At peak capacity, bare metal pays off soon. Even when they run at near 100% of CPUs, weather models never run 24/7 but only fraction of the time. Examples: A n

At peak capacity, bare metal pays off soon. Even when they run at near 100% of CPUs, weather models never run 24/7 but only fraction of the time. Examples: A national weather service (any country) may run at 20-50% capacity. In a research setting, 5-10% is more common (personal experience). Consulting businesses are somewhere in between. The lower the fraction of capacity used, the longer the time after which bare metal investment pays off. For businesses that need to do a one-off feasibility study or to work on a few projects/events, upfront investment of > $100K for bare metal is prohibitive. And this is before hiring a sysadmin and a weather modeler to run the thing. I appreciate your comment, it motivates me to take some time soon to write up the unit economics in more detail.