As artificial intelligence booms, data centers—the physical infrastructure driving AI—have become a topic of concern for towns across the United States. Research shows that even the industry’s claims of modest economic benefits are overstated. They’re worsened by the staggering health and environmental costs.
UCS has documented the wide range of negative ecological impacts associated with artificial intelligence, as well as the financial burdens placed-on ratepayers. These factors, coupled with health impacts, can feel overwhelming. But, cumulative impacts analysis, a framework used for evaluating all the way the decisions may affect the local environment and health, could help. Data-center-specific cumulative impacts analyses could give decision makers a more complete understanding of the plethora of potential health impacts from a proposed center. Ultimately, this could help us avoid, minimize, and mitigate some of these impacts, preventing them from further exacerbating existing inequities.
Understanding Data Centers’ Public Health Impacts
One of the most prominent environmental stressors associated with data centers is air pollution. Across the United States, around 60% of data centers’ exorbitant energy demands are met through fossil fuels, emitting massive quantities of particulate matter, sulfur dioxide, and heat-trapping greenhouse gases like carbon dioxide and nitrous oxides. In fact, it was estimated that last year hyperscale data centers were responsible for up to 54 million metric tons of CO2 emissions—as much as 11.7 million cars.
Many data centers also rely on diesel backup generators, which are even more polluting, releasing 200-600 times as much nitrogen oxide, for example, than even gas power plants. Air pollutants harm nearly every system in the body and have been linked to atherosclerosis, neurodegenerative disease like Alzheimer’s, decreased intelligence and cognitive processing, and increased stroke and cardiac mortality, among other conditions.
But data centers are doing more than polluting the air. In fact, we’ve identified over six distinct pathways by which data centers threaten the health of their host communities.
One of the most overlooked yet potentially deadly environmental health impacts of a data centers is thermal pollution. A preliminary study found hyperscale centers raised surface temperatures up to ten kilometers away by up to 16 degrees Fahrenheit. For reference, a 16-degree temperature change is the same difference as that between the average US temperatures in July versus October. In Utah, a proposed data center known as the Stratos Project, could release thermal energy equivalent to the heat output of roughly 23 Hiroshima-scale atomic bombs every single day. This excess heat can accelerate already worsening extreme heat resulting in more rapid and frequent heat illnesses (heat stroke or heat exhaustion) on hot days. It would also worsen chronic conditions aggravated by heat including asthma, acute kidney injury and preterm and still birth rates.
Between loud cooling systems and brightly lit perimeters, data centers can also be significant source of noise and light pollution. In fact, research suggests a hyperscale data center can emit as much light as a small town. Both noise and light pollution have been linked to a plethora of health challenges as they trigger the body’s stress response which causes inflammation and high blood pressure, the leading contributor to heart disease, stroke, and kidney failure.
In addition to air, light, noise, and thermal pollution, most data centers rely on toxic “forever” chemicals called PFAS to cool their computers, for fire suppression, and in the computers’ semiconductors. These chemicals can escape data centers as gas or enter waste-water or ground-water. Some PFAS are known carcinogens, and they’ve also been linked to reduced fertility and disrupted fetal neurodevelopment.
Data centers can also indirectly threaten public health by increasing energy bills, which often forces many to make costly and difficult trade-offs such as keeping their homes at unsafe temperatures or foregoing healthcare to afford energy costs. In addition to contributing to energy insecurity, most hyperscale centers consume millions of gallons of water each day, worsening droughts and contributing to water scarcity
Addressing and preventing these health harms
Without proper guardrails and regulations, data centers could become a public health nightmare. The limited research on how simultaneous exposure to these environmental health threats will impact communities is of deep concern. We know, for example, that exposure to noise and light pollution have separately been linked to hypertension—but how would exposure to both compound the health impacts?
While we need more research to better understand how these threats interact simultaneously, communities should take immediate action to address independent impacts. Across the country, we’ve also already started to see data centers harm human and ecological health. Moreover, given the exponential rise in data center demand, we need frameworks for considering and protecting human and ecological health as new data centers are proposed.
