Jensen Huang, CEO of Nvidia, unveiled the collaboration during his keynote address at the Consumer Electronics Show (CES) 2025 in Las Vegas and said “The Autonomous Vehicles (AV) revolution has arrived.” “Today, Toyota and Nvidia are going to partner to create their next-generation AVs,” he added.
Huang further added that self-driving cars have transformative potential, predicting they will become the “first trillion-dollar robotics market.”
This announcement is part of Nvidia’s larger unveiling of its Cosmos AI platform, which focuses on enabling robotics and autonomous systems. Cosmos employs generative world foundation models trained on 20 million hours of video data to help AI systems better understand and navigate the physical world.
Huang referred to Cosmos as “the world’s first world foundation model” and highlighted its role in revolutionising how robots and autonomous vehicles learn and operate.
Cosmos works in conjunction with Nvidia’s Omniverse, a physics simulation tool. Omniverse generates realistic simulations that Cosmos converts into photorealistic video imagery, enabling training at an unprecedented scale.
Huang explained this synergy as “Take thousands of drives and turn them into billions of miles.” He also revealed that Cosmos is available under an open-source license on GitHub, offering accessibility to developers and researchers.
Toyota’s adoption of Nvidia’s automotive technology represents an expansion of their longstanding relationship. Toyota has used Nvidia’s DGX systems for training AI models for self-driving vehicles.
The new agreement extends this collaboration to include the integration of Nvidia’s onboard AI computers, specifically the advanced “Thor AGX” chip, which Huang described as 20 times more powerful than its predecessor, the Orin model.
Huang also detailed how Cosmos and Omniverse can be applied beyond autonomous vehicles. For example, warehouse robots can undergo extensive training in simulated environments before deployment. Humans could use devices like Apple’s Vision Pro headset to demonstrate tasks, with their movements captured as data for training robots.
In addition to the Toyota partnership, Huang announced updates to Nvidia’s suite of AI tools. This includes Llama Nemotron, an adaptation of Meta’s Llama model tailored for enterprise applications.
He also discussed the rise of “agentic AI,” where large language models or multi-modal AI systems can integrate with external programs to perform complex tasks. Nvidia is collaborating with startups developing frameworks like LangGraph and Llama Index to advance this technology.
