Open models and the future of Physical AI with NVIDIA

Open models serve as a foundational catalyst for AI innovation, enabling developers to move beyond the constraints of closed APIs to create highly customized, efficient solutions. Ming-Yu Liu, Vice President of Cosmos Lab at NVIDIA, highlights how these models are essential for the advancement of physical AI—systems integrated into hardware like humanoid robots and autonomous vehicles. Unlike digital-only applications, physical AI demands real-time, local inference and specialized architectures capable of processing diverse sensor inputs. World models, such as NVIDIA’s Cosmos, bridge the gap between simulation and reality by fusing world understanding with predictive simulation. This approach facilitates faster iteration cycles and enables the development of complex multi-agent systems where specialized agents collaborate to perform intricate tasks. Ultimately, fostering an open ecosystem empowers the industry to build safer, more capable autonomous systems that operate effectively within the physical world.