Lightning Talks

Green AI Summit 2025: Lightning Talks - Emerging Innovations in Green AI

Session Date: April 25, 2025 Location: Harvard University Gutman Conference Center

Overview:

This dynamic session on Day 1 featured a rapid-fire series of presentations showcasing cutting-edge ideas and novel research directions shaping the future of Green AI. Researchers and industry pioneers offered glimpses into emerging innovations, sparking curiosity and highlighting diverse approaches to tackling AI's sustainability challenges.

Highlights from the Talks:

• KJ (Kaushik) Joshi (Hitachi): Provided an industry perspective on the significant shifts in AI data center design. He discussed key drivers like viewing compute as future labor and energy constraints, leading to trends such as moving to Tier 2/3 locations, massive increases in scale (up to gigawatts), on-site power generation, evolving electrical systems (higher rack density, AC to DC conversion), advanced cooling technologies (RDHx, direct-to-chip, immersion), and the rise of smaller, modular edge data centers.

• Jennifer Turliuk (MIT): Addressed the complex question of AI's overall climate impact by introducing a framework to calculate the Net Climate Impact of AI. This involved balancing AI's climate harms (e.g., energy consumption estimated to potentially reach 21% of demand by 2030) against its potential climate benefits (estimated at 5-10% emissions mitigation). She emphasized considering factors like the installed base of hardware, the time value of carbon, Jevons Paradox (efficiency vs. usage increase), and load flexibility to guide AI development towards projects with a net positive climate outcome.

• Monica Wang (Shanghai Jiaotong University): Focused on AI's transformative role in education, innovation, and entrepreneurship. She discussed how AI is enabling decentralized work, fostering leaner companies, and empowering "super individuals." Using her daughter's experience creating AI-powered games without prior coding knowledge, she argued that AI lowers innovation barriers, accelerates timelines, and that education must evolve to cultivate not just technical skills but also creativity, passion, and critical thinking for the AI era, ensuring technology serves humanity.

• Can Hankendi (Boston University): Presented Carbon Meter, an open-source tool designed to estimate the carbon footprint of data centers using minimal input data (power capacity and area) to overcome common data gaps. He explained how the tool breaks down embodied (manufacturing) versus operational footprints, showed case studies demonstrating how this ratio varies significantly based on server type (especially storage), and highlighted its potential for analyzing carbon reduction strategies like workload migration across different geographic locations with varying energy costs and carbon intensities.

Speakers:

• KJ (Kaushik) Joshi: Chief Business Officer & Head, Data Center Business, Hitachi

• Jennifer Turliuk: Practice Leader, Climate and Energy AI, Martin Trust Center for MIT Entrepreneurship

• Monica Wang: Deputy Director, the Institute of Industrial Innovation, Shanghai Jiaotong University (I3, SJTU)

• Can Hankendi: Post-doctoral Researcher, Boston University

This session provided a valuable snapshot of diverse, emerging fronts in the Green AI landscape, from large-scale infrastructure design and impact analysis to educational transformation and practical footprint calculation tools.