Nvidia Introduces Cosmos 3 Foundation Model to Power Robotics and Physical AI

Nvidia unveiled Cosmos 3, a new foundation model designed to support physical AI applications including robotics, autonomous vehicles, and industrial automation. The announcement, made in early June 2026, signals a shift in the AI industry from software-focused chatbots toward systems that interact with the physical world.

Physical AI refers to artificial intelligence that operates in and responds to the real world, rather than processing text or images in isolation. Robots, self-driving cars, and automated factory systems all require AI that can perceive physical environments, plan actions, and respond to unexpected conditions in real time.

Cosmos 3 is designed to serve as a foundation model for these applications, providing a base layer that developers can fine-tune for specific physical tasks. Nvidia says the model is trained on large amounts of video and sensor data from real-world environments, giving it a richer understanding of physical dynamics than models trained primarily on text.

The announcement came as Nvidia expanded its partnerships in Asia, securing agreements with South Korean companies including SK Hynix, LG, and Hyundai to build what the company calls "AI factories," large-scale infrastructure installations designed to run AI workloads at industrial scale.

TechStartups reported on June 8 that the broader industry is increasingly using the term "MANGO," standing for Meta, Anthropic, Nvidia, Google, and OpenAI, to describe the leading companies in AI infrastructure. Nvidia's position in that group is anchored by its dominance in the GPU market, which remains the primary hardware for training and running large AI models.

The shift toward physical AI is expected to drive significant demand for new types of sensors, actuators, and edge computing hardware. Analysts say the market for physical AI applications could be substantially larger than the market for software-only AI tools, given the breadth of industries that rely on physical processes.