AI Data Centers Threaten Public Health Through Energy Strain

Energy demands of AI pose hidden health risks

The computing infrastructure powering artificial intelligence systems is creating significant strain on electricity and water systems that underpin public health, according to two leading epidemiology professors writing in STAT.

Sten H. Vermund, dean of the University of South Florida College of Public Health, and Patricia J. Kissinger, associate dean at Tulane University's School of Public Health, argue that the hundreds of new data centers planned across the United States require extraordinary amounts of electricity and water for cooling—at a scale that many communities and power grids cannot accommodate.

The consequences extend far beyond inconvenience. Hospitals, clinics, emergency services, and home medical equipment all depend on stable, affordable power. When energy systems face stress, vulnerable populations—including older adults, people with disabilities, and those with chronic illnesses—face heightened risks from service disruptions and equipment failures.

Why it matters

As AI adoption accelerates in healthcare and other sectors, the infrastructure supporting these systems could paradoxically undermine the health outcomes the technology aims to improve. The strain on power grids affects not just data center operators but entire communities, with health burdens falling disproportionately on lower-income and historically marginalized populations.

Current carbon accounting falls short

Technology companies typically rely on carbon offsets or annual renewable energy purchases to claim environmental responsibility. These approaches do not ensure clean energy is available when and where data centers actually consume power, the authors note.

When data centers draw electricity from fossil-fuel-dependent grids while claiming "net zero" status on paper, they contribute to emissions of fine particulate matter and other pollutants linked to cardiovascular disease, respiratory illness, and premature mortality. Communities near power plants and transmission infrastructure—disproportionately lower-income areas—bear the greatest health burden from this pollution.

Growing electricity demand also threatens health system resilience. Grid instability increases outage risks with direct consequences for patients requiring uninterrupted power for life-sustaining care. Data centers' water demands for cooling systems further exacerbate local water stress, affecting sanitation and heat mitigation.

A measurable alternative

The authors propose requiring every kilowatt-hour consumed by AI data centers to be matched by new clean energy added to the same grid at the same time—an approach called "hourly, location-based" matching or 24/7 carbon-free energy.

This framework ensures increased demand pairs directly with increased clean supply. Emerging research suggests aligning large electricity loads with renewable generation, storage, and real-time grid management can simultaneously reduce costs, emissions, and system strain while decreasing real-time air pollution exposure in communities where electricity is generated.

Unlike carbon offsets, which can be difficult to verify and disconnected from local conditions, kilowatt-hour matching is measurable and enforceable. Data center operators gain more predictable energy costs and improved reliability, communities gain new clean energy infrastructure rather than additional pollution exposure, and regulators gain clearer oversight frameworks.

The authors emphasize that AI will play a central role in healthcare's future, but its growth should not compromise the energy systems on which that future depends. Linking AI's electricity use directly to clean energy development represents a practical step to ensure technological progress strengthens rather than strains public health foundations.

These details were first reported by STAT.