Artificial intelligence and ecological destruction

By Ilker Kalayci

Translated from the Turkish journal Teori ve Eylem, No.62, Spring 2026

Introduction

The relationship between humans and nature has always been the subject of narratives since ancient times. For example, one of the fundamental dimensions of Gilgamesh, one of the oldest written texts, is related to the human-nature-culture triad, as Gezgin emphasizes. ^[1] Gilgamesh tells the story of the struggle of man, who lives as a part of nature, against his own nature, and the fate of man who is not content with what nature gives him and wants more. Throughout history, humanity has created another nature, expanding its own natural universe along with labour processes: culture and technology. But no period in human history has created such a destructive human-nature conflict and, in Marx’s words, such a great “metabolic rift” ^[2] as capitalism, which is actually a very short period historically. Perhaps today’s stories will consist not of myths created by man himself, but of the myths created by technology and, in particular, artificial intelligence, which is the “secondary nature” that man created.

When discussing technology within capitalism, it’s impossible to speak of it as a neutral and autonomous phenomenon. Since the emergence of capitalist production relations, it has always served functions such as increasing surplus value, attempting to exclude and replace labour power, and controlling labour power. We must also mention the creation of new areas of commodification where it believes it can generate profit, intra-capitalist class competition, and the drive to increase accumulation. Machines and automation have been the carriers of these movements since the inception of capitalism. Artificial intelligence is a contemporary representative of this trend.

The increasing prevalence of artificial intelligence (AI) is also raising debates about its environmental impact. Recently, the environmental impact of AI and related technologies has become undeniable. Yet, there are claims that its environmental impact is minimal or that, as a technology affecting everything, it will have positive ecological contributions. However, the direct environmental impact of AI is now undeniable and is steadily increasing. In this article, we will examine these concrete environmental footprints, claims of positive contributions, and their links to capitalism.

Where did this artificial intelligence come from?

The prevalence of artificial intelligence today and its penetration into every aspect of life has increased significantly with the release of tools known as generative AI. Large language models that generate text; models that generate images and videos have become widely used. For example, ChatGPT, one of the best-known large language model tools, reached 1 million users within 5 days of its release in November 2022. It currently has 900 million weekly active users and will likely be the earliest digital tool in history to surpass the 1 billion user mark. When using these tools, we feel like we are living in worlds with intelligent beings similar to humans, like in the stories told in science fiction, and that we are conversing with them. We are faced with something disembodied but artificial, something we don’t even realize is real. In this magical world, we may not see the background of the tools we use, the scenery behind their appearance, the labour processes, and their impact on nature. Yet, behind their history lie enormous networks of labour and energy.

The lightness of writing a ‘prompt’

The environmental impact of artificial intelligence is often underestimated in statements, reports, and scientific studies. Generally, the focus is on the individual and average environmental impact of the texts written by the user when making a request to the AI ​​tool, known as “prompts.” ^[3] Most studies perform calculations based only on simple text-based queries. However, longer questions and answers; requests to create images or videos; using different models such as reasoning, in-depth research, or different scenarios can also increase energy consumption. Artificial intelligence is not only found in chatbots; it has spread to every aspect of the digital world, including social media, online shopping, online publishing, translation, ticket searching, and banking. All of these can significantly increase environmental impact.

These are simply functions of using a trained model, the process of generating output from a pre-trained model based on a given input command, called inference . The responses returned by the tools are based on calling previously trained models. In training a model, the hardware works for months, receiving, processing, and performing calculations on training data. This is a time-consuming, expensive, and resource-intensive process. For example, it is estimated that training OpenAI’s GPT-4 model cost more than $100 million and consumed 50 gigawatt-hours of energy. ^[4] Model creators only release the model after this training, and they hope to recoup their enormous costs and eventually make a profit when users “infer” outputs from AI models.

It’s not just about inference; even when considering model training, individual and average values ​​don’t actually tell us much. Average values ​​might be an indicator of efficiency in capitalism. But for nature, not just the average, but the total quantity; how resources are used, where they come from; how energy is produced; and the state within the ecosystem are also important. Furthermore, this perspective, which reduces the issue solely to individual average values, confines everything to the logic of quantifying nature, while attributing the problem to individual consumption habits and concepts like “carbon footprint,” thus distancing the view from the social sphere and concealing the underlying total capital accumulation process and the real structural problems.

