Powering the Future: How the AI Boom Sparks a Nuclear Renaissance for Low-Carbon Energy

To go back to this article, head to My Profile and then click on View saved stories.

The Surge in Artificial Intelligence Sparks Renewed Interest in Nuclear Energy

After being inactive for half a decade, the first reactor at Pennsylvania's Three Mile Island nuclear facility is poised for revival. A partnership with Microsoft is set to reactivate the reactor by 2028, with the sole purpose of providing the technology giant with an abundant supply of eco-friendly power.

The relationship between major technology companies and nuclear energy continues to evolve. In March, Amazon Web Services committed to purchasing a data center that receives its power from the Susquehanna nuclear power plant in Pennsylvania. During a presentation at Carnegie Mellon University on September 18, Alphabet's CEO Sundar Pichai highlighted small modular nuclear reactors as a possible energy solution for data centers. Moreover, the connections extend further, with OpenAI CEO Sam Altman leading the boards of nuclear energy startups Oklo and Helion Energy.

The surge in artificial intelligence has prompted tech firms to urgently seek out eco-friendly energy options for their data centers. According to the International Energy Agency, the electricity needs for AI, data centers, and cryptocurrency could potentially increase by over 100% by 2026. Their conservative projections indicate that the additional electricity required could match the current consumption levels of Sweden, while their more extreme scenarios suggest it could equal the usage in Germany.

The increase in demand for energy is being warmly welcomed by those in the nuclear power sector. For years, the need for electricity in the US has remained relatively unchanged, but the rapid expansion and impact of artificial intelligence are shifting this trend. A report from a power industry advisory group in December 2023 announced the end of the era of steady electricity consumption, attributing this change to the rising needs of data centers and industrial operations. This report predicts that by 2028, the peak demand for electricity in the US will rise by 38 gigawatts, which is the equivalent of the output of reactor one at Three Mile Island multiplied by 46.

"Artificial intelligence (AI) is rapidly gaining momentum and drawing significant interest within the energy sector," states John Kotek, who holds the position of senior vice president for policy development and public affairs at the Nuclear Energy Institute, a trade association for the nuclear industry. Kotek highlights a crucial aspect of national security. "There's a genuine perception of AI as an arena where the US is competing with international rivals." The concern that the US might lag in AI advancements due to insufficient energy resources "is prompting increased focus," he notes.

Nuclear energy is appealing to technology firms because it offers a continuous supply of low-carbon power, a feature not shared by solar or wind energy, which depend on variable conditions and typically need energy storage solutions to ensure reliability. By rebooting its first reactor, Microsoft will secure 835 megawatts of clean energy for a duration of 20 years under the agreement. With a commitment to achieving a carbon negative status by 2030, Microsoft faces a significant challenge in meeting its environmental goals due to the increasing power demands of artificial intelligence. This challenge became evident in 2023 when the company's carbon emissions saw a 29 percent rise from 2020 levels, largely due to the expansion of its data center infrastructure.

The Three Mile Island nuclear facility includes two reactors. Reactor number two became widely known for experiencing a partial meltdown in 1979, and has not been operational since that time. Meanwhile, the first reactor continued to operate smoothly up until 2019, when it was decommissioned due to economic challenges, particularly from the rise of gas and wind energy sources. Kotek points out that there are only a limited number of dormant reactors that could be restarted with relative ease. However, he also notes a significant interest among power plant operators in seeking license extensions for their currently active reactors, aiming to capitalize on the surge in demand for electricity driven by advancements in artificial intelligence.

The excitement among operators of power plants can partly be attributed to governmental incentives aimed at promoting the generation of low-carbon electricity. The Inflation Reduction Act offers tax benefits for the electricity generated by existing nuclear facilities. However, according to Kotek, the sector must also focus on constructing new reactors to meet the anticipated increase in energy demand. The highest number of nuclear reactors in operation in the United States was 112 in 1990, but this figure dropped to 92 by the year 2022. Moreover, the construction of the most recent reactors at the Vogtle power plant in Georgia stretched over 14 years, exceeding the initial budget by more than two-fold.

Todd Allen, who leads the nuclear engineering and radiological sciences department at the University of Michigan, commented on the United States' challenges in constructing nuclear facilities, using the Vogtle project as an example of inefficiency. However, Allen noted that China demonstrates a faster pace in erecting nuclear power plants, suggesting that the United States could potentially accelerate its processes. He also mentioned that the rising energy needs of data centers might make the development of new nuclear plants a more appealing choice in the near future.

The extended timelines associated with traditional nuclear reactors have led Microsoft to explore the possibilities of small modular reactors, seen as faster and less expensive alternatives. However, according to Sasha Luccioni, who leads AI and climate initiatives at Hugging Face—a firm specializing in creating machine learning application tools—tech companies generally focus more on identifying new energy sources rather than enhancing the efficiency of their AI systems. Luccioni suggests that implementing regulations, such as obligatory disclosure and openness for enterprises offering AI solutions, could encourage better efficiency practices within the industry.

During an event at Carnegie Mellon University, Pichai acknowledged that efforts to enhance the energy efficiency on the consumption end of AI technology are just beginning. He admitted that the current process of training these models is highly inefficient. However, he also highlighted that the process of inference—where an AI model is tasked with executing an operation—has the potential to become significantly more energy-efficient in the future. Google's carbon footprint in 2023 saw a 48 percent increase compared to its 2019 levels, mainly due to higher energy use in data centers and greater emissions from its supply chain. This surge challenges Google's ambition to achieve net-zero emissions by the year 2030. Luccioni pointed out that although the demand for energy by AI systems is climbing, the availability of renewable or low-carbon energy sources to power these systems isn't growing fast enough to meet the demand.

The idea of utilizing the location of the most infamous nuclear incident in the United States to fuel the advancement of artificial intelligence may cause discomfort for some individuals. However, Allen emphasizes that the closure of reactor one was not due to functional problems. According to him, the primary considerations for reactivating the reactor involve confirming its operational readiness and ensuring a sufficient number of skilled personnel are available to operate it efficiently.

Discover More…

Direct to your email: A curated selection of WIRED’s most fascinating and unusual tales.

Elon Musk poses a threat to national security

Conversation: Meredith Whittaker Aims to Disprove Capitalism

What's the solution to a challenge such as Polestar?

Occasion: Don't miss out on The Major Interview happening on December 3rd in San Francisco.

Additional Coverage from WIRED

Assessments and Manuals

© 2024 Condé Nast. All rights reserved. Purchases made via our website may result in WIRED receiving a share of the sales, due to our Affiliate Partnerships with various retailers. Reproduction, distribution, transmission, storage, or any form of usage of the content on this site is strictly prohibited without express written consent from Condé Nast. Advertisement Choices

Choose a global website