AI Data Centers Linked to 1,300 Annual Deaths by 2030

Artificial intelligence infrastructure poses a deadly but overlooked threat to public health, according to research published in December 2024 by scientists at Caltech and the University of California, Riverside. The study, first reported by Futura Sciences, reveals that surging electricity demand from AI data centers could cause up to 1,300 premature deaths annually in the United States by 2030—a toll roughly double that of the entire American steel industry.

The problem isn't what happens inside data centers, but what happens at the power plants feeding them. Most AI facilities draw electricity from grids still dependent on coal and natural gas, while backup diesel generators kick in during demand spikes. Both sources release fine particulate matter and nitrogen oxides directly linked to respiratory disease, cardiovascular illness, and early death.

Why it matters

Tech companies publish detailed sustainability reports tracking carbon emissions and water consumption, but they systematically ignore air quality impacts—leaving communities near data centers without reliable information about what they're breathing. As AI workloads drive data center electricity consumption from 3.7% of US demand in 2023 to a projected 11.7% by 2030, this transparency gap represents a growing public health blind spot worth tens of billions in annual healthcare costs.

The Virginia pollution corridor

Virginia hosts the world's largest concentration of data centers, making it an ideal case study for regional impact. The research team found that pollution from Northern Virginia's dense network of backup generators regularly crosses state lines into Maryland, West Virginia, Pennsylvania, New York, New Jersey, Delaware, and Washington, D.C.

This transboundary pollution already generates between $190 million and $260 million in annual health costs. If operators ran backup generators at maximum legal capacity, those costs could reach $2.6 billion yearly—a tenfold increase.

The cost of training one model

To illustrate AI's energy intensity, researchers calculated the pollution footprint of Meta's Llama-3.1 large language model, launched in July 2024. Training this single model consumed enough electricity to generate air pollution equivalent to more than 10,000 round-trip car journeys between Los Angeles and New York.

Using statistical models from the US Environmental Protection Agency, the team projected total healthcare costs from AI-driven air pollution could reach $21.5 billion annually by 2030. These costs encompass cancer treatment, asthma care, chronic illness management, and lost worker productivity.

The transparency problem

The researchers aren't calling for immediate emission caps. Instead, they urge regulators to mandate standardized reporting requirements that force tech companies to disclose air quality impacts from their electricity consumption. Without this data, communities lack the information needed to understand local health risks.

The findings were first detailed by Futura Sciences based on the December 2024 study from Caltech and UC Riverside researchers.