Enter cumulative impacts analyses.
Cumulative impacts are defined as the totality of exposure to chemical and non-chemical stressors and their effects on health, well-being, and quality of life outcomes. Cumulative impacts analyses (CIAs) are a well-suited framework for addressing the health concerns associated with data centers. CIAs would not only facilitate the consideration of the total burden from the range of chemical (PM2.5, NOx, PFAS) and non-chemical (light, noise, heat) stressors from the proposed data center, but they’d also allow for the consideration of existing burdens such as current pollution levels from other data centers or nearby infrastructure.
Currently, only a few states have cumulative impacts legislation, and none have data center-specific frameworks. Existing CIA legislation is imperfect and contains loopholes, allowing many data centers to avoid the process entirely. In Massachusetts, a state which already uses cumulative impacts in air quality permitting, residents of the city of Lowell were forced to take a data center to court after the center’s proposal to add eight additional backup generators and new cooling towers was approved without adequate consideration of cumulative impacts.
In Minnesota, CIAs are only required if projects are proposed in environmental justice (EJ) communities in three of the state’s major cities. While a focus on communities that have already shouldered more than their fair share of impacts makes sense, data center-specific CIA requirements would help ensure that data centers don’t create new overburdened areas.
Data centers are also often built in rural communities that may not be classified as EJ communities but still are more vulnerable due to limited access to healthcare. These communities have a set of unique considerations, such as being some of the only areas with minimal light pollution, where stars aren’t drowned out by urban lights. Under current CIA frameworks, these considerations often go overlooked.
What would a Data-Center Specific Cumulative Impact Process Look Like?
We can draw on existing CIA frameworks to imagine what a data-center-specific process could look like.
First, a data center developer would be required to outline all potential and estimated chemical and non-chemical stressors based on a predetermined list that includes on-site stressors and stressors released elsewhere, including from secondary sites such as power plants.
Second, for each category of stressor, the estimated impacts would be compared against current conditions. If the CIA finds there is already a higher density of pollution and health burdens, then that data center could be subject to more stringent regulations. This could help address the problem of multiple data centers compounding to worsen public health, ensure that overburdened communities aren’t further exposed to harm, and that unique ecological and cultural resources aren’t overlooked. Other elements of a community’s risk, including access to healthcare, linguistic barriers, and ecological importance, could also be considered here.
Surprisingly, simply requiring companies to disclose emissions can motivate them—particularly if publicly traded via shareholder pressure—to minimize pollution. Beyond being publicly available, though, the completed CIA could be used by permitting agencies to deny or “condition” permits. “Conditioning” could mean that agencies approve permits only if data centers take additional specific steps, such as using battery backup options rather than polluting generators to avoid pollution. Conditioning could also mean requiring mitigation strategies, such as heat reuse, where excess thermal energy from data centers is redirected to heat nearby buildings. This could not only mitigate the “data center heat island” effect, but it could also lower community electricity costs. CIAs could also lead to discussions about community benefit agreements (CBAs).
CBAs are agreements between developers and community members that guarantee specific local benefits for the community.For many states, this wouldn’t require a whole new set of laws. States like New Jersey already use CIAs in the permitting process for certain facilities (including major air polluters, landfills, and sewage plants), ensuring that decisionmakers are thoroughly informed. Expanding these requirements to encompass data centers could be a step towards ensuring public health is considered in data center related permitting and decisions. In states without CIA legislation, the creation of data center CIA laws—inspired by other states’ general CIA frameworks—could catalyze a broader advancement of environmental justice by bringing CIA’s to these states.
The unmitigated growth of hyperscale data center development and the associated public health impacts are not inevitable. A data center-specific CIA framework would ensure that decisionmakers and community members are aware of the many, simultaneous exposures associated with a proposed data center. Moreover, it would highlight the severity of potential health impacts by detailing baseline and existing community risk factors that could compound the data centers’ individual impacts.
Ultimately, such a framework could facilitate the avoidance, minimization, and mitigation of data center pollution, improve public health, and could even foster more sustainable community relationships with industrial developers.