An Atomic Solution to AI’s Energy Needs
A surge in artificial intelligence technology has spurred tech companies to find alternative energy sources, with many turning towards nuclear power to meet their demand. But this highly reactive energy source can bring more headaches than benefits.
The growth of artificial intelligence in the past year has been explosive. AI is now so heavily integrated into everyday technology—search engines, emails, text messages—that it is almost impossible to avoid. Yet as AI skyrockets nationwide, its large energy demand poses problems for prominent tech firms, leading some to turn to a previously shunned energy source in America: nuclear.
Expansions in generative AI require building large data centers: facilities designed to handle the computing needs of AI models and algorithms. Companies implementing AI technology are now constructing hyperscale data centers, which can have power demands upwards of 100 MW, an electricity demand comparable to over 400,000 electric vehicles.
Many businesses are attempting to power these data centers using clean energy sources to reduce their carbon emissions, but renewable power production has not yet proven reliable enough to act as their sole energy source. Solar and wind plants are weather-dependent and therefore volatile. And at their peak, they can only meet 80% of the data centers’ demand. Nuclear power still produces zero carbon emissions, but in contrast to renewables, nuclear can provide a consistent source of electricity that is necessary to keep data centers running around the clock.
Recent legal battles, however, have dulled the high hopes of the tech industry. In Berwick, PA, Amazon signed a deal with the Susquehanna Steam Electric Station to power their nearby data center with an additional 180 MW, in an agreement known as “co-location.” This setup has raised strong concerns about equity with neighboring communities, as large data centers will circumvent the electric fees that go towards maintaining grid systems. Additionally, if a data center purchases all the energy from an already-established nuclear plant, it essentially takes that plant offline for all nearby customers, forcing them to pay higher prices for more distant energy sources. For those customers powered by PJM, the regional transmission organization that serves Pennsylvania and surrounding states, the Amazon deal would result in an additional $140 million in costs.
The other major obstacle for nuclear-powered data centers comes from a lack of availability. Goldman Sachs estimates that 85-90 GW of nuclear capacity will be needed by 2030 to meet the demand of data centers, yet only 10% of that is predicted to be available. Due to safety concerns and fears of repeat incidents like those at Three Mile Island, the Nuclear Regulatory Commission implements many barriers to gain approval for new reactors, dampening large companies’ ambitions.
Despite these hurdles, businesses are still looking confidently towards nuclear energy for their AI initiatives by investing in small modular reactors, or SMRs. These advanced nuclear reactors are a fraction of the size of their conventional counterparts (“small”) and are able to be assembled and transported from different locations to a final installation location (“modular”). SMRs have gained traction because of their siting flexibility—tech companies can construct them close to their data centers to reduce transmission costs—as well as their reduced demand for water as a coolant. While the first SMRs are not expected to be operational until 2029, major corporations have signed development deals aimed at accelerating implementation. Google has partnered with Kairos Power, which was granted the first U.S. permit in 50 years to build a new type of nuclear reactor, and Amazon has signed deals with Dominion Energy and X-energy to deploy upwards of 5 GW of SMR projects by 2039.
As the future of domestic clean energy remains uncertain, nuclear power may be a viable option for the tech industry to expand its artificial intelligence capacity without escalating its carbon emissions. Yet, as new investment arises, we must remember the lessons of nuclear power catastrophes from decades past and be careful to balance innovation with environmental responsibility.
Matthew Barotz
Undergraduate Seminar FellowMatthew Barotz is a sophomore in the College of Arts and Sciences studying politics, philosophy, and economics with minors in sustainability and data science. Barotz is also a 2025 Undergraduate Student Fellow.
