Panel 1

Green AI Summit 2025: Panel 1 - AI and Data Center Energy Usage and Environmental Impact

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

Overview:

Kicking off the technical discussions on Day 1 of the Green AI Summit 2025, Panel 1 dove deep into the critical environmental costs associated with the backbone of modern artificial intelligence: data centers. Moderated by Professor Le Xie of Harvard University, this session brought together experts from academia and industry leaders like Hitachi, Meta, and Tencent to dissect the challenges and explore innovative solutions for a more sustainable AI infrastructure.

Key Discussion Points:

The panel explored the multifaceted environmental impact of AI technologies and the data centers that power them, moving beyond just operational energy use:

• Broadening the Impact Scope: Experts highlighted often-overlooked environmental costs, including the significant water consumption required for cooling, the supply chain implications and environmental effects of mining rare earth minerals for AI chips, and the substantial embodied carbon associated with manufacturing hardware and constructing facilities. The distinction between power (instantaneous rate) and energy (total consumption) and its impact on grid reliability was also discussed.  

• Technological Innovations: Solutions focused heavily on increasing efficiency. Advances in cooling technologies were a major topic, including waterless cooling methods, direct-to-chip cooling, full immersion cooling, and innovative ways to redistribute waste heat for other uses like heating buildings. The shift towards more efficient DC power distribution within data centers was also noted.  

• Industry Efforts & Strategies: Companies like Meta shared insights into their efforts to instrument and understand energy consumption across the AI lifecycle (training, inference, hardware manufacturing) and the importance of tackling embodied carbon through supply chain engagement. The trend towards building smaller format or edge data centers located closer to power sources or end-users was discussed as a strategy to manage latency, sovereignty, and large-scale power constraints.

• Challenges & Barriers: Panelists acknowledged significant hurdles, including supply chain bottlenecks for capital equipment (like substations) and GPUs, the limited availability of power hindering new projects, the complexity of achieving grid flexibility and developing effective demand response mechanisms (especially for co-location facilities), and the difficulty of scaling solutions due to varying data center types and data access/virtualization limits. The NIMBY ("Not In My Backyard") phenomenon and community backlash against data center construction were also raised as significant business and regulatory challenges.  

Panelists:

• Le Xie (Moderator): Gordon McKay Professor, Electrical Engineering, Harvard John A. Paulson School of Engineering And Applied Sciences

• Francesca Dominici: Clarence James Gamble Professor of Biostatistics, Population, and Data Science at the Harvard T.H. Chan School of Public Health; Director, Harvard Data Science Initiative at Harvard University  

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

• Carole-Jean Wu: Director, AI Research at Meta

• Noman Bashir: Computing & Climate Impact Fellow, MIT Climate & Sustainability Consortium (MCSC)

• Jiaqi Liang (Online): Director, Data Center Operations Planning and Energy, Tencent

This panel set a crucial foundation for the summit, clearly outlining the environmental stakes of the AI revolution and highlighting the diverse technological, operational, and policy pathways needed to align data center growth with sustainability goals.