“I’ve met a lot of people who say, ‘Oh my God, are you using AI?’ But you’re the climate man! Don’t you know that AI queries have a high carbon footprint?” says Roling, clinical assistant professor at Kellogg and executive director of the Abrams Climate Academy. “It made me think, ‘Am I underestimating the climate impact of these new tools?’
Self-reflection led Roling to explore how the carbon emissions associated with using AI apps compare to those of other everyday activities, from streaming a movie to eating meat. What he found was a familiar pattern in climate and sustainability debates, where individual behaviors are misunderstood compared to the environmental impact of corporate decisions.
As tech companies scramble to build the infrastructure that powers the age of artificial intelligence, the real question lies not in how people use chatbots, but in the policy decisions of state and local regulators. By setting standards for energy efficiency, water use and sustainable siting of hyperscale data centers, regulators can ensure that these massive projects deliver better outcomes for communities and the environment.
The focus, Roling says, should be on guiding overscalers toward responsible consumption of resources, rather than worrying about the marginal impact of someone asking for a prescription online.
“It’s not that AI doesn’t have environmental impacts. But if that’s your concern, AI is not the straw that breaks the camel’s back as it relates to a person’s carbon footprint,” Roling says. “Yet AI is still putting enormous strain on energy systems in the United States. And we’re seeing demand for electricity in a way we haven’t seen in decades.”
Roling describes the energy impact of using AI, from individuals to companies to data centers, and what actions can make a big difference to the technology’s carbon footprint.
Putting the use of artificial intelligence in context
The average American produces about 20 tons of carbon dioxide per year, mostly through transportation and household energy use. A flight between the US can emit more than a quarter ton of CO2 per passenger. a 20-mile round trip in a gas-guzzling car can contribute up to five tons of carbon per year.
In comparison, what you do on your computer generally leaves a much smaller carbon footprint. A typical AI query—when you ask a question on ChatGPT or when Google provides you with an AI overview—produces about five grams of carbon dioxide. That’s ten times higher than a regular Google search — but about 300 times less than an hour’s video stream.
Roling’s favorite point of comparison is the hamburger, which generates between 2 and 3 kilograms of carbon from farm to table—the equivalent of hundreds of AI queries. While creating images or videos requires more energy, these are less routine tasks.
“So for the person who uses AI sporadically over the course of a week to help them with their work or their personal life, it doesn’t even compare to some of the other things we do and it’s not judged as environmentally careless,” Roling says. “All I have to do is eat two fewer hamburgers a year and I’m good.”
Data center energy usage
On the other hand, the infrastructure required to perform these individual AI tasks can have far greater consequences. As companies scramble to build “hyperscale” data centers to power tomorrow’s AI applications, massive, energy-intensive construction projects are creating new environmental pressures.
After 15 years of relatively flat energy demand growth, the US set an all-time high in electricity consumption in 2024 — and is expected to grow rapidly in the coming years. Currently, 4.4 percent of all US energy goes to data centers, and that’s it projected to triple by 2028reaching between 165 and 326 terawatt hours per year. That’s enough to power nearly a quarter of American homes.
While other trends, such as electric cars and manufacturing upgrades, are contributing to this increase in U.S. demand, the big driver is artificial intelligence and the big tech companies behind it, Roling says. This puts these previously “green” leaders in an unfamiliar position.
“The hyperscalers—Microsoft, Google, Meta, Amazon, etc.—are companies that have historically been at the forefront of setting emissions reduction goals,” says Roling. “And now you’re seeing a lot of them go back to those goals — or be a little more cautious about them — because they’re in the middle of this arms race and they need to build as many of these data centers as quickly as possible.”
Transparency around resource usage
As demand growth outstrips grid bottlenecks and policy or permitting delays slow down renewable energy deployment, hyperscalers have had to rely on existing fossil fuel infrastructure to provide the 24-hour reliable power their data centers require. As a result, many of these companies are suddenly investing in natural gas plants and even coal-fired power plants as a short-term solution. Others are looking at nuclear power, although regulatory and technological hurdles remain high.
“On the fringes, you see these companies trying to do innovative things to show off their green credentials,” Roling says. “But because it’s such a race, they’re going to get electricity where they can, as cheaply as they can. And that’s often in communities where it’s coal or natural gas, which is very carbon-intensive and very bad for air pollution and human health.”
Another concern is water, which is used by data centers to cool the heat generated by hundreds of thousands of computer chips. Some estimates say hyperscale data centers require five million gallons of water per day—an amount that’s particularly worrisome given the amount of construction planned in already water-stressed areas like Texas and the Southwest.
A caveat to these figures is that most of them are estimates because, in the hypercompetitive AI boom, companies rarely share information about their resource needs and impacts with regulators or communities. As local resistance to planned data centers intensifies, Roling hopes to see more information released by tech companies about energy and water consumption and how they plan to account for it.
“Disclosure and transparency are paramount, especially because of the speed at which this freight train is moving,” says Roling. “I’m sympathetic to the need to protect trade secrets, but at the same time, our clean air, clean water and atmosphere are common goods that must be protected.”
The impact of corporations
When considering the environmental impact of artificial intelligence, it is also important to recognize that the customer base for many tools is not individuals, but corporations. As more organizations incorporate AI into their operations, they will pay big fees to technology vendors.
That can give those companies some leverage in keeping data center energy requirements in check, Roling says. There is historical precedent in the rise of cloud computing, where companies in various industries suddenly found themselves buying remote server space. At the time, Roling worked with a video game producer that was considering choosing a cloud vendor based on the carbon intensity of its server farms.
“If you’re making a lot of money out the door going into data centers, as a corporate leader, you have an opportunity to work with that vendor and, if necessary, shift your spend to the lowest-carbon provider,” Roling says.
And as electricity bills rise for both companies and households, Roling also sees a potential investment as utilities and technology companies have new price incentives to innovate and integrate cheap renewables.
“My hope is that to address the demand challenges that AI brings, we’ll start to see states and utilities looking at options that are more modular, that are more incremental, like wind, solar, and storage,” Roling says. “There are a lot of technologies that I think we’re going to see emerge as we tackle this problem. Because if there’s one lesson we’ve learned in the last five years, it’s that inflation is not popular. We haven’t felt it in energy in the last five years, but we’re starting to feel it now.”
