Odyssey raises $310M at a $1.45B valuation to transform AI model simulation

Odyssey, an artificial intelligence lab building world models, today announced it raised $310 million in early funding at a $1.45 billion valuation led by Natural Capital.

Amazon.com Inc., AMD Ventures, GV, EQT, IQT and others also joined the Series B round. Existing investors also joined the round, including Elad Gil; Jeff Dean, Google LLC’s chief scientist; Qasar Younis, co-founder and CEO of Applied Intuition Inc.; Garry Tan, president and CEO of Y Combinator Inc.; Guillermo Rauch, founder and CEO of Vercel Inc.; and Kyle Vogt, founder of Cruise LLC.

In addition to the funding, Odyssey announced a new deal with Amazon Web Services Inc., which provides access to the company’s Tranium custom silicon and instances as its preferred cloud provider. Tranium is AWS’s family of silicon designed to assist with the training and deployment of AI models at scale.

“We believe world models represent a new class of foundation model—AI that can understand and simulate the world itself,” said co-founder and Chief Executive Oliver Cameron.

The last few years have seen numerous breakthroughs in scaling, interactivity, multimodality and physics simulation for AI models. This has come with a similar scale for generative AI model size, compute and performance needs. Cameron said partners across the industry are needed to push the frontier for general world models to achieve the GPT-3 moment for the field.

“World models represent one of the most demanding workloads in AI—they require massive compute throughput with tight latency constraints,” said Ron Diamant, vice president and distinguished engineer at Amazon. “Odyssey’s team has been pushing the boundaries of what’s possible in this space, and Trainium is purpose-built for exactly this kind of scale.”

World models are AI systems designed to learn and simulate how the physical and digital world works. Instead of simply operating by recognizing patterns or predicting the next word, like a large language model, they build an internal representation of reality. This includes physics, causality and time. So, they are capable of simulating or predicting outcomes of actions before taking them.

The resulting AI models created to predict environments and simulate physics are necessary for training physical AI agents, robotics, autonomous vehicles and other AI models that directly interact with the physical world. In many cases, having a world model to simulate the environment can increase the safety of deploying physical AI agents in robotics and cars, train them in days rather than months, and prepare them for unexpected conditions.

Over the past three years, Odyssey has built its own world models to push the limits of research in the field. It released Odyssey-2 Max, which advanced the state of the art in physics accuracy for general world simulation. Starchild-1 introduced the first real-time multimodal world model and Agora-1 brought multi-agent interactivity within a shared world simulation.

The category for world models is growing. In recent months, AI developer Decart.ai Inc. also raised $300 million at a nearly $4 billion valuation to develop a pair of world models, named Lucy and Oasis. AI pioneer Yann LeCun’s startup AMI Labs raised $1.03 billion in March to train world models, and world model startup Runway AI Inc. raised $315 million in February.

All of these startups face competition from the technology giant Google LLC, which introduced a tool called Project Genie. It allows developers to rapidly generate three-dimensional virtual environments using prompts, combining the company’s powerful Nano Bana Pro image generator and Genie 3 world model into one package.

Image: Odyssey