Supply of GPU servers for generating large model training and demonstration