What exactly is artificial intelligence?

“Artificial intelligence” is a comprehensive set of technological tools that can automate mental tasks and perform functions that humans do mentally, such as learning, decision-making, and pattern recognition, encompassing various aspects of life. Technically, these tools are complex mathematical methods that establish correlations between data and often create patterns. To establish these correlations and create patterns, they need to go through a process called model training, using massive amounts of data. This data is produced and categorized by people all over the world; that is, there is a very large-scale labour behind it. ^[5] But beyond being a technical method, artificial intelligence, as Kate Crowford emphasizes, “is an idea, an infrastructure, an industry, a way of using power and a way of seeing the world; it is also a tangible manifestation of highly organized capital, supported by a vast system of resource extraction and logistics with supply chains that span the entire planet.” ^[6] The supply chain of artificial intelligence is a complex, global, and opaque mechanism that extracts, produces, and distributes the components that play a key role in the making of artificial intelligence today.

Based on the approach of media theorist Jussi Parikka, we need to think of technology not only as an extension of humans but also as an extension and component of the earth’s materials. ^[7] Information systems are the totality of geological, ecological and labour processes. These systems, in contrast to the volatility or eco-friendly appearance implied by the cloud metaphor, are based on a vast physical infrastructure deeply rooted in the earth, different types of labour forms and energy sources.

Excavation of the earth and minerals

Many materials are used to produce the hardware that forms the physical infrastructure for artificial intelligence and related digital technologies. At the heart of these materials lies a complex mineral ecosystem that enables digital performance, data storage, and high-speed connectivity. Rare earth elements, including silicon, cobalt, copper, gold, tungsten, lithium, and 17 types of metals, are of fundamental importance for various devices such as semiconductors, microchips, data storage, connectivity and cabling, power supplies, and batteries. Mining these minerals comes with significant environmental costs, including chemical waste, soil pollution, and the leakage of radioactive byproducts. It has been shown that 54% of these technologically critical materials are extracted from indigenous or peasant lands, and 62% from drought-prone regions. ^[8] These elements are essential not only for artificial intelligence and related technologies but also for electric vehicles and renewable energy systems. This means there is increasing demand and pressure for the extraction of these minerals. So much so that even proponents of capitalism are expressing concerns that this is an unprecedented attempt to scale up digitalization, decarbonization, and new technologies simultaneously, all fed by the same limited resource base. ^[9] Supply chains are complex and often not transparent. For example, Intel has more than 16,000 suppliers in over 100 countries; these suppliers provide materials directly for the company’s production processes, tools and machinery for its factories, and logistics and packaging services. ^[10] The secrecy and untraceability of these processes make it difficult to conduct a full lifecycle analysis of devices such as the hundreds of thousands of graphics processing units used to train AI models and make inferences.

‘Energy Vampires’ ^[11] Data Centres

Data centres, in their simplest form, are physical facilities that house an organization’s digital technology infrastructure, run applications, and manage data. Traditionally consisting of in-house hardware such as servers, these systems have evolved into large, integrated facilities. While data centres are not solely used for artificial intelligence tools, they are increasingly taking on this function, and even dedicated data centres are being built for this purpose. These AI-specific data centres are designed to run large-scale, computationally intensive systems such as those used for training and using large language models and computer vision systems. This requires high-performance hardware, energy to operate it, and cooling systems to absorb the heat generated.

Data centre energy consumption

In addition to the resources consumed during the construction of data centres, the amount of energy spent on manufacturing processors and training massive models over months is high. According to the International Energy Agency’s 2025 report ^[12] , data centres are responsible for approximately 1.5% of global electricity consumption in 2024. A typical AI-focused data centre consumes more electricity than approximately 75,000 households, and a large portion of this energy is generated from non-renewable sources: Currently, 30% of the electricity supplying data centres comes from coal, 26% from natural gas, 27% from renewable energy (wind, solar, hydro), and 15% from nuclear energy. Although companies state in their reports that they mostly operate on renewable energy by citing contract-based green energy purchase agreements, the actual energy mix that data centres draw from the grid is mostly generated by burning coal and gas because of their need for uninterrupted power.

