Indeed and it seems attainable now, if it weren't for the expensive hardware and massive energy required for general pre-trained transformers. Don't want my car to call home just to run a neural network on Azure, it needs to run locally.