An estimated $580 billion will be spent on AI data centers in 2025, creating a demand for energy that is quickly outpacing America’s existing infrastructure. The average data center uses as much electricity as 100,000 homes. And some of the largest facilities under construction today will use 20 times that, according to the International Energy Agency. All told, data centers consumed more than 4 percent of the United States’ electricity in 2023, and a study released by the Department of Energy projected the AI boom could push that figure to 12 percent by 2028.
All of this energy consumption is not only bad for the planet, but also the areas surrounding the data centers themselves. In many regions, utility companies are scrambling to expand their power plants, upgrade their transmission lines and shore up the grid — and those costs are being passed down to their customers. Households near major data center hubs in states like Ohio and Virginia have reportedly seen their electricity bills skyrocket.
That financial strain — combined with grid stress and environmental concerns — is prompting major political pushback. In fact, a group of environmental organizations recently called for a moratorium on new data center construction, but no action has been taken yet. In the meantime, tech companies and data centers operators are no longer relying solely on electricity from the grid — and have begun hunting for the fastest, cheapest, cleanest energy sources available to them.
How Are AI Data Centers Powered?
Data centers in the U.S. mostly rely on natural gas, followed by renewables, nuclear and coal. To meet AI’s energy demands and their green energy goals, companies are seeking new forms of electricity. They’re reopening shuttered nuclear plants, installing solar projects and investing in small modular reactors. They’re also experimenting with clean-energy alternatives like geothermal, hydrogen and concentrated solar power.
More than 40 percent of today’s data centers are powered by natural gas, according to the IEA report. Solar and other renewable sources deliver 24 percent of the power supply, while nuclear contributes 20 percent and coal produces 15 percent. This energy mix is likely to change, though, as tech companies continue to seek energy sources that can meet their needs while also fulfilling their carbon emissions targets. By 2030, Microsoft has pledged to be carbon negative, and Google aims to reach net-zero emissions. Amazon has committed to reach net-zero emissions by 2040.
Solar and nuclear energy have emerged as popular solutions, but each have their limitations.
Solar and wind facilities can be built relatively quickly and cheaply, but they only produce power when the sun’s shining and the wind’s blowing — which can pose issues for AI data centers that run around the clock. Nuclear power plants are emissions-free and provide a constant source of energy, but these facilities probably won’t be ready until the 2030s. To meet their short-term needs, some tech companies have been seeking out less conventional sources of clean energy, like hydrogen, geothermal and using the sun to heat up rocks in a shipping container.
Below, we take a look at the wide variety of energy solutions being explored to power America’s growing supply of AI data centers.
Nuclear Fission
Nuclear power plants have been on the decline over the past 30 years, but that tide has turned since the start of the AI boom. Most of the major AI players have partnered with nuclear energy companies, and President Donald Trump has signed executive orders to boost nuclear energy, with a goal of increasing the United States’ capacity from 100 gigawatts to 400 gigawatts by 2050.
In perhaps the most symbolic example of nuclear energy’s renewed popularity, Three Mile Island — where a partial meltdown of one of the reactors in 1979 caused radioactive materials to be released into the environment — may reopen in 2028 to power Microsoft’s growing number of data centers. The reactor that experienced the meltdown has long been shuttered, but a second reactor was closed down in 2019 due to lack of demand. That reactor will undergo a $1.6 billion restoration with help from a $1 billion government loan.
Google, meanwhile, has partnered with NextEra Energy to reopen the Duane Arnold nuclear plant in Iowa. That plant, which closed in 2020, is slated to be reopened by the end of 2028. Amazon plans to power its Pennsylvania data center with nuclear power from the nearby Susquehanna power plant through 2042, and Meta signed a 20-year purchase agreement with an Illinois nuclear plant that was facing potential closure.
Building a new nuclear plant can be time consuming, though. The last nuclear reactor launched in the United States took more than 14 years to build and cost more than double what was budgeted. As a result, AI companies are also looking into small modular reactors, which are cheaper, easier to build and able to be located on-site.
There are only two small modular reactors in the world — neither of them in the U.S. — but several startups have risen to the challenge. Google has purchased 500 megawatts of electricity from SMR startup Kairos, and Amazon has invested in SMR startup X-Energy. Switch, one of the top data center operators, has agreed to buy 12 megawatts of power from Oklo, a SMR startup backed by OpenAI CEO Sam Altman. Microsoft cofounder Bill Gates launched SMR startup TerraPower, which has partnered with Sabey Data Centers. These small modular reactors are expected to start coming online in 2030.
Natural Gas
Natural gas is the dominant power source in the U.S., powering more than 40 percent of data centers. The IEA report anticipates natural gas will play the largest role in meeting the energy demands of data centers in the United States.
Some of the most high-profile facilities in the country run on natural gas. For example, xAI’s Colossus in Memphis is fueled by 35 gas-powered turbines, which has sparked environmental concerns due to a history of air pollution and higher cancer rates in the area. And a Stargate data center in Texas will tap into more than a gigawatt of electricity generated by natural gas from the nearby Permian Basin, an area rich with oil and gas.
Coal
Coal powers 30 percent of data centers worldwide, but only 15 percent of those facilities are in the U.S., according to the IEA report. Coal-burning plants emit mercury, sulfur dioxide and other harmful pollutants, so efforts have been made to transition away from it. As a result, more than 500 coal plants were decommissioned during the 2010s. Trump has promoted coal as a potential energy source for data centers, and his Energy Department has ordered several dozen coal plans to delay closure to meet the United States’ energy needs. Coal is unlikely to be adopted by major tech companies with ambitious emissions targets, though, especially with the promise of new nuclear reactors coming online in 2030.
Solar
Solar power has become incredibly popular in the rush to find data center energy solutions. Since the beginning of the year, the world’s top AI companies have signed at least a dozen solar deals that would each add more than 100 megawatts of capacity.
Solar projects are relatively cheap and can be installed faster than other energy infrastructure — less than two years, according to one study. But solar power, like wind power, is limited by intermittency — they don’t generate power when the sun’s not shining or the wind’s not blowing — so battery storage is critical to provide a continuous stream of power to AI data centers, which run 24 hours a day.
Wind
Like solar, wind projects can be built relatively cheaply and quickly, but its adoption has been hindered by intermittency. Microsoft’s data center in Cheyenne, Wyoming is powered entirely by wind energy, and Google has purchased wind energy for its data center operations in Belgium. Still, wind projects have fallen out of favor in the U.S. under Trump, who has ended tax credits for wind and solar energy projects and issued stop work orders on wind projects that were nearly completed.
Nuclear Fusion
Unlike nuclear fission, which creates energy by splitting atoms, fusion creates energy by fusing them together. The technology is still in its experimental phase, but numerous startups are working to crack the code. While the sector is still in early innings, Microsoft has already agreed to purchase at least 50 megawatts of power from fusion startup Helion Energy. The company, which has raised more than $1 billion so far — including at least $375 million from Sam Altman — plans to open its first fusion plant in 2028.
The development of fusion may be aided through the use of artificial intelligence under President Donald Trump’s Genesis Mission, which aims to create a platform that could tap into the government’s scientific data, build scientific foundation models and accelerate scientific breakthroughs. The Energy Department has been tasked with identifying potential research topics for the mission, and nuclear fusion has been floated as a potential area of interest.
Concentrated Solar Power
Concentrated solar power, sometimes known as thermal solar solar power, uses a lens or mirror to reflect sunlight onto a receiver that is filled with a substance that holds heat.
This type of energy is not typically used in data centers, but one startup, Exowatt, believes its variation on concentrated solar power could deliver a cheap, consistent stream of power that would be a game-changer for data centers. Exowatt uses a lens to focus sunlight into a beam that heats a special brick inside of a shipping container, where a fan carries the heat to another box containing a Stirling engine, which converts the heat into mechanical energy, and a generator. The company told TechCrunch that each thermal battery can hold the heat for five days, and that its energy cost could drop to one cent per kilowatt-hour if production scales to 1 million units per year.
Hydrogen
Hydrogen has not been widely used to power data centers, as the technology is still being developed for commercial use. Its cleanest form, green hydrogen, uses solar or wind energy to split water molecules — but this is a small fraction of hydrogen energy.
Microsoft, which has expressed enthusiasm about hydrogen, launched a green hydrogen pilot project last year to power a data center building in Dublin. Most hydrogen projects are classified as grey hydrogen, which generates steam by reforming natural gas. But several companies, most notably ECL, are experimenting with blue hydrogen, which decarbonizes hydrogen production through carbon capture. ECL powers its Silicon Valley data center with 1 megawatt of blue hydrogen through fuel cells that combine hydrogen and oxygen to create heat and water vapor, which are then converted into electricity.
Geothermal
Geothermal energy is a form of renewable energy that taps into deep wells of hot water below the earth’s surface, releasing steam that can then power a turbine and generate electricity. Less than half a percent of the United States’ electricity comes from geothermal, according to federal data, but a Department of Energy report projected it could generate more than 16 percent of the country’s electricity by 2050.
Some tech companies have already adopted geothermal energy to power its data centers. Google, for instance, partnered with geothermal startup Fervo to power the grid that feeds its Nevada data centers. And Meta has agreed to purchase 150 megawatts of geothermal energy from Sage Geosystems starting in 2027.
Frequently Asked Questions
Why do AI data centers use so much energy?
Data centers require massive computational power to train and run AI models, processing enormous datasets simultaneously. Unlike traditional data centers, AI facilities must maintain intensive processing 24/7, with servers running complex calculations that generate significant heat and require extensive cooling systems.
Are AI data centers contributing to climate change?
Currently, many data centers rely on fossil fuels like natural gas and coal, contributing to emissions. To meet their carbon emissions goals, major tech companies are driving investment in clean energy solutions like nuclear, solar and geothermal energy sources.
Can renewable energy alone power AI data centers?
Solar and wind face only generate power when conditions are favorable, so they require battery storage or backup power sources to provide AI data centers with power 24 hours a day. That’s why many companies are pursuing a hybrid approach that combines renewables with other energy sources.
