The AI boom (or bubble, depending on your perspective) has spurred renewed interest in building new nuclear reactors to supply the enormous amount of energy data centers use. In fact, artificial intelligence and nuclear power have a lot in common. Both require a tremendous amount of costly infrastructure and natural resources to produce a product that can easily be made more cheaply and with far fewer environmental impacts.
On average, a generative AI query using ChatGPT needs nearly 10 times as much electricity to process than a simple Google search, a process that now occurs automatically whether we want it or not. In a similar “overkill” scenario, the Point Beach nuclear power plant uses about 1,100 million gallons of Lake Michigan water per day for summer cooling, while wind turbines and solar panels use—none.
“It’s like cutting butter with a chain saw”
It’s worth noting here that electricity can be made in many ways, but they all work on the same basic principle: a fuel source creates motion, which spins a turbine connected to a generator or stimulates a flow of electricity. Coal and gas burn the fuel source to make steam or pressure, which spins a turbine.
Nuclear fission, on the other hand, is a process in which radioactive atomic nuclei break down and release energy in the form of radiation. The chain reaction, if not controlled, can produce so much heat that the reactor core can melt and release dangerous radiation. The energy produced from nuclear fission heats water, which produces steam, simply to turn a turbine. Which raises the obvious question: if you can heat water to make steam to turn a turbine in safer ways, why would you do it with radioactive materials?
Or maybe you don’t need to make steam at all…
Consider wind energy. The wind turns blades, which in turn spin a turbine to create electricity. And when sunlight hits a solar panel, it energizes the atoms inside, and electrons break free, creating a flow of electricity. Wind and solar can be licensed and constructed quickly, use no water, and are the least expensive form of energy generation available. Unlike nuclear power, wind and solar don’t require a ten-mile “evacuation zone” around them in the event of a disaster. Nevertheless, proponents of nuclear power claim that this technology is the key to meeting data center demands for electricity, and some Wisconsin lawmakers are scrambling to offer nuclear developers as many incentives as possible to locate in our state.
What do they want? Power. When do they want it?
Data centers are being built now. Nuclear projects face lengthy approval processes, construction delays, and financial challenges that don’t match data centers’ rapid growth timelines. High up-front costs and first-of-a-kind deployment make advanced nuclear designs economically risky. Although the Trump administration recently announced an $80 billion-plus commitment to help fund new reactors, there are plenty of reasons to be skeptical that these projects will materialize.
The fiscal estimate for a sales tax exemption bill for nuclear power being circulated in the Wisconsin legislature notes that “Regarding construction timelines, according to the International Atomic Energy Agency, between 1991 and 2024 the United States constructed four new nuclear reactors. The median construction time (measured from the first pouring of concrete to grid connection) for those 4 reactors was 216 months, or 18 years. The reactor with the shortest construction time was 121 months, or 10 years.” And that doesn’t count planning and licensing time.
Two of those reactors, at the Vogtle site in Georgia — came in at more than $21 billion over budget and seven years behind schedule.
So many ways to pay (too much) for nuclear power
For decades, the U.S. has poured billions of taxpayer dollars into research and development for nuclear reactors, which are far and away the costliest and most risky form of energy production. Federal taxpayers also pay $800 million per year to cover government payments to utilities—the result of the government’s failure to remove highly radioactive waste from reactor sites across the country as the law requires. Even with those payments, owners of the retired Kewaunee nuclear station spend over $10 million per year on spent fuel management.
Taxpayers pick up the tab for the millions of dollars’ worth of subsidies being paid to the nuclear industry, and utility customers pay a premium for the energy they generate. WE Energies contracts with the Point Beach nuclear facility to purchase power that imposes an excess cost of almost $5 billion on ratepayers for the period ending in 2030, which works out to over $3,000 per electricity customer, or $300 per year. According to a recent report, solar currently costs $38-86 per megawatt hour; wind costs $37-86 per megawatt hour, and nuclear costs $141-220 per megawatt hour.
What do we have to show for our tax dollars? Promises of “advanced” and “next generation” reactor designs that are somehow never quite ready for commercial operation after decades of research and promises of safe deep geologic storage for radioactive waste that never materializes.
So how will we power those data centers, then?
Clean Wisconsin has called for a pause on new data center construction until practical questions such as this can be answered. How much energy infrastructure can we afford, both economically and environmentally? How many new power plants, how much water use, and how much air pollution are we willing to trade for the sake of AI and “economic development”?
Nuclear power cannot be built fast enough, or affordably enough, to supply data centers with the power they need now. Newer designs will be even more expensive than the existing reactors.4 So-called Small Modular Reactors (SMRs) have not been proven to be safe or economical at commercial scale (and are not necessarily small).
To be responsible stewards of our air and water, and to guard against sky-high electric rates, Wisconsin must evaluate all the costs and benefits of data centers and how they are powered. Before we invest our hard-earned tax dollars and electric bills in nuclear technology that has proven itself to be wildly overpriced and unable to keep its promises, we need to ask ourselves: for what?
