Not even about improving bottom line. The costs of inference at scale are staggering for something like GPT-4. Inference costs rise exponentially with model siz

Not even about improving bottom line. The costs of inference at scale are staggering for something like GPT-4. Inference costs rise exponentially with model size. Until everyone stops paying the Nvidia tax, there is no service based on a very large model like GPT-4 that can scale and be profitable. Beyond that, they are constrained by Azure’s GPU infrastructure as well.