{"id":15485,"date":"2026-05-29T15:24:11","date_gmt":"2026-05-29T13:24:11","guid":{"rendered":"https:\/\/wide-vision.pl\/bez-kategorii\/how-does-artificial-intelligence-affect-the-environment\/"},"modified":"2026-05-29T15:24:11","modified_gmt":"2026-05-29T13:24:11","slug":"how-does-artificial-intelligence-affect-the-environment","status":"publish","type":"post","link":"https:\/\/wide-vision.pl\/en\/green-vision-en-2\/how-does-artificial-intelligence-affect-the-environment\/","title":{"rendered":"How Does Artificial Intelligence Affect the Environment?"},"content":{"rendered":"\n

Artificial intelligence, also known as AI, is not “intangible.” Each query for models like ChatGPT, Gemini, or Grok requires data centers, which consume electricity and vast amounts of water to cool the infrastructure. The problem isn’t the individual query itself, but the scale of AI’s global use.<\/strong> <\/p>\n\n\n\n

AI Does Not Work “in the Cloud”, but in Data Centers<\/strong><\/h2>\n\n\n\n

Users often perceive AI as a digital service without a material cost. After all, they can use it without paying for a monthly subscription. Some companies deliberately encourage users to use their AI services, such as Google, which displays the Gemini chatbot at the top of search results, above human-written articles. In practice, however, using AI is not free – generative models are powered by energy-intensive data centers that have a negative impact on the environment.  <\/p>\n\n\n\n

AI doesn’t work “in the cloud,” but in physical buildings full of servers that need to be constantly powered and cooled. AI models like ChatGPT, Gemini, and Grok run on specialized computers equipped with  Graphics Processing Units<\/strong> (GPU) \u2013 graphics processing units.  Although GPUs were originally designed for graphics and computer games, today they are crucial for AI because they can perform a huge number of calculations simultaneously. It is this parallelism that makes them suitable for training and running large language models. <\/p>\n\n\n\n

This can be compared to the difference between a standard CPU (Central Processing Unit), which works like a single expert solving tasks one by one, and a GPU, which works like thousands of specialists working in parallel. In practice, AI data centers contain thousands of GPUs connected in clusters, running 24 hours a day. Modern GPUs can draw hundreds of watts per processor, and entire clusters can draw megawatts. <\/p>\n\n\n\n

Environmental Importance of GPUs<\/strong><\/h2>\n\n\n\n

The larger the AI \u200b\u200bmodel and the more users use it, the more GPUs must work simultaneously, increasing power consumption and the need for cooling. AI servers produce enormous amounts of heat \u2013 GPUs perform trillions of mathematical operations while running, converting a significant portion of electrical energy into heat. Without cooling, the hardware’s temperature could quickly exceed safe levels and lead to failure. <\/p>\n\n\n\n

That’s why data centers utilize extensive cooling systems. Traditionally, air cooling was used – fan systems, industrial air conditioning, and cold air flow between server rows.  Today, however, water cooling is increasingly used in data centers.  Water or special liquids  remove heat from servers,  transport it to cooling systems, and  maintain a stable temperature. <\/p>\n\n\n\n

This is where the environmental cost of AI comes in.  In many data centers, some water  evaporates,  needs replacement, or is constantly drawn from local water supplies.  Therefore, the development of AI infrastructure may increase pressure on regions exposed to water stress.<\/strong><\/p>\n\n\n\n

AI and Water and Abiotic Stress<\/strong><\/h2>\n\n\n\n

Water stress is a condition in which water demand exceeds water availability or its quality deteriorates. AI can exacerbate this condition by: <\/p>\n\n\n\n

    \n
  • increase in water consumption by AI infrastructure<\/li>\n\n\n\n
  • local concentration of technological infrastructure<\/li>\n\n\n\n
  • increased energy demand<\/li>\n<\/ul>\n\n\n\n

    Overexploitation of water resources can exacerbate natural phenomena affecting living organisms, including abiotic stress<\/strong> \u2013 the response of plants and ecosystems to unfavorable natural conditions, such as drought or water shortage.
    Learn more on “Stop Drought! Start Retention!” website. <\/em><\/p>\n\n\n\n

    AI Models Consume Energy Even for Learning<\/strong><\/h2>\n\n\n\n

    The AI \u200b\u200b”learning” process, i.e. the so-called model training, <\/strong> is the process by which AI begins to recognize patterns in massive datasets. For language models, this means analyzing books, websites, articles, code, and much more. During training, the model repeatedly predicts future words and corrects its own errors. This is a very long process, not only computationally intensive but also requiring a huge number of GPUs. It can take months, even with thousands of processors working simultaneously. <\/p>\n\n\n\n

