The need for materials with improved or new properties is often at the heart of the challenges society faces. Despite massive expansion in experimental capabilities and data, knowledge- and data-centric challenges prevent prediction-driven materials discovery. The NSF AI Materials Institute (AI-MI) aims to propel foundational AI research past the limitations of existing AI algorithms by pursuing materials discovery and conquering knowledge- and data-centric challenges. AI-MI will advance the foundations of artificial intelligence while accelerating the discovery of next-generation materials essential for sustainable energy, electronics, environmental stewardship, and quantum technologies. AI-MI brings together computer scientists, materials researchers, and data scientists to leverage advances in AI technology and materials data to drive use-inspired progress in fundamental AI, catalyzing prediction-driven materials discoveries. By tightly integrating data generation, AI inference, and rapid experimental feedback, AI-MI aims to reduce discovery cycles from months to days and to establish reproducible, reusable workflows for the broader community. AI-MI plans to create the AI Materials Science Ecosystem (AIMS-EC)- an open, cloud-based portal. AIMS-EC will couple a science-ready large-language model with multimodal data streams (experimental measurements, simulations, images, and textual literature). This platform will allow researchers to pose natural-language queries and