The fast-paced evolution of artificial intelligence has led to an exponential growth in data centers. Amazon, Google, and Microsoft are building their data centers at an accelerated pace, causing emissions from fossil fuel usage to increase to alarming levels (Grochulska-Salak et al. 2025). With increasing reliance on artificial intelligence, data centers will continue to consume energy and water at rates that cannot be sustained, and fossil fuel emissions will only continue to rise. This issue is often overlooked due to the techno-optimism associated with these data centers. However, if we think more long-term now, we can reduce some of these costs.

Growing Environmental Crises: The Energy Demand Crisis

Among the most energy-intensive buildings ever built, data centers stand out. On an international level, data centers, crypto mining, and artificial intelligence use accounted for about 460 TWh of electricity usage in 2022, representing close to 2% of global electricity use (Sheng et al., 2026). This number is expected to rise further – to consume as much electricity as Germany uses as a whole – by 2026 (Sheng et al., 2026).

The demand for energy from data centers is estimated to be about 945 TWh by 2030, growing by an average annual rate of about 15%, which is over four times faster than any other industry (Sheng et al., 2026). Due to the concentration of data center infrastructure in selected locations like Northern Virginia, Phoenix metropolitan area, the Netherlands, and Ireland, the pressure on the electricity infrastructure is substantially higher in some areas than national averages (Sheng et al., 2026).

With rising electricity consumption, the cost of generating electricity and constructing infrastructure increases for utility companies, ultimately reaching consumers at higher rates. It is believed that the end consumer will bear the brunt of these rate hikes, as they are powerless to avoid them (Grochulska-Salak et al., 2025). 

The rapid increase in data center energy requirements driven by AI technology threatens to overload power grids and home energy budgets alike. Taken together, the energy and water demands of AI infrastructure represent not just an environmental challenge, but a social one. The costs are real, they are growing, and they are being passed on to communities that have no say in where these facilities are built or how they operate, which is causing a burden on the environment.

Growing Environmental Crises: The Hidden Toll on Water Supply

The issue of water shortage forms a second, equally important aspect of this problem. Most major data centers make use of evaporative cooling systems that function by circulating water through cooling towers and letting it vaporize into the atmosphere. One such major center may require up to one to five million gallons of water each day, similar to a small town (Kaack, et al. 2022). As the water evaporates, instead of returning to the source of rivers or underground sources, it is effectively taken out of the local ecosystem (Scherer, 2025).

Consumption by such facilities becomes particularly harmful in areas already experiencing problems with water scarcity. Within the American Southwest, areas experiencing shortages due to drought for many years, there have been significant allocations of water made to the operation of data centers to the detriment of agriculture and natural systems. Similar tensions have also become a problem in some regions of Spain, Chile, and South Africa. In certain cases, despite the objections of those farming communities relying on the same source for irrigation purposes, governments in affected areas have authorized water licenses for the use of such facilities (Kaack et al. 2022). This does not occur in a vacuum, however; there are multiple underlying factors in addition to data center activity contributing to water scarcity.

The Path Forward: What Must Change

Recognizing these problems is necessary, but not sufficient. There are concrete, technically viable strategies that can meaningfully reduce the energy and water footprint of AI infrastructure, if they are pursued with genuine commitment rather than treated as basic improvements.

• Waste heat recovery: Up to 90% of a data center’s electrical energy is ultimately converted to heat. Technologies including heat pumps, thermal storage systems, and advanced cooling architectures can capture and redirect that energy to district heating networks or nearby buildings, cutting net carbon output significantly (Zhang, et al., 2025).
• Renewable scheduling: Scheduling energy-intensive computing tasks to align with periods of high renewable generation on the local grid reduces reliance on fossil fuel backup capacity (Sheng et al., 2026).
• Responsible siting for water use: Locating new facilities in regions with abundant and sustainably managed water supplies, rather than in drought-stressed areas, prevents extraction from competing with existing community needs. For facilities operating in drier climates, dry cooling systems can dramatically reduce or eliminate freshwater withdrawal. These technologies exist today; the barrier is not technical feasibility but rather whether developers are required or incentivized to use them (Li et al., 2023).
• Efficient AI model design: Careful feature selection, removing unnecessary model complexity, and more efficient training protocols can reduce the energy consumption of model training by more than 90% in some cases (Grochulska-Salak, et al., 2025). Treating efficiency as a primary design goal from the outset, rather than optimizing for raw performance and addressing energy use as an afterthought, would produce compounding environmental benefits as these standards spread across the industry.
• Policy frameworks to enforce accountability: Efficiency gains alone are not sufficient. The rebound effect, a well-documented economic pattern, means that as AI becomes cheaper and faster to run, demand expands rapidly, and aggregate energy use can climb even as per-unit efficiency improves (Kaack et al., 2022). Carbon pricing, mandatory disclosure requirements, and enforceable regulatory standards are essential if technical improvements are to translate into actual reductions in climate impact.

Data centers are long-lived assets, typically operating for 20 to 30 years, and the choices made today about where and how they are built will be difficult and costly to undo. There is still time to get this right, but only if developers, funders, and regulators treat energy demand, water use, and environmental justice as core design constraints, not footnotes.

References

Grochulska-Salak, M.; Maciejewska, E.; Bujak, P.; Płoszaj-Mazurek, M.; Pękalska, M.; Amiri, O.; Rybak-Niedziółka, K.; Wężyk, T. 2025. “Green Data Centres: Sustainable Solutions with Green Energy and Green–Blue Infrastructure.” Energies 18(24):6592. doi:10.3390/en18246592.

Kaack, Lynn H.; Donti, Priya L.; Strubell, Emma; Kamiya, George; Creutzig, Felix; Rolnick, David;. 2022. “Aligning Artificial Intelligence with Climate Change Mitigation.” Nature Climate Change 12(6):518–527. doi:10.1038/s41558-022-01377-7.

Li, Chuang; Wang, Zhijia; Wang, Liping. “Factors affecting firms’ green technology innovation: an evolutionary game based on prospect theory.” Environmental monitoring and assessment, vol. 195, no. 1, 2023, pp. 227. ProQuest, 

Scherer, Glenn. “AI data center revolution sucks up world’s energy, water, materials.” Mongabay News [BLOG]Newstex, 2025. ProQuest, 

Sheng, Y.; Zhang, C.; Zhu, Z.; Xu, H.; Wen, J.; Wang, R.; Yang, J.; Wang, Q.; Bu, S. 2026. “Power for AI Data Centers: Energy Demand, Grid Impacts, Challenges and Perspectives.” Energies 19(3):722. doi:10.3390/en19030722.

Zhang, L.; Zhao, Z.; Chen, B.; Zhao, M.; Chen, Y.  2025. “Zero-Carbon Development in Data Centers Using Waste Heat Recovery Technology: A Systematic Review.” Sustainability 17(22):10101. doi:10.3390/su172210101.