In Pando, a city of 30,000 located 30 kilometers from Montevideo, Uruguay’s capital, a data center has been in operation since 2016. It belongs to Antel, the state-owned telecommunications company. Amenaza Roboto analyzed 25 years of satellite imagery of the area and found something that until now had never been measured before in Uruguay. The installation generates its own heat island: a space that maintains a higher temperature than the surrounding area.

Unlike a regular building, thousands of servers inside a data center are converting electricity into heat, which a cooling system expels into the environment. The resulting heat signature is visible from a satellite orbiting the planet over 700 kilometers above Earth.

This data center has a maximum capacity of 12 megawatts, a fifth of what Google is building 11 kilometers away.

Amenaza Roboto used satellite imagery from Landsat, operated by NASA and the United States Geological Survey (USGS), which has recorded data on the surface temperature of the Earth since 2000. The satellite’s thermal sensor has a resolution of 100 meters, which with standard processing can be refined to a 30 meter resolution: enough to distinguish a building from the street surrounding it.

The satellite measures the temperature of surfaces: the ground, roofs, pavement. It allows us to know how much hotter it is on one surface than another. The temperature that it detects is not what a person would feel, but it does help us locate which areas are producing heat.

The temperature of an area seen from space depends on multiple factors like vegetation, the type of ground, pavement or the time of year. If the land where the data center was built lost vegetation and gained asphalt, its temperature would go up not because of the center’s activity, but because of its development. To separate one effect from the other, Amenaza Roboto built a statistical model that considers all of these variables and compares the same piece of land in two periods: before the data center existed (from 2000 to 2013) and after operations began (2017 to 2025). The years of construction were excluded to ensure the results were not distorted.

To rule out any sort of temperature rise due to general factors like climate change, the urban growth of Pando, or natural variations, the model accounts for any changes experienced by two nearby industrial buildings: a post office depot and a textile warehouse, each with neither processing machinery nor industrial cooling.

In total, the analysis is based on over 32,000 temperature measurements taken via satellite across 25 years.

Inside Antel’s data center there are rows of servers running without interruption. Each server uses electricity to process information, and this process generates heat as an inevitable byproduct. It’s not an inefficiency of the design, but a consequence of thermodynamics.

To remove heat from the building, cold water flows through the rooms in a closed circuit, absorbs the heat given off by the servers and returns warm to the chillers, industrial machines that cool the water and return it to the circuit. The retained heat leaves the building through condensers, rows of fans that push it to the outside air. This is the main cooling system in the three server rooms currently operating.
In addition to this, four large Kyoto wheels spin in one of the rooms. These heat exchangers use outside air to cool the air inside without mixing the two airflows. When the outside temperature allows it, the system activates. Up to three wheels operate at a time, with a fourth always on standby. If they aren’t enough to cool the space, the chillers finish the job. The industry calls this “free cooling” or natural cooling: It doesn’t replace the main system, but complements it while saving energy—the biggest expense for a data center. Amenaza Roboto inspected this equipment during a visit to the facility.

The heat generated by the servers must be vented out of the building. It is released through cooling equipment installed outside: the chillers’ condensers, the Kyoto wheels and each point where the system extracts heat from the inside of the building into the environment. No cooling system eliminates heat: It merely moves it outside.

The statistical model isolated a part of the heat that only appears when the servers are turned on. That kind of emission would likely leave a physical mark on the area. To find it, we compared the data center pattern with the nearby reference buildings. Any differences would be the signature of the condensers and the Kyoto wheels.

Normally, the heat of a roof comes from the sun, and that is where it remains. The temperature is at its height on the building itself, and it decreases with distance. That is what occurs on the control warehouses: a gradual, steady decline, with a drop in temperature between the building and the first 150 meters and another similar drop up to 300 meters away.

All of the aforementioned corresponds to a data center with a 12 megawatt capacity which, during the period analyzed, never worked at full capacity: During our visit to the facility, Amenaza Roboto was able to confirm that one of the four server rooms was empty.

