Artificial Intelligence: Implications for Energy and the Environment

The Environmental and Energy Study Institute (EESI) held a briefing discussing the intersection of artificial intelligence (AI) and climate change in federal policy-making. While AI can aid in climate resilience and boost economic competitiveness, it is also on a trajectory to increase energy demand, greenhouse gas emissions, and water usage. This paradox presents an important opportunity for discussion on how to best minimize the negative impacts of AI on the environment and harness its powers for a sustainable future. 

This briefing provided a foundational understanding of AI’s role in the climate and energy arena. Panelists discussed the massive energy and water needs of data centers that run AI algorithms. They also layed out how the technology is already being put to use—from precision agriculture to resilient grid infrastructure and improved weather forecasting. The briefing highlighted the frontiers of AI, including the federal government’s role in research and development at the Department of Energy’s National Laboratories, and explored Congress’s role in aligning the rapid rise of AI development and usage with global goals to reduce greenhouse gas emissions and adapt to climate impacts.

Highlights

KEY TAKEAWAYS

Artificial intelligence (AI) has introduced new uncertainties around energy demand, energy supply, and energy prices. A Lawrence Berkeley National Lab report predicts that data centers could consume between 6% to 12% of total U.S. electricity demand by 2028.  AI can also be a helpful tool to modernize the grid and respond to disasters, including by predicting extreme weather events to make it possible for people to get out of harm’s way.  Frugal AI models—those that achieve high performance without excessive complication or energy costs—can be beneficial and outweigh their costs.

 

Rep. Chuck Fleischmann (R-Tenn)

• Rising U.S. energy demand calls for bipartisan solutions, including government funding of National Laboratories, modernizing energy infrastructure, and securing diverse sources of American energy production.
• The Energy and Water Development Subcommittee of the U.S. House of Representatives Committee on Appropriations routinely meets with energy stakeholders to ensure grid stability.
• Nuclear energy is safe, secure, clean, and reliable, and it is experiencing a renaissance. Nuclear energy can be seen as an important fuel to advance artificial intelligence (AI).

 

Sen. Brian Schatz (D-Hawaii)

• The popularity of AI increases energy demand, causing energy costs to rise.
• AI can also be a helpful tool to modernize the grid and respond to disasters, including by predicting extreme weather events to make it possible for people to get out of harm’s way.
• The Transformational Artificial intelligence to Modernize the Economy against (TAME) Extreme Weather and Wildfires Act (S.1378) is a bipartisan bill designed to advance the use of AI models in weather forecasting.

 

Mike Sexton, Senior Policy Advisor for AI and Digital Technology, Third Way

• Artificial general intelligence is a hypothetical version of AI, with broad problem-solving capabilities, which some people would equate to the level of a high-functioning human.
• According to the singularity theory, increases in artificial general intelligence would yield a self-improving AI that could create solutions beyond human capabilities and comprehension.
• When ChatGPT was publicly released in November 2022, it was a simple large language model, using predictive decision-making to answer queries.
• Now, ChatGPT also uses reasoning models that are more intensive versions of large language models, using more energy, resources, and time to “think” through answers. When asking ChatGPT 5 (the latest publicly available version at the time of this briefing) a query, the chatbot now decides whether to use a large language model or a reasoning model to answer the question based on its complexity.
• When ChatGPT puts little “thought” into a query, it is susceptible to providing incorrect and careless answers. It can also “hallucinate,” or make up completely false sources or facts.
• Data centers processing AI use a lot of water. Closed-loop cooling systems can reuse the same water repeatedly to cool devices, reducing overall water use. Not all data centers are using closed-looped systems.
• The United States is in a competition with China to build the best AI, which includes competition to generate an energy supply for AI. China is ahead of the United States in terms of investments in renewable energy to power AI, so the United States needs to scale up its renewable energy capacity.

 

Ahmed Aziz Ezzat, Assistant Professor, Department of Industrial and Systems Engineering, Rutgers University

• AI has introduced new uncertainties around energy demand, energy supply, and energy prices. A Lawrence Berkeley National Lab report predicts that data centers could consume between 6% to 12% of total U.S. electricity demand by 2028.
• The escalating rate of extreme weather events is disrupting both renewable and traditional forms of energy supply.
• AI-powered tools can also contribute to addressing these uncertainties by complementing and replacing the traditional methods of forecasting to help us better understand extreme weather and energy markets.
• In the case of Winter Storm Elliott in 2022, the post-storm report from PJM Interconnection, the regional transmission organization, found that the traditional weather models used did not accurately predict the severity of the storm. Also, the cold temperatures experienced during the storm were outside the range of temperatures used to train the utility’s load forecast models.
• AI can be used to improve forecasts because it can consider multiple modes of data (e.g., images, text, time series, etc.) and multiple data sources (e.g., satellites, smart meters, weather stations, etc.). AI can also generalize and make rapid adaptations.
• Better forecasting tools and operations can mean lower costs, higher reliability, and a smaller environmental footprint.
• Rutgers University researchers are making advancements in AI to help improve energy system operations.
• AI-powered weather forecasting models are able to take in both the traditional models and additional information. The AI models extrapolate data for places where there is no information. Rutgers researchers found that this modeling could reduce the forecasting error by up to 8% relative to traditional numerical weather prediction models.
• Rutgers researchers integrated these AI-powered weather forecasts into a model for the New York power grid that assesses how much additional capacity is needed on a daily basis for contingencies. The AI-powered forecasts demonstrated that New York could reduce reserve requirements by up to 5%, which is equivalent to $3.2 million in annual cost savings.
• Frugal AI models—those that achieve high performance without excessive complication or energy costs—can be beneficial and outweigh their costs.
• It is important to train and invest in the next generation of AI-literate engineers to accelerate innovation and broaden the impact of AI.

