3 ms·
LLM inference uses on the order of 1 Wh per query. That's under 10 meters of driving on an EV or running air conditioning for under 5 seconds. https://hannahri
by jampekka 7mo ago
LLM inference uses on the order of 1 Wh per query. That's under 10 meters of driving on an EV or running air conditioning for under 5 seconds.
https://hannahritchie.substack.com/p/ai-footprint-august-2025 https://hannahritchie.substack.com/p/ai-footprint-august-202...
- deadbabe 7mo agoOr a human riding a stationary bike for 36 seconds.
- bluefirebrand 7mo agoOne query is not going to be a useful benchmark when people are deploying AI swarms in loops to solve simple problems
- AngryData 7mo agoHow many querys are people going to be using though when they don't see the upfront costs? And how many querys would have been just as solved with a simple google search or wikipedia page that will instead/also be put through an llm and use extra power? Like if I search for the chocolate tempering temperature or the atomic number of tantalum, isn't all that energy just wasted on the companion llm results? It seems like it will add up.