(Konstantin K4/Shutterstock)
The story of AI is often told in terms of GPUs, breakthroughs, and billion-parameter models. However, behind every chatbot and AI application lies a more fundamental force: electricity. Without it, even the most advanced AI comes to a grinding halt. At the blinding pace at which AI is scaling, power is becoming both a currency and a constraint.
Faced with rising demand and tightening grid constraints, policymakers and tech leaders are scrambling for solutions that scale. For years, nuclear energy remained on the sidelines, pushed out by political risk and cost volatility. Now, as AI turns power from a background utility into a front-line strategic concern, nuclear power is making a comeback. It’s not surprising, after all, that very few options match nuclear’s combination of energy density and reliability.
To see why nuclear is back on the table, it helps to first look at the sheer scale of energy that AI systems now demand. Just last week, Google shared a technical report on how much energy it takes to process a single AI prompt. It is the most transparent estimate yet from a Big Tech company with a popular AI product. While it reflects just one type of interaction, it offers a clear glimpse of the kind of energy demand we are moving toward as AI becomes increasingly embedded in daily life.
According to Google, processing a single median Gemini prompt requires 0.24 watt-hours of electricity. This figure reflects not only the power consumed by the AI chips themselves, but also by associated components such as CPUs, memory, backup systems, and cooling.
(Shutterstock AI Generator)
Google reports that its AI models are becoming more efficient. That may be true on a per-query basis, but it does not change the bigger picture. As AI use spreads across nearly every industry, the total number of queries is skyrocketing. The energy footprint is growing faster than any efficiency gains can offset.
Multiply that single prompt number from Google across billions of queries a day, and you start to see why energy planners are sounding alarms and why AI companies are saying that power is the real bottleneck to growth.
A recent analysis from Goldman Sachs estimates that global data center power demand could rise by as much as 165% by 2030. McKinsey predicts consumption in the U.S. could reach 400 terawatt-hours a year, up from fewer than 100 terawatt-hours in 2020. That would exceed the current electricity demand of an entire country like Mexico.
Tech companies are placing long bets on nuclear, but energy demand from AI is rising on a much shorter timeline. Many firms will need reliable, large-scale power within the next three to five years. Conventional nuclear plants can take close to a decade to build, and even next-generation options, such as small modular reactors (SMRs), are still years away from their first commercial deployments for AI use.
A few SMRs are already operating overseas. Russia’s floating Akademik Lomonosov has been generating power since 2020, and China’s Linglong‑One went live in late 2024. Canada recently approved construction of a next-gen BWRX‑300 at Darlington, with hopes of switching it on by 2030. However, none of these reactors are powering AI infrastructure in the U.S. For now, SMRs remain more of a long-term bet than a near-term solution.
In reality, most AI companies will still rely on fossil fuels for years. They are keeping coal plants running and building new gas ones. Building nuclear facilities might be the long-term fix, but it is not ready yet. In the meantime, tech giants are turning to existing nuclear assets, signing deals with existing plants and backing restarts to get clean power sooner.
(Ole.CNX/Shutterstock)
Over the past year, Microsoft, Google, Amazon, and Meta have all announced nuclear energy initiatives. Some are straightforward agreements to purchase power from existing plants, while others are focused on SMRs and other advanced designs that remain in development.
Microsoft has already moved ahead with a 20-year agreement with Constellation Energy to draw power from the planned restart of the Three Mile Island Unit 1 reactor. Amazon recently signed a contract to purchase 1.9 gigawatts of nuclear-generated electricity through 2042 from Talen Energy’s Susquehanna facility, a scale that makes it one of the largest corporate nuclear deals on record.
Meta has also turned to nuclear, securing a long-term contract to source nuclear power from the Clinton Clean Energy Center in Illinois beginning in 2027, securing a steady supply of carbon-free power for its data centers.
Google recently signed a deal with Kairos Power to build SMRs to meet its energy needs. Michael Terrell, Google’s senior director of energy and climate, framed the deal as part of a broader push to secure reliable and clean power for the company’s data centers. “We feel nuclear can play an important role in helping us to meet our demand, and helping us to meet our demand cleanly and round the clock,” he said in an interview with the Financial Times.
Some companies are going even further, targeting old nuclear plants that have been shut down. The Palisades reactor in Michigan, for example, is on track to restart in 2025 with help from a $1.5 billion federal loan. In Iowa, the closed Duane Arnold plant may be revived with a new SMR hybrid project.
Others are working to stretch the life and output of the plants still running. By 2035, 24 U.S. reactors will be up for license renewal. Each one could get a 20-year extension. The Department of Energy also sees room for capacity upgrades at current plants, estimating that between 2 and 8 GW of extra power could be added just by boosting efficiency. These stopgap arrangements may not be headline-grabbing, but they are some of the few nuclear options available in the next few years.
Another challenge with nuclear power is less technical and more rooted in the past and reputation. Old disasters, waste fears, and weapon associations still cloud public understanding. That hesitation can slow approvals, delay investments, and make even safe projects politically risky.
“I am really excited about nuclear, but this is a technology that has a lot of myths and misinformation around it,” said Anna Erickson, Woodruff Professor in the George W. Woodruff School of Mechanical Engineering (ME), and leader of the Consortium for Enabling Technologies and Innovation (ETI), which is focused on nuclear technology.
“Concerns about nuclear weapons, accidents, and waste have overshadowed nuclear energy’s potential as a clean, carbon-free technology,” she added.
However, the AI power needs are not going anywhere. If AI is going to keep growing, the grid needs to grow with it. One way or another, tech companies and goverments will have to solve the power problem, and right now, nothing else offers the same mix of reliability and scale as nuclear power.