 

Fatima Ahmad, Founder and CEO, AI For Energy

• The National AI Research Resource aims to provide resources for all universities—including historically black colleges and universities and minority-serving institutions—to have access to computing in order to conduct groundbreaking research.
• The development and deployment of AI require energy, and, at the same time, AI has the potential to transform the energy sector.
• Data centers are located nationwide, but they are clustered in particular areas, including the Mid-Atlantic. Data centers used about 4% of total electricity in the United States in 2024.
• We do not know what the demand for AI is going to look like in the future. It is likely that as the chips and models become more efficient, they will become cheaper, which means that consumer use will increase. This will increase total energy use despite the efficiency gains made.
• The introduction of air conditioning in buildings was the last time the grid experienced this scale of load growth.
• Players involved in managing the grid are the Federal Energy Regulatory Commission (FERC), regional grid operators (e.g., PJM, MISO (Midcontinent Independent System Operator), and ERCOT (Electric Reliability Council of Texas)), each state’s energy regulatory commission, and the utilities. All these entities have a role to play in managing the energy demand of AI.
• In the short term, building renewable energy and storage can increase grid capacity. Enhancing the existing electric grid through advanced conductors, energy efficiency, and demand response can also help the United States make the most of the grid as it is today. In the longer term, nuclear energy and next-generation geothermal energy are likely to provide a greater share of energy to the grid.
• Co-location means putting a power plant next to a data center and using that power to directly run operations. FERC rejected an application from Amazon to co-locate a data center with a nuclear power plant in Pennsylvania in 2024.
• Data center flexibility means shifting data processing to areas near abundant renewable energy sources, even if the place where the work is being done is not nearby. The challenge here is higher latency, meaning the time it takes to process data is longer.
• Turning on backup power for data centers when the grid is seeing high demand from other users may be a way to offset stress on the electrical grid, though there are air pollution implications if the backup power is diesel.
• The White House AI Action Plan includes steps to create streamlined permitting for data centers and updating the grid.
• On permitting, it will also be important that these data centers have a “social license” to operate, meaning that communities allow them and they meet high environmental standards.
• The Department of Energy’s Grid Deployment Office has an active request for information, “Accelerating Speed to Power/Winning the Artificial Intelligence Race: Federal Action To Rapidly Expand Grid Capacity and Enable Electricity Demand Growth.”
• The National Labs are on the front lines of AI for energy. A 2024 report recommends using AI for permitting processes. Pacific Northwest National Lab is running a pilot, Permit AI, to test using AI to collect data from past permit reviews and to compile public comments.

 

Q&A

Q: How close is the United States to properly implementing small modular reactors, and is that the solution when it comes to increasing energy supply?

Ahmad

• Small modular reactors can be part of the solution to AI’s increasing energy demand, but it is a long-term solution that will take time to further develop and deploy.
• In the meantime, policymakers need to explore other ideas, like increasing the efficiency of the existing grid.

 

Q: Can you elaborate on why FERC rejected the co-location of the data center and nuclear plant in Pennsylvania?

Ahmad

• There is a lack of guidance surrounding the proper conditions for co-location, so approvals and rejections are on a case-by-case basis.
• Because there is potential in co-location, experts expect that FERC will provide additional guidance on siting criteria for future data centers with existing power plants.

 

Q: What is being done at Rutgers University to inform the next generation of engineers about AI?

Ezzat

• Multiple new courses at Rutgers emphasize and encourage proper use of AI. Universities are figuring out how to shift from the perspective of AI as a tool for cheating to AI as a tool for learning.

 

Q: How do you see AI changing or reshaping the policy-making process itself? Are you seeing any federal agencies integrating AI into their workflows?

Sexton

• NotebookLM, by Google, is an AI tool that makes reviewing legislation easier by allowing users to make hard-to-read texts more digestible through visuals.

Ahmad

• The General Services Administration has efforts underway to modernize how the federal government uses technology. The federal IT system is very complex, and it will likely take a long time before AI is integrated across it.
• Historically, the general public did not have access to technology that would allow individuals to easily understand legislation, especially large packages of legislation that were moving quickly through Congress. Now, with AI, the average person can find specific topics within thousands of pages of text in minutes, which makes the policy-making process more accessible, but also might make it more challenging for legislators to reach consensus.

 

Compiled by Olivia Benedict and Hailey Morris and edited for clarity and length. This is not a transcript.