Eleven kilometers away, in Ciudad de la Costa—within the Montevideo metropolitan area—Google is finalizing the construction of the Teros Project, its hyperscale data center, the largest category of data centers in the industry. At full capacity, it will consume 560 gigawatt-hours per year, according to the project description in the environmental impact study released by the Uruguayan Environment Ministry. This is equivalent to the electric consumption of around 200,000 homes, more than any city in the country except Montevideo. Its maximum energy consumption implies a median potency of around 64 megawatts, five times that of the Antel data center.

Google already operates a data center in Chile, the first built by the company in South America. The Teros Project will be their second in the region and the first in Uruguay.

According to the environmental impact report, none of the studies presented by Google address the heat island effect. Uruguayan legislation defines environmental impact as any alteration caused by “matter or energy,” which in principle applies to heat. But no regulation requires it to be addressed, and the environmental authorities never required it either. Reviewing the file, it became evident that the heating effect of a data center on its surroundings had not been documented. The heat from a facility that will effectively function as an open-air heating system, 24 hours a day, was never evaluated.

There are no studies that measure the health consequences of the heat emitted by data centers into their surroundings. Regarding heat and health generally, however, there are decades of evidence.

Scientific literature associates prolonged exposure to heat with increases in cardiovascular and respiratory mortality, heat stroke, kidney disease and mental health decline. The most affected are the elderly, pregnant women, young children and outdoor workers.

This investigation doesn’t just cover our findings about a building in Pando. With 25 years of satellite data, we were able to see how much heat a data center produces, even below its full capacity and with efficient cooling systems. Google’s data center will use the same cooling technology, but at five times this level in an area with different characteristics and a higher population. If a medium-sized data center that doesn’t even work at full potential produces a heat island detectable from space, a facility five times that would produce an even larger signature. All that remains is to measure the magnitude of its effects, and we now have a proven way to do this.

The key contribution of this investigation is having demonstrated that, with free and public data, it is possible to distinguish between the heat generated by the building and that generated by the servers. The method is replicable.

DuringIn the course of this investigation, in March 2026, the first study connecting data centers to heat islands was published (Marinoni et al., “The data heat island effect”, arXiv). This “preprint” (a preliminary version released before going through peer review) reports an average increase of 2°C around the analyzed data centers.

...The study drew methodological criticism from independent science communicator Andy Masley, the senior researcher for Omdia, Vlad Galabov and University of Bristol researcher Chris Preist, all of whom coincided on a single point: The study does not separate the heat produced by the servers from that which accumulates from simply replacing vegetation with roofs and pavement. Masley also pointed out that the paper does not compare other types of commercial buildings and that the resolution of the MODIS sensor used in the study (1 km per pixel) prevents one from distinguishing the data center from the surrounding area.

Amenaza Roboto’s analysis used Landsat, with a resolution 33 times greater, controlling for all of these factors. According to the literature reviewed, it is the first study of its kind to isolate both components.

The global data center industry consumes more energy than many entire countries. Unlike other industrial infrastructure, a data center converts virtually all of the electricity it consumes into heat and dissipates it from a single source, continuously, 24 hours a day, 365 days a year.

Uruguay has positioned itself as a regional hub for this type of infrastructure. Antel operates three data centers—Pando, Pocitos and Lezama. It will equip a fourth server room at the Pando location and build two new centers in Aguada and at a yet-undetermined location. Google is building its own center. According to sources familiar with the industry, three additional private installations are projected to be built. Supporters like to point out that these projects bring economic investment, jobs and development of digital infrastructure.

When discussing the environmental impact of data centers, public debate usually focuses on the consumption of water and energy. The thermal effect on the environment—the heat that these places release into the surrounding communities—does not appear in the conversation. The data demonstrates that this impact does exist, is measurable and yet no one is keeping track of it.

Now there is a way of measuring this impact and a baseline for what comes next. The heat behind the cloud is no longer invisible—and it’s detectable from 700 kilometers high.