The carbon emissions of global health care activities make up 4–5% of total world emissions [4], placing the health care industry on par with the food sector [2]. Health care carbon in between countries varies. The United States health care industry is the second largest in the world, expending an estimated 479 million metric tons (MMT) of carbon dioxide per year; nearly 8% of the country’s total emissions. Compare this with China’s health care carbon, at 600 MMT, or 6.6% of the country’s emissions, or India’s health care carbon, at 74.1 MMT, or 3.5% of the country’s emissions [2].

Health care carbon emissions come from health care infrastructures and health care delivery. Health care infrastructures include use of heating, lighting, cooling, food services, transportation, and water waste [5]. Health care delivery includes medical activities which have a carbon impact, such as doctor’s visits, medical procedures, prescription of drugs [6], and hospitalization [7]. For instance, inpatient admission to a hospital, based on admission intake plus 3.6 bed days, emits 380 kg of CO2 per patient [8].

Beyond carbon emissions of health care, medical waste is a significant environmental problem. While disposable (single-use) medical devices increase safety and confidence, they also use resources, create waste and emit carbon [9]. Not only are landfills occupied with medical waste, but also the contaminated materials in landfills get released in the environment, creating more noxious gasses in the atmosphere. In 2018, 24% of all municipal waste generated in the EU was landfilled according to the European Commission [10].

Given that carbon emissions contribute to poor health conditions related to climate change health hazards, the medical industry should have an interest in carbon reduction generally—as a means to reduce disease burden—and specifically—as health care carbon is counterproductive to patient health. To be sure, the destructive cycle of health care carbon and climate change has been recognized by a number of health care initiatives and publications. In 2021, the National Academy of Medicine (NAM) launched their Action Collaborative on Decarbonizing the United States Health Sector and wrote the Biden Administration advocating decarbonization as part of the US COP26 climate commitments. The 2021 Lancet Countdown on Health and Climate Change Report indicates both the seriousness of climate change as a health threat and the culpability of the health care industry in contributing to carbon emissions and climate change [11]. The same year the New England Journal of Medicine ran an article on “Decarbonizing the U.S. Health Sector—A Call to Action,” which opined tackling the carbon emissions of the health care supply chain is central to decarbonization and pointed to the carbon impact of the biopharmaceutical, biotechnology, and medical device industries [12].

We support these and other efforts to make health care more sustainable and extend the call for health care climate action to the often overlooked area of artificial intelligence (AI) in health care, which is not only widely used but also has a significant carbon cost.

Modern, technological health care has advanced through the emergence of larger data sets and data sources, which has paved the way for the use of artificial intelligence (AI), autonomous and intelligent systems (AIS), prescriptive and predictive analytics, bioinformatics, and even the Internet of Things (IoT) in health. These technological innovations have advanced and improved health care delivery significantly. Yet, these same technologies also contribute to the negative effects on the environment through carbon emissions in the development, deployment, dissemination, and disposal of medical devices, digital infrastructure and medical services [13].

The energy impact of AI is not limited to its use phase. The computational infrastructure that enables AI systems has significant additional environmental implications. Notably, the largest AI models are doubling in energy necessary to compute every three to four months, thereby severely outpacing the increasing efficiency of hardware [14]. The energy impact of AI is not limited to its use phase. The computational infrastructure that enables AI systems has significant additional environmental implications. For instance, forty days of training Google’s AlphaGo Zero game generated the equivalent of 1,000 h of air travel or a carbon footprint of 23 American homes [15]. Google’s biggest AI model “The Switch Transformer” now has more than 1.6 trillion parameters—measure that refers to variables in computer programming language used to pass information between functions or procedures [16]. Each training run of a giant transformer like this can generate 626,155 pounds of CO2 emissions, the equivalent of 17 American life-years (at 36,156 CO2 emissions / per year), or the “lifetime” of five cars (at 126,000 CO2 emissions per car) [17]. As health care relies more on AI, so do carbon emissions expand.

Moreover, AI is raising demands for metals and plastics, thereby also generating a lot of electronic waste. The environmental impact of cobalt mining is known to be substantial [18]. Most of the minerals necessary for electronics are mined in conflict areas and mining often takes place under poor labor conditions. The extractive effects on the environment of AI reliant technologies extend well beyond fossil fuel extraction, and include mineral mining for chips, exploitative human labor for labelling training datasets, and the significant waste produced by products designed for planned obsolescence and inevitable upgrades.. Health care AI use is also complicit in these social and environmental externalities, each of which demands ethical evaluation. These associated concerns will be bracketed, as we now move to a focus on the use of AI in health care specifically, and the environmental impacts.

It is pivotal to acknowledge that any discussion around the environmental impacts of AI software services and the computational infrastructures on which these run cannot be resolved without addressing the role of concentration of power over these platforms and infrastructures, as well as over the organizations that depend on them, by a small number of technology firms. In absence of forced transparency, these companies have no incentive to help users understand their digital footprint. More worryingly, the political economy of AI and computational infrastructure seems to be at sharp odds with what we need to curb its environmental impact. As Meredith Whittaker argues, “tech firms are startlingly well positioned to shape what we do—and do not—know about AI and the business behind it, at the same time that their AI products are working to shape our lives and institutions.”[19]