Contributed by Lauren Gossett
Lauren Gossett is a Senior Water Quality Technician at Denver Water. She moved from Tennessee to Colorado after earning a B.S. in Geology, started in the field, and transitioned to the lab in 2024. She’s still growing her experience on instruments but has fully embraced her unofficial role as the lab’s social director. When she’s not in the lab, Lauren is chasing live music and will show up for almost any artist if there’s a stage and sound.
Disclaimer: This blog post was generated with the assistance of Artificial Intelligence (AI).
As the holidays approach, many of us cozy up with hot cocoa, decorate trees, and—more than ever—turn to AI helpers for gift ideas, recipes, or even writing blog posts like this one! But did you know that every time you chat with an AI, there’s a hidden water footprint behind the magic?
Why Does AI Use Water?
Large language models like ChatGPT and Copilot run on powerful servers housed in data centers. These servers generate heat as they process your requests, and just like we cool down after a brisk winter walk, servers need cooling too. That’s where water comes in.
Most data centers use evaporative cooling systems, which rely on water to keep equipment from overheating. The type of water varies:
- Potable water: Clean drinking water, often used in regions without alternatives.
- Non-potable or recycled water: Increasingly common in sustainable facilities.
- Closed-loop systems: Advanced setups that minimize water loss by recirculating water.
How Much Water Are We Talking About?
It’s less than you might think—but it adds up across billions of queries. Here’s a festive way to picture it:
Quick Facts (from least to most water):
- Writing this blog post (short conversation): 20 mL → about four teaspoons → enough to fill a tiny jingle bell.
- Reviewing a large file or attachment: 25 mL → about half a shot.
- Generating an Excel or PowerPoint document: 40 mL → about two tablespoons, or enough to make icing for one gingerbread cookie.
- Average-length holiday chat (~20 prompts): 50 mL → enough to fill a small Christmas ornament.
- Generating a festive AI image: 200 mL → almost a full mug of hot cocoa.
- Generating 190 AI images: 3.8 liters (1 gallon) → Watering a Christmas tree for one day!
Note: Estimates vary widely. Some studies include the entire lifecycle of AI models (training + inference), which can make per-query water use appear much higher—hundreds of milliliters or more. The numbers above reflect real-time inference only, which is the water used to process your request in the moment.
Why It Matters
As professionals in the drinking water and wastewater industries, we are stewards of our natural resources. While AI tools can be incredibly helpful at work and in our personal lives, we must be mindful of their environmental impact. By using these tools responsibly, we set an example and help protect the very resource we all work for—water.
So next time you ask an AI to write your holiday poem or design a festive card, remember there’s a little splash of water behind every word.
Water Used for This Blog Post:
Including writing, calculations, planning, and sourcing, this interaction likely consumed about 20 mL (four teaspoons). Just enough to fill a jingle bell!
Sources and Further Reading:
Wishing you a joyful, sustainable holiday season! May your celebrations be merry, bright, and mindful of the resources that make life possible.