Pacific's GPUs run as a managed cluster for your term, and you are the only tenant. Each pod has one customer and runs as single-tenant bare metal, so you get the whole machine. Day to day the pod is run remotely, with contracted field service in your region.