[] Title: Realizing Artificial Intelligence: Edge-to-Cloud-to-Exascale Abstract: Foundational models with trillions of parameters are being trained. Multi-modal GenAI and Inference Serving services are being deployed for a variety of use cases. To meet the computational demands of these AI workloads, we now have infrastructure with larger than ever GPUs and networks with ever increasing bandwidths. In this presentation, I will talk about challenges of running today’s AI workloads on extreme scale infrastructure. Hewlett Packard Labs is pursuing different research directions for building resilient, scalable and sustainable AI infrastructures. I will discuss how we are tackling the complexities of orchestrating AI/ML workloads by leveraging AI Workload simulations, GPU virtualization, performant communication collectives and novel accelerators. Virtual: https://events.vtools.ieee.org/m/462458