Microsoft's Bold Bet on Nuclear: Powering AI and Climate Goals with Three Mile Island Deal

As artificial intelligence (AI) technology rapidly advances, its substantial power consumption has become a major concern. Last week, Microsoft announced a significant move by signing a 20-year agreement with Constellation Energy to purchase power from the Three Mile Island nuclear plant. This move comes amidst a surge in demand for clean energy driven by green economy and climate considerations, with new nuclear projects expected to proliferate across the globe, potentially propelling nuclear energy development further with the involvement of tech giants.

Constellation Energy disclosed that they have entered a long-term agreement with Microsoft to supply energy to the company’s data centers. This agreement includes plans to restart a reactor at the Three Mile Island facility in Pennsylvania, located adjacent to a reactor that was closed following a 1979 accident.

The restart of the Three Mile Island plant requires federal, state, and local approvals. In 1979, a partial meltdown of an independent unit occurred, marking a significant industrial accident in the nation's history. The U.S. Nuclear Regulatory Commission (NRC) spokesperson mentioned that Constellation must clarify the rationale behind the restart, and the agency is ready to engage with the company for subsequent steps.

It is noteworthy that AI may hinder tech giants' climate objectives. Microsoft had previously pledged to reduce its greenhouse gas emissions to zero by 2030, yet recent sustainability reports indicate rising emissions, with AI being cited as a contributing factor.

Microsoft’s VP of Energy, Bobby Hollis, stated: “This agreement is a significant milestone in helping to decarbonize the grid and support our carbon commitment. Microsoft continues to work with energy suppliers to develop carbon-free energy to help meet the grid's capacity and reliability needs.”

In May of the previous year, Microsoft signed a power purchase agreement with nuclear fusion technology firm Helion Energy, anticipating the operation of its first nuclear plant by 2028 with a 50 MW target output. Helion Energy's seventh-generation machine, Polaris, is expected to go live this year, utilizing high-powered electromagnetic pulses and helium-3 fuel for fusion.

Much like Microsoft, other major players in AI and computational power competition, such as Amazon and Google, are also turning towards nuclear energy. Microsoft founder Bill Gates revealed plans in June to invest billions more into next-generation nuclear plants in Wyoming, motivated by an unprecedented "electricity storm" sparked by AI’s emergence.

The demand for nuclear energy is likely to further escalate. Despite solar and wind energy expanding at record speeds, they face challenges such as weather dependency, equipment costs, and unresolved storage solutions. Western countries are increasing nuclear power investments, with plans in France to delay reactor closures and Sweden's government aiming to build ten large nuclear units by 2045. Meanwhile, nations like India, Turkey, and Brazil are actively constructing new nuclear facilities.

Last year's COP28 in Dubai saw 24 countries, including the US, Japan, Canada, the UK, and France, pledge to triple their nuclear energy output by 2050. AI's evolution might further drive demand, with the International Energy Agency forecasting that electricity consumption from data centers, AI, and cryptocurrency could grow to 4% of global demand by 2026, double its 2022 level.

As power companies phase out coal plants, transitioning from fossil fuels, grid stability could face demand-side shocks. In July, PJM Interconnection, the largest U.S. grid operator, warned that with coal plant closures outpacing the construction of new plants, system reliability is increasingly at risk.

Benjamin Fowke, interim CEO of American Electric Power, projected to investors that data center demand might surpass 15,000 MW by century's end, with infrastructure costs anticipated to reach hundreds of billions of dollars. Traditionally, a large manufacturing facility might need 100 MW, equivalent to about 100,000 households, but now a single data center can require 3 to 15 times that amount.