As the world keeps warming and electricity bills take center stage in national politics, the data center boom will drive up US carbon emissions and electricity costs. But a few simple policies could help bring both emissions and prices back down.
That’s the message of a new analysis from the Union of Concerned Scientists released Wednesday, which models a variety of scenarios for how to fuel the coming AI boom. The US is poised to see a 60 to 80 percent increase in electricity demand through 2050, with data centers alone making up more than half of the increase by the end of this decade, the analysis finds. If policies stay the same as they currently are—with attacks on renewable energy being embedded into regulatory regimes and few significant national policies restricting carbon emissions from power plants—we could see between a 19 and 29 percent increase in CO2 emissions from US power plants tied just to the energy needs of data centers over the next 10 years.
There are answers, though: Bringing back tax credits for wind and solar, which were political targets in last year’s One Big Beautiful Bill, would cut CO2 emissions by more than 30 percent over the next decade, even if data centers eat up a significant chunk of new demand for electricity. They could also make wholesale electricity costs go down by about 4 percent by 2050, after a slight rise over the next decade.
Power plants are the second-largest source of greenhouse gas emissions in the US, making up about a quarter of the country’s overall emissions. Last year, emissions from the US power sector rose slightly, marking the first increase since 2023; commercial buildings like data centers, a separate analysis released last week from the Rhodium Group found, were the main drivers of that demand.
Predicting the amount of energy the US is going to need for AI in the future is an incredibly tricky project. Many of the public estimates we have are provided by utilities that are wrangling a number of requests for new capacity from data centers; data center companies often take their requests to a number of utilities as they shop for the best price, which inflates estimates of overall actual need. Technological advances over the next few years could also make data centers and AI much more energy efficient. Some of the craziest numbers splashed over headlines or trumpeted out by tech and energy executives are probably exaggerations. (Earlier this month, PJM, one of the largest regional transmission organizations in the country, downgraded its projections of how much energy the grid is going to need over the next couple of years after more carefully vetting some data center proposals.) In order to get an accurate read on this, UCS modelers used middle-range electric growth scenarios and assumed that just half the projects publicly announced in the pipeline would actually be built.
But the Trump administration has moved so aggressively against both renewable energy and climate policies in the past year that the analysis likely underestimates how high emissions from data center demand could actually be. While the UCS modeling accounts for some policy changes, including backpedaling on regulations on coal-fired power plants, doing away with renewable tax credits, and delaying some offshore wind projects, it didn’t take others into account. An Interior Department policy that has mandated review of all wind and solar projects on federal lands, for instance, has created a whopping bottleneck of 22 gigawatts of projects—enough to power more than 16 million homes. In December, the administration issued stop work orders for five East Coast wind farms under construction, citing national security concerns. (On Friday, three different judges ruled that construction could proceed.)
“The Trump administration is doing this with projects that have already been approved and are under construction,” says Steve Clemmer, the lead author of the analysis and the director of energy research at UCS. “It sends a chilling signal to the industry, and to efforts to power data centers and meet electricity demand. We need to build as much as we can, as fast as we can.”
The Trump administration has made it extremely clear that it wants fossil fuels—specifically coal, the dirtiest form of energy—to power the AI boom. Energy secretary Chris Wright has ordered at least two coal-fired power plants to stay online past their retirement dates, while the White House and the Environmental Protection Agency have pushed a number of policies that benefit the coal industry.
Despite the administration’s clear preferences for fossil fuels, they’re not always the cheapest or easiest option, especially as demand for gas turbines has skyrocketed and experts project yearslong waits to build new natural gas plants. Some power providers are starting to push back against the administration’s anti-renewables moves. Earlier this month, PJM filed a brief supporting a Virginia offshore wind developer that is suing the Trump administration over blocking the project. In the brief, PJM argued that the wind project, which will provide enough power for more than 600,000 homes, “is important to meet a rapidly increasing demand for electric power.”
Big Tech, meanwhile, made a variety of pledges in recent years to cut emissions and be more climate-friendly; many of these pledges were promptly derailed by the growth of AI. It’s not clear whether these companies are ready to push directly against the Trump administration to advocate for more renewable generation. Last week, with encouragement from the White House, Microsoft rolled out a set of commitments for its data centers to be “better neighbors” to communities where they’re located. There were no mentions of emissions or climate policies in the set of commitments provided by Microsoft, which did not respond to a request for comment.
The UCS estimates do not include estimates from energy generation that happens at the data centers themselves, which is becoming an increasingly popular option for projects that are waiting to connect to the grid. These installations can include renewable generation, usually solar and batteries; different forms of nuclear energy—from restarting entire power plants to investing in small modular reactors—are also popular choices for tech companies. But many of the biggest hyperscalers have also announced plans to build their own onsite gas plants.
The prices modeled in the UCS analysis are just for wholesale electricity—the cost of delivering electrons to the grid—and not necessarily what Americans see on their bills. Consumer electric bills are shaped by a whole host of factors beyond just what utilities pay for the energy itself, including upgrades to equipment and buildouts of transmission lines. Renewable energy, which is intermittent and often located farther from where power needs to be provided, needs more transmission than traditional fossil sources, which means more financial investment.
“What’s so crazy about renewables is [that] both political arguments are true,” says Pier LaFarge, a cofounder at Sparkfund, a utility services company. “They are the cheapest power at the source of generation—but they are also raising rates because of downstream upgrades to the distribution grid.”
Just reintroducing tax credits for wind and solar wouldn’t be enough to stave off the worst impacts of climate change. The UCS study also modeled the costs of policies that would more seriously decarbonize the US grid as demand rises from AI. This includes more stringent power plant regulations and more investment in the transmission upgrades that renewable energy needs. This scenario, the analysis finds, would slightly raise wholesale electricity costs through 2050, by about $412 billion—a 7 percent increase. However, the analysis finds, it would avoid up to $13 trillion in climate costs: damages incurred by floods, wildfires, droughts, and other extreme weather worldwide, as well as the local health costs associated with dirty power plants. (Earlier this month, the EPA announced that it would no longer factor in the costs of lives saved from excess pollution when considering pollution policies around power plants.)
Much of the US grid is in serious need of upgrades, especially if the country becomes serious about moving off fossil fuels. Part of the challenge of the next few years is going to be ensuring that the upgrades the grid needs—with or without more renewables—are not unfairly pushed on to consumers.
“There definitely needs to be much stronger guardrails in place for data centers themselves, as well as for making sure that we have enough electricity capacity and generation in place to power those data centers, and that it doesn’t take away from other customers,” Clemmer says.
Despite the Trump administration’s aggressive attacks on renewables and eye-watering figures for energy demands from AI, there’s some reason to hope. LaForge believes that utilities’ increasing deployment of batteries, coupled with contracts that make data centers pay for infrastructure and other associated costs, will help drive electric rates down for regular consumers. (Unlike credits for wind and solar, tax credits for batteries mostly made it through the One Big Beautiful Bill negotiations.) In this scenario, the US could look more like Texas: tons of cheap wind and solar on the grid, a few gas plants, and installing a lot of batteries.
“The good news is that, just like the Biden administration couldn’t control the fate of the universe, neither can the Trump administration,” he says, pointing out that solar, wind, and storage made up more than 90 percent of new power put on the grid last year. “We’re building more renewables more quickly in more places for purely economic reasons.”
The post The AI Boom Will Increase US Carbon Emissions—but It Doesn’t Have To appeared first on Wired.
