Panel 6

Green AI Summit 2025: Panel 6 - AI and the Grid: Complex Interactions

Session Date: April 26, 2025 Location: Boston University Duan Family Center for Computing & Data Sciences

Overview:

This crucial panel on Day 2 delved into the intricate relationship between artificial intelligence, the electric power grid, and the burgeoning energy demands of data centers. Moderated by Professor Ayse Coskun of Boston University, the session featured experts from academia, research labs, and utilities, exploring the challenges and opportunities in optimizing grid operations, integrating renewables, and enhancing reliability with the help of AI, while managing the impacts of large AI loads.

Key Discussion Points:

The panelists navigated the complex dynamics at the intersection of AI and power systems:

• AI Applications for the Grid: While forecasting (load and generation) remains a primary AI use case, the panel discussed broader applications. AI can potentially assist in optimizing grid operations, improving stability analysis, identifying optimal interconnection strategies, and potentially even accelerating breakthroughs in clean energy like fusion. Foundation models (beyond just LLMs) were highlighted by Hendrik Hamann as a promising approach to overcome the limitations of bespoke, non-scalable AI solutions currently common in the sector. Junwei Cao provided a specific example from China involving using AI and multi-source data to diagnose complex power quality issues affecting industrial customers.

• Data Challenges: A significant barrier identified by multiple panelists is data. Utilities possess vast amounts of data, but it is often siloed in bespoke, inconsistent systems, difficult to access due to confidentiality regulations (Critical Energy Infrastructure Information - CEII), and not structured for AI applications. Solutions discussed included developing better protocols for synthetic data generation, leveraging initiatives like EPRI's for data anonymization and sharing, and deploying technologies like Advanced Metering Infrastructure (AMI) to gather cleaner, more granular data.

• Data Center Flexibility & Demand Response: The immense potential of data center flexibility was reiterated. Tyler Norris emphasized that the grid is significantly underutilized most of the year, and flexible loads could absorb vast amounts of energy during off-peak times, mitigating the need for fossil fuel expansion and buying time for cleaner resource deployment. Elli Nkatou stressed that flexibility must be bidirectional and predictable to be useful for utility planning and operation, preventing instability from either under- or over-generation relative to demand.

• Implementation Hurdles: Despite the technical potential, significant challenges remain in implementing widespread flexibility. Junwei Cao noted that in China, the high cost of AI hardware and the lack of established market mechanisms currently disincentivize flexibility. Elli Nkatou pointed to the workforce gap within utilities, needing operators who understand and trust AI tools beyond a "black box" level. Tyler Norris and Hendrik Hamann discussed the complexity arising from different types of data centers with varying flexibility potential and Service Level Agreements (SLAs), the issue of data locality (data needs to be near compute), and the misaligned incentives for utilities and transmission providers, who are often rewarded for building capacity rather than optimizing utilization. The adversarial nature of regulatory proceedings was also cited as a barrier to collaborative solutions.

Panelists:

• Ayse Coskun (Moderator): Professor, Electrical and Computer Engineering & Systems, Engineering, Boston University

• Hendrik Hamann: Chief Science Officer, IBM Research for Climate and Sustainability

• Elli Nkatou: Manager, System Resilience and Reliability, Eversource Energy

• Junwei Cao: Professor, Beijing National Research Center for Information Science and Technology, Tsinghua University

• Tyler Norris: J.B. Duke Fellow & PhD Student, Duke University

This panel highlighted the critical need for bridging the gap between the rapidly evolving AI/data center sector and the more traditional, regulated power grid industry. Success hinges on addressing data barriers, developing appropriate market signals and incentives, fostering a capable workforce, and enabling deeper collaboration to harness AI for a truly smarter, cleaner, and more flexible grid.