True, and that's why there is a project that is using volunteered, distributed GPUs to run BLOOM/BLOOMZ: https://github.com/bigscience-workshop/petals, http://chat.petals.ml.