With AI, How Are We Going to Keep the Lights On?
The growth of generative AI usage is astounding. Researchers from Harvard have found that roughly 40% of the US population between 18 and 64 have used generative AI. The report notes that “adoption of generative AI has been faster than adoption of the personal computer and the internet.” And there isn’t good reason to think that this rapid growth is going to slow down anytime soon. In August 2024 — less than two years since it was unleashed on the public — OpenAI announced that ChatGPT has 200 million active weekly users. That’s twice as many active weekly users as in November 2023.
According to the Electric Power Research Institute (EPRI), ChatGPT uses about 0.0029 kWh of electricity to answer each of your questions — about 10 times more electricity than what’s required to power your typical Google search. It doesn’t seem like much. But at scale, experts at BestBrokers estimate that ChatGPT’s energy consumption could be upwards of 450 million kWh per year – that’s enough energy to power over 40,000 homes for a year in the US.
And that’s just AI usage. We also have to account for other power-intensive industries that are taking hold. Take electric vehicles (EVs), for example. The number of EVs on the road has nearly tripled since 2021. But as costs come down, battery ranges are extended, the charging network is expanded, and regulators mandate a higher percentage of EV sales, adoption will only continue to increase. The number of EVs on the road is expected to eclipse the 30 million mark by the end of the decade.
And what about powering the nation’s fleet of EVs? The folks at BestBrokers estimate that this will require ~240 million kWh to charge each car once — or about 50% of the annual AI power requirements (since EV owners are charging their cars dozens of times a year, the annual requirements to keep EVs running are astronomical. Oh, and this is just the US, by the way).
We also have to consider the increasing number of data centers — not just the ones that are dedicated to AI — that will be required to support the ever-growing deluge of apps and cloud services coming to market. According to Citi Research, the data center market size will grow from 33 gigawatts (GW) in 2023 to 100 GW in 2030 at a CAGR of 17% (GW is used to measure demand because it is a proxy for how many servers a data center can hold, according to analysts at McKinsey & Co.).
Tally it all up, and you get an exponentially larger number than what we can currently produce and distribute. Our grid can barely keep up with electrical needs now. Can we generate enough power to keep up with demand? If we can scale our grid to keep up with demand, then how will it impact the environment? Is it possible to increase our capacity without pumping more carbon into the atmosphere?
The nuclear option
When I was at the Executive Connection Summit in January of this year in Scottsdale, AZ, I attended a really interesting session led by Kirk Offel, CEO of Overwatch (he’ll be speaking at the 2025 event as well). The session was titled “The Fifth Industrial Revolution” and Offel discussed a solution that has been around for nearly three-quarters of a century: nuclear energy. He isn’t alone in thinking that nuclear power is a viable option to meet our energy needs. Tech behemoths are looking to nuclear energy to power their operations. Earlier this year, Amazon acquired a nuclear powered data center and last month Microsoft announced plans to reopen the infamous nuclear power plant on 3 Mile Island to power one of its data centers. Not to be outdone, Google has been toying with the idea of using small modular reactors (SMRs) to provide the 1GW of electricity required to power their new data centers.
Despite its ability to help energy intensive industries to meet their energy needs without increasing carbon output, there are many who oppose nuclear power for environmental reasons. In their view, any non-zero chance of an environmental disaster isn’t worth the risk. In the logic of Fight Club, “on a long enough timeline, the survival rate for everyone drops to zero” — indicating that Murphy’s Law is bound to strike. The nuclear disasters at Chernobyl and Fukushima happened, even after the proponents for those projects promised they would not.
With that said, Offel, who operated nuclear subs while in the Navy and remained in the industry after his Navy career, said that the Navy has never lost a nuclear sub. When you rule out natural disaster, Chernobyl is the only major disaster due to human error (smaller disasters have occurred, but none of them have left the environment uninhabitable for 20,000 years like Chernobyl).
As with many other obvious solutions to society’s problems, economic forces have also inhibited the proliferation of nuclear power. Even though nuclear power plants are more cost effective once you get them off the ground, it is extremely costly to get them off the ground. For example, a recent project to construct a pair of nuclear reactors in Georgia cost $35 billion. Add to that, and power providers prefer the flexibility that comes with fossil-powered plants which can scale up and down quickly to keep up with fluctuating demand for electricity, and you can see why nuclear is currently responsible for only 30% of the power we generate (even though it’s capable of producing much, much more).
As we look to the future, it’s clear that meeting the energy demands of our AI-driven, electrified, and increasingly digital world will require rethinking our energy sources and infrastructure. While nuclear energy presents a compelling, low-carbon solution with the potential to support our power-intensive industries, it faces social, economic, and logistical hurdles. No matter where you stand in the debate, one thing is certain: We are going to need to produce and distribute a lot more power if we want to keep up with demand. Nuclear power could offer one of the best solutions to ensuring your ChatGPT query on when the Yankees last won the World Series is answered (spoiler alert – NOT in 2024).
Patricia Ames
Patricia is president and senior analyst for BPO Media, which publishes The Imaging Channel and Workflow magazines. As a market analyst and industry consultant, Ames has worked for prominent consulting firms including KPMG and has more than 15 years experience in the imaging industry covering technology and business sectors. Ames has lived and worked in the United States, Southeast Asia and Europe and enjoys being a part of a global industry and community.