An Nvidia B300 GPU running inference at 100 percent utilization can generate approximately 40 dollars per hour in revenue