The adoption of artificial intelligence is not slowing down. In fact, researchers at Epoch AI found that AI training compute has been quadrupling annually over the last several years. Even if models become smaller or more efficient, the sheer volume of compute required to train them is likely to cause a major power crunch as more and more data centers go up and demand more and more energy.
Ramping up quickly has its shortcomings. As tech companies’ energy needs keep growing, power providers have tried to keep up by ramping up whatever energy source is readily available, including coal plants that release significant amounts of carbon emissions into the atmosphere.
The solution of relying on dirty-burning fuel sources is bad for the planet and bad for the reputation of tech firms that have made major pledges to cut their emissions to zero in the coming years. As a result, companies are increasingly looking for ways to continue to support AI development while meeting their sustainability goals. And there's one solution in particular that seems to be attracting a lot of buzz: nuclear energy.
Nuclear power generation has been effectively stagnant in the United States for the better part of the last three decades, largely as a result of growing costs and safety concerns following several high-profile disasters. But as the need for energy grows, popular clean energy options present a challenge: they don’t always provide the capacity needed to meet demand.
Nuclear power, by contrast, can be more stable and predictable. “Nuclear facilities are capable of generating power over 90% of the time,” says Maksim Sonin, Hydrogen Projects Fellow at Stanford University’s Precourt Institute for Energy. And for data centers in particular, which are typically designed to be always on and in operation, sustained access to power is important, Sonin says.
Of course, getting nuclear reactors online takes time. On average, it takes between six to eight years to get a reactor up and running, according to an analysis from University of Oxford researcher Hannah Ritchie. That lead time is the result of “a very prolonged approval cycle, extended schedules to bring online, and high capital expenditure,” Sonin says.
In recent years, though, small modular reactors (SMRs) have become more feasible thanks to development in the space and efforts to unwind some regulatory red tape. SMRs generate about one-third the power of a traditional nuclear reactor, though the upstart cost and buildout time are considerably less. For that reason, companies like Google, Amazon, and Microsoft are eyeing SMRs as an option for clean energy generation that could ramp up faster than traditional options.
That doesn’t mean that SMRs are without their challenges. “Social, safety and other matters can pose roadblocks at any stage of these technologies' commercialization,” Sonin says. Because of those potential concerns, Sonin believes that SMRs will likely only contribute around 10% of nuclear energy growth in the coming years.
Even as AI development drives renewed interest in nuclear power, it’s possible that AI itself could help remedy the problem of its own increased energy needs by accelerating the development and efficiency of clean energy solutions.
“While rapid AI development requires more and more energy capacity to become available, it can also help us to disrupt the energy sector with more technologies to generate this power in a faster and more sustainable way,” says Sonin.
According to IBM's 2024 State of Sustainability Readiness report, renewable energy use and total energy consumption top the list of key performance indicators business leaders look at when assessing sustainability goals. And AI may be able to point to potential solutions, from helping in the development of new energy generation technology to improved optimization and load management across grids.
“Strategically, I very much positively view the pace of such innovation in tech, despite it raising infrastructure-related concerns in the short term, as it is something we can deal with eventually,” Sonin says.
While attention on nuclear power solutions is growing, it isn’t the only option available. In the US, clean energy like wind and solar were the fastest-growing sources of electricity in the first half of 2024. Hydropower growth continues as well, though it has faced challenges, such as drought and water scarcity, that could cause companies to view it as less reliable going forward.
Sonin says that when it comes to investing in energy resources, most companies will avoid putting all of their eggs in one basket. So while nuclear power is having a moment, it won’t be the only option companies explore.
“This is the right pragmatic approach, I believe, in terms of long-term viability, as there are so many variables, unknowns and different scenarios,” Sonin says. “It absolutely doesn’t imply that one particular type of clean energy is viewed as being better than others.”
Sonin stresses that nuclear plants and other energy solutions aren’t mutually exclusive, but rather components of a broader, comprehensive solution in support of a clean energy transition.
“I believe nuclear power is not purely competing with hydro, geothermal, solar, wind, hydrogen, ammonia and other clean energy sources, as all of them are complementary,” he says. “They all work best in different situations. Therefore, none of them [is likely to become] the sole winner for the big tech corporations, with a portfolio approach generating more value.”