Typeost

Tech news in plain type

Latest / Science

ScienceReported by 3 outlets

New study tallies how much US power hyperscale data centers consume

A Nature Sustainability paper estimates hyperscale data centers use about 2% of US electricity, and fossil fuels still supply much of that power.

According to Ars Technica, a new study attempts to tally the electricity use and emissions of hyperscale data centers, the very large facilities that power cloud services and many artificial intelligence tools. The outlet reports that these facilities account for about 2 percent of electricity used in the United States.

The Verge, citing a paper in the journal Nature Sustainability, reports that fossil fuels remain the largest source of electricity for hyperscale data centers in the country. It says the facilities likely consume more power collectively than many nations do, which translates into substantial greenhouse gas emissions. The study focuses on hyperscale sites rather than the full range of smaller server rooms and enterprise facilities.

Wired offers a more local comparison, reporting that roughly 400 US data centers together use more electricity than the state of Michigan. The three outlets frame the same underlying question in different ways: a share of the national total, a comparison with a country, and a comparison with a single state.

The reporting does not fully spell out every methodological detail, including exactly how each facility was identified and how energy use was estimated. Readers should treat the headline figures as the best current tally from one study rather than a settled national accounting.

Background

Hyperscale data centers are the largest category of computing facilities, built and operated by major cloud and technology companies to run enormous banks of servers. They host everything from video streaming and email to the training and running of large AI models. Because these sites are concentrated in certain regions and can draw hundreds of megawatts each, their demand can be significant for local utilities and grid operators.

Electricity used by a data center does not carry a single fixed carbon footprint. The emissions depend on the mix of generation that serves the local grid at a given time, which is why fossil fuels such as natural gas and coal can dominate the picture in some regions even when a company has purchased renewable energy credits elsewhere. Accounting methods differ, and the distinction between location-based and market-based reporting often shapes the headline numbers that companies and researchers publish.

Measuring this demand is hard. Many operators do not disclose facility-level energy data, and public estimates often rely on modeling, utility filings and assumptions about server utilization. Independent researchers have had to piece together estimates from partial information, which makes any national total approximate and easy to contest.

Demand has grown quickly in recent years as cloud computing expanded and AI workloads moved into mainstream products. Building new generation, transmission lines and transformers can take years, so the pace of data center construction and the pace of grid upgrades do not always match. That gap is at the center of many debates about how the sector should be planned and paid for.

Why it matters

The figures matter because they move a technology industry conversation onto physical ground. Software and digital services often seem weightless, but they depend on power plants, substations and cooling water. When a single category of facility accounts for a noticeable share of national electricity, questions about climate targets, grid reliability and consumer electric bills become harder to separate from questions about cloud and AI growth.

The fossil fuel finding is especially significant for companies that have set public climate goals. If the grids serving these sites remain dependent on gas and coal, the gap between a company's stated ambitions and the emissions attributable to its operations may be a central point of scrutiny. Investors, regulators and local communities may all press for clearer data.

For policymakers, the study provides a reference point for planning. Decisions about permitting, transmission investment, power purchase agreements and where new sites should be located all depend on accurate estimates of demand. Better disclosure rules could make those decisions less dependent on guesswork.

The comparison with entire countries and states also shifts the framing of the debate. It invites readers to ask not only whether the sector is efficient, but whether its growth is compatible with the energy system that already exists.

What to watch

Watch for responses from the companies that operate the largest facilities, particularly whether they publish updated or more detailed energy and emissions data. Watch also for independent researchers who may test the study's methods or offer alternative estimates, since national totals of this kind often change once assumptions are revisited.

Regulators and utility commissions may take up the question of how data center load is forecast and who pays for grid upgrades. Local permitting fights over specific projects can offer early signals of where public pressure is building.

Finally, watch the mix of generation in regions with heavy data center growth. Changes in gas, coal and renewable capacity there will shape whether the emissions picture improves or worsens.

Viewpoints

Commentary from our columns: three angles on this story.

The Skeptic open questions and weak points

Every headline number in this story deserves a second look before anyone builds a policy on it. A single study that tallies a sprawling, largely private industry is working with incomplete data, and the estimates depend heavily on which facilities get counted and how their power use is modeled. Two percent of national electricity sounds precise, but the boundaries of "hyperscale" are not always clear, and companies rarely publish the detail needed to check the work. The comparison to entire countries is vivid, yet it can make the sector look more uniform than it is. Some data centers run efficiently and some do not, and the same building can have very different footprints depending on the local grid. The useful question is not whether the number is shocking, but whether the methods hold up when independent researchers try to reproduce them. Until disclosure improves, treat these totals as a starting point for argument rather than a final verdict on the industry's climate impact.

The Builder the engineer's view

From an engineering standpoint, a data center's electricity draw is shaped by decisions made long before the first server boots: cooling design, power conversion losses, how aggressively workloads are scheduled, and how much capacity is kept idle as a safety margin. Those choices are where real efficiency gains come from, and studies like this one are valuable if they push operators to measure more consistently. The fossil fuel finding is less about the building and more about the grid around it, which engineers can influence through siting, flexible workloads that shift to cleaner hours, and on-site or contracted generation. Teams that treat energy as a first-class constraint, alongside latency and cost, tend to make better architectural trade-offs. The hard part is that most software has never been designed to report its own carbon intensity. Better tooling, standardized metrics and honest telemetry would let engineers see the consequences of a deployment in the same dashboard where they see its performance, which is where this conversation needs to go next.

The Rulebook privacy, safety and regulation

Numbers like these tend to arrive before the rules do, and that gap is worth watching closely. When a sector's electricity use is large enough to appear in national totals, questions about disclosure, permitting and cost allocation move from abstract to concrete. Who should be required to report facility-level energy and emissions, and in what format? Should new large loads be made to bear the expense of the grid upgrades they require, rather than spreading that cost across ordinary ratepayers? Local communities may also have a stake in where these facilities are sited and how they affect water and air. Existing frameworks were not written with hyperscale campuses in mind, and regulators will need data that is consistent enough to compare across states and companies. The study does not answer those questions, but it sharpens them. Any approach should protect the public's ability to understand the impact of infrastructure that increasingly sits at the center of daily digital life.

More in Science