Data centre water consumption

Data centres consume water through three main mechanisms: direct consumption via on-site cooling, indirect water consumption for electricity generation at power plants that supply energy to data centres, and hardware (semiconductor chips) manufacturing. High-performance devices operating on servers, necessary for AI training and inference, generate significant amounts of heat. Therefore, cooling systems are required to maintain optimum operating temperatures and prevent overheating. The majority of AI water consumption – 60% according to the International Energy Agency – occurs at power plants that supply electricity to the data centre. These plants use water to generate steam for turbines and require large amounts of chilled water for recondensation. In the semiconductor industry, ultra-pure water is used in almost all stages of manufacturing processes. According to the International Energy Agency’s AI and energy report, data centres require water consumption equivalent to that of 6,500 to 10,000 households. The water consumption of large data centres can be equivalent to that of a settlement with a population of 10,000 to 50,000. ^[13]

Data centre carbon emissions

Some AI-focused companies state in their sustainability reports that they are investing in renewable energy and committing to net-zero emissions. However, direct greenhouse gas emissions from AI appear to be increasing. Many AI companies have reported significantly higher greenhouse gas emissions compared to previous periods. This is due to the increased development and use of generative AI. For example, Google, in its 2024 annual environmental sustainability report, wrote of a 48% increase in greenhouse gas emissions since 2019. It attributes this “primarily to increases in data centre energy consumption and supply chain emissions.” ^[14] Microsoft also states in its 2025 report ^[15] that there has been a 23.4% increase in total emissions since 2020, and that this is “due to growth-related factors such as AI and cloud expansion.” In fact, an analysis conducted in 2024 revealed that actual emissions from Google, Microsoft, Meta, and Apple’s “on-premises” or company-owned data centres between 2020 and 2022 may have been approximately 662% higher than officially reported emissions. ^[16] This massive discrepancy between officially reported emissions and actual emissions stems primarily from methods known as “creative accounting.” Even if companies physically consume electricity based on fossil fuels, they can make their emissions appear to be zero on paper thanks to “Renewable Energy Certificates” purchased from clean energy facilities elsewhere in the world.

Beyond the companies’ own operations, the supply chain for AI hardware, including semiconductors and cloud server equipment, is a significant contributor to AI-related emissions. In 2024, supply chain emissions from AI giants accounted for a significant share. More than 80% of the total emissions of AI chip designers such as AMD, Nvidia, Qualcomm, and Broadcom originated from their supply chains. ^[17]

Indirect effects of Artificial Intelligence

Artificial intelligence and related technologies have direct environmental impacts as well as indirect effects on ecological destruction. Holly Alpine, who resigned from her position as senior program manager for Environmental Sustainability at Microsoft’s Data Centre Community, warns that the greater climate risk associated with AI stems from its indirect uses. ^[18] AI is enabling oil and gas companies to accelerate fossil fuel extraction and increase production by increasing precision and efficiency. Fossil fuel companies are using tools such as AI, cloud computing, and the Internet of Things at every stage of the oil and gas lifecycle. ^[19] AI tools are being used for tasks such as discovering more oil and gas, extracting hard-to-reach reserves, optimizing pipeline maintenance, and forecasting oil and gas demand. This leads to faster exploration, increased fossil fuel production, and ultimately, allows for the development of an industry with a criminal record for the climate crisis. It is thought that tools like artificial intelligence could help “build dozens of new facilities”, “drill thousands of new wells” and increase production “by up to 15 percent”. ^[20]

Companies like Microsoft, Google, and Amazon have established extensive partnerships with the world’s largest oil and gas companies. According to a 2022 study, Microsoft holds nearly 60% of all cloud service partnerships with the fossil fuel industry, including major companies such as BP, Shell, Chevron, Total, Equinor, and Repsol. Microsoft develops and provides AI and cloud tools to oil and gas producers to extend asset lifecycles, reduce costs, and unlock new supply opportunities. For example, a 2019 partnership between ExxonMobil and Microsoft aimed to increase oil production by up to 50,000 barrels of oil equivalent per day by 2025 using technologies such as cloud computing and artificial intelligence. ^[21] This goal was estimated to potentially lead to an annual emission reduction of 6.8 million metric tons of CO2 equivalent.

