ChatGPT's water usage primarily stems from the evaporative cooling systems in data centers powering its AI models, where processing queries generates heat that requires water-intensive cooling. Studies consistently peg a typical ChatGPT query—such as generating around 100 words—at about 500ml of water , equivalent to a standard bottle, though this varies by model efficiency and data center tech.

Key Figures

  • Per Query : 519ml for a 100-word response (UC Riverside/Colorado 2023 study).
  • GPT-4 Specific : Up to 3 bottles (~1.5L) for complex 100-word outputs in newer reports.
  • Daily Scale : ChatGPT's global ops could guzzle 2–160 million liters daily, depending on query volume.

Breakdown by Usage

Scenario| Water Used| Real-World Equivalent
---|---|---
1 query (100 words)| ~500ml 13| 1 plastic bottle
10–50 queries| ~2L 5| Half a gallon jug
1 email/week for a year| 27L 7| 1.5 large jugs
1 in 10 U.S. users emailing weekly| 435M+ liters/year 7| Massive reservoir strain

Why Water?

Data centers like those for Microsoft (OpenAI partner) use cooling towers where water evaporates to dissipate heat—not drunk by the AI , but a hidden environmental cost. A single query might draw 0.01–0.34 watt-hours of power, translating to water via cooling (1.8–2.5L per kWh). Infrastructure trumps individual use: Switching to immersion cooling cuts water by 31–52% over 10 years.

Trending Context (2025–2026)

As of early 2026, amid California wildfires and droughts, this topic exploded—forums like Reddit buzz with alarm over AI's "hidden footprint," while Sam Altman confirmed low per-query energy (~0.34Wh) but massive scale. Latest news : Fortune highlighted wildfire risks to water-guzzling AI just last week (Feb 2025), pushing calls for policy.

"If 1 in 10 people used AI for one email a week, it'd match Rhode Island's daily water use."

Multiple Viewpoints

  • Critics : AI rivals household energy (500kW+ daily for 200M requests) and strains arid regions.
  • Defenders : Local AI (e.g., Ollama) uses zero water (runs on your device), slashing impact 400x for routine tasks; infrastructure upgrades matter more than quitting AI.
  • Optimists : Efficiency gains—new cooling tech drops usage 15–50%; still dwarfed by crypto mining or streaming.

Reduce Your Impact

  1. Batch prompts: Combine tasks into one query.
  2. Use local models for simple edits (zero water).
  1. Opt for efficient AIs or downtime during peak heat.
  2. Advocate: Push data centers for liquid cooling.

TL;DR : Around 500ml per typical query , scaling hugely with popularity—think bottles adding up to pools daily. Go local to sidestep it.

Information gathered from public forums or data available on the internet and portrayed here.