Cosmos World Foundation Model Platform for Physical AI is a platform developed by NVIDIA to address the challenges of training GenAI models in physical fields. The platform enables training and fine-tuning of physical AI models in a digital twin environment, allowing them to be useful for various applications such as autonomous driving or robotic manipulation. Cosmos consists of several components, including video curation, tokenization, pre-training, post-training, and guardrails, which work together to transform raw video data into a lower-dimensional representation that can be processed by AI models. The platform facilitates the development of AI models that can predict future observations and generate realistic video outputs for physical applications. While the models are able to generate high-quality videos with good 3D consistency, there is still a challenge in improving their adherence to the laws of physics in the generated videos.