According to the “Amazon Unsustainability Report,” created by Amazon Employees for Climate Justice to expose Amazon’s sustainability lies, Amazon earns billions of dollars by selling AI services to fossil fuel companies, helping them extract more oil and gas. Amazon Web Services (AWS) estimated that it could generate $9.6 billion in annual revenue from the oil and gas sector in 2025. This represents approximately 10% of AWS revenue. ^[22]

Destruction covered by a curtain

Companies can evade transparency regarding data centres not only through accounting tricks but also through their relationships with local governments and bureaucracy, or by claiming “trade secrets.” Leaked data from Amazon, the company with the most data centres worldwide, reveals that Amazon has or uses far more data centres than previously known. ^[23] For example, while open sources show Amazon having 4 data centres in Germany, leaked data indicates this number is 50. According to the leaked data, the electricity used by these data centres in 2023 (1.3 million MW) is more than the energy needs of all households in Frankfurt, where most of these data centres are located. ^[24]

In Spain, three Amazon data centres are located close to each other in the cities of Zaragoza, Gállego and Huesca. ^[25] Although EL PAÍS newspaper requested data on changes in industrial water consumption from all three municipalities using the transparency law, only the municipality of Huesca provided data. According to this data, industrial water consumption in the city increased by 62 million litres per year after the Amazon data centre became operational. This increase is far above the 36 million liters per year estimate that the company predicted in its pre-construction reports.

A lawsuit in Oregon, USA, exposed massive water consumption and tax incentives imposed on the public by large data centres. ^[26] Google (first in 2006) and Amazon (first in 2011) had established large data centres in the area due to its energy infrastructure, water access, and ample land. When Google wanted to expand in 2021, it refused to disclose its water consumption; the local government also refused to disclose this information, citing it as a “trade secret.” When The Oregonian newspaper requested the records, the city sued the newspaper. Ultimately, the court ruled that the documents were “not a trade secret” and ordered their disclosure. Journalists discovered that Google used approximately a quarter of the city’s total water, about three times the amount used a few years earlier. It was also revealed that some public officials had engaged in irregularities in processes such as illegal tax advantages and land sales.

Can Artificial Intelligence save the Earth?

The argument that artificial intelligence will provide ecological benefits is based on the view that these technologies can be used in sustainable practices to combat the climate crisis. Generative AI methods, such as large language models, have limited applications related to sustainability or ecology. However, AI techniques have been shown to be applied in areas such as prediction, optimization, and data analysis in a broader sense. It is predicted that these methods will make a positive contribution to sustainability thanks to their ability to measure, monitor, predict, and optimize complex systems. Optimization applications are particularly effective in areas such as emission reduction, biodiversity conservation, energy, agriculture, traffic, and water management. Thanks to these applications, it is predicted that large-scale AI applications can reduce global greenhouse gas emissions by 0.9-2.4 gigatons of CO2 equivalent by 2030. ^[27] It is also stated that it can support the achievement of 79% of the targets under the United Nations’ 17 Sustainable Development Goals (134 targets). ^[28] As shown in an article that examined nearly 800 scientific studies and addressed artificial intelligence in sustainable development research, ^[29] there are very few studies that deeply integrate artificial intelligence methods with research related to the Sustainable Development Goal. The intersection applied to contribute to achieving the UN’s Sustainable Development Goals has been almost never realized to date. Even when used effectively in environmental applications, it is not sufficiently used in areas of social sustainability such as social equity.

Whether these positive contributions will offset the total environmental damage and emission increases that artificial intelligence will cause is still unclear. However, it is misleading to think of this as a plus-minus or a race. First, these studies are conducted within the existing socioeconomic horizon. This horizon includes an instrumentalist perspective that views technology as a neutral tool and above class. Current discussions on sustainable artificial intelligence are largely confined to this. It ignores the social dimension and class context of the problems. As in Bögner’s Key-and-Lock Model ^[30] , some technologies give the impression of fitting perfectly and frictionlessly to a challenge or task, like a “lock” and a “key”. Artificial intelligence is also presented as a “magic wand” for problems that are actually complex and rooted in production relations, such as climate change. This creates the perception that technology is the most suitable and effortless tool to solve the problem. This seemingly smooth fit actually excludes other valid options, namely socio-economic, political or behavioral solutions.

This perspective, a form of technological problem-solving, is based on the belief that complex social, political, and ethical problems can be solved through technological solutions, applications, or algorithms. Complex problems are reduced to engineering challenges that can be “solved” with an application or algorithm, shifting the focus from the root causes to the symptoms. By drawing attention to these symptoms, it prevents discussion of more fundamental systemic changes. In this respect, AI studies for sustainability play a critical role in creating an expectation that ecological, social, or economic problems can be solved with algorithmic optimizations and technological patches, rather than radical changes.

