How do large AI models actually worsen air pollution?
Large AI models require enormous amounts of electricity to run their data centers, and much of that electricity still comes from burning fossil fuels like coal and natural gas. A 2023 study measured PM2.5 (fine particulate matter) levels at transportation stations and manufacturing sites in Addis Ababa, finding concentrations between 63.46 and 104.45 μg/m³—well above the World Health Organization's safe limit of 15 μg/m³ [1]. These are exactly the kinds of industrial sources that power AI data centers, and the study linked them directly to health risks: 57.92% of 480 surveyed residents reported pollution-related illnesses confirmed by medical evidence [1].
The problem is global. Another 2023 paper notes that 99% of the world's population now breathes air exceeding WHO limits, driven by increased fossil fuel use and extreme climate events [2]. AI models are a growing part of that fossil fuel demand—each large model training run can consume as much electricity as hundreds of homes use in a year. The result is more PM2.5, more CO2, and more strain on local power grids, which often respond by firing up older, dirtier power plants.
Who is most affected by this renewed pollution?
The health impacts fall hardest on vulnerable populations. A 2023 review found that pregnant women exposed to high levels of air pollutants face increased risks of hypertensive disorders, postpartum depression, low birth weight, preterm birth, and infant mortality [2]. Children and the elderly are also at higher risk because their bodies are less able to filter out toxins. The same study emphasizes that oxidative stress and immune system damage are key mechanisms, meaning even short-term exposure can have lasting effects [2].
Workers in pollution-heavy zones are especially exposed. The Addis Ababa study identified drivers, street vendors, and manufacturing employees as the groups most affected, with PM2.5 levels reaching 104.45 μg/m³ at industrial sites—nearly seven times the WHO safe limit [1]. These are often low-income workers who have little choice about their environment, making the issue one of environmental justice as well as public health.
Why is this becoming a public issue again now?
Two factors are converging: better monitoring technology and growing public awareness. A 2021 study demonstrated an AI-powered IoT system that uses sensor arrays to detect eight pollutants (including PM2.5, CO2, and NO2) in real time, sending alerts within 5 to 60 minutes when levels become hazardous [4]. This makes pollution from AI data centers instantly visible to regulators and the public, rather than being an invisible problem.
At the same time, public perception is shifting. A 2022 interview study in Brussels found that people categorize air pollution based on five mental schemes: source, health impact, climate impact, functionality, and sensory perception [3]. As AI data centers become more visible (both physically and in news coverage), more people are connecting the dots between their digital activities and the smokestacks powering them. The study also found that people often underestimate harm when they feel they can avoid the pollution—but as AI infrastructure spreads into residential areas, that sense of avoidability is shrinking [3].
About These Sources
This answer is built on 5 peer-reviewed studies — published from 2021 to 2024, 1 from 2024 or later, 3 in Q1 journals, collectively cited 421 times — selected as the most relevant from 5 studies that passed quality screening, drawn from 48 papers retrieved from a database of over 500 million.
Sources used in this answer
Urban Air Pollution and Greenness in Relation to Public Health
Measured PM2.5 levels in Addis Ababa ranged from 50.13 to 96.84 μg/m³ at transportation and industrial sites, with 57.92% of 480 respondents reporting pollution-related health issues confirmed by medical evidence.
Air pollution and pregnancy
99% of the global population breathes air exceeding WHO limits; pregnant women and neonates face increased risks of hypertensive disorders, preterm birth, and infant mortality due to air pollution.
The Public’s Perceptions of Air Pollution. What’s in a Name?
In 51 interviews in Brussels, public perception of air pollution was found to be diverse and subjective, often underestimating harm; people categorize pollution by source, health impact, climate impact, functionality, and sensory perception.
IoT enabled environmental toxicology for air pollution monitoring using AI techniques
An AI-enhanced IoT system using sensor arrays and Elman Neural Networks can detect eight pollutants (including PM2.5 and CO2) in real time, with alerts generated within 5 to 60 minutes of hazardous levels.
Next-Generation Air Pollution Forecasting: Integrating AI, Spatiotemporal Dynamics, and Privacy-Ensuring Approaches for Urban Areas
Proposes integrating AI with spatiotemporal models and privacy-preserving methods for next-generation air pollution forecasting, aiming for high precision in urban areas.