AI Ethics: Enironmental & Labor Impacts

Power Consumption:

The servers that operate ChatGPT and other generative AIs require staggering amounts of power due to the complex computing involved. This power usage is predicted to explode in the second half of the decade as more & more AIs come online, demand increases, and more & more datacenters are built.

• Currently, 4.4% of all electricity in the United States is used by AI datacenters.
• By 2028, it is estimated that generative AIs will use as much electricity as 22% of all American households.
• The International Energy Agency estimates that, by 2030, AI datacenters will consume more power annually that the entire nation of Japan.

Water Usage:

• In 2022, Google's data centers consumed 5.2 billion liters (1.4 billion gallons) of water to cool its AI operations. 
• ChatGPT uses approximately 700,000 liters of water during training, the same amount an American family uses in 20 years.
• ChatGPT uses 500 ml of water for every 20-50 queries. Given that there are currently over 100 million users of ChatGPT, that water usage is astronomical.
  • By way of comparison, a simple Google search uses 2 ml of water.
• Researchers estimate that ChatGPT, Copilot and Gemini combine to use between 3 million & 6 million cubic feet of water per year, the equivalent of annual water consumption of 300,000 to 600,000 people.

Carbon Footprint:

Because generative AIs are so new, our energy grid has not yet adapted to account for their enormous power requirements. Major tech companies such as Google and Amazon have vowed to triple America's nuclear power capabilities to power their AIs, but that may take decades to accomplish, and in the meantime, AIs are simply being powered by dirtier fossil fuel energy, such as natural gas. At a time when we must be doing everything we can to reduce our carbon footprint to fight climate change, generative AIs are doing the exact opposite, vastly increasing greenhouse gasses.

• AI data center's carbon intensity is 48% higher than the average American power usage.
• It is estimated that ChatGPT generated approximately 552 tons of carbon during the training of GPT-3, equivalent to the amount of carbon produced 123 gas-powered cars in one year.
• In early 2023, shortly after its public launch, ChatGPT is estimated to have had the same carbon footprint as 605 Americans, a number that will have soared given the vast increase of users.

Ref: George, A. Shaji, George, A.S. Hovan, & Martin, A.S. Gabrio. "The Environmental Impact of AI: A Case Study of Water Consumption by Chat GPT." Partners Universal International Innovation Journal (PUIIJ) 1, no. 2 (2023), 91-104. www.researchgate.net/publication/370202417_The_Environmental_Impact_of_AI_A_Case_Study_of_Water_Consumption_by_Chat_GPT.

International Energy Agency. "AI Is Set to Drive Surging Electricity Demand from Data Centres While Offering the Potential to Transform How the Energy Sector Works." www.iea.org. 10 April, 2025. AI is set to drive surging electricity demand from data centres while offering the potential to transform how the energy sector works.

O'Donnell, James & Casey Crownhart. "We Did the Math on AI's Energy Footprint. Here's the Story You Haven't Heard." MIT Technology Review, 20 May, 2025. We did the math on AI’s energy footprint. Here’s the story you haven’t heard. | MIT Technology Review.