Furthermore, the construction process of artificial intelligence systems is based on the relentless effort to collect and quantify data about the world. “Dataification” of the world by bringing together all available resources has become a central necessity for the capitalist system, aiming to organize production in the most efficient way possible, with the goal of generating profit. In this context, artificial intelligence technologies are direct outputs of this “dataification culture.” ^[31] However, data has never been neutral. It will never be neutral, both because it originates from existing production relations and because raw data always requires at least a minimal level of human interpretation to gain meaning. ^[32] Data collection is a concrete manifestation of the desire to structure and control it, shaped by historical capitalist developments, and is guided by the logic of capital accumulation. ^[33]

Artificial Intelligence and Jevons Paradox

Making technical processes more efficient is also presented as a solution to reduce the ecological impact of artificial intelligence. However, as the phenomenon called Jevons Paradox shows, technological developments that increase the efficiency of using a resource lead to an increase, rather than a decrease, in the total consumption of that resource. In 1865, economist William Stanley Jevons observed that technological developments that increased the efficiency of coal use led to an increase in coal consumption in a wide variety of industries. This paradox shows that efficiency and technological developments can actually worsen our energy expectations. This paradox is based on a fundamental principle: every time you reduce the cost of consuming a valuable resource, people respond by consuming more. ^[34]

As John M. Polimeni et al. emphasize, Jevons Paradox shows that market-based solutions cannot solve today’s energy or related environmental problems. ^[35] We can observe Jevons paradox concretely in artificial intelligence as well. Google’s 2025 report states that they are innovating in hardware and software to enable data centres to offer six times more processing power per unit of electricity consumption compared to five years ago. However, the 2025 report states that there has been a 48% increase in carbon emissions compared to the target base year of 2019. They present the main reason for this increase as “increases in energy consumption and supply chain emissions in data centres”. As efficiency in AI training and inference increases, we can predict that these tools will be used more frequently and will permeate more places.

Resistance and struggles against data centres

The environmental impact of data centres established in different parts of the world has led to local resistance in many places. ^[36] In Chile, valuable groundwater resources are threatened with depletion. In South Africa, where power outages have long become routine, data centres are further straining the national grid. Ireland is one of the most data centre-dense areas in the world: in 2024, data centres consumed 22% of the country’s energy. ^[37] In this context, resistance has emerged in Brazil, India, the Netherlands, the United Kingdom, Spain, Malaysia, Mexico, and Singapore. Opposition to data centres largely stems from local concerns. While varying from region to region, some common points of contention include water depletion, excessive energy load, high bills, noise, and the protection of green spaces.

A study conducted by Data Centre Watch between May 2024 and March 2025 ^[38] examined 28 states in the US where large-scale data centre companies had data centre projects in the planning or development phase. In these states, 142 organizations/movements actively organizing to block new data centre developments were identified. During the study period, it was reported that $18 billion worth of data centre projects were blocked and $46 billion worth of projects were delayed due to local struggles. According to Data Centre Watch ^[39] , in the second quarter of 2025 (March 2025 – June 2025), an estimated $98 billion worth of projects were blocked or delayed. This is more than the total for all previous quarters since 2023. During this period, 66% of the protested projects were blocked or delayed. This is becoming an intensifying trend as political resistance strengthens and local organizing becomes more coordinated. Under pressure from these struggles, companies planning to build data centres were forced to make statements and commitments to be “good neighbors.” In its statements, Microsoft committed to not increasing electricity prices in the regions where data centres would be built, creating job opportunities for the local people, and minimizing its own water usage. ^[40]

