AI infrastructure, encompassing compute, data, and models, will determine who is able to build, deploy, and govern AI over the coming decade. At present, access to these foundational resources remains uneven and highly concentrated, constraining meaningful participation across much of the Global South. This session, organised by Factly and Meta, examines how an AI Commons can be developed through shared compute frameworks, AI-ready open data systems, and interoperable model and tooling ecosystems aligned with globally recognised standards.