You’re right about the time not being free and certainly hardware failure. Although HW is way more reliable than people let on.

But aws is extremely expensive per unit of raw compute performance. Obviously I concede that the stability and availability is unmatched compared to DIY solutions.

For my specific workload I need lots of memory and cpu cores. (Lots meaning my single 32c/64t 32Gb Xeon tower in my home). I almost fully utilize my resources and STIL need to pay aws for storage/tunneling/DNS/etc.

Consuming equivalent compute in aws would be hundreds of dollars per month.

Hybrid is just my preference!