Artificial Intelligence and the growth ‘mpasse of capitalism

The driving force of the capitalist mode of production is “ competitive accumulation of profit that will produce new capital formations in order to create ever more profit and accumulation .” ^[41] The system constantly seeks to overcome its structural limits through technological developments and spatial-market expansions. The ultimate goal of this cycle is to realize a higher surplus value and profit than the initial commodity-money relationship. This intrinsic dynamic of capital imposes the imperative of “grow or die.” Since there is no “saturation point” of accumulation in the capitalist movement, there is no amount of consumption that can be described as “sufficient” or “excessive.” Therefore, the ecological environment appears not as a common living space or a space with natural boundaries, but as an area of ​​exploitation that is contained and commodified in the process of the expanded reproduction of capital. We can see that artificial intelligence is also a part of the capital movement in capitalism, and is not free from the contradictions and dynamics created by capitalist production relations. Therefore, artificial intelligence will be a driving force in appropriating, using, and owning renewable resources and energy such as wind and solar power, as well as world resources like minerals and water, and indeed the entire natural world, just as it appropriates socially and historically produced data. This is, and will continue to be, an arena where capital’s relentless expansion and accumulation of wealth to seize a larger share of total surplus value is accompanied by fierce competition among capitalist classes.

This growth can also be seen in the International Energy Agency’s ^[42] projections: Demand for data centres is expected to almost triple by 2030. Data centre electricity consumption is projected to more than double by 2030, reaching approximately 945 TWh. This is slightly more than Japan’s total electricity consumption today. By 2030, companies worldwide are projected to invest approximately $7 trillion in building and upgrading data centres. Moreover, this growth is occurring for a new technology that has not yet fully found its use case. It may be the wrong tool for many applications, or at least there may be a less energy-intensive alternative. Artificial intelligence is a gamble that relies on a speculative and unproven hypothesis ^[43] called scaling, that is, the idea that AI models can be improved by training them with more data and using more hardware.

Conclusion

As stated in the book “Limits to Growth: A 30-Year Update” ^[44] , if a society’s implicit goals are to exploit nature, enrich capital, and disregard long-term consequences, then that society will develop technologies and markets that destroy the environment, increase inequality, and optimize short-term gains. In short, capitalism will develop technologies that accelerate destruction rather than prevent it, and will create markets to accommodate it.

Gilgamesh, whom we mentioned earlier, is also a symbolic representation of the transition from hunter-gatherer to settled agricultural life, marking a new era in the relationship between nature and humanity. ^[45] We need to imagine and struggle for a world where the alienation of human-to-human relations, which emerged in class societies, and the human-nature conflict/alienation, which reached its peak with capitalism, are eliminated. If we liken capitalism’s boundless pursuit of growth (infinity) to Gilgamesh’s pursuit of immortality, we should consider taking the “herb of youth” from Gilgamesh and condemning him to death. We must remember that we are now at a life-or-death turning point for nature and humanity, and that only we can make this turn. Perhaps then a new story of humanity can be written, not the myths written by artificial intelligence.

