Odyssey Raises $310M to Build the GPT-3 Moment for World Models

Odyssey, the Palo Alto AI lab building general-purpose world models, has raised a $310 million Series B at a $1.45 billion valuation. The round was led by Natural Capital and brings the company's total funding to $337 million , a dramatic leap from the $27 million it had raised across its seed and Series A. The signal here is not just the size of the check; it's who signed it.

Two Self-Driving Veterans Bet on the Next Foundation Model

Odyssey was founded by CEO Oliver Cameron and CTO Jeff Hawke. Cameron served as VP of Product at Cruise, where he led the launch of the world's first fully driverless product in San Francisco , he joined Cruise after it acquired Voyage, a self-driving startup he co-founded. Hawke was the founding engineer at Wayve, where he pioneered the use of end-to-end models to drive cars on complex roads.

Their thesis is simple but ambitious: the same core problem that powers self-driving cars , predicting what happens next in a physical environment , can be generalized to everything. Instead of simply recognizing patterns or predicting the next word like a large language model, a world model builds an internal representation of reality, including physics, causality, and time, so it can simulate or predict the outcomes of actions before taking them.

What Is a World Model, Exactly?

A world model is an AI system trained to understand and simulate how the physical world works. Think of it less like ChatGPT and more like a physics engine that learned from reality rather than being hand-coded. These models are necessary for training physical AI agents, robotics, autonomous vehicles, and other AI systems that directly interact with the physical world. A robotics company, for example, could run thousands of training scenarios inside Odyssey's simulation instead of on a real factory floor.

World models are the next big thing in AI beyond text- and chat-based large language models , they gather data from the physical world and simulate it with accurate physics. In Odyssey's case, it mimicked how Google Earth gathered data, sending people out with cameras strapped to their backs. Odyssey developed an advanced camera capture system equipped with six cameras, two lidar sensors, and an inertial measurement unit, capable of capturing the environment in 360 degrees at 3.5K resolution.

The Model Stack They've Built

Over three years, Odyssey has shipped a series of research milestones that back up the fundraising narrative:

• Odyssey-2 Max , their most powerful general-purpose world model, advancing the state of the art in physics accuracy for general world simulation.
• Starchild-1 , introduced the first real-time multimodal world model , going beyond visual-only learning.
• Agora-1 , added multi-agent interaction to a shared-world simulation , letting multiple humans or AI agents coexist in the same generated environment in real time.
• PROWL , demonstrated how world models can improve through active exploration , using reinforcement learning to find and fix model weaknesses.

The team has made significant contributions to language models (DeepMind Gemini), video models (DeepMind Veo), world models (Wayve GAIA), and autonomous systems (Tesla FSD).

The Chip Plot Twist

The investor list tells a story within the story. NVentures, Nvidia's venture capital arm, backed Odyssey's Series A in February 2026 as part of Nvidia's broader strategy to fund world-model startups. Nvidia is nowhere in the Series B. Trainium is Amazon's in-house answer to Nvidia's grip on AI computing, built for the fast, high-volume workloads that real-time world simulation demands. Pairing it with money from AMD Ventures, Nvidia's main chip rival, makes the round read like a vote against the market leader.

Alongside the funding, Odyssey announced a new agreement with Amazon Web Services, which will become the company's preferred cloud provider , the partnership will support the compute demands of world model development, including workloads requiring high throughput, low latency, and strong price-performance. Odyssey is also collaborating with Amazon's Annapurna Labs to optimize its world models on AWS Trainium chips.

The honest caveat is that this is diversification, not a divorce. It is unclear whether the switch reflects real conviction in Amazon's chips or simply better terms in a fierce market , and "preferred cloud" is not "exclusive." But the optics matter in a market where everyone is hunting for Nvidia alternatives.

A Crowded but Nascent Race

Odyssey is not operating in a vacuum. The world model space has attracted a wave of capital in the past year:

• Runway, valued at $5.3 billion after a $315 million Series E, launched its first world model product and gained $40 million in annual recurring revenue.
• World Labs, Fei-Fei Li's spatial AI startup, launched its Marble product after raising $230 million.
• Decart.ai raised $300 million at a nearly $4 billion valuation to develop a pair of world models named Lucy and Oasis.
• Yann LeCun's startup AMI Labs raised $1.03 billion in March to train world models.
• Google DeepMind's Genie model is already being used at Waymo.

While most competitors focus on video generation or gaming, Odyssey sees itself as a pure research lab, similar to what OpenAI did with language models. That positioning , research-first, applications-second , is a deliberate bet that the foundational technology is still being written.

Who Wins, Who Watches Nervously

The winners here are clear: existing angel investors including Jeff Dean (Google's chief scientist), Elad Gil, Qasar Younis (Applied Intuition), Garry Tan (Y Combinator), Guillermo Rauch (Vercel), and Kyle Vogt (Cruise) all see their stakes grow substantially with a unicorn valuation. Natural Capital, which led the round, made this its largest investment to date.

The more interesting question is what this means for Nvidia. It lands as buyers everywhere hunt for Nvidia alternatives, from China's carmakers designing their own chips to startups turning Nvidia GPUs into a commodity. Every high-profile deal that routes compute spend to Trainium or AMD is a data point in that narrative.

The presence of In-Q-Tel, and Odyssey's own mention of defence as a target use, is a reminder that whoever can convincingly simulate the world will have customers well beyond gaming. IQT is the CIA-affiliated fund, and its participation adds a dimension that goes well beyond robotics and film.

What Becomes Possible Now

The company's stated ambition is to reach what it calls the "GPT-3 moment" for world models , the inflection point where the technology transitions from a promising research direction into a breakthrough foundational technology that developers can build on top of. "This round provides the compute, infrastructure, and partners to push the frontier of general world models, and to achieve a GPT-3 moment for the field,