# AI Carbon Footprint

Artificial intelligence offers incredible capabilities, but it also relies on heavy compute infrastructure. Use this guide to understand what drives AI emissions and compare the tools you actually use.

## Training and everyday use are different questions

Training headlines show the scale of model development, while most people really want to understand everyday inference: repeated use of tools like ChatGPT.

## Hardware and cooling still matter

AI workloads sit on top of accelerator-heavy servers, networking, and cooling systems, so the visible product is only part of the footprint story.

## What you'll find in this hub

### Start with everyday AI use

For most people, the main question is the repeated use of text and image tools like ChatGPT, Claude, or Midjourney, rather than one-off training headlines.

### Connect prompts to infrastructure

Prompt volume, model size, accelerator hardware (especially GPUs for image generation), and data center cooling all shape the footprint together.

### Open the closest matching tool page

After the overview, jump to the calculator page for the AI tool you use most, whether it's for text analysis or generating art.

## Highest estimated emitters

### Good pages to start with

1. [Midjourney 250g CO2 / hour](/content/carbon-footprint/midjourney/index.html)  
2. [ChatGPT 150g CO2 / hour](/content/carbon-footprint/chatgpt/index.html)  
3. [Gemini 145g CO2 / hour](/content/carbon-footprint/gemini/index.html)

## Explore related calculators and guides

**Claude**  
Claude's carbon footprint involves large-scale inference processing in data centers. Each prompt contributes to electricity demand for both processing and cooling.  
140g CO2 / hour  
[Open](/content/carbon-footprint/claude/index.html)

**Midjourney**  
Midjourney requires massive GPU power to generate high-resolution images. Image generation models typically consume more energy per request than text models.  
250g CO2 / hour  
[Open](/content/carbon-footprint/midjourney/index.html)

**Gemini**  
Gemini's multimodal capabilities mean it processes text, images, and audio, all of which require significant data center compute power.  
145g CO2 / hour  
[Open](/content/carbon-footprint/gemini/index.html)

**ChatGPT**  
ChatGPT's carbon footprint comes from every request, response, and supporting system behind the model. If you're asking how much CO2 ChatGPT produces, the answer depends on model size, response length, and how often you use it.  
150g CO2 / hour  
[Open](/content/carbon-footprint/chatgpt/index.html)

## Compare related apps and habits

[**ChatGPT vs Claude**  
See two related digital habits side by side.  
150g CO2 / hour vs 140g CO2 / hour](/content/carbon-footprint/compare/chatgpt-vs-claude/index.html)

[**ChatGPT vs Midjourney**  
See two related digital habits side by side.  
150g CO2 / hour vs 250g CO2 / hour](/content/carbon-footprint/compare/chatgpt-vs-midjourney/index.html)

[**ChatGPT vs Gemini**  
See two related digital habits side by side.  
150g CO2 / hour vs 145g CO2 / hour](/content/carbon-footprint/compare/chatgpt-vs-gemini/index.html)

## Common questions

### Why isn't there one simple AI number?

Because emissions vary with model size, hardware, prompt volume, and how often you use the tool.

### What should I do after reading this guide?

Start with the biggest drivers, then open the calculator page that matches the AI tool you use.