1. Gezgin, İ. (2009) Gilgamesh – The Mythic Hero of the Acculturation Process, Alfa Publications, Istanbul. ↩
2. Although Marx did not use this as a concept, Foster et al. developed it, attributing it to Marx. Foster et al. state: “…Karl Marx coined the term metabolic rift, or the idea of ​​a gap opened in metabolic exchange between humanity and nature.” According to Marx, the drive for capital accumulation “ reduces the agricultural population to a progressively decreasing minimum and confronts it with a constantly growing industrial population crammed together in large cities; in this way it produces conditions that cause an irreparable rift in the interconnected process of social metabolism (a metabolism formed by the natural laws of life itself). The result is the waste of the vitality of the land by being transported far beyond the borders of a single country through trade. ” Source: Foster, JB et al. (2021) The Ecological Rift: Capitalism’s War Against the Planet , Marx21 Publications , Istanbul. ↩
3. OpenAI’s ChatGPT, which is projected to generate $2 billion in monthly revenue and nearly 1 billion weekly active users by the end of March 2026, has an environmental impact as explained in parentheses in a blog post by OpenAI CEO Sam Altman: “ People often wonder how much energy a ChatGPT query consumes; an average query consumes approximately 0.34 watt-hours of energy. This is roughly equivalent to the energy consumed by an oven in a little over a second or a high-efficiency light bulb in a few minutes. It also consumes approximately 0.000085 gallons [yn, roughly 0.322 ml] of water; which is roughly one-fifteenth of a teaspoon .” Source: Altman, S. (2025) “The Gentle Singularity”, https://blog.samaltman.com/the-gentle-singularity . OpenAI does not have any publicly available emissions data or climate commitments. Source: “OpenAI Sustainability Profile”, DitchCarbon , https://ditchcarbon.com/organizations/openai . Google also reported a median energy consumption of 0.24 Wh and water consumption of 0.26 mL for Gemini text-based prompts. Source: Elsworth, Cooper et al. “Measuring the environmental impact of delivering AI at Google Scale.” arXiv preprint arXiv:2508.15734, 2025. EcoLogits calculations estimated that a small prompt could consume between 1.83 Wh and 6.95 Wh of energy. (Rincé, S. and A. Banse (2025) “Ecologits: Evaluating the environmental impacts of generative AI”, Journal of Open Source Software , 10(111):7471. ↩
4. James O’Donnellarchive and Casey Crownhart, “We did the math on AI’s energy footprint. Here’s the story you haven’t heard.”, May 20, 2025, https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/ . ↩
5. Cant, Callum et al. Feeding the Machine The Hidden Human Labour Powering AI , Bloomsbury Publishing, 2024. Gray, Mary L. and Suri, Siddharth. How to Stop Ghost Work Silicon Valley from Building a New Global Underclass , Houghton Mifflin Harcourt, 2019. ↩
6. Crawford, Kate. Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence , Yale University Press, 2022. ↩
7. Parikka, Jussi. A Geology of Media , University of Minnesota Press, 2015. ↩
8. Owen, John R et al. 2023. “Energy transition minerals and their intersection with land-connected peoples”. Nature Sustainability 6, 2, 2023, 203–211. ↩
9. Hebbar, Priya Agarwal. “Critical minerals: Why innovation begins beneath the Earth’s surface,” World Economic Forum, January 16, 2026, https://www.weforum.org/stories/2026/01/critical-minerals-innovation-earth-surface/ . ↩
10. Crawford, Kate. Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence , Yale University Press, 2022. ↩
11. This is the slogan of activists fighting against data centres in Ireland. Source: Climate Camp Ireland, https://climatecampireland.ie . The activists’ description of data centres as “vampire data centres” is also reminiscent of Marx’s view of capital in Volume 1 of Capital as ” dead labour that survives by sucking up living labour, and only by sucking up more and more . “
12. International Energy Agency, Energy and AI , April 2025. https://www.iea.org/reports/energy-and-ai . ↩
13. Yañez-Barnuevo, Miguel. “Data Centres and Water Consumption,” Environmental and Energy Study Institute, June 25, 2025, https://www.eesi.org/articles/view/data-centres-and-water-consumption . ↩
14. Google. Google Environmental Report 2024 , 2024, https://www.gstatic.com/ gumdrop/sustainability/google- 2024- environmental- report.pdf ↩
15. Microsoft, Environmental Sustainability Report 2025 , 2025, https://www.microsoft.com/en-us/corporate-responsibility/sustainability/report/ ↩
16. O’Brien, Isabel. “Data centre emissions probably 662% higher than big tech claims. Can it keep up the ruse?”, The Guardian, 15 September 2024, https://www.theguardian.com/technology/2024/sep/15/data-centre-gas-emissions-tech ↩
