Large data centers in the US already consume as much energy as the whole of Chile
U.S. hyperscale data centers used an estimated 68 to 99 terawatt-hours of electricity. That is roughly 1.8% of total U.S. electricity consumption, a striking share for one kind of facility. The range reflects uncertainty because operators do not publish complete, transparent figures. The comparison with Chile makes the scale easier to grasp. Chile’s annual electricity use is 83.2 TWh, according to the U.S. Energy Information Administration. The study’s highest estimate for U.S. hyperscalers exceeds that amount; its lower estimate does not. The researchers also say the total approaches the electricity use of states such as Indiana, Michigan, or Tennessee. These are estimates for the largest data centers, not a complete count of every U.S. facility. Researchers mapped 403 hyperscalers using observations made between May 2024 and April 2025, and say the count is an undercount. New construction and expansion since then could mean actual use is already higher.
How much electricity do U.S. hyperscale data centers use, and how does that compare with Chile’s annual use?
U.S. hyperscale data centers used an estimated 68 to 99 terawatt-hours of electricity. That is roughly 1.8% of total U.S. electricity consumption, a striking share for one kind of facility. The range reflects uncertainty because operators do not publish complete, transparent figures.
The comparison with Chile makes the scale easier to grasp. Chile’s annual electricity use is 83.2 TWh, according to the U.S. Energy Information Administration. The study’s highest estimate for U.S. hyperscalers exceeds that amount; its lower estimate does not. The researchers also say the total approaches the electricity use of states such as Indiana, Michigan, or Tennessee.
These are estimates for the largest data centers, not a complete count of every U.S. facility. Researchers mapped 403 hyperscalers using observations made between May 2024 and April 2025, and say the count is an undercount. New construction and expansion since then could mean actual use is already higher.
What is a hyperscale data center, and what happens inside one?
A hyperscaler is one of the largest data centers. These facilities are a backbone of digital services because they process, store, and distribute vast quantities of data. Their scale matters: they need large amounts of electricity not only to run computing equipment but also to keep it cool.
Inside, tightly packed processors handle data used by online services and artificial intelligence. A facility may support AI training or inference, data hosting, or other cloud services. Those activities do not all use the same amount of energy: the article notes that AI-focused computing can consume far more than hosting data. A smaller AI center can therefore use more power than a larger center doing other work.
The study focused on U.S. hyperscalers, rather than every type of data center. Researchers identified 403 from satellite observations and other information, estimating an average installed capacity of 40 megawatts and an average footprint of about 2,000 square meters. They warn that this count is incomplete and that the AI boom is driving continued construction and expansion.
How did researchers estimate data-center energy use when companies do not publish complete data?
The researchers faced a basic information gap: companies do not disclose complete data-center energy use, and public figures are aggregated rather than facility by facility. There is also no public register showing every U.S. hyperscaler and its location. Without those details, a direct count and energy tally are difficult.
The team first sought information from operators, working company by company and state by state. Then they used satellite images to spot buildings or plots with the distinctive appearance of hyperscalers, including identifiable cooling systems and proximity to energy infrastructure. They estimated each facility’s installed capacity from the area it occupied, then estimated electricity use based on its size.
This approach produced a map of 403 U.S. hyperscalers, but the authors call it an undercount. They also modeled low, moderate, and high consumption scenarios because operators rarely reveal what each center does. For emissions, the researchers considered the energy mix in each locality. Their figures are therefore estimates, and the article notes that new and expanded facilities could already have raised the totals.
Why has electricity demand from data centers started rising again, and what role does AI play?
The study describes a change in the pattern of data-center electricity demand. Use rose substantially in the early 2000s, then stayed steady through the 2010s even as the industry grew. The researchers point to better hardware efficiency and cooling systems as possible reasons that growth in facilities did not bring a matching rise in energy use.
Demand has grown again in recent years because of the computing needed to support commercial generative AI models. AI training and inference can require much more energy than data hosting or other cloud services. As a result, a smaller center dedicated to AI can consume more electricity than a larger facility doing different work.
The article says the recent AI boom has sparked a construction frenzy, and the study’s mapped facilities were observed only through April 2025. Since then, new centers have been built and existing campuses expanded, while demand for AI computing continues to grow. The researchers therefore warn that their energy and emissions estimates may already be considerably higher.
How does the mix of electricity sources in a region affect a data center’s carbon emissions?
A data center uses electricity, and producing that electricity can release different amounts of carbon dioxide depending on the energy sources in the local mix. The article says the researchers accounted for the mix used in each locality when estimating emissions. It does not provide regional source shares in the text provided here.
As general background, coal and natural gas release carbon dioxide when burned to generate electricity. Electricity from sources that do not burn fossil fuels generally has lower direct carbon emissions during generation. So, all else being equal, a facility supplied by a more fossil-fuel-heavy mix would be associated with more power-sector emissions than one supplied by a less carbon-intensive mix. Actual estimates also depend on how much electricity the facility uses.
The researchers estimated hyperscaler emissions at about 25.7 million metric tons of CO₂. They say this is two to three times higher than a previous estimate for 2018. Because their facility count and consumption figures are conservative, and because electricity sources vary by locality, the figure should be read as an estimate rather than a precise inventory.
Why are so many of the detected data centers concentrated in Iowa, Ohio, Oregon, and Virginia?
The researchers found that four states—Iowa, Ohio, Oregon, and Virginia—accounted for more than half of the energy use they detected. This is an important feature of the estimate because data-center electricity demand is not evenly spread across the country. The article does not give a reason for the concentration or explain why these particular states have so many facilities.
The finding came from the team’s effort to locate hyperscalers despite the lack of a public register. Researchers gathered information from operators where possible and examined satellite images for the distinctive buildings and cooling systems associated with these sites. They then estimated facility capacity based on the area occupied and calculated energy use from size, using different consumption scenarios.
The result should be treated as a detected pattern, not a complete map of all U.S. data centers. The researchers located 403 hyperscalers from observations made between May 2024 and April 2025, and describe that number as an undercount. The article says new facilities and campus expansions since then may have raised both electricity use and emissions, but it does not update the state breakdown.
How does burning coal or natural gas to generate electricity produce carbon dioxide?
Burning coal or natural gas is a chemical reaction called combustion. Coal contains carbon, while natural gas is made mostly of methane, which also contains carbon. When these fuels burn in oxygen, the carbon atoms combine with oxygen atoms to form carbon dioxide, or CO₂. This explanation is established scientific background, not a mechanism described in the article.
In electricity generation, the fuel is burned to release energy that can be used to produce electricity. If a power plant burns a carbon-containing fuel, the resulting carbon dioxide is released as part of the generation process. The amount associated with a data center therefore depends in part on how much electricity it uses and the sources supplying that electricity.
The researchers estimated emissions for U.S. hyperscalers at about 25.7 million metric tons of CO₂, taking account of the energy mix in each locality. The article says that figure is two to three times higher than a previous estimate for 2018. It does not break down the reported total by fuel source or describe combustion chemistry.
This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.
Read more in the JupiteX app
Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.
Or read more news on the web