The total cