In the midst of a competition where silicon giants ally with automation pioneers, Odyssey ML emerges, a company that with $310 million in new funding aims to surpass the limits of textual AI. Valued at $1.45 billion, the company attracts giants like Amazon, Nvidia, and AMD in a technological challenge to map the physics of the real world through so-called world models.
BEYOND LANGUAGE: SIMULATING THE PHYSICAL WORLD
Odyssey defines itself as an artificial intelligence research lab focused on developing general world models, that is, multimodal and causal systems capable of predicting and interacting with the physical environment over long time horizons.
Unlike traditional language models, these technologies learn physics, body language, and object dynamics, aiming for a “much more complete understanding of the world” that includes all those aspects of reality that “language alone cannot capture.”
ODYSSEY’S MODELS
The flagship product, Odyssey-2, is a simulator that generates interactive streams lasting several minutes starting from textual or image inputs, initiating streaming in just 50 milliseconds. The startup’s research also includes Starchild-1, a model capable of simultaneously simulating sounds and images in real time, and Agora-1, which allows multiple participants, human or AI, to interact within the same simulation. Another pillar is PROWL, a reinforcement learning framework where agents explore game environments to improve the world model’s performance.
WHO’S BEHIND IT
Odyssey’s leadership is entrusted to co-founders Oliver Cameron, who serves as CEO, and Jeff Hawke, both with backgrounds in autonomous driving.
The company currently has 55 employees distributed across offices in London, Zurich, and Palo Alto. Many team members come from prestigious AI labs and startups specialized in autonomous navigation such as Tesla, Waymo, Wayve, DeepMind, Meta, and Apple.
With their expertise, Odyssey aims to revolutionize sectors ranging from robotics, where models allow robots to try complex tasks before acting, to video games, healthcare, and logistics.
THE NUMBERS OF A BILLION-DOLLAR CLIMB
The recent $310 million funding round, writes the Financial Times, has brought Odyssey ML’s valuation to $1.45 billion, capital inclusive of the new funds raised.
Despite technological efficiency, managing these models involves high operating costs. Cameron revealed that using the tools on Nvidia H200 and B200 chips incurs costs between $2 and $4 per hour per user, a figure that does not include the substantial expenses required for training the models.
Jay Zaveri, partner at Natural Capital, described these world models as a “second wave of artificial intelligence,” stating that “a structure similar to the human brain” has been taken, to which, until now, only language had been taught. And according to Luna Schmid of GV, the current phase of massive investments could represent a “GPT-3 moment” for this technology, referring to OpenAI’s breakthrough that sparked the sector’s boom.
THE ALLIANCE WITH HARDWARE TITANS
Financial support for Odyssey includes investment arms of Nvidia and AMD, as well as the CIA-affiliated fund IQT, Natural Capital, and GV (formerly Google Ventures).
Among individual investors stand out then Google chief scientist Jeff Dean, investor Elad Gil, Guillermo Rauch, and Garry Tan.
A central role is played by Amazon, which participated in the round as part of a collaboration aimed at expanding the use of its proprietary Trainium chips. Under this agreement, Odyssey will use Amazon Web Services (AWS) as its primary cloud provider, actively contributing to improving Amazon’s accelerators in competition with market leaders. Ron Diamant, AWS vice president, expressed the ambition that this synergy will build “the best artificial intelligence accelerator in the world.”
Partnerships with chip manufacturers are fundamental for Odyssey to offset the costs of real-world simulation, in a context where Amazon is heavily investing in AI, with commitments reaching $83 billion in companies like Anthropic and OpenAI.
