Nuclear Energy

Data Centers and AI – The Leading Edge for a Nuclear Revival? 

“All of the above” has been part of the energy conversation in the United States since at least the Obama Administration. But now, 15 years later, “all of the above” may be reaching fruition in an economic sector probably no one foresaw: data centers including artificial intelligence (AI). Perhaps even more surprising, the data center/AI (DC/AI) sector may be on the leading edge of a nuclear power revival. 

As discussed in OnLocation’s recent Energy Horizons long-term forecast for the U.S. energy economy, overall demand for electricity is expected to increase by more than half of current levels by 2050 (Figure 1). One of the main components of this demand growth is the DC/AI sector, which is the subject of another recent OnLocation study. As shown in Figure 2, in the OnLocation DC high case, by 2050, DC/AI demand is expected to reach about 800 terawatt-hours (tWh). To put this in perspective, this almost matches total industrial demand for electricity in 2024 (expected to be about 1000 tWh).

  

Figure 1: Electric Sales by Sector from OnLocation’s Energy Horizons Report

DC/AI demand has several characteristics that help to explain the sector’s interest in nuclear and other forms of dispatchable power. An underlying factor is the ravenous demand for power associated with the AI component. AI requires large amounts of electricity to train the AI systems and then to respond to queries. This results in an AI server rack requiring on the order of 10 times more electricity than a conventional data server. The impact can be seen in Figure 2, which includes a counterfactual projection that assumes no increase in the AI component of demand (the bottom line in the graphic). In this scenario, the data center sector has only modest demand growth through 2050, illustrating how the main driver of demand growth in the DC/AI sector in the low and high scenarios comes from artificial intelligence. 

Figure 2: Electricity Demand by Data Centers from OnLocation’s Data Center Demands & Impacts on the Energy System Report

A common characteristic of conventional DC and the new AI demand is a requirement for an absolutely reliable and continuous electricity supply, in order to provide 24/7/365 service to customers. This has driven many DC/AI vendors and the utilities that service them toward dispatchable natural gas-fired generation. To avoid lengthy utility interconnection queues, some DC/AI facilities are aiming to locate near natural gas pipelines that can supply fuel to dedicated combined cycle plants. DC/AI demand has also contributed to decisions to delay the retirement of some coal-fired generators, at least for a few years. Examples include plants in Maryland, Nebraska, Georgia, and possibly Mississippi and Utah. But at the same time, the DC/AI vendors would like to be seen as environmentally conscious, especially in regard to climate and greenhouse gas emissions. This is obviously inconsistent with reliance on fossil fuels, the future of which may be uncertain in any case. Variable output renewables (solar and wind) linked with batteries is another option to provide firm, clean power. DC/AI vendors such as Google and Meta (Facebook) are also exploring conventional and novel geothermal technologies as a source of dispatchable electricity.  

But perhaps most surprising has been the interest DC/AI developers have shown in nuclear power, including constructing reactors dedicated to DC/AI facilities. Several projects have been announced in 2024: 

  • In addition to the resurrection of Three Mile Island, Google has struck an agreement with small modular reactor (SMR) developer, Kairos Power to deploy a 500 MW fleet of reactors
  • Amazon is pursuing multiple nuclear projects. It is one funder of a small modular reactor (SMR) project in Washington State that envisions the initial deployment of four 80 MW units using technology from the firm X-energy. Amazon is also examining, in partnership with Dominion Energy, the construction of a 300 MW SMR project near the site of the North Anna nuclear station in Virginia. And it is seeking to co-locate a data center in Pennsylvania at Talen Energy’s Susquehanna nuclear station. 

Why are the DC/AI developers interested in nuclear power? In addition to the quest for clean dispatchable energy noted above, it is likely that they aim to reduce future supply risks by diversifying their electricity sources. The prospects for natural gas may be particularly uncertain if the effects of climate change become more severe. Another factor is that they have the money to play the nuclear game. As discussed earlier, Microsoft is willing to pay far above market prices for nuclear-generated electricity. Google has estimated that each of the Kairos SMRs may cost as much as $1 billion, but by DC/AI standards this is, if more than pocket change, not a financial stretch. In the second quarter of 2024 alone, the largest DC/AI companies – Alphabet (the parent of Google), Microsoft, Meta, Apple, and Amazon – spent $59 billion on capex. By one estimate, capital spending on DC/AI worldwide will exceed $400 billion; even by nuclear power standards these are immense expenditures. 

The DC/AI developers may also be counting on federal incentives, in part to help fund their own projects and in part to kickstart a revival of the nuclear industry that will rebuild the American nuclear construction infrastructure. In October, the U.S. Department of Energy (DOE) announced that it would begin taking applications for up to $900 million in funding to help subsidize the development of SMRs. The agency is currently considering requests for up to $65 billion in loans to support nuclear development. The federal government had previously provided a $1.52 billion loan guarantee for the restart of the Palisades nuclear plant in Michigan, plus grants of more than a billion dollars to two rural cooperatives, Wolverine and Hoosier, to buy capacity from Palisades. DOE is also funding a program to keep at-risk nuclear plants in operation, including $1.1 billion to support the Diablo Canyon plant in California. The outgoing Biden administration recently released a blueprint for tripling American nuclear capacity by 2050 and has even been financially supporting the deployment of U.S. nuclear technology abroad, including in Poland and Romania

It is important to remember that the ambitious plans of the DC/AI developers are just that – plans. Not a shovel of earth has been turned yet. Many of the details remain nebulous; for example, the 500 MW of Kairos reactors that Google has announced have no set locations, only an intent to place them in “relevant service locations.” The DC/AI firms have also begun to encounter the regulatory complexities inherent in nuclear development. A plan by Meta to co-locate a data center at an undisclosed existing nuclear plant site was canceled because of environmental (endangered species) impacts and other unspecified regulatory issues. Amazon’s plan to build a data center that will draw dedicated power from the Susquehanna nuclear plant has run into a roadblock at FERC because of concerns that diverting firm power to a data center will have deleterious impacts on other customers. Nonetheless, even with all the uncertainties and challenges, the picture for nuclear power in the United States is more favorable today than it has been in many years, with demand from DC/AI leading the way. 

Book Consultation
Cookie Notice

This site uses technical and analytical cookies to ensure we give you the best experience on our website. Privacy Policy