Artificial Intelligence (AI) carries much promise for people with disabilities, particularly in their ability to address unmet needs by enabling the design of novel accessible technologies, scaling deployments, and creating personalized experiences. Yet, this rapid proliferation, like prior large-scale technological shifts, risks leaving people with disabilities behind, resulting in what I term the AI divide. In this talk, I outline my work exploring the causes of the AI divide. I use case studies with people with visual impairments in India and draw from critical disability studies to highlight how the AI divide is the outcome of (1) ableist technology gatekeepers, (2) culturally misaligned models, and (3) ineffective governance mechanisms. I highlight the need for a multi-pronged approach that centers community engagement and inclusive policy development to bridge the AI divide. I conclude with my future research agenda, which outlines opportunities to further engage with the AI divide.