Inference cost at scale with napkin math

An article discusses the cost of inference at scale for AI models, particularly language models, using napkin math to estimate the number of users a GPU can serve and the dollar cost per user.