    The model learns before responding to users. This is one of the most energy-intensive stages of AI development. Its environmental costs include  energy for servers and  cooling, and  indirectly, water used by the power system and data centers. <\/p>\n\n\n\n

    Carbon Footprint Left After a Simple Inquiry<\/strong><\/h2>\n\n\n\n

    Once training is complete, the model moves on to the inference <\/strong>stage \u2013 responding to users daily.  This is what happens when someone:<\/p>\n\n\n\n

      \n
    • asks ChatGPT for a recipe<\/li>\n\n\n\n
    • uses Gemini to search for information<\/li>\n\n\n\n
    • generates text or image in AI<\/li>\n<\/ul>\n\n\n\n

      The model then does not learn from scratch \u2013 it uses already trained knowledge to generate an answer.  Inference is  less energy-intensive than training a single model,  but it takes place incomparably more frequently.  And this is where the scale effect comes into play.<\/p>\n\n\n\n

      A single AI response may require only a small amount of energy and cooling, but millions of users, billions of prompts, and 24\/7 service operation mean that total energy and water consumption become significant. Daily processing of the massive number of queries requires continuous operation of data centers. It is inference <\/strong>that makes AI a permanent component of the global demand for energy and water resources. <\/p>\n\n\n\n

      Energy and Water Consumption of the Most Popular Chatbots<\/strong><\/h2>\n\n\n\n

      First of all, we must point out that the data is partially confidential, and companies use different methodologies to calculate their carbon footprint. <\/p>\n\n\n\n

      ChatGPT<\/strong><\/h3>\n\n\n\n

      OpenAI’s first research suggested that 5 to 50 ChatGPT queries may require approximately 500 ml of water <\/strong> (directly and indirectly through cooling and energy production). However, more recent OpenAI declarations indicate that  the average ChatGPT query requires: <\/p>\n\n\n\n

        \n
      • approx. 0.34 Kwh of energy<\/strong><\/li>\n\n\n\n
      • approx. 0.32 ml of water<\/strong> <\/li>\n<\/ul>\n\n\n\n

        It is worth emphasizing here that the data concern a single query and do not include the full costs of model training and infrastructure.<\/strong><\/p>\n\n\n\n

        Gemini <\/strong><\/h3>\n\n\n\n

        Google has published its own calculations for Gemini:<\/p>\n\n\n\n

          \n
        • approx. 0.24 Kwh of energy<\/strong><\/li>\n\n\n\n
        • approx. 0.26 ml of water per prompt<\/strong> (“approximately five drops”) <\/li>\n<\/ul>\n\n\n\n

          However, experts point out that the methodology may not include the total water footprint, particularly water used indirectly by power plants and the full energy infrastructure. <\/p>\n\n\n\n

          Grok <\/strong><\/h3>\n\n\n\n

          xAI doesn’t release full environmental data for Grok. According to analyses, xAI’s Memphis data center can draw about 1.3 million gallons of water per day<\/strong> (about 4.9 million liters) to cool its infrastructure.  <\/p>\n\n\n\n

          The Impact of AI on the Local Environment<\/strong><\/h2>\n\n\n\n

          Iowa (USA) \u2013 When AI Developments Impact Local Water Resources<\/strong><\/h3>\n\n\n\n

          Iowa, and West Des Moines in particular, has become one of the most frequently cited examples of the environmental cost of AI. It’s home to Microsoft’s databases used to develop ChatGPT-related technologies, as well as Google’s infrastructure. The servers running their language models generate enormous amounts of heat and require intensive cooling, often using water. <\/p>\n\n\n\n

          Research on Microsoft\/OpenAI infrastructure indicated that generating AI responses could be related to water consumption for cooling and energy production, and Microsoft itself reported a  34% increase in global water consumption between 2021 and 2022<\/strong>, which the company partially attributed to AI developments. In Iowa, this has sparked a debate about how large water centers impact local water resources – especially during periods of high temperatures and increasing drought risk. “How much technological infrastructure can a local water system support?” is a question local residents are asking themselves as water shortages and droughts begin to weigh on their concerns. <\/p>\n\n\n\n

          Memphis, Tennessee (USA) \u2013 Grok and the Social Conflict Surrounding the xAI Data Center<\/strong><\/h3>\n\n\n\n

          Memphis has become one of the most controversial examples related to the development of artificial intelligence and the activities of xAI, the company developing Grok. The Colossus supercomputer<\/strong>, which runs the AI \u200b\u200bmodels, requires enormous amounts of energy and intensive cooling. According to analyses and media reports, the facility can consume approximately 1-1.3 million gallons of water per day<\/strong> (approximately 3.8-4.9 million liters), and its energy demand is estimated at up to 1.1 GW, <\/strong>comparable to that of a large city.  <\/p>\n\n\n\n

