The point is still valid. If the big companies could save money running multiple small specialised models on cheap hardware, they wouldn't be spending billions on the highest spec GPUs.