As AI systems increasingly move from cloud infrastructure into robots, autonomous vehicles, and other physical systems, a new generation of models and approaches is taking shape. NVIDIA's Ming-Yu Liu, Vice President of Cosmos Lab, discussed how open models are becoming essential for unlocking innovation in this space. Speaking on the Practical AI podcast, Liu emphasized that open-source model strategies can democratize access to AI capabilities, allowing researchers and companies to develop and customize solutions for physical AI applications without depending solely on proprietary cloud services. World models—AI systems trained to understand and simulate the physical world—are emerging as crucial infrastructure for this transition. These models enable AI systems to predict how physical environments will respond to actions, a capability essential for training robots and autonomous systems safely and efficiently. NVIDIA's Cosmos Lab is actively developing these technologies, positioning the company at the center of a broader industry shift toward edge-based AI that brings inference and control closer to where physical systems operate. The conversation highlights a competitive inflection point in AI infrastructure: as capabilities mature and deployment moves from cloud to embedded systems, the industry is recognizing that open models can accelerate development and adoption. The stakes extend beyond robotics to entire categories of autonomous systems, making decisions about model openness and accessibility increasingly strategic for major AI companies.