no they dont because inference stacks are getting more efficient and models are getting more intelligent per parameter.