Science Vs

AI: Is It Ruining the Environment?

November 13, 2025

Key Takeaways Copied to clipboard!

  • The energy consumption of a single AI query varies widely based on the model's size (number of parameters), with text generation potentially using more energy than image generation for certain model comparisons. 
  • AI data centers' water usage is complex, with only a small fraction being direct drinking water, but the overall impact is regional and tied to the water used by the power plants supplying the data centers. 
  • The primary environmental concern regarding AI is not the technology itself, but the continued reliance on fossil fuels to power the rapidly increasing energy demands of data centers. 

Segments

AI Energy Use Introduction
Copied to clipboard!
(00:00:00)
  • Key Takeaway: Concerns exist that AI strains power grids and increases emissions due to high electricity and water consumption.
  • Summary: Recent public discourse highlights accusations that AI is extremely power-hungry and thirsty, leading to protests against new data centers. One claim suggests asking ChatGPT to write one email is equivalent to pouring out an entire water bottle. Data centers, housing the necessary servers, are identified as the major culprits behind this resource strain.
CPU vs GPU Energy Difference
Copied to clipboard!
(00:06:47)
  • Key Takeaway: AI relies on GPUs, which perform many tasks simultaneously, requiring significantly more energy than CPUs that process tasks sequentially.
  • Summary: Normal computing uses CPUs, while AI relies on GPUs, analogous to a single paintball gun versus 200 bound together firing at once. This parallel processing capability makes GPUs the powerhouse for machine learning but necessitates greater energy consumption.
Measuring Single Query Energy Use
Copied to clipboard!
(00:09:13)
  • Key Takeaway: Energy use per query is highly variable, but a text prompt to a large language model is estimated to be equivalent to microwaving something for less than 10 seconds.
  • Summary: Journalists James O’Donnell and Casey Crownhart measured open-source models, finding the smallest used 114 joules (0.1 microwave seconds) and the largest (50 times bigger) used 8 seconds. Proprietary models like ChatGPT are estimated to use between one and two seconds in the microwave per text query.
Image and Video Energy Costs
Copied to clipboard!
(00:16:05)
  • Key Takeaway: Generating a short, low-quality AI video can consume significantly more energy than text or image generation, equating to over an hour in a microwave.
  • Summary: Making an AI image was found to be equivalent to about five and a half seconds in the microwave, less than the eight seconds estimated for a large text model query. In contrast, a five-second, 16-frame-per-second AI video required over an hour of microwave time for comparison.
Total AI Energy Impact
Copied to clipboard!
(00:19:10)
  • Key Takeaway: Data center electricity consumption tripled between 2014 and 2023, and projections suggest AI data centers will consume as much electricity as 25% of U.S. households by 2028.
  • Summary: OpenAI reports receiving 2.5 billion prompts daily, and 78% of surveyed organizations are using AI to some extent. Because the U.S. grid relies heavily on fossil fuels (only 9% renewable), this increased energy demand contributes significantly to planet-warming emissions. Building new clean energy sources like nuclear plants is too slow to meet immediate AI demand, leading some companies to use gas-burning generators.
AI Water Consumption Analysis
Copied to clipboard!
(00:28:17)
  • Key Takeaway: A back-and-forth conversation of about 30 messages with ChatGPT3 uses the volume of a half-liter water bottle, but only about 12% of that water is direct drinking water used for cooling.
  • Summary: Data centers use water evaporation in cooling towers, drawing water from municipal systems, which is concerning because evaporated water is unevenly redistributed globally. Overall, data centers consume 0.3% of the nation’s water supply, roughly equivalent to Rhode Island’s total usage, though this is projected to double.
Final Verdict and Personal Use
Copied to clipboard!
(00:36:32)
  • Key Takeaway: The primary environmental villain is the failure to switch to renewable energy, as AI’s drawbacks are often perceived as less valuable than the benefits of traditional energy-intensive activities like flying or eating meat.
  • Summary: Experts interviewed still use AI for tasks like trip planning or technical help, but they emphasize the need for thoughtful usage. Over 61% of U.S. people believe AI has more drawbacks than benefits, making it an easy target for environmental criticism. The core issue remains the dependency on fossil fuels to power the growing energy demands of data centers.