These days, you can find someone presenting a slide about artificial intelligence’s potential to save the planet at any major tech conference. The typefaces are tidy. The images are green. The optimism is as thick as a knife.
However, the image of the data centers humming in the distance—massive concrete blocks drawing electricity at a scale that would have seemed ridiculous ten years ago—is more difficult to ignore when standing outside one of those locations. The conference slides don’t fully convey the tension that exists here.
| Topic | AI and Climate Change — Promise vs. Environmental Cost |
| Key Organizations | UNEP, WRI (World Resources Institute), Grantham Research Institute, Google DeepMind, IEA |
| Key Figures | Roberta Pierfederici (Policy Fellow, Grantham Research Institute), Golestan “Sally” Radwan (Chief Digital Officer, UNEP) |
| Core Debate | Whether AI accelerates climate solutions or deepens the environmental crisis |
| Key Data Point | AI could reduce global emissions by 3.2–5.4 billion metric tons annually by 2035 |
| Major Risk | A single AI data center consumes as much electricity as 100,000 households |
| Water Concern | AI infrastructure may soon consume six times more water than all of Denmark |
| Electronic Waste | Data centers use rare earth elements, often mined in environmentally destructive ways |
| Policy Status | Mostly non-binding recommendations; formal guardrails remain limited |
| Reference | United Nations Environment Programme — AI & Environment |
The current AI boom’s environmental cost is no longer a specialized issue. Energy-intensive data centers are growing more quickly than most power grids can handle them, and locals are resisting tech companies constructing new facilities in areas where industrial pollution is already a problem. A typical AI-focused data center uses as much electricity as 100,000 homes, according to the International Energy Agency.
Twenty times as much will burn through the bigger ones currently being built. Peak electricity demand is expected to increase by 128 gigawatts in the US alone by 2029, primarily due to the computing infrastructure needed for AI. That is a seismic shift, not a footnote.
However, not everyone who works in climate research is concerned about AI’s potential in the future. Surprisingly, some of them are cautiously optimistic. Last year, Roberta Pierfederici, a policy fellow at the Grantham Research Institute on Climate Change and the Environment, co-authored a study in npj Climate Action that defied the general pessimism. By concentrating on just three industries—transportation, meat and dairy, and light road vehicles—the researchers calculated that by 2035, AI-driven developments could reduce greenhouse gas emissions worldwide by 3.2 to 5.4 billion metric tons annually.
If that amount came to pass, it would surpass the total anticipated emissions from data centers around the world during that time. It’s a powerful refutation. It also has a lot of requirements.
Crucially, the estimate is predicated on the idea that these tools will be used for purposes other than extraction and profit. That’s a significant assumption that isn’t at all assured. Currently, the majority of AI being developed and expanded is intended to produce income rather than lower carbon emissions.
The difference between what AI is actually being used for and what it could do for the climate is huge, and it seems to be getting wider every day. However, there are early warning signs to be aware of.
Already, some of the most important apps are operating quietly and unnoticed in the background. In order to reduce idling and exhaust emissions in ways that add up significantly at scale, AI models are being used to real-time adjust traffic signals throughout cities. AI-powered systems are being used in supermarkets and commercial kitchens to monitor and minimize food waste, one of the more neglected sources of emissions worldwide.
In order to model how ecosystems might change under various climate scenarios, conservation scientists are combining satellite imagery with machine learning. This type of ecological foresight was not previously computationally feasible. They’re not moonshots. These subtle, useful applications imply that the technology has actual operational value that goes beyond the hype.
AI may also have a real impact on the energy transition. Over the past ten years, renewable energy capacity—led by solar—has increased dramatically, and between now and 2030, it is expected to grow at a rate that is almost three times faster than it did in the preceding six years. However, it is truly challenging to integrate all that additional capacity into aging, fragmented power grids. Because renewable energy sources are sporadic—the sun doesn’t always shine and the wind doesn’t always blow—it has always been difficult to predict supply fluctuations precisely enough to maintain grid stability.
Pierfederici contends that AI could enhance demand forecasting and handle fluctuating supply in ways that facilitate a more seamless transition. She might be correct. It is another matter entirely whether governments and utilities will truly implement these tools at the necessary scale.
There is also a longer arc to take into account. Recently, 2.2 million new crystal structures were predicted by Google DeepMind’s GNoME project, a research tool that uses AI for material discovery. Of these, hundreds of thousands were stable enough to potentially power next-generation batteries and superconductors. These kinds of materials could significantly lower the emissions found in renewable infrastructure and electric vehicles, both of which presently rely on supply chains plagued by environmental and human rights issues, if they can be developed and produced on a large scale. Even measured skeptics are put on hold by this kind of outcome.
However, the risks are as real as the opportunities, and they should be given the same sincere consideration. Data centers generate electronic waste that is contaminated with lead and mercury. Approximately 800 kilograms of raw materials, many of which are mined in ways that destroy local ecosystems, are needed to build a single two-kilogram computer. In a world where 25% of people already lack consistent access to clean water, global AI infrastructure may soon use six times as much water as Denmark.
Then there are what researchers refer to as “higher-order effects,” such as AI being used to spread false information about climate change or self-driving car technology subtly encouraging people to switch from public transportation and bicycles to private vehicles, thereby increasing emissions through a door that no one was aware was open.
Golestan Radwan, Chief Digital Officer at UNEP, stated, “There is still much we don’t know about the environmental impact of AI, but some of the data we do have is concerning.” She pointed out that governments are racing to develop national AI strategies without giving environmental safeguards much thought. That’s a pattern with repercussions that often manifest later, subtly, and in difficult-to-reverse ways.
It’s difficult to ignore the fact that we’ve been here before. Every significant technological advancement is accompanied by claims about the problems it will solve, and sober accounting typically follows years later. This time, it feels different because of how quickly AI is being incorporated into the way we build, govern, eat, and power our homes.
Technology is not waiting for legislation to catch up. It is already present, embedded, and powered by grids that, in many places, still rely on natural gas and coal.
Can climate change be solved by AI? Definitely not automatically, and most likely not on its own. Can it worsen the situation? In certain respects, it already is. It might be easier to put it this way: AI is a catalyst, and what it accelerates depends almost entirely on what we point it at. There are no values in the technology. They are the ones who construct and implement it. Depending on how closely you’ve been observing, that distinction can be either reassuring or frightening.
