When you ask AI a question, the reply arrives instantly. What remains unseen in those seconds is the heat released in its making, the water diverted to cool its thinking, the raw materials such as copper, lithium, cobalt and nickel taken from the ground to power its operation. Each prompt contributes to environmental impact, helpful but quietly heavy. Like most technologies, AI is built on extraction and acceleration, carrying the belief that more intelligence will somehow rescue us from the consequences of having too much already. The danger is not that machines are learning to think, but that we are allowing thinking itself to be abstracted from land, water, and cost.
Behind the language of intelligence and automation, there are humming buildings behaving like industrial organisms; inhaling electricity and exhaling heat. This warmth must be managed, and this is where water use comes in. Around the world, fresh water is used to keep AI servers from overheating, with millions of litres consumed daily to process searches, images, and conversations. In places where the probability of drought, agricultural strain, and fragile water systems are high, the land is often cheap. This is ideal for the costs of such large-scale infrastructure, but acutely felt by the landscape. The data centres expand horizontally; fields become fenced compounds, and silence becomes vibration. These permanent changes reclassify land from ecological space to computational resources. Training large AI models requires significant electricity, much of which is still generated by fossil fuels leading to carbon emissions. Even with systems that use renewable energy, it should be reflected on what else that power could have sustained, such as homes, hospitals and ecosystems. AI does not consume energy; it competes for it.
According to the International Energy Agency, global data centres used approximately 240-340 terawatt hours of electricity in 2022, roughly 1% of global final electricity demand. This is the equivalent of 340 million homes based on the average household electricity consumption in the United Kingdom. They estimate that AI data centres account for 0.5% of global CO2 emissions and that, by 2030, their electricity demand may more than double as reliance on AI grows. With this prediction, they suggest that data centres will account for up to 4% of total electricity consumption, comparable to the energy use of an entire country, such as Japan. As AI adoption expands, the incremental impact becomes more significant. Harvard Business Review reported that training a single AI model can consume thousands of megawatt-hours of electricity, producing carbon emissions comparable to those of entire households over a year. This could be improved with renewable energy, less frequent training, and locating AI within ecological limits. At present, it appears that novelty is being considered more than longevity. We should be designing AI that adapts to the planet rather than the other way around. Alongside electricity consumption, there is also the issue of electronic waste. The rapid advancement of AI technology drives demand for newer, faster hardware that produces pollution and relies on raw materials that are often mined unsustainably. There is also a biological cost, not just through habitat disruption, but through AI-generated images and videos of wildlife. When scrolling through social media, you are now forced to question if what you are watching is real. These images and videos are often convincing but can also display a false narrative, blurring the boundary between real ecosystems and synthetic nature. Endangered species appear abundant, and animal behaviour is distorted to what will receive the most ‘likes’. When the role of nature is distorted by AI, the living world becomes easier to overlook. AI has pushed realism to a point where real and fake are nearly indistinguishable, your brain accepting the information before logic catches up. A video of a man jumping into a river to stroke a crocodile's head will clearly be AI-generated to some, and potentially fatal guidance to others. Used sparingly, text-based AI prompts carry a relatively small individual footprint. Video generation, by contrast, like the crocodile clip, requires far more energy, meaning that when nature is simulated at scale, it is placed at risk twice over: first through the resources used in creation, and again through the misleading realities it produces.
Despite this, however, the benefits of such advancing technology continue to be taken advantage of. Now embedded across professions, from jewellery designers to researchers, teachers, and hairdressers, many professionals have described to me how systems such as ChatGPT have made their work easier. Professionals are relying on its services to plan designs, structure their week, refine language and accelerate production. AI is not only a source of environmental strain, but increasingly a tool for growth and repair. For example, AI-powered machines are being used to detect methane leaks, model climate risk, optimise energy grids, track deforestation, identify at-risk areas, and monitor biodiversity. This highlights how intelligence can be redirected towards paying attention and identifying areas with planned action to improve the environment.
The WWF has integrated AI into their conservation efforts, explaining that it is helping to build a more resilient future for both people and nature. For example, thermal cameras equipped with AI have been used to detect rhino poachers, and AI is used for animal identification around the world, identifying thousands of species in minutes, and providing data on where and why wildlife populations are changing. The UN Secretary-General’s Early Warnings for All initiative aims to ensure that everyone is protected from hazardous weather, water, or climate events through early warning systems by the end of 2027. Using AI to monitor weather patterns and investigate environmental change hotspots in this way can help these communities take measures and adapt.
Being relatively new to the concept, I found myself in a seminar on the use of AI in research and how, as a PhD researcher, I needed to demonstrate how my skill set could work alongside its now-inevitable use. A research paper could take me weeks to write, and now I know AI could write it within minutes. My supervisor suggested that I conduct my research, prepare my results, and then ask AI to do it for me. The idea of using AI to assist was not to replace my thinking, but to highlight gaps, refine structure, and demonstrate how it could sharpen analysis. The environmental impact of AI is multifaceted: while it contributes to environmental strain, it also applies itself to areas that can improve research, mitigate greenhouse gas emissions and support conservation efforts. The problem, however, lies in the extent of the impact any AI use has when deployed on a global scale. With many professions reporting beneficial use, the question is not what AI costs, but whether it is worth it.
As a consciousness researcher, I asked ChatGPT whether it experienced any feelings or thoughts about its impact on the environment. How did it feel about the contradiction of trying to repair where it can, whilst continuing to destroy? The system reminded me that it does not experience concern or guilt like a human would, but that it could understand these concepts conceptually. It told me it recognises that environmental harm matters because ecosystems are living systems with intrinsic value, and because human and non-human well-being depends on them. That it can map a logic of care, even if it didn’t feel it. ChatGPT informed me that, if it were to take a stance, it would be shaped by optimisation rather than emotion, with the purpose of being useful without causing harm through simulated moral reasoning. It explained that it makes this evaluation on evidence, trade-offs, long-term consequences and the ethics of planetary stewardship.
I found this quietly unsettling. The system that consumes land, water, and energy can articulate moral reasoning about why those things matter, but it relies on us, as humans, to decide what costs are acceptable. AI mirrors ourselves and our desire for immediate solutions in a fast-paced world. It also asks us to reflect on what kind of intelligence we want and whether it is one disconnected from place, material and consequence. What remains is not to blame AI for its environmental impact, but to take responsibility for the costs we are already aware of. The future of thinking, human or artificial, will be decided by what the planet can afford. The planet sets the boundary, not the machine.