Technological progress at what cost? How AI is draining our planet’s resources
We live in a world where, if we are lucky (or unlucky, depending on your perspective), we have access to a phone and the internet. This means, every day, we use some form of artificial intelligence (AI). From the algorithm curating your perfect Spotify playlist, to Google Lens identifying the flower you had to know the name of, to the facial recognition unlocking your phone, AI is embedded into our daily lives. You can hop onto ChatGPT or any other large language model (LLM) for everything from writing love poems to having deep, existential late-night conversations about the state of the world. Want to ask Gemini about the weather or binge cat videos? AIs got you. AI has undoubtedly changed our world. But while we enjoy this digital renaissance, we must consider a cost of our silicon utopia: its environmental impact.
Humanity is already on the edge of several overlapping environmental crises – climate change, biodiversity loss, ocean acidification, and ozone degradation. We are all too familiar with the Big Oil and Big Agri(tech) narratives driving these crises, but the contributions of AI and Big Tech might not be on your radar.
So, what is the environmental impact of AI?
AI is a massive energy consumer and greenhouse gas emitter. The training of a typical LLM consumes around 626,155 kilowatt-hours (kWh) of electricity, approximately the annual electricity usage of 57 US households. Training ChatGPT-3, for example, consumed approximately 1,287,000 kWh of electricity and spewed out roughly 552 metric tons of carbon dioxide equivalents (CO₂), about the same as 123 gasoline-powered cars emit in a year. The operational electricity cost of ChatGPT amounts to around 4 million kWh every month. And every query you send to ChatGPT could emit anywhere from 0.2 to 0.9 grams of CO₂e.
Data centres, where AI models reside, also have massive energy use. In 2023, Google and Microsoft each consumed 24 terawatt-hours (TWh) of electricity. The total annual power consumption of global AI data centres in 2022 amounted to 460 TWh of electricity; if data centres were a country, this would make them the 11th largest electricity consumer in the world, behind France. Meanwhile, Big Tech is responsible for about 4% of global greenhouse gas emissions.
The electricity consumption of Big Tech is expected to rise considerably. In an attempt to address this issue, Big Tech companies are relocating to places like Iceland, where they can tap into renewable geothermal energy and use the naturally cool weather to chill out their systems. But this has its own problems, straining local energy infrastructure and potentially outpacing supply for local communities. Even places like London are feeling the strain from Big Tech’s energy-hungry data centres, while in the US, electricity utilities are overloaded with demands for new power capacity that they are struggling to keep up with.
What about water? AI data centres need water for cooling. If you think your laptop gets hot after a few hours of use, imagine that multiplied by millions of servers. Training ChatGPT-3, for example, is estimated to have used around 700,000 litres of water. A 20-50 question chat with an AI bot uses the same amount of water as a bottle of mineral water. Google’s US-based data centres alone consumed 12.7 billion litres of water in 2023, 90% of which was potable, while collectively data centres are estimated to use up to 5 million gallons per day and consume six times more water that Denmark annually.
This level of water consumption is especially problematic in regions already facing water scarcity. Currently, around 28 active data centres, operated by the world’s three largest tech companies, are located in water-stressed areas. Plans are underway for a 78% global increase in data centres, many of which are being proposed in some of the driest regions of our planet. For example, Amazon’s proposed data centres in Spain’s Aragon region are expected to consume approximately 755,720 litres of water annually, despite the area already being at risk of desertification. This expansion has triggered backlash from activist groups like Tu Nube Seca Mi Río (“Your cloud is drying my river”).
And of course, there is electronic waste, or E-waste, a massive environmental and health issue which we will save for another article.
As AI’s energy and water demands increase, even Trump recently acknowledged the issue, proposing that more coal is the solution (smart) and pledging to support the expansion of tech giants.
Now, imagine a dystopian world not driven by an AI uprising, but by resource depletion. Maybe the singularity will occur, maybe we will see rogue robots take over, or a nation-state hack its way into our data centres (again) but let us not ignore the insidious threat AI poses to our environment. As AI continues to demand more and more energy and water, could it create a future where human overlords buy up all available resources, starving us out and wiping out populations in a race to keep their servers running? Sure, we can hope for self-driving cars and virtual reality utopias, but at the end of the day, humans cannot eat their Spotify playlist or drink cat videos.
Before you run off screaming about a digital apocalypse, however, let us look at what Big Tech is doing to mitigate these impacts. In 2017 Google claimed to have achieved 100% renewable energy for its operations through corporate power purchase agreements. In 2020, Google, Microsoft, Apple, Amazon, and Meta collectively procured 7.2 giga-Watts of renewable electricity through power purchase agreements, contributing an additional 3.5% of global renewable energy capacity. China’s DeepSeek, a model that recently shook up the US market, seems to be an impressively efficient AI that suggests not all models need to devour the planet to function. Improving the location of data centres, smart scheduling of AI training, efficient model design, and green data centres run on renewable energy, are all approaches that can be adopted to reduce the environmental impact of AI.
AI can also be part of the environmental solution. AI can optimise energy efficiency in buildings and improve the efficiency of supply chains. For example, by using AI, one Manhattan building reduced energy consumption by 15.8%, saving $42,000 annually and cutting 37 metric tons of CO₂e emissions. AI can predict weather patterns and extreme weather events. For example, Google’s Flood Hub provides early flood warnings in multiple countries. AI can adjust energy distribution from renewables and monitor deforestation and illegal habitat destruction using satellite imagery. It can track endangered species (like the Geoffrey’s spider monkeys in Costa Rica), monitor ocean litter, and improve waste recycling efficiency. In agriculture, AI can analyse soil data, predict crop yields, and track disease outbreaks.
As we stand at the crossroads of technological progress and environmental catastrophe, we need to ask: Are we using AI to truly benefit humanity, or are we just indulging in digital parlour tricks while filling the pockets of a few? Beyond its environmental impact, the centralisation of AI in the hands of Big Tech raises questions about equity and access. As these companies expand globally, the benefits of AI risk being hoarded by the few, deepening socio-economic inequalities. It is time to ask how AI can serve the greater good, not just the interests of a handful of corporations. This calls for transparent policies, fair distribution of AI’s benefits, and involvement with local communities to avoid negative impacts. To date, the environmental impact of AI has been grossly neglected in legislation.
While most of AI’s environmental toll lies in massive data centres, individual digital habits still matter. Every conscious choice counts and helps steer us toward a more sustainable, equitable path in this new age of AI. So, what can you do to reduce your contribution to AI’s environmental impact? Here are some suggestions:
- Be a conscious digital consumer: Optimise your queries, avoid unnecessary interactions, and be mindful of how much you are using resource-intensive AI for trivial tasks. If you are going to ask an AI something, try scheduling those queries during off-peak hours when grids might have more renewable energy to spare.
- Choose energy-efficient devices: Pick laptops and smartphones that consume less energy. Digital minimalism, which involves reducing the number of devices you upgrade and maintaining them for longer, can make a big difference.
- Support green tech initiatives: Choose providers committed to sustainability, get involved with advocacy groups, and use your voice to raise awareness about AI’s environmental impact.
Ultimately, when it comes to the environment, AI has a lot of potential to radically change the way we monitor, control, and mitigate our critical environmental crises. But we must consider the flip side to this potential, the very real environmental impact of AI itself, and how this often affects the most vulnerable societies. The dystopic future of AI might not be environmental, but we would be remis not to consider and act on its impacts, steering our technological progress towards a more sustainable and equitable path.
