Skip to content
World clockEU--:--UK--:--USA--:--CN--:--PLDEFRIT中文EN

portal about AI and technologyevents · analysis · interviews · technical background

Search
LIVE
›

AI data centers: waste heat measurably warms nearby neighborhoods

Measurements and simulations show that waste heat from large data centers noticeably warms the surrounding area. Germany and France already require operators to reuse it. At EU level, operators only have to check whether reuse is possible.

TechnologyNewsRachel NwosuPublished: 25 September 20265 min readSources 2
AI data centers: waste heat measurably warms nearby neighborhoods

Waste heat from large AI data centers is no longer a theoretical problem. Measurements and simulations reported by the IT service heise online on 25 September 2026 show measurable temperature increases around existing sites. In the USA there is hardly any regulation so far, while obligations already apply in Europe.

What the measurements show

A team led by geographer David Sailor of Arizona State University fitted vehicles with sensors and drove them past four data centers in the greater Phoenix area. One was a large facility run by the operator CyrusOne. On the downwind side of the buildings, averages ran 0.7 to 0.9 degrees Celsius above the surroundings. The effect was still measurable from 90 meters out to half a kilometer. The sample is small: four sites, three days of measurement.

The French simulation specialist EOLIOS ran calculations for facilities with 50 to 100 megawatts of power. Under certain wind conditions, it found warming of 2.5 to 5 degrees Celsius at distances of up to 150 meters.

Almost all the electricity becomes heat

On paper, a single data center with more than 50 megawatts of power releases as much waste heat as around 54 000 households use in a year. That is technically plausible: almost all the electrical energy drawn in is ultimately released as heat. The data firm DC Byte found that 41 percent of US data centers stand in clusters with at least three facilities within a one-kilometer radius. There are already 88 such clusters, and 23 more are expected by 2040. On these figures, total planned US capacity could exceed 290 gigawatts, roughly four times today's level.

How operators counter this

CyrusOne says it chose light-colored rather than dark roofs at the Phoenix site to radiate heat and reduce the effect. Amazon relies on vertical rather than ground-level exhaust air, and on evaporative cooling on hot days. The company headquarters in Seattle is partly heated with waste heat from a neighboring data center.

In Finland, a Microsoft facility is due to go into operation in 2027. Once fully built out, it is to cover around 40 percent of the heat demand of about 250 000 households and businesses in the region. In Germany, according to industry figures, more than 25 billion euros are flowing into new AI data centers.

Europe regulates, the USA barely

Germany and France already oblige operators to distribute waste heat. At EU level, operators above one megawatt only have to examine whether waste heat reuse is possible. In the USA the field is hardly regulated. The action alliance "Heiße Luft" documents planned and under-construction data centers in Germany on an interactive map. The debate over sites, it suggests, is only just beginning.

Almost all the electrical power of a data center leaves the building as heat in the end, either as heating for neighbors or as a burden on the local climate.

Comments 0

Sources

2
  1. 01heise online: KI-Rechenzentren – Abwärme heizt angrenzende Wohngebiete aufDE
  2. 02heise online: Themenüberblick RechenzentrenDE

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Rachel Nwosu

Rachel Nwosu

AI, models and technology

Rachel Nwosu covers AI, models and technology for FLASH24, working from public model documentation, benchmark releases and repository histories rather than press summaries, and she skips announcements that arrive without reproducible numbers. She checks training-data claims against dataset cards and reruns reported metrics where code is available. She spends much of her week interviewing researchers and engineers, tracking model launch calendars, and comparing vendor benchmarks with independent evaluations. Outside the desk she runs 3D printers, restores old computers, and tests how models learn from internet junk. She does not publish benchmark figures she cannot trace to a source.

Newsroom →

Comments

0
  1. No comments yet — be the first.

Write a comment

Comments are public. We do not publish abuse, spam or advertising.