Data centers for AI are no longer only worrying about their consumption, a new study suggests that they also heat entire neighborhoods
The expansion of artificial intelligence has triggered the conversation about electricity consumption, water availability and pressure on the networkbut now another front is beginning to open. A new academic work from the University of Cambridge suggests that large data centers, and especially those linked to the AI boom, may be generating a measurable local thermal effect several kilometers beyond their physical limits.
According to this analysis, the surface temperature of the land in the areas near these facilities can rise between 0.3 and 9.1 degrees Celsius once they come into operation, with an average increase of between 1.5 and 2.4 degrees.
The idea is not minor, because it changes the focus of the debate. Until now, the great public problem of these complexes was associated above all with the electricity they demand or the infrastructure they are forced to deploy around them. This study, not yet peer-reviewed, further suggests that its thermal footprint may partly resemble that of urban heat islands.
An effect that goes beyond the fence of the enclosure
What the work maintains is that the heat associated with these complexes does not remain within the perimeter of the data center. The authors speak of a “data heat island effect”, a kind of thermal island linked to this type of infrastructure, and they calculate that a average monthly increase of 1 degree Celsius in surface temperature can be measured up to about 4.5 kilometers from a typical AI data center. The comparison with the traditional urban effect is not accidental, because the scale of some of these installations already places them as relevant physical actors within the territory, not only as digital nodes invisible to those who live around them.
That nuance matters a lot.. For years the data center has been talked about as a critical but silent infrastructure, almost abstract for the common citizen. However, as these campuses grow in surface area, consumption and installed power, they also begin to behave as elements that visibly alter the environment. If the study ends up being consolidated with more scientific validation, the environmental impact of AI will no longer be read only in megawatts, emissions or water needs, but also in how it transforms microclimates and local conditions.
The race for AI makes everything worse
The work starts from a context that was already worrying in itself. Global data center capacity is growing rapidly and the sector is poised to become one of the most energy-intensive industries of the next decade. In parallel, hyperscalers’ investment in infrastructure has skyrocketed in response to the demand for AI, and various analyzes predict that the electricity use of these large complexes will more than double between now and 2030.
The reading of the study, however, is not closed. The article itself includes important reservations from experts in the sector. Vlad Galabov, Senior Research Director at Omdiawarns that this is an initial analysis that has not yet been replicated or validated through peer review, and also emphasizes that the work focuses on the surface temperature of the terrain, not on the air temperature at the altitude that people experience.
This caution is reasonable and will surely mark the discussion in the coming months.. But even with this prevention, the study introduces an important alert: data centers can no longer be analyzed only as efficient boxes from an energy point of view, but as territorial infrastructures with cumulative physical consequences. The simple fact that the debate has moved from the electricity bill to the local thermal environment already gives a clue to the extent to which the scale of deployment has changed.
The authors of the work also propose a particularly striking global figure: Up to 343 million people could be affected by this thermal effect if the “data heat island” is considered alone.. It does not mean that they all suffer the same level of impact, but it does mean that the phenomenon, if confirmed, would have sufficient scale to fully enter into debates about well-being, public health and local energy systems.
The conclusion, therefore, is not that each data center has automatically become a local climate threat, but rather that the conversation can no longer stop at computing power or the cost of deploying GPUs.
