Part 8 of 13 in Thirteen Records. The long article, Data Centers Don’t Have an Environmental Record. They Have Thirteen of Them., rates thirteen kinds of environmental complaint against data centers. This series takes them one at a time.
I work at a local ISP in Northern Illinois and co-founded FD-IX, an internet exchange. I have no customer, vendor, or interconnection relationship with any facility or developer named here and no financial position in any of them beyond broad-market index funds. My industry grows with infrastructure investment in a general way, which is an indirect interest; the full disclosure is at the top of the complete article.
Every watt a data center draws leaves as heat, and an air-cooled campus puts that heat into the air at the fence line. Whether neighbors feel it is a newer question than the others in this series, and it has two very different answers in circulation: a peer-reviewed Arizona State University study that drove sensors upwind and downwind of four Phoenix facilities and measured up to 4°F in summer, and a satellite preprint claiming 2°C for miles around AI data centers worldwide. [14a][14b]
The two are not the same kind of evidence, and the second has a problem its method cannot resolve. This post rates three thermal claims: the Phoenix measurement, the satellite preprint, and the claim that a heat island is coming to a corridor that switched to air cooling to save water. The Yorkville projects use the design the ASU study measured. What differs is the climate, and nobody has measured an Illinois July.
How the ratings work. Each complaint below is stated the way it usually arrives at a hearing and rated on the scale the full article uses. Systemic: documented at compliant facilities across multiple geographies. Regional: real, but only under specific geographic or operational conditions. Isolated: documented at one or few facilities due to misconduct or atypical design. Exaggerated: the concern is real but the figures are permit ceilings or worst cases, not operating norms. Selective: a legitimate concern applied only to data centers, not to comparable land uses. Misattributed: the harm is real but the source, geography, or cause is wrong. Contested: a plausible mechanism exists but causal documentation at compliant facilities does not. Fabricated: no credible evidentiary basis exists. Where I say I found no record of something, that means a search of public sources through October 2026, not paid databases or records requests.
Thermal Effects
Claim: Data centers heat the air around them. Tier: Regional. Documented at air-cooled facilities in Phoenix.
Air-cooled data centers in Phoenix measurably warm the air downwind, by up to 4°F (2.2°C) in summer. A peer-reviewed Arizona State University study measured ambient air temperature at four Phoenix-area facilities from June through October 2025 with vehicle-mounted sensors driven simultaneously upwind and downwind. Average downwind increases ran 1.3 to 1.6°F, detectable up to about a third of a mile from the perimeter. It is the first study to measure air temperature directly rather than satellite-derived land surface temperature. [14a]
Claim: AI data centers raise temperatures 2°C for miles around. Tier: Contested. A satellite preprint, not peer reviewed.
A preprint first posted to arXiv in March 2026 (Marinoni et al.) reports that satellite-measured land surface temperature rises about 2°C on average after an AI data center starts operating, with effects detectable up to 10 km away and an average 1°C increase out to 4.5 km, and estimates that more than 340 million people could be affected. [14b] It measures ground surface temperature, not the air people experience. The authors say they filtered out seasonality and “other types of anthropogenic activities,” but the paper does not describe a control for the facility’s own conversion of fields to roofs and pavement, which by itself raises satellite-measured surface temperature and is already in place when operations start. [14b][21] The result may be real, overstated, or largely an artifact of construction, and the method as described cannot tell which.
Claim: A data center will create a heat island here. Tier: Contested. Unstudied in a cold climate, not shown absent.
The Yorkville projects moved to air cooling to cut water use, so the corridor uses the design the ASU study measured. What differs is the climate. Whether a 4°F downwind effect measured in a Phoenix summer appears at all in an Illinois July has not been studied, and no published evidence supports predicting a neighborhood heat island here.
What to ask for. This is the record with the least to require and the most to measure. Ask the applicant to state the cooling design and the heat rejection capacity at full build-out, which is already needed for the noise and power questions. If a council wants a thermal baseline, it is the cheapest of the baselines in this series: ambient air temperature at the property line and at the nearest homes across a summer before construction, taken the way the ASU team took theirs, so that a later claim in either direction can be checked against a number rather than a satellite image. [14a] No approval I reviewed asks for one, and no rule requires it.
What nobody knows. Whether a 4°F downwind effect measured in a Phoenix summer appears at all in an Illinois July has not been studied. Northern Illinois has far more cool-weather hours, which changes the fan load for much of the year, and no published evidence supports predicting a neighborhood heat island here. It also does not support ruling one out on design-day afternoons.
This is part 8 of 13 in Thirteen Records, a series that takes the long article Data Centers Don’t Have an Environmental Record. They Have Thirteen of Them. one topic at a time. Previous: construction. Next: land and habitat, where the real question is whether to rezone farmland at all.
Citations
[14a] Sailor, D.J., Abolhassani, S.S., and Martin, E., “Data Center Waste Heat as an Emerging Urban Thermal Hazard: First Field Measurements of Neighborhood-Scale Air Temperature Impacts,” ASME Journal of Engineering for Sustainable Buildings and Cities, Vol. 7, No. 2, May 2026. https://asmedigitalcollection.asme.org/sustainablebuildings/article/7/2/024501/1233035/Data-Center-Waste-Heat-as-an-Emerging-Urban (archived: https://web.archive.org/web/20261005124524/https://asmedigitalcollection.asme.org/sustainablebuildings/article/7/2/024501/1233035/Data-Center-Waste-Heat-as-an-Emerging-Urban) The journal page sits behind a bot check and was not read directly; study details are confirmed through Interesting Engineering, May 18, 2026: four facilities from a 36 MW Mesa building to a 169 MW Chandler campus; June 18 to October 25, 2025; vehicle-mounted sensors with simultaneous upwind and downwind readings; up to 4°F, average 1.3 to 1.6°F; primarily air-cooled, condenser discharge 14 to 25°F above ambient; detectable up to a third of a mile; “the first to directly measure air temperatures both upwind and downwind of data centers in real time.” https://interestingengineering.com/science/data-center-phoenix-temperature (archived: https://web.archive.org/web/20260519113752/https://interestingengineering.com/science/data-center-phoenix-temperature)
[14b] Marinoni, A., Cambria, E., Lin, W., et al., “The data heat island effect: quantifying the impact of AI data centers in a warming world,” arXiv:2603.20897, submitted March 21, 2026, revised April 21, 2026 (v3). Not peer-reviewed. Abstract: land surface temperature “increases by 2°C on average after the start of operations”; “more than 340 million people could be affected.” Full text: impact “reaches up to 10 km distance”; “an average monthly LST increase of 1 °C can be measured up to 4.5 km”; filtering for “seasonality effects” and “other types of anthropogenic activities in the surrounding of AI hyperscalers”; “over 6733 data points.” https://arxiv.org/abs/2603.20897 (archived: https://web.archive.org/web/20261013004906/https://arxiv.org/abs/2603.20897)
[21] Andy Masley, “Data centers’ heat exhaust is not raising the land temperature around where they’re built,” The Weird Turn Pro (Substack), March 31, 2026. [Commentary, not peer-reviewed.] https://blog.andymasley.com/p/data-centers-heat-exhaust-is-not (archived: https://web.archive.org/web/20261013004907/https://blog.andymasley.com/p/data-centers-heat-exhaust-is-not)