It depends. There isn’t an easy answer to this, but we’re going to try and weigh the pros and cons of why using programs like ChatGTP or DeepSeek so that you have the information you need to decide if or even how much you use AI-fueled programs to get your work done.
The short of it is that the data centers behind AI programs use a surprising amount of energy and water for each single use. If you’re using the programs, that energy is not something you can easily perceive, and that makes it easy to ignore.
To me there’s a parallel in using credit cards to pay for everything. The less you physically experience a transaction, the less you register its scale. That’s why paying $100 for something online feels a lot differently than paying $100 in cash at the store.
For those of you that dislike writing, think of it as all the soul-sucking energy you’ve used in crafting that sentence has now been offloaded into a data center. Instead of struggling to get a single sentence out, you can just shoot out a request and there the piece of writing appears magically.
Data centers are energy hungry – server energy usage in the U.S. more than tripled between 2014 and 2023, leaping from 30 terawatt-hours (TWh) to nearly 100 TWh in 2023 (2024 United States Data Center Energy Usage Report). To put that in perspective, 100 TWh is the equivalent of the electricity usage of more than 8 million homes for an entire year.
To bring that further down to scale, a single 100-word email request uses 0.14 kWh of electricity, the equivalent of powering 14 lightbulbs for 1 hour (Washington Post).
A second problem arises from the way these data centers use electricity. Training a data center requires a significant amount of electricity but that usage is predictable and stable. It’s difficult to predict the electricity demand from every day use since it depends on the number of users at any given time. The strain on the grid from spikes in electricity demand is a cause for concern. We already see power outages during extreme weather days when heating or air conditioning use rises precipitously.
Writing is an incredibly complex activity. For every 100-word email you request, ChatGPT requires thousands of calculations to determine the best words to use. Just as when your computer overheats if you overload it, so do these data centers. Electricity or water is used to cool data centers – essentially a large building filled with computers, generating heat.
That really depends on the companies and the electricity sources feeding into the electricity mix. Some companies have signed power purchase agreements with renewable energy companies to source clean energy. But that practice is losing momentum. In reality, many companies are turning to fossil fuels to help power these data centers or they are pushing utilities to turn to fossil fuels.
Not only hungry, these data centers are also thirsty but how thirsty depends on where they located. If they are in a hot area, it takes a lot of electricity or water to cool down. Regardless, they are often the heaviest users of water in the places where they are located. Microsoft’s data center used in West Des Moines, Iowa, used around 6 percent of the district’s water according to the water department’s records.
Data centers are often the heaviest users of water in the places where they are located.
This is concerning for areas prone to drought. If large data centers are being built in vulnerable areas, this poses a risk to the local ecosystem. One option is to regulate where data centers are being built and increase transparency around industrial and commercial water consumption. A worse option is to let it be decided at a grass roots level by the people living in the area.
The simple answer is to restrict our usage of them or don’t use them at all. However, it’s nearly impossible to put it back in the box after we’ve unleashed them.
A better option would be to take steps to address the increasing demand for electricity by accelerating our rollout of clean energy sources. Wind and solar, as two of the most cost-effective sources of electricity, and some of the fastest to build, it makes sense to include both of those in the planned new capacity.
The reality is that some utilities are scrambling to meet the rising demand for electricity and have rolled back clean energy goals. Georgia Power is one example of how data centers are contributing to decisions to keep burning fossil fuels.
Georgia Power had originally planned to retire most of its coal plants by 2028. Citing the rising demand of electricity, in no small part from data centers, they decided to keep a Mississippi coal plant open past its retirement date. The reason they gave was the increasing demand for electricity – far outstripping their predictions.
All in all, AI servers in data centers are still a relatively modest factor in the projected increase in electricity demand. The adoption of electric vehicles was in fact a larger driver in the demand for electricity. We’re all in the midst of a seismic shift in society to electricity and at this point we have a choice. Do we rely on fossil fuels and push off problems to the not-so-distant future? Or, do we push for cleaner sources of energy despite the headwinds.
Bravery is doing what is right despite potential social cost. We don’t need to stop progress and enjoying the fruits of what we’ve created, but we do need to take full responsibility for their effects. And lastly do what every we can to leave a positive legacy behind.
This article was written without the use of artificial intelligence – just plain old intelligence.
World Energy Outlook 2024. iea: https://iea.blob.core.windows.net/assets/140a0470-5b90-4922-a0e9-838b3ac6918c/WorldEnergyOutlook2024.pdf
What’s driving electricity demand? It isn’t just AI and data centers. MIT Technology Review: https://www.technologyreview.com/2025/02/20/1112119/electricity-demand-2025/
2024 Long-Term Reliability Assessment December 2024. North American Electric Reliability Corporation (NERC): https://www.nerc.com/pa/RAPA/ra/Reliability%20Assessments%20DL/NERC_Long%20Term%20Reliability%20Assessment_2024.pdf
2024 United States Data Center Energy Usage Report. Berkeley Lab: https://eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report.pdf
A bottle of water per email: the hidden environmental costs of using AI chatbots. Pranshu Verma and Shelly Tan. The Washington Post: https://www.washingtonpost.com/technology/2024/09/18/energy-ai-use-electricity-water-data-centers/
Artificial intelligence technology behind ChatGPT was built in Iowa — with a lot of water. Matt O’Brien and Hannah Fingerhut. AP News: https://apnews.com/article/chatgpt-gpt4-iowa-ai-water-consumption-microsoft-f551fde98083d17a7e8d904f8be822c4