Plenty of attention in bioethics is being directed toward regulating AI in health care. Grants and conferences are booming regarding privacy, transparency, bias, ownership, and reliability. Plenty of other conferences are examining whether and when AI will pose an existential threat to humanity. What has not gotten sufficient attention is the ethical challenge posed right now by building the computing power that cutting-edge AI demands. AI may not kill us, but using the computing power now available to fuel its ever-expanding appetite might.
Massive data center construction is booming in the U.S. and globally. Major expansions by Google, Amazon, Meta, and Microsoft are underway in Northern Virginia, Texas, Ohio, and the Pacific Northwest. Thousands more are planned worldwide.
These data centers are not your father’s traditional networked computers, which would quickly be overwhelmed by today’s AI workloads. Today’s data centers require huge numbers of high-performance graphics processing units (GPUs) and the IT infrastructure storage to support them. These require massive amounts of energy and cooling capabilities.
The sheer number of GPUs necessary for current AI uses requires far more square footage. How much? The biggest data center by area is currently the China Telecom Data Center in Inner Mongolia, which spans nearly one million square meters. Other projects underway are OpenAI’s ‘Project Stargate’ in Texas, 8 million square feet and Fermi America’s west Texas campus, which will be 18 million square feet. That is an area covering 270 American football fields, which is huge even by Texas standards. This is the start of a trend toward larger and larger “giga-scale” computing facilities.
Many tech and business types, economic pundits, and science media are gurgling on about the need for huge investments in these behemoth facilities. But, before the entire globe is converted into computer parks, two questions need to be asked: are colossal data centers good for human health, and does it make sense to build gigantic ‘machines’ that may quickly go out of date?
AI data centers pose public health risks by releasing ambient air pollutants—“digital smog”—both directly and indirectly. Direct risks stem from the release of harmful air pollutants by on-site backup generators. The existing power grid in many states can’t support the needs of these huge centers, so backup diesel generators are used. Emissions from diesel engines and diesel generators have several impacts on human health. Diesel engines emit CO, which impairs oxygen transport in the blood and leads to cardiovascular problems. Unburned hydrocarbons and fine particulates from diesel exhaust can penetrate deep into the lungs, causing respiratory diseases and exacerbating existing conditions. Their emissions contribute to smog formation and respiratory issues. Diesel emissions are associated with lung cancer, cardiovascular diseases, and premature mortality.
Indirect risks arise from reliance on electricity sources that are highly polluting. According to the Lawrence Berkeley National Laboratory, U.S. data centers are projected to account for 6.7% to 12.0% of America’s total electricity consumption by 2028, up from 4.4% in 2023. As American data centers continue to rely on the existing inefficient power grid as their primary energy supply, their soaring energy consumption imposes a substantial public health burden in the form of the mining to obtain more coal for outdated plants, pollution from fracking and burning coal, chemical exposures due to shipping accidents, and, in the case of nuclear power, waste disposal.
Growing AI workloads demand more powerful chips, which generate more heat, necessitating more cooling, thus increasing water demand. Data centers often tap into freshwater resources to achieve this cooling, which puts what water is available to nearby and downstream communities at risk. A large data center can consume up to 5 million gallons per day, equivalent to the water use of a town of 10,000 to 50,000 people. Diverting this water means less water for people near the centers.
Maybe all this health and environmental impact is worth it. But what if the current push to build is actually an investment in crude albatrosses, which ought to be replaced as rapidly as possible by more efficient, less polluting, and less disruptive computing power engines? If we sink a pile of resources into the current massive data centers, are we discouraging efforts at needed improvement just to satisfy short-term profit?
Data centers in their current form scream white elephants. Overbuilding today’s giant data centers that demand too much power and water may not be good public policy. Sometimes the tech you have is not yet ready for the goal you dream of—consider Elon Musk’s drive to colonize Mars using rocket and space travel technology that is far from being up to the task.
Sadly, the gigantic data centers now being built or planned are not good for public health or efficient resource use. In their current form, they suck up enormous amounts of energy, land, and water, pollute, are extremely noisy, involve a lot of maintenance, and increase global warming. True, they power a hugely important tool, but at a price that needs to be fully and transparently calculated and then openly debated.
Arthur Caplan PhD, NYU Grossman School of Medicine