          The development has sparked protests from local communities, particularly in South Memphis, where residents are concerned not only about water resource extraction but also about air quality and environmental  justice. Community organizations have accused xAI of operating dozens of gas turbines without full environmental permits, citing additional pollutant emissions.  <\/p>\n\n\n\n

          Virginia, Oregon, and Ireland \u2013 AI Infrastructure Straining Power Grids<\/strong><\/h3>\n\n\n\n

          The impact of AI is not limited to individual investments. A growing body of research indicates that the concentration of data centers<\/strong> in specific regions is becoming a problem. Virginia and Oregon in the US, as well as Ireland, are among the locations particularly vulnerable to the growing energy pressures associated with AI. Forecasting models suggest that the development of AI-generative infrastructure in these areas could lead to local grid congestion, increased competition for energy, and an increased risk of conflicts over water and energy resources. <\/p>\n\n\n\n

          Ireland is a particularly striking case in this regard. According to data from the Irish Office of Statistics, AI infrastructure is responsible for  22% of the country’s total measured electricity consumption in 2024<\/strong>, compared to only 5% in 2015. Energy consumption by this sector increased by  531% between 2015 and 2024<\/strong>. This scale means that data centers are beginning to be perceived not as a marginal element of the digital economy, but as infrastructure with a real impact on energy security and national climate policies. <\/p>\n\n\n\n

          Artificial Intelligence and Sustainable Development<\/strong><\/h2>\n\n\n\n

          The growing impact of AI on the environment doesn’t mean that artificial intelligence is at odds with sustainable development. It’s increasingly emphasized that what matters most is not the use of technology itself, but how it’s designed and communicated. A responsible approach includes, among other things, greater transparency from technology companies regarding energy and water consumption, choosing more efficient infrastructure, developing renewable energy sources, and consciously reducing the digital carbon footprint. AI can support environmental protection \u2013 helping with climate data analysis, managing energy networks, and monitoring water resources \u2013 but its development should be coupled with genuine responsibility for the resources used. <\/p>\n\n\n\n

          This approach is also in line with the activities of Green Vision,<\/strong>an initiative implemented by the Wide Vision Marketing Agency, which focuses on communicating ecological projects, environmental education, and building public awareness around water management and climate change. Our agency’s portfolio includes activities related to social campaigns and projects related to retention and counteracting the effects of drought, such as “Stop Drought! Start Retention!” and “Water Matters.” In this context, the conversation about AI and the environment is not limited solely to criticism of technology, but to the question of how to address growing problems more responsibly – taking into account limited natural resources and the long-term effects on ecosystems.<\/p>\n\n\n\n

          For Green Vision, sustainable development means not only promoting pro-ecological activities but also addressing topics that are only just beginning to gain wider attention \u2013 such as the ecosystem costs of digitization and artificial intelligence. In a world where data centers are becoming as crucial an infrastructure as power plants or water networks, an informed debate is needed about how to reconcile technological innovation with the protection of natural resources and water security for future generations. <\/p>\n\n\n\n

          AI’s Carbon Footprint and the Hidden Cost of Language Models<\/strong><\/h2>\n\n\n\n

          The dynamic development of artificial intelligence is increasingly being analyzed not only through the prism of innovation, but also through the prism of the impact of AI on the environment<\/strong>. Although AI-based<\/strong> tools support business, education, and science, their operation is associated with increasing energy consumption and emissions. Experts point out that AI’s carbon footprint<\/strong> is becoming a significant challenge for sustainable development,<\/strong> especially as global energy consumption<\/strong> continues to rise and a significant share of energy production still relies on fossil fuels<\/strong>. The question is no longer just what AI models<\/strong> can do, but also how their operation impacts the environment<\/strong> and our planet’s resources.<\/strong> <\/p>\n\n\n\n

          One of the most energy-intensive steps is training a large model. <\/strong>During this process, the algorithm <\/strong>analyzes large datasets and learns to predict subsequent words or patterns.  The more data the model has, the  greater the number of parameters<\/strong>, the greater the computational requirements.  A good example is GPT-3<\/strong>, ChatGPT’s predecessor.<\/strong> According to widely cited analyses, training <\/strong>it consumed approximately 1,287 MWh<\/strong> of electricity and generated up to 550 tons of CO\u2082,<\/strong> hundreds of tons of carbon dioxide <\/strong>equivalent to the annual emissions of many households<\/strong>. This demonstrates that training a single large AI model <\/strong>can result in a huge carbon footprint,<\/strong> even before the system reaches users. Therefore, the topic of greenhouse gas emissions<\/strong>, carbon dioxide emissions<\/strong>, and the total scale of greenhouse gas emissions<\/strong> generated by the development of digital infrastructure is increasingly emerging in the AI \u200b\u200bdebate. <\/p>\n\n\n\n

