Techno-economic assessment of energy solutions to power U.S. AI infrastructure and the policy implications for broader electrification

Expanding data center infrastructure to train and deploy artificial intelligence (AI) models is driving a surge in electricity demand in the U.S., and associated impacts for costs, emissions, and resource use. The electric infrastructure buildout from this AI growth will influence the capabilities and feasibility of broader electrification. We develop an open-source end-to-end techno-economic model of energy solutions to power AI infrastructure across the continental U.S. and compare grid-connected and off-grid data center energy infrastructure using AC-coupled solar and storage, DC-coupled solar and storage, behind the meter natural gas, and conventional grid power across the continental U.S. To estimate opportunity costs of construction and equipment delays, we quantify the deployment speed value of AI data centers using GPU spot market rental prices. We discuss the technical, policy, and environmental implications of these different pathways to power U.S. data centers. 

Speaker: Costa Samaras, Carnegie Mellon University

Attend in person or watch online (see weblink)