AI Factories and Gigawatt Datacenters

The rapid emergence of AI factories is driving a paradigm shift in datacenter design, pushing infrastructure toward unprecedented gigawatt-scale deployments. In this panel, we examine the converging technological, energy, and architectural challenges that define this new class of computing systems. Central to this discussion is power: as availability becomes the primary constraint, we explore strategies for grid integration and co-design with utilities, alongside on-site generation models spanning small modular reactors (SMRs), natural gas, and renewables. We further consider the evolving role of datacenters as energy hubs, balancing supply, demand, and resilience at scale.
Beyond power provisioning, the panel addresses the co-optimization of hardware and workloads in increasingly heterogeneous environments. Advances in hardware-aware inference, mixed-precision techniques, and multi-objective scheduling are enabling new trade-offs across performance, energy, and cost. At the physical layer, extreme rack densities exceeding 100 kW are accelerating the transition to direct liquid cooling, fanless system architectures, and large-scale heat reuse, with emerging reference designs illustrating pathways to gigawatt-scale integration.
Equally critical are software-level innovations that elevate energy efficiency as a first-class objective. From energy-aware compilers and runtime power management to carbon-aware scheduling, the focus is shifting from peak performance to sustained, system-level efficiency. Finally, we examine the architectural spectrum between monolithic gigawatt facilities and federated multi-site deployments, highlighting trade-offs in scalability, resilience, and grid impact. By bringing together perspectives across infrastructure, systems, and policy, this panel aims to define the design principles and research frontiers underpinning the next generation of AI-driven computing at planetary scale.