AI video generation can be thousands of times costlier
A recent study from Hugging Face reveals that making AI-generated videos can consume thousands of times more electricity than generating text or images.
The longer and sharper the video, the more power it drains—sometimes hundreds of watt-hours for just one clip.
This heavy energy use is now a big part of rising emissions, with Google linking AI workloads to a 13% jump in carbon output last year.
Energy use and wait times grow way faster
The study found that energy use and wait times grow way faster as videos get longer or higher-res—a six-second video can need nearly four times the power of a three-second one.
Most of this load (over 80%) falls on GPUs, and bigger models can use thousands of times more power than smaller ones.
Tech giants like Microsoft and OpenAI are now betting
Experts like ex-Google CEO Eric Schmidt warn that AI's future could hit a wall because of its huge electricity needs—US demand alone could rival 92 nuclear plants.
Tech giants like Microsoft and OpenAI are now betting on nuclear and fusion energy to keep up.
Hugging Face suggests smarter tech tricks—like diffusion caching and cutting out waste—to help shrink AI's carbon footprint.