The long-stagnant U.S. power infrastructure is beginning to be stretched thin in the face of skyrocketing electricity use due to the rise of artificial intelligence (AI). As a result, flexible, round-the-clock nuclear power generation has begun to become a new target for tech majors who continue to expand data centers to develop AI. However, can long-dormant nuclear power really land in data centers and fuel AI development in the U.S.?
From robbing chips to robbing electricity
The demand for electricity spawned by the AI boom has led U.S. tech companies to search for power supplies across the country. Now, they’re zeroing in on a key target: America’s nuclear power plants.
The Wall Street Journal reported on the 1st that about 1/3 of nuclear power plant owners across the United States are negotiating with technology companies on the supply of electricity for new data centers to meet artificial intelligence needs. According to people familiar with the matter, Amazon Web Services (referred to as AWS) will reach an agreement with the largest U.S. nuclear power plant operator Constellation Energy by the latter located in the U.S. East Coast of a nuclear power plant direct power supply.
Previously, in another deal in March, AWS had acquired a nuclear-powered data center in Pennsylvania for $650 million. The move also raised the alarm of Patrick Cicero, a consumer advocate in the state, who claimed, “Never before has anyone been able to say to a nuclear power plant, we want as much energy as you can provide.”
The Wall Street Journal describes that of the green energy sources, solar and wind require geographic and weather conditions, and that there are problems with power supply and storage when there is no sunlight or wind. And round-the-clock nuclear power would be far more reliable than other types of power generation, and could help tech companies fulfill climate protection commitments while powering artificial intelligence.
A paper published in the sustainable energy journal Joule says that if Google were to incorporate generative AI into every search, its electricity use is estimated to rise to around 29 billion kilowatt-hours per year. That’s more electricity than is used in Kenya, Croatia, and many other countries. The FT says that data center electricity use will rise from 4% to 6% of total U.S. electricity demand by 2026.
Not only that, each big factory is also continuing to “roll” chip, “roll” arithmetic big model, in the future, AI power consumption or will face exponential growth. According to the International Energy Agency forecast, by 2026, global data centers, artificial intelligence and other power-consuming industry power demand will be about twice as much as in 2022, equivalent to Germany’s total power demand for four years.
In fact, the “nuclear power plant and data center” model of energy supply cooperation has already begun, especially in areas where data centers are concentrated. In addition to Amazon, OpenAI has also reached an agreement with Oaklow to build a small nuclear power plant in southeastern Idaho to provide power for its data centers. Last summer, Microsoft also signed a partnership agreement with Constellation Energy, which will power Microsoft’s data centers in Virginia through its nuclear facilities.
Notably, instead of open-sourcing the construction of new power facilities, these tech companies are buying power from existing nuclear plants and diverting existing power resources, a move that the Wall Street Journal says raises concerns about grid reliability. In addition, the grab for power resources will force other power plants to have to increase generation from natural gas to replace diverted nuclear power, hindering the goal of the U.S. energy transition.
A special “job posting”
In an effort to explore future power supply solutions, Microsoft opened a job posting in 2023 looking for a nuclear technologist to evaluate and integrate small modular nuclear reactors and microreactors “to power data centers that support Microsoft’s cloud and artificial intelligence.” In May of the same year, Microsoft also signed a power purchase agreement to buy power from nuclear fusion startup Helion in 2028, a startup in which OpenAI CEO Altman was a key early investor.
In response to the prospect of nuclear power-fed data centers, Christopher Loss, Innovation Technology Leader for the U.S. Department of Energy’s Nuclear Accelerated Innovation Portal Program, said, “We continue to study the prospects for the economic and power benefits that can be generated by combining nuclear power with data centers.” Recently, the U.S. also passed the Nuclear Recovery Act, which hopes to pave the way for large-scale new nuclear power plants and advance the commercialization of new nuclear power technologies.
U.S. power supply faces bottlenecks
In addition, the U.S. power grid plan is still very old, even if the nuclear power into the AI data center power supply source, the U.S. power tension can not be relieved. According to the latest data from the Lawrence Berkeley National Laboratory, the surge in overall demand has created a backlog of generation and transmission projects that are lining up across the U.S. to connect to the grid.
Huatai Securities, a researcher recently in the research report that, in the face of the explosion of demand, the U.S. power supply has become a new “bottleneck”. Affected by the approval process and construction capacity factors, the United States from the submission of new power projects to the grid application for grid connection and commissioning of an average of 5 years, the new transmission projects require 7.5-13 years for the entire process. Therefore, under the background of the continuous growth of new loads, the United States power supply and demand tensions, rising electricity prices or will become a long-term continuing trend.
