markets.

The strategic partnership with Amazon Web Services is the other headline from this announcement. AWS becomes Odyssey’s preferred cloud provider, and Odyssey will deploy AWS Trainium chips — Amazon’s custom AI accelerator — to train and run its models. That is a meaningful hardware commitment that indicates Odyssey expects to scale training compute significantly with this capital.

AMD Ventures’ participation is also notable. AMD has been aggressive in positioning EPYC and Instinct hardware as alternatives to NVIDIA’s data center stack. An early-stage bet on Odyssey — a company that will consume vast amounts of GPU and AI accelerator compute — is consistent with that strategy.

In-Q-Tel’s involvement is a reliable signal that U.S. intelligence and defense agencies see world-model simulation as strategically relevant. The firm typically co-invests alongside leading commercial rounds when the underlying technology has dual-use applications.

At $1.45B, Odyssey now sits in serious-startup territory. The next question is whether its world models can achieve the fidelity and generalizability needed to actually transfer learning from simulation to real hardware at scale — a problem self-driving has fought for a decade without a clean solution.

Sources

Related reading