Quick Overview

In recent years, society has begun to normalize blaming consumers, rather than the producers, for the results of consumerism. We’ve all seen this in prior years with regards to recycling and fossil fuels, but in the past year or two I’ve noticed the same ideologies when discussing LLMs (large language models).

The Issue

While protecting the environment is everyone’s duty, it is unreliable to expect each person to care about their individual contribution. As a single person, its easy to abstract yourself away from global issues, but the symptoms of these issues begin to arise when the population is conglomerated to upwards of three hundred million (or billions, when regarding the entire planet).

According to the United States Environmental Protection Agency, 30% of greenhouse gas emissions are from industry, and 31% are from a combination of residential & commercial, while the other 39% are a sum of transportation and agriculture.

A reasonable estimate for commercial and industrial is between 40-50% of the total greenhouse emissions in the United States, but as individuals we’re expected to carpool to cut down on emissions. It’s ridiculous that the blame is placed on the consumer rather than the producers.

While the issue with fossil fuel emissions has existed for as long as I can remember, the usage of fossil fuels and coal has become a staple in all modern societies, and thus has been deemed a necessary evil. I can only hope that better alternatives arise in the near future.

Water Consumed

In the past few years, particularly with the creation of ChatGPT, LLMs have become incredibly popular. When ChatGPT was first released, it was met with widespread usage, as it was essentially a fancy chatbot, relatively up to date with current events, and could assist with some light critical thinking such as explaining complex topics and writing code.

After its debut, it began attracting negative attention due to the amount of energy it uses during its training and usage. According to Shaolei Ren’s team, they estimate that GPT3 “consumes” 500mL of water, and a the common consensus on the internet is that a google search uses approximately 5% of the energy of a GPT3 query.

The word “consume” here doesn’t mean that the water is destroyed, but instead that it’s dispersed through the system and reintroduced to the environment. An estimated 20% of this “consumed” water evaporates, which means it can move to different areas on the globe. New Mexico is a culprit of this issue, as data centers are using it’s already limited water supply, which then evaporates out of New Mexico.

It has been estimated that the new and improved GPT4 uses ten times the amount of energy that of which GPT3 uses, which directly correlates to ten times the amount of water needed to cool it.

Energy Consumed

Frequently online I’ll see complaints about the water usage, but most of the water is returned into the environment it was taken from. Obviously this isn’t ideal, but the primary issue that arises from the usage of giant LLMs is the sheer amount of energy it uses.

In the United States, only 20% of the power grid is fueled using nuclear, which leaves the majority of the energy being a product of fossil fuels and coal.

The United States contains 45% of the world’s data centers. In 2024, these data centers used approximately 200 million terawatt-hours, with AI specific servers making up between 53 and 76 terawatt-hours - enough to power 7.2 million US homes for an entire year - for reference, Florida has about 8.5 million households, with a population of 22.3 million. Researchers are estimating that this number will go up to between 165 and 326 terawatt hours. Using a high estimate, that will be enough to power nearly 25% of all households in the United States.

A 2024 report from Virginia legislation estimates that residential taxpayers could pay about $38 a month in additional energy bills due to increasing data center costs. Why are corporate costs being placed on people that aren’t affiliated with the corporations? Note that this price isn’t even being attached to the end user.

Trillion dollar corporations, such as NVidia, Apple, Google, and Amazon are stealing all of American resources, increasing the scarcity of energy, and then sticking the costs to citizens of America, all while destroying the environment.

Individual Effects

I don’t have an issue with corporations training AI models or consumers using them, but I do have an issue with this concept being nearly unregulated. As someone who lives and works in northern Virginia near the large masses of data centers, I shouldn’t be subject to higher electricity bills simply because Apple spent another $500 billion dollars investing in AI.

I see people online boycotting content creators for their AI usage, and while I don’t condone the substitution of creativity for AI, I don’t believe these creators should be shamed. As an example, Markiplier has faced backlash for asking ChatGPT questions during his videos.

Solution

My request is that rather than punishing the ignorant consumer, please contact your legislation and express your concerns. Fossil fuels and coal are already a terribly destructive means to energy production, and energy usage will continue going up until a large breakthrough in LLM hardware is made.

Published by tygutowski on Thu Aug 07 2025 00:00:00 GMT+0000 (Coordinated Universal Time)