Across northern Virginia, central Ohio, and the outskirts of Phoenix, hyperscale data centers have been rising at a pace that has reshaped local power grids and water supplies. Now a research team says those facilities may be reshaping local temperatures, too. A working paper revised in April 2026 reports that land surrounding large data centers warms by roughly 2 degrees Celsius on average once the servers start running, based on satellite imagery captured before and after construction. The finding has not yet been peer-reviewed, but it draws on some of the most trusted thermal datasets in earth science and arrives at a moment when the U.S. data center footprint is expanding faster than at any point in the industry’s history.
What the satellite data actually show
The paper, titled “The data heat island effect,” compares multi-decade satellite readings of land surface temperature at hyperscale data center sites against readings from the same locations before construction began. The pattern the authors describe is consistent across facilities: cooler baseline temperatures, a transitional phase during site clearing and building, and a higher, relatively stable thermal plateau once operations reach scale.
Two federal data pipelines anchor the analysis. The first is the Landsat surface temperature record, spanning Missions 4 through 9 under Collection 2, which the U.S. Geological Survey maintains and explicitly recommends for urban heat island research. The second is NASA’s MODIS/Terra Land Surface Temperature and Emissivity product (MOD11A2), which delivers global coverage at roughly 1-kilometer resolution in 8-day composites. Both instruments have been validated across thousands of peer-reviewed studies over more than two decades.
A complementary USGS program strengthens the analytical framework. The agency publishes annual Landsat-derived surface temperature products and Surface Urban Heat Island intensity metrics for 71 U.S. metropolitan areas, tracking thermal changes from 1985 to the present. That long baseline is exactly what researchers need to separate a new, localized heat source from the background trend of regional warming. The working paper applies a similar before-and-after satellite comparison, but focuses the lens on data center campuses rather than city cores.
The authors note that the size of the warming signal varies by setting. A facility built on previously forested land registers a different thermal signature than one erected next to an existing industrial park, suggesting that both waste heat from servers and the physical replacement of vegetation with concrete and steel contribute to the observed change. The 2-degree figure is a mean across a diverse set of locations, not a blanket prediction for every future project.
Why the number still carries caveats
A working paper posted to arXiv is a public draft, not a finished product. The statistical methods, site selection criteria, and choice of control areas have not been independently vetted by other climate or remote-sensing scientists. That distinction matters because a single temperature estimate could ripple into zoning fights and infrastructure policy if it is treated as settled fact before replication.
Neither USGS nor NASA has publicly endorsed or challenged the paper’s specific claim. The USGS urban heat island products were designed to monitor thermal dynamics across entire metro regions, not to attribute warming to a single industrial campus. Scaling those methods down introduces questions about spatial resolution and confounding factors: tree removal, grading, and paving during construction all change how a surface absorbs and radiates heat, independent of any waste heat the servers produce.
Satellite instruments add their own noise. Land surface temperature readings shift with cloud cover, sensor viewing angle, and differences in rooftop materials. The researchers describe filtering steps and multi-year averaging, but without independent replication it is hard to know how sensitive the 2-degree estimate is to alternative data-cleaning decisions or different definitions of a facility’s thermal footprint.
There is also a gap between what satellites measure and what people feel. Land surface temperature is not the same as air temperature at street level. A 2-degree rise on a rooftop or parking lot does not automatically translate to the same increase in the ambient heat experienced by someone living a quarter-mile away. Without ground-level air temperature stations or building-energy models at the studied sites, the health and comfort implications for neighboring communities remain an educated inference.
Missing pieces: water, mitigation, and industry response
The working paper focuses tightly on thermal signatures and does not examine water consumption, even though large data centers can draw millions of gallons per day for evaporative cooling. In drought-prone regions like central Arizona or west Texas, the combination of added heat output and heavy water withdrawal could compound environmental stress in ways the current analysis does not capture.
Equally absent is any assessment of how much the observed warming could be reduced through design choices already gaining traction in the industry. Several major operators have begun deploying liquid cooling systems that reject less heat directly into the atmosphere. Meta has piloted waste-heat recovery at European facilities to supply district heating networks. Google has publicly committed to improving cooling efficiency across its global fleet. None of these efforts are evaluated in the paper, which means the 2-degree average may overstate the thermal impact of newer, more efficiently cooled campuses while understating the impact of older, air-cooled ones.
No major data center operator has publicly commented on the paper’s findings as of May 2026. Industry trade groups have generally emphasized the economic benefits of data center investment, including jobs and tax revenue, while pointing to voluntary sustainability commitments. Whether those commitments extend to localized thermal effects, as opposed to carbon emissions and water use, remains an open question.
What communities and regulators can do with this now
For local planning boards weighing new data center proposals, the practical value of the paper is not the specific 2-degree number but the method behind it. The same freely available Landsat and MODIS records can be used to establish a thermal baseline for any proposed site before construction begins. If the warming signal holds up under independent review, regulators would have a quantitative benchmark for requiring heat mitigation measures: reflective roofing, vegetated buffers, waste-heat capture, or setback distances from residential areas.
Some jurisdictions are already moving in that direction. Loudoun County, Virginia, home to the densest concentration of data centers on the planet, has tightened zoning rules around noise and visual screening in recent years. Thermal impact has not yet entered those ordinances, but the political pressure is building as residents in adjacent neighborhoods raise questions about cumulative environmental effects.
The broader significance of the paper is that it widens the environmental conversation around AI infrastructure. For years, the public debate has centered on electricity consumption and carbon emissions. Water use entered the discussion more recently. Localized heat output is the next layer, and it is one that affects the people living closest to these facilities most directly. Whether the final, peer-reviewed numbers land at 2 degrees, 1 degree, or something higher, the core question the researchers have surfaced will not go away: when a data center pumps tens of megawatts of waste heat into the air around the clock, what happens to the neighborhood next door?
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
