Recently, Google published a report on the environmental impact of its Gemini AI model,24 commenting that it chose not to report indirect water consumption because it “does not fully control the water consumption in electricity generation.”25 This statement highlights a lack of willingness to disclose this metric, ignoring that this water consumption is a direct consequence of Google’s electricity demand

surpassing the amount of electricity used for Bitcoin mining in 2024

between 312.5 and 764.6 billion liters

The study highlights that from one state to the next, the environmental cost of AI looks radically different.

Our findings reveal that AI systems emit between 130 and 1500 times less CO2e per page of text generated compared to human writers, while AI illustration systems emit between 310 and 2900 times less CO2e per image than their human counterparts.

this article is the first time we are aware of where researchers have compared the carbon footprint of AI to that of humans.

the findings presented here suggest that concerns about the emissions generated by AI systems should be tempered by recognition that, even relying on cautious assumptions, humans produce far more emissions when engaging in some of the same tasks.

An article in The Writer magazine states that Mark Twain’s output, which was roughly 300 words per hour, is representative of the average writing speed among authors21. Therefore, we use this writing speed as a baseline for human writing productivity.

The findings above demonstrate that the environmental footprint of AI completing two major tasks is substantially lower than that of humans completing those same tasks.

It is estimated that training the AI model for the first version of ChatGPT, launched in November 2022, used roughly the same amount of electricity as 130 average U.S. households consume in a year, highlighting the significant power demand.

Therefore, a day of refrigerator use is equivalent to approximately 5000 online searches or 500 generative AI prompts.‍

According to a study in Nature, using generative AI to create text or images can produce 130 to 2,900 times less CO2 than humans doing the same tasks.

If that same teacher were to sit in the office for hours writing a quiz with the lights and air conditioner running the whole time, it could be argued that using AI is a better and more environmentally sound option.

In this context, we have conducted a first-of-its-kind comprehensive study to quantify the environmental impacts of our LLMs. This report aims to provide a clear analysis of the environmental footprint of AI, contributing to set a new standard for our industry.

This claim about water use has been republished in dozens of other outlets. It is probably the most influential single statistic when talking about AI’s impact on the environment. Anyone who believes it is true will be trying to solve a problem that doesn’t exist.

The article, compellingly, centers on the morality of the customer’s actions. You, the end user, are held responsible for consuming half a liter of water every time you use an LLM. If this were true, you could, clearly, have a meaningful impact by boycotting ChatGPT. It would also be important to try to prevent other people from using ChatGPT, since they are directly responsible for using up a lot of water.

The takeaway is that it’s impossible to know with any certainty, because companies don’t disclose what they’re building.