2025 Second Green AI Summit Concludes, Charting a Roadmap for Sustainable AI Development

Day 1: Harvard University Hosts Opening Discussions

The summit commenced at Harvard's Gutman Conference Center with introductions and Opening Remarks setting a tone of urgency and opportunity. Keynote speakers Senator Edward Markey (via video), Secretary Tomas Lamanauskas (ITU), Vice Provost James Stock (Harvard), David Havelick and Sarah Craig (Harvard Sustainability), and Jerry Huang (Green AI Institute) discussed AI's dual environmental role, its economic potential, the worsening climate crisis, university sustainability initiatives, and the need for policy and collaboration. Mr. Huang also launched the Institute's second white paper on global data center siting.  

Panel 1: AI and Data Center Energy Usage and Environmental Impact, moderated by Professor Le Xie (Harvard), delved into the environmental costs of AI infrastructure. Panelists Francesca Dominici (Harvard), KJ Joshi (Hitachi), Carole-Jean Wu (Meta), Noman Bashir (MIT), and Jiaqi Liang (Online, Tencent) explored energy and water consumption, carbon emissions, the impact of embodied carbon in hardware, cooling system innovations (waterless, immersion), grid reliability concerns, efficiency efforts at major tech companies, and the challenges of scaling data centers sustainably, including supply chain and power availability constraints.

Following the panel, Zhaoyang (Roger) Wang (Online, Alibaba Cloud) delivered a Keynote Speech detailing Alibaba Cloud's efforts in open-sourcing AI models, achieving carbon neutrality goals through sustainable infrastructure (PUE improvements, clean energy adoption), and deploying AI for sustainability applications like grid resilience forecasting and industrial decarbonization, referencing their ANGEL research collaboration.

The Initiatives Launching session highlighted efforts to scale Green AI. Jeremy Renshaw (EPRI) introduced the Open Power AI Consortium, a global collaboration focused on developing efficient, domain-specific AI models for the power sector using open-source data and sandboxes. 

Professor Le Xie (Harvard) presented Harvard SEAS's "Power and AI Initiative" (PAI), researching the crucial two-way interaction between AI capabilities and power system needs. 

Paulo Carvao (Harvard Kennedy School) discussed the US AI policy landscape and proposed a dynamic governance model involving standards, compliance, and accountability to foster innovation responsibly.

After lunch, Panel 2: Towards a Low-Carbon AI and Data Center Ecosystem, moderated by Andreas Klaube (EPRI), explored solutions for reducing the carbon footprint of the digital backbone. Speakers Junhui Zhao (Eversource), Varun Sivaram (CFR), Peng Gao (Harvard), and Shaolei Ren (Online, UC Riverside) discussed energy-efficient infrastructure, renewable energy adoption, demand flexibility, the role of AI in utility operations (like asset inspection and outage management), policy frameworks, environmental justice considerations related to infrastructure siting, and the importance of system-wide efficiency thinking.

Panel 3: Next-Generation Computing Hardware, moderated by Professor Gage Hills (Harvard), looked beyond current silicon. Panelists Nate Gemelke (QuEra Computing), Professor Ajay Joshi (BU & Lightmatter), and Olivier Ezratty (Online, Quantum Energy Initiative) discussed the potential and challenges of advanced electronic integration (3D stacking), optical computing and interconnects (photonic advantages in efficiency and embodied carbon), and quantum computing (unique capabilities, but current limitations in size, speed, energy use, and fault tolerance). The discussion emphasized the need to integrate sustainability considerations early in the design of these future technologies.

A series of Lightning Talks showcased emerging innovations. KJ Joshi (Hitachi) detailed AI data center design trends (location, scale, power, cooling). 

Jennifer Turliuk (MIT) presented a framework for analyzing AI's net climate impact, balancing harms and benefits.

 Monica Wang (SJTU) discussed AI's transformative role in education and fostering decentralized innovation. 

Can Hankendi (BU) introduced the Carbon Meter tool for estimating data center footprints.

The final panel of the day, Panel 4: Green AI and Data Center Policy and Investment, moderated by Karen Clopton (formerly CA PUC), focused on accelerating sustainable AI ecosystems. Panelists Ari Peskoe (Harvard Law), Galen Nelson (MassCEC), Vernon Turner (Cognizant), Benjamin Sovacool (BU IGS), and Klára Talabér-Ritz (Online, European Commission) explored public-private partnerships, financial incentives (like green bonds), regulatory frameworks (including the EU AI Act), the role of state versus federal policy in the US, environmental justice, and defining/measuring sustainable AI practices. Day 1 concluded with a Launching Ceremony and networking.