Keynote Address

The special keynote on AI’s energy challenge highlighted how data center and power grid coordination could pave the way to a more sustainable future, focusing on the rapidly rising energy demands of AI and data centers. Professor Ayse Coskun from Boston University opened the discussion by emphasizing the critical strain that global data centers, now consuming approximately 500 TWh annually, are placing on power infrastructure. She projected an additional 10 GW demand over the next decade driven by AI, equivalent to the output of multiple large nuclear plants. To meet this demand sustainably, Coskun proposed demand response strategies where data centers could dynamically adapt to power grid conditions, thus supporting both operational reliability and environmental goals. Her presentation underscored the need for enhanced infrastructure and grid coordination, especially for renewable energy integration to meet these evolving demands.

Following, Devesh Tiwari, Associate Vice Provost from Northeastern University, explored methods to reduce the carbon footprint of AI operations through what he termed “Lean AI,” which emphasizes efficient resource utilization without sacrificing performance. He advocated for extending the hardware lifecycle, shifting to serverless architectures, and creating incentive structures that encourage sustainable practices. Tiwari suggested that adopting carbon-based budgeting, rather than the traditional core-hour budgeting for supercomputers, could reshape how data centers prioritize energy use. He also proposed a carbon-focused ranking system, enabling stakeholders to compare the environmental impacts of computing facilities. 

Concluding the keynote, Professor Andrew A. Chien of the University of Chicago, speaking online, presented a vision for balancing AI power needs with environmental sustainability through flexible, grid-aware operations. Chien highlighted the concept of “Zero Carbon Cloud,” in which adaptable, virtualized data centers prioritize renewable energy sourcing by sharing data in real-time with power grids. By optimizing power usage and minimizing emissions, his model aimed to decouple data center growth from environmental strain, addressing the dual demands of carbon and water sustainability in modern cooling systems. This keynote collectively emphasized the potential of strategic, coordinated actions between data centers and power grids to drive meaningful change in sustainable AI infrastructure.