Our Deadly Addiction

When we speak of the widespread use of AI today, we typically refer to large language models (LLMs). ChatGPT and other GenAI models with wide-ranging abilities are a result of recent advancements in machine learning. Specifically, the massive increase in the number of parameters that can be run in parallel has allowed the models to process a much larger (almost infinite) amount of data. This is what differentiates them from the AI of the past, like earlier versions of Siri. The more tasks a model can do, the more processing power and data it needs, and, to access so much data all the time, very resource-intensive large language models rely on external data centers. It can be difficult to conceptualize data in this way, so imagine that an AI model is a car and data is the gas; more complex models are like giant gas-guzzling trucks, and as our demand and reliance skyrockets, our need and consumption of gas/data increase. This metaphor of data as oil goes even further, because our reliance on these two resources follows a very similar pattern: a new technology that revolutionized efficiency; reports that they’re bad for the planet and our health, which we ignore because we have all forgotten that it was ever possible to live without it. Both cater to our obsession with efficiency, driven by corporations that exploit our desire for instant gratification while selling us products made at the direct expense of our communities’ welfare. They are like Pandora’s box: once it’s been opened, there is no going back. The AI boom is scarily similar to that of fossil fuels, a reliance that has single-handedly destabilized the entire climate of the planet and cost millions of lives, both human and animal. We are at a crucial crossroads, where we get to decide if we let LLM reliance be the next climate catastrophe.

  So why is the rapid expansion of AI so problematic? As previously mentioned, the complex models we have become so reliant on require a massive amount of data and processing power, which is currently handled externally in data centers. Data centers are made up of large-scale storage and processing systems, housing the most important resource of the digital age: computing power. Due to the constant demand for data, these centers operate at very high intensity, requiring a massive amount of energy. Not only has this driven up overall energy demand, but it has also strained power grids and caused skyrocketing energy prices. In 2024 alone, the world’s data centers consumed an estimated 415 terawatt hours of electricity, enough energy to power 38 million American households for a year. A 2023 government-funded report by the Lawrence Berkeley National Laboratory projected that data centers could account for anywhere between 6.7 and 12% of US Energy Consumption by 2028. In order to meet this explosive demand, many data companies have quietly doubled back on their renewable energy goals. While 2024 saw massive growth in global renewable energy capacity, it also saw record-high energy-related CO2 emissions. With data centers getting an estimated 56% of their energy from fossil fuels, it is no surprise that our energy emissions keep rising. Thus, we see our two deadly dependencies, data and fossil fuels, intertwined further. 

A second concern with data-center expansion is the immense amount of water used to cool processor chips that heat up due to their intense processing load. In 2023, data centers directly used 64 billion liters (17 billion gallons) of water for cooling, as well as an additional 800 billion liters (211 billion gallons) indirectly. This consumption will only increase as demand for processing power grows and data centers continue to expand. As if it wasn’t enough for data centers to massively consume one of our most valuable resources–a resource that millions worldwide do not have consistent access to–many of these data centers are located in already water-scarce regions. Analyses found that Amazon, Google, and Microsoft (“the big three” of the data world) have 38 active data centres in water-scarce regions, with 24 more in development. Slowly, we begin to see why AI use is so much more than a personal ethical issue, or even merely an environmental one: it is a question of human rights, of who gets access to the world’s resources and who doesn’t. 

A final aspect of data centers that is not discussed nearly enough is their heat emissions. While a massive amount of water is used to cool the technology, a considerable amount of heat still escapes from data centers (by the nature of thermodynamics, though I won’t bore you with physics). In 2016, studies suggested that waste heat from EU data centers alone had reached 56 TWh, enough to heat almost 6.8 million EU households for a year; now imagine what this value would be on a global scale, 10 years later. Not only is this problematic for communities living nearby who have to deal with the adverse effects of thermal pollution, but this also has implications for global climate change as a whole. While data centers’ use of fossil fuels for power drives greenhouse gas emissions that accelerate global warming, they have also found a way to skip a step and heat the planet directly.

Beyond just data centers, the rapid expansion of the AI industry also means more electronic waste due to the pace of innovation that makes technology quickly obsolete, and the fact that AI chips are harder to recycle. Most high-consuming nations don’t deal with their own E-waste: up to 75% of the world’s E-waste is shipped to countries in Africa and Asia. Thus, people who contribute little to the global flow of E-waste are forced to deal with the environmental and health repercussions.

With all these harmful impacts of data centers laid out, we must ask ourselves: which communities are going to bear the brunt of AI infrastructure? Whose energy bills will spike? Whose water supply will be depleted? In what communities will sprawling heat-emitting data centers be built? If the fossil fuel industry is any indication, these impacts will fall most heavily on already disadvantaged communities: on people of color, lower-income communities, and indigenous peoples. Once again, the luxury of efficiency for the more privileged comes at the expense of the marginalized. We’ve seen it all before, and we cannot fall for it again. 

So how can we avoid making these same mistakes? First, we must all recognize the harmful effects of AI expansion and refuse it whenever possible. Individual usage may not seem to make such a big difference, but it is our demand that is leading to AI’s integration into every part of our lives. Besides the fact that recent research suggests that relying on generative AI harms our ability to retain information and dampens our ability to think critically, any time saved using ChatGPT is quite irrelevant if the planet becomes unlivable. So make the choice for yourself and the planet to reject generative AI usage. A second important lesson we can learn from the damages caused by the fossil fuel industry is that we must regulate the harmful impacts before it is too late. To put it bluntly, the current exponential model of data center expansion is incompatible with the renewable energy transition–they are too energy-intensive, and companies simply do not care about sustainability goals when there is profit to be made. Policy can make them care. This means restrictions on water use, electricity consumption, and all the other harmful and extractive practices related to the data industry. At the same time, not all AI is created equal, and the development of more sustainable models must be incentivized. There is still the possibility that with more careful planning and community input, data centers can be powered with renewable energy sources and that water consumption and heat emissions can be minimized. Importantly, we must curb our reliance on extremely resource-intensive large language models and shift our innovative focus to small language models. There are strong forces, backed by world leaders and big money, opposing AI regulation, but just like with the fossil fuel industry, united community power can be stronger. Advocating for robust regulatory policy is the only way to make AI advancement compatible with a livable future. 

When I started my internship, I could not have imagined the extent of the data industry’s implications on environmental justice and human rights. It was truly an eye-opening experience, a revelation I hope to bring to others. While learning about these destructive impacts was disheartening, it also imbued me with a renewed commitment to living my life ethically and sustainably. Once I was faced with the hard truths, I was given the chance to reaffirm my rejection of the ruthless demand for efficiency that got us into this mess, a decision I make over and over again every time I refuse to use generative AI. Whether or not the AI economic bubble “bursts” in the coming years, 2026 must be the year where we burst it culturally. I just hope that as a society we can find the strength to overcome the inertia of our efficiency obsession, and instead strive to create a better world that will outlast us.

Photo Credit: https://www.instagram.com/p/CEXCeWqF4CN/?short_redirect=1

Graphic by Emma Weibel

Other posts that may interest you:

• We Are All Epstein Victims
• No, Caring About the Planet Is Not A Hobby!
• The Freedom We Take For Granted
• The Worst Sport of Them All