AI/Data Centers

This is a CTEN Meeting Summary where we were lucky to have Chris Rivers present on

AI Energy Demands and Nuclear Power

Andy introduced Chris, a consultant with experience in AI and data centers, to discuss the intersection of AI and energy demands. Chris shared his background, including his work in the military, government, and consulting, and explained the different approaches to AI development in China, the EU, and the US. He highlighted the Trump administration's push for both nuclear energy and AI development, noting the potential environmental and regulatory concerns. The presentation aimed to increase knowledge and debate on how to address the growing energy demands of AI, with a focus on the role of nuclear power.

AI Energy Consumption and Solutions

Christopher discussed the significant energy consumption of AI, highlighting that training models like ChatGPT requires substantial electricity, with estimates for ChatGPT-4 being 1,750 megawatt hours. He explained that using AI models also consumes energy, with a single response taking 3-30 watt hours, equivalent to running a 10 Watt LED bulb for 18 minutes to 3 hours. Christopher noted that data centers, which host AI models, use large amounts of electricity and water for cooling, and their energy demand is expected to increase, potentially reaching 8-11% of U.S. power demand by 2030. He suggested that nuclear energy might be a viable option to meet the growing energy demands of AI, as it offers a large-scale, clean energy solution.

Nuclear Power: Prospects and Challenges

Christopher discussed the pros and cons of nuclear power, highlighting its role in job creation and energy production but also addressing concerns about spent fuel storage and high construction costs. He explained the challenges of building new nuclear power plants, including the lengthy approval process, and introduced the concept of small modular reactors (SMRs) as a potential solution. Christopher noted that while SMRs offer advantages like standardized designs and potential cost savings, they are still in the developmental stage and face regulatory and safety challenges. He concluded by mentioning the Biden Administration’s Inflation Reduction Act's funding for nuclear and fusion research, while expressing uncertainty about the Trump administration's commitment to basic science research.

Data Center Energy Challenges

Christopher discussed the challenges of powering data centers with renewable energy, noting that while battery storage is sufficient for most homes and commercial uses, it may not meet the around-the-clock needs of data centers and AI models. He highlighted the high costs, regulatory challenges, and public perception issues associated with nuclear energy, emphasizing the need for nuanced conversations on the topic. Christopher also addressed questions about the power characteristics of AI and data centers, mentioning the growing demand for data storage and the potential for using locations with cheap electricity, such as Chile. Peter raised concerns about small modular nuclear reactors being a 10-year solution for a rapidly emerging two-year problem, to which Christopher responded that the U.S. might experience increasing electrical costs and brownouts in the near future, potentially leading to a resurgence in coal and natural gas usage while continuing to invest in renewables and small modular reactors.

Data Centers as Thermal Resources

The group discussed the potential for data centers to serve as thermal resources for energy networks, with Peter inquiring about reciprocal interest from data center developers. Christopher expressed interest in the idea but noted that data centers are often secretive. Bernard and Andy suggested that data centers, being profitable, should be encouraged to contribute to community benefits. The conversation also touched on the intermittency of wind and solar energy, with Bernard advocating for wind power due to its lower cost and quicker implementation compared to nuclear options. Diane inquired about water usage by data centers and zoning regulations to protect drinking water sources, to which Christopher agreed to research and provide information.

AI Data Centers and Water Usage

The group discussed the potential conflict between AI data center water usage and human needs during droughts, with Chris explaining that smaller, more efficient AI models could reduce power demands. They explored tax incentives for data centers in Connecticut, with Chris noting that while these incentives help struggling rural towns, data centers often try to strong-arm local officials. The discussion concluded with Peter raising questions about infrastructure upgrade costs for data centers, with Chris explaining that while some states require businesses to pay for grid upgrades, this approach might not be practical for data centers and could lead to concentrated power usage near production sites.

Data Centers: Benefits and Challenges

The meeting focused on the value and implications of data centers, with Christopher explaining their importance for storing and processing data and highlighting the benefits of having them in the U.S. for data protection and reduced latency. Andy raised concerns about data privacy and the need for updated laws to protect personal information, noting that current regulations are outdated. The group discussed the potential for new legislation regarding AI and data centers in Connecticut, with Andy emphasizing the need to address environmental impacts and generate a list of legitimate questions about data center installation and use.

AI can make mistakes. Review for accuracy.

Policy tools being used regarding Data Center Water use

(will be posted soon)