17. Wu, Katrin et al. “Supply Change: Tracking AI Giants’ Decarbonization Progress,” Greenpeace East Asia, 2025, https://www.greenpeace.org/static/planet4-eastasia-stateless/2025/10/128afb32-supply-change-2025.pdf ↩
18. Alpine, Holly. “Enabled Emissions: How AI is Being Used to Expand Fossil Fuels,” Climate & Water Risk, Rethink & Resilience, November 24, 2025, https://cwrrr.org/interviews/enabled-emissions-how-ai-is-being-used-to-expand-fossil-fuels/ . ↩
19. Sellien, Lars and Sharpe, Hannah. “Enabled emissions: How AI helps to supercharge oil and gas production,” Global Witness, January 27, 2026, https://globalwitness.org/en/campaigns/digital-threats/enabled-emissions-how-ai-helps-to-supercharge-oil-and-gas-production/ , ↩
20. Alpine, Holly. “Enabled Emissions: How AI is Being Used to Expand Fossil Fuels,” Climate & Water Risk, Rethink & Resilience, November 24, 2025, https://cwrrr.org/interviews/enabled-emissions-how-ai-is-being-used-to-expand-fossil-fuels/ . ↩
21. Microsoft, “ExxonMobil to increase Permian profitability through digital partnership with Microsoft,” February 22, 2019, https://news.microsoft.com/source/2019/02/22/exxonmobil-to-increase-permian-profitability-through-digital-partnership-with-microsoft/ . ↩
22. Amazon Employees for Climate Justice, “Burns Trust: The Amazon Unsustainability Report,” 2024, https://www.amazonclimatejustice.org/our-research . ↩
23. “Amazon’s hidden data centres”, Sourcematerial and Bloomberg, November 24, 2025, https://www.source-material.org/amazon-data-centres-energy-water-usage/ . ↩
24. age ↩
25. Arandia, Pablo Jiménez. “Descifrando el consumo de agua de la IA: así oculta Amazon cuánto bebe su nube en España”, EL PAÍS, 7 March 2025, https://elpais.com/tecnologia/2025-03-07/descifrando-el-consumo-de-agua-de-la-ia-asi-oculta-amazon-cuanto-bebe-su-nube-en-espana.html . ↩
26. Iruoma, Kelechukwu. “The Oregonian exposes Google and Amazon’s massive water use for data centres,” November 20, 2023, https://businessjournalism.org/2023/11/oregonian-data-centres/ .
    Rogoway, Mike. “The Dalles sues to keep Google’s water use a secret,” The Oregonian/OregonLive, November 1, 2021, https://www.oregonlive.com/silicon-forest/2021/11/the-dalles-sues-to-keep-googles-water-use-a-secret.html ↩
27. Herweijer, Celine et al. “How AI can enable a Sustainable Future,” PwC UK and Microsoft, 2021. ↩
28. Chaudhary, Gyandeep. “Environmental Sustainability: Can Artificial Intelligence be an Enabler for SDGs”, Nature Environment and Pollution Technology , Vol. 22, No. 3, 2023. ↩
29. Gohr, C. et al. “Artificial intelligence in sustainable development research”. Nature Sustain , vol. 8, 2025. ↩
30. Bögner, Frieder. “The Lock-and-Key Model of Technosolutionism: the Case of AI and Sustainability Challenges,” Philosophy & Technology , vol. 39, No: 22, 2026. ↩
31. Schütze, Paul. “The Problem of Sustainable AI: A Critical Assessment of an Emerging Phenomenon,” Weizenbaum Journal of the Digital Society , 2024. ↩
32. D’Ignazio, Catherine and Klein, Lauren F. Data Feminism . The MIT Press, 2020. ↩
33. Sadowski, Jathan. “When data is capital: Datafication, accumulation, and extraction”. Big Data & Society , 6(1), 2019. ↩
34. Polimeni, John M. et al., The Jevons Paradox and the Myth of Resource Efficiency Improvements , Earthscan, 2008. ↩
35. age ↩
36. Mozer, Paul et al. “From Mexico to Ireland, Fury Mounts Over a Global AI Frenzy,” October 20, 2025, https://www.nytimes.com/2025/10/20/technology/ai-data-centre-backlash-mexico-ireland.html ↩
37. The Central Statistics Office, Data Centres Metered Electricity Consumption, 2024, https://www.cso.ie/en/releasesandpublications/ep/p-dcmec/datacentresmeteredelectricityconsumption2024/ ↩
38. “$64 billion of data centre projects have been blocked or delayed amid local opposition”, Data Centre Watch, https://www.datacentrewatch.org/report ↩
39. “Q2 2025 UPDATE: 125% Surge in Data Centre Opposition”, Data Centre Watch, https://www.datacentrewatch.org/q22025 ↩
40. Brad Smith, “Building Community-First AI Infrastructure,” Microsoft, January 13, 2026, https://blogs.microsoft.com/on-the-issues/2026/01/13/community-first-ai-infrastructure/ . ↩
41. Foster, John Bellamy and Magdoff, Fred. What Every Environmentalist Should Know About Capitalism , Patika Kitap, 2014. ↩
42. International Energy Agency, Energy and AI , April 2025. https://www.iea.org/reports/energy-and-ai . ↩
43. Marcus, G. (2025) “The Dream of a Superintelligent Mind is Finally Shattered”, trans. TE Kalaycı, https://www.iyimserirade.org/eli-kulaginda-super-zekâ-hayali-sonunda-bozuluyor/ ↩
44. Meadows, Donella et al. Limits to Growth The 30-Year-Update , Chelsea Green Publishing, 2005. p: 223. ↩
45. Traveler, age. ↩