The Environmental Costs of Artificial Intelligence

Artificial intelligence (AI) is advancing at breakneck speed, and researchers are still working out what this technology means for humanity. What they're finding isn't all good news. Alongside its benefits, AI has a less-discussed side: its heavy environmental costs. Here's what you need to know.

The world is watching

At its core, AI's value lies in its ability to spot patterns in data and use historical information to predict future outcomes. Given those capabilities, some governments and environmental groups once expected AI to be a major asset in tackling the climate emergencies the world faces. And it has proven useful for mapping and monitoring environmental damage. However, experts are increasingly clear-eyed about the trade-offs that assistance comes with.

In December 2025, the UN Environment Assembly held its seventh session in Nairobi, where the environmental impact of AI took center stage. Member countries adopted the first-ever UN resolution on the environmental sustainability of AI systems—a landmark step, though a more ambitious provision requiring monitoring of AI's water and power use across its full lifecycle was dropped after objections from Russia, Saudi Arabia, and the United Arab Emirates during negotiations. The resolution still asks the UN Environment Programme (UNEP) to produce a report on AI's environmental benefits, risks, and impacts, and encourages governments, industry, and researchers to work together on the issue.

AI servers and electronic waste

AI runs on servers housed in large data centers, and the hardware behind them generates electronic waste that can be hazardous, releasing substances like lead and mercury into the environment. As computing power grows and more data is gathered to train AI models, the volume of e-waste grows with it.

Servers, graphics processing units (GPUs), and related equipment typically last only two to five years before needing replacement. If that pace of turnover continues, research published in Nature Computational Science projects that generative AI alone could produce between 1.2 and 5 million metric tons of e-waste by 2030.

Some organizations are working to offset this. The National Center for Electronics Recycling (NCER) in the United States, for example, helps recycle or refurbish electronics that businesses replace. Programs like this can divert a meaningful share of e-waste from landfills while supplying schools and community groups with refurbished computers, laptops, and monitors they might not otherwise be able to afford.

Data centers demand large amounts of natural resources

AI's environmental footprint primarily shows up in its appetite for natural resources.

Water is a resource data centers consume heavily, both during construction and, once operational, to cool electrical components. Projections from UC Riverside and UT Arlington researchers estimate the water needed to support AI infrastructure worldwide could reach four to six times the annual water consumption of Denmark's roughly six million residents by 2027—though estimates like this vary depending on methodology. Even if this estimate proves to be an overstatement, water use by AI data centers represents a drain on local resources.

Electricity is a similar story. AI consumes vast amounts of electricity—even compared to the power demands of the technology industry over the past 10-20 years. In a May 2025 report in MIT Technology Review, researchers found that from 2005 to 2017, power use by data centers did not increase significantly, even as the rise of cloud computing and digitization sparked a construction boom in new facilities. Efficiency gains kept electricity use by data centers relatively flat.

Beginning in 2017, this changed with the increase in energy-intense hardware built to accommodate AI’s computing demands. By 2023, data centers had doubled their electricity use. The following year, data centers in the US represented 4% of the country’s total electricity consumption, an amount roughly equivalent to the power usage of Pakistan.

A 2025 Pew Research Center look at the energy use of the largest data centers is also illustrative. A “hyperscale” data center refers to a warehouse-sized facility with at least 5,000 servers—though many contain far more—that support massive computing workloads. A hyperscaler consumes as much electricity as 100,000 households per year, and larger ones under construction may use 20 times that amount.

If these data centers were powered by renewable energy, the environmental impact might be less concerning. However, most of that electricity still comes from burning fossil fuels. The MIT Technology Review article noted that the carbon intensity of energy consumed by data centers was 48% higher than the US average. This means more greenhouse gases released into the atmosphere as AI use scales up.

Other future problems

Outside of its environmental impact, there are of course huge concerns about AI’s impact on labor, the economy, education, the media, democracy, and even how we think, reason, and perceive reality. There's concern that AI could be used to spread misinformation about issues like climate change, leading the public to underestimate real threats. And as AI-powered self-driving vehicles become more available, some speculate they could encourage more driving rather than more use of public transit or cycling—working against efforts to cut greenhouse gas emissions.

In closing

AI represents a major shift in human history. The technology has benefits, but its environmental consequences are significant and just beginning to be understood. As AI increasingly becomes a part of everyday life, we owe it to ourselves—and our planet—to stay informed about its true costs.

Next
Next

Why Mountains Are Warming Faster Than the Rest of the Planet