Skip to main content
Carsten Felix Draschner, PhD

Unmasking AI’s Footprint - The Real Cost of Large Language Models

The Hidden Environmental Cost of AI

Image 1

TL;DR ⏱️

Background

🌍 Mistral AI just released the first full life-cycle analysis (LCA) of an LLM (Mistral Large 2).
The numbers are eye-opening:

This raises questions about sustainability, usage patterns, and model selection.

What have I done:

I reviewed the Mistral post and analysed where the environmental impact of LLMs comes from:

  1. Training once, paying for years – the up-front carbon loan dominates unless utilisation is high.
  2. Hardware & supply chain – producing GPUs and cooling systems can rival the energy cost.
  3. Inference at scale – small grams per query add up massively with millions of requests.
  4. Utilisation ratio – measuring "total inference ÷ total lifecycle" helps evaluate if training is “earned back.”

IMHO:

🔬 Key takeaways for sustainable AI:

AI has the potential to optimise logistics, drug discovery, and code generation — but its own footprint is real and must be addressed.
At Comma Soft AG, we always validate the most reasonable approach to solve a problem and optimise resource spend, especially as we host LLMs ourselves within Alan.de.

❤️ Feel free to reach out and like if you want to see more of such content.

#artificalintelligence #llm #oneearth