Artificial intelligence (AI) is reshaping the world—but at a growing cost to public health. A study from the University of California, Riverside, shows that the energy needed to power AI’s massive data centers is driving up air pollution, with deadly consequences.
By 2030, pollution from power plants and backup diesel generators used by these centers could cause 1,300 premature deaths a year in the United States. The public health costs—tied to illnesses like asthma and cancer, as well as missed work and school—could reach $20 billion annually.
Ignoring the problem
Tech companies often highlight their progress in reducing carbon emissions, but the study points out they largely ignore another problem: the harmful air pollutants their energy use creates. The researchers call for companies to report these pollutants and to compensate communities most affected by the pollution.
Low-income areas, often located near power plants or data center generators, are hit hardest. But the effects spread far beyond these neighborhoods. Airborne pollutants, including fine particles and nitrogen oxides, can travel across state lines, burdening distant communities with health problems and costs.
For example, emissions from data centers in Northern Virginia drift into nearby states, causing $190 million to $260 million in annual health costs. If these generators operate at their maximum allowed levels, those costs could rise to over $2 billion.
Huge health costs
In some regions, the health costs of pollution from AI processing exceed what tech companies spend on electricity. By 2030, the air pollution from AI data centers could rival that of all cars, buses, and trucks in California and double the burden caused by the U.S. steel industry.
Training advanced AI models is a particularly energy-intensive process. The study estimates that training Meta’s Llama-3.1 model—on par with OpenAI’s GPT-4—produced air pollution equal to over 10,000 car trips from Los Angeles to New York City.
The researchers used Environmental Protection Agency methods to estimate the health toll of this pollution, predicting between 940 and 1,590 premature deaths each year by 2030.
AI’s energy demands are growing faster than any other sector, the authors note, and the resulting pollution is set to rise sharply. They warn that unless tech companies act responsibly, the rush to build AI tools could create a public health crisis.
“If you have family members with asthma or other health conditions, the air pollution from these data centers could already be affecting them,” the study concludes. “This is a problem we need to address now.”