          Sustainability Challenges and Environmental Costs<\/strong><\/h2>\n\n\n\n

          The problem doesn’t end with training, however. Once AI systems<\/strong> are deployed, they begin handling millions of queries daily, or inference, which accounts for the AI’s ever-increasing energy needs<\/strong>. This also leads to an increase in the demand<\/strong> for water to cool data centers. This is known as the “secret water footprint of AI<\/strong><\/em>” and  “water footprint of AI models<\/strong><\/em>.”  <\/p>\n\n\n\n

          AI infrastructure generates a real water footprint, often in the millions liters of water<\/strong> consumed every day. Data published by technology companies shows that in 2021 <\/strong>and 2022<\/strong>, the increase in AI-related investments coincided with increased consumption of water and energy resources. Globally, data centers are responsible for a growing amount of the world’s electricity<\/strong>, and forecasts suggest that AI development could require hundreds of terawatt<\/strong>-hours of energy annually. <\/p>\n\n\n\n

          At the same time, AI development doesn’t have to mean only threats. Actions that reduce energy consumption<\/strong> and environmental impact<\/strong> are becoming increasingly important. The technology industry is investing in more efficient processors, intelligent server optimization<\/strong>, heat recovery, and electronic component recycling<\/strong>. There’s also growing talk of using AI for environmental protection <\/strong>– monitoring air quality, forecasting droughts, and managing power grids. The paradox, then, is that the use of AI <\/strong>can simultaneously help reduce greenhouse gas emissions<\/strong> and generate new climate burdens. Therefore, the AI \u200b\u200bindustry<\/strong> today faces the challenge of finding a balance between innovation and responsibility – so that the rapid development of AI<\/strong> doesn’t come at the expense of the natural environment.<\/strong> <\/p>\n\n\n\n

          The Problem Is Scale<\/strong><\/h2>\n\n\n\n

          A single question to AI doesn’t cause a water crisis. However, billions of queries per day and the technological race between AI companies translate into increasing energy and natural resource consumption. As AI becomes more efficient, the number of applications and users grows faster. AI requires physical data centers, and locations with the largest number of them are most congested and at risk of increased local water stress. Chatbots like ChatGPT, Gemini, and Grok generate water and energy footprints that can be hazardous to local environments. The key is to design more responsible AI infrastructure and transparently inform the public about its risks.  <\/p>\n\n\n\n

          This is just one example of the topics we cover as part of Green Vision<\/a> and our environmental communication activities. You can read more about our approach, projects, and services here<\/a>. <\/p>\n\n\n\n

          Check out also our social media profiles:\u2013 Facebook<\/a>\u2013 Instagram<\/a>\u2013 LinkedIn<\/a>There, we share our projects, social campaigns, and topics related to water, climate, and responsible communication.<\/p>\n","protected":false},"excerpt":{"rendered":"

          Artificial intelligence, also known as AI, is not “intangible.” Each query for models like ChatGPT, Gemini, or Grok requires data centers, which consume electricity and vast amounts of water to cool the infrastructure. The problem isn’t the individual query itself, but the scale of AI’s global use. <\/p>\n","protected":false},"author":7,"featured_media":15480,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2549],"tags":[2281,2742,2744,2740,2745,2743,2746,2747,2749,2739,2741,2748],"class_list":["post-15485","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-green-vision-en-2","tag-ai-en-2","tag-ai-carbon-footprint","tag-ai-water-footprint","tag-artificial-intelligence","tag-chatgpt-en","tag-data-centers","tag-energy-consumption","tag-gemini-en","tag-grok-en","tag-sustainable-development-en-2","tag-the-impact-of-ai-on-the-environment","tag-water-stress"],"_links":{"self":[{"href":"https:\/\/wide-vision.pl\/en\/wp-json\/wp\/v2\/posts\/15485","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wide-vision.pl\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wide-vision.pl\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wide-vision.pl\/en\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/wide-vision.pl\/en\/wp-json\/wp\/v2\/comments?post=15485"}],"version-history":[{"count":0,"href":"https:\/\/wide-vision.pl\/en\/wp-json\/wp\/v2\/posts\/15485\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wide-vision.pl\/en\/wp-json\/wp\/v2\/media\/15480"}],"wp:attachment":[{"href":"https:\/\/wide-vision.pl\/en\/wp-json\/wp\/v2\/media?parent=15485"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wide-vision.pl\/en\/wp-json\/wp\/v2\/categories?post=15485"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wide-vision.pl\/en\/wp-json\/wp\/v2\/tags?post=15485"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}