thirsty-ai

"AI isn't magic. It drinks."

Thirsty AI is a minimalist, open-source project designed to visualize the hidden environmental cost of modern Artificial Intelligence. It calculates the water consumption (Scope 1 Cooling + Scope 2 Energy Generation) of LLM queries and media generation.

The Science

Calculations are based on 2024/2025 environmental reports (UC Riverside / Microsoft).

Action	Cost (ml)	Context
Chat	15 ml	Standard GPT-4o class query.
Reason	150 ml	Chain-of-Thought (o1/Gemini). 10x compute heat.
Image	500 ml	SDXL / DALL-E 3. One full water bottle.
Video	3000 ml	High-compute video gen. A bucket of water.

Why "Waste"? The water used to cool Data Center GPUs evaporates into the atmosphere. It is removed from the local watershed and does not return to the ground immediately.

Web App

The web interface features a fluid WebGL background that reacts to device orientation (Gyroscope) and visualizes the water level rising in real-time.

Development

# Clone
git clone https://github.com/mgks/thirsty-ai.git

# Run Local Server
npx http-server .

NPM Package (thirsty-ai)

The core logic is available as a zero-dependency package for developers.

Installation

npm install thirsty-ai

Usage

import { ThirstyCalculator } from 'thirsty-ai';

const calc = new ThirstyCalculator();

// Add Interactions
calc.add('QUERY');      // +15ml
calc.add('IMAGE', 2);   // +1000ml (2 Images)

// Get Statistics
const stats = calc.getStats();
console.log(stats.formattedString); 
// Output: "Equivalent to 1.5 Bottles"

// Get a "Reality Check" Fact
// Returns a string based on the severity of usage
console.log(calc.getShockFact()); 
// Output: "One Liter. The amount required to keep a human alive for 8 hours..."

License

MIT

{ github.com/mgks }