Generative AI’s Environmental Impact
- Katrina Armalovica
- Dec 18
- 3 min read
The first mention of Artificial Intelligence (AI) as a concept can be found in Jonathan Swift's fantastic novel “Gulliver's Travels”. In the novel, a powerful computing device, known as The Engine, is used to aid scholars in developing new bodies of literature. However, it was not until 1914 that El Ajedrecista was introduced at the Exposition Universelle in Paris. The chess-playing machine played simple endgames of king and rook versus king, requiring no human intervention to complete legal moves. In 1959, Arthur Samuel pioneered the concept of machine learning from experience. Samuel’s developed machine specialized in playing checkers. Samuel introduced the concept of machines being programmed to learn from past experiences by documenting the machine’s checkers-playing ability. It wasn’t until 2020, however, that OpenAI introduced GPT-3. At the time, the bot was the largest ever created, with 175 billion parameters.
Since the release of GPT-3, countless technology companies have launched their own AI bots. Among the most famous are Gemini by Google, Chat-GPT by OpenAI, and DeepSeek by High-Flyer. Two years after the release of GPT-3, ChatGPT experienced 264.7 million visits. By July of 2025, the number of visits had grown to 5.2 billion. On average, ChatGPT has 800 million weekly users and processes 2 billion queries daily. According to the Pew Research Center, 58% of adults under 30 in the United States of America have used AI, a 15% increase from 2024. The 21st century is marked by milestones in AI, and while many praise the development of digital computing, its environmental burden might push humanity over the climate change tipping point into the point of no return.
Despite popular belief that AI is developed solely in devices or apps, computing intelligence is maintained and located in physical AI Data Centers. Data centers host a large number of file servers and networking equipment that store, process, and analyze images, code, text, and other media of communication, all of which draw immense power and water. Controlled by large language models (LLMs) and machine learning, data centers can act autonomously for each individual inquiry. Furthermore, these centers are equipped with powerful hardware, such as graphics processing units (GPUs) and central processing units (CPUs), which run AI models and algorithms.
However, to continue operating, data centers hosting AI require substantial electrical power and water for storage and processing. In 2023, such data centers consumed around 4.4% of America’s electrical power, and that percentage is predicted to rise substantially over the next decade. To prevent centers from overheating, many facilities previously relied on air-cooling systems to maintain low temperatures. However, after the popularization of AI, data centers began requiring more substantial cooling methods. Hence, many facilities began adopting water cooling systems. Water-cooling systems use liquid, primarily water, to transfer heat away from electronic components, such as the CPU and GPU, to prevent overheating and damage to the system. Because water conducts heat faster than air, water-cooling systems achieve higher cooling rates, enabling continuous operation.
On average, large data centers can consume 5 million gallons per day, the equivalent of the water use of a town populated by 10,000 to 50,000 people. According to scientists at the University of California, Riverside, each 100-word AI prompt is estimated to use roughly one bottle of water. Many still believe, however, that the issue is not urgent, as 71% of Earth’s surface is covered in water. However, the cooling systems in the data facilities cannot use saltwater due to its corrosiveness and potential contamination. Hence, such systems require freshwater, which accounts for a mere 2.5% of Earth’s total water supply. Data centers usually withdraw water for cooling systems from the same municipal systems that supply homes and businesses. Furthermore, approximately 80% of the water involved evaporates during the process, with the remaining water being discharged to municipal wastewater facilities.
—
Comments