It greatly depends on the requirements that the model has. Some models can even run off of cpu and system ram. But, there are many GPU VMs available with AWS, Azure, etc. and you’d deploy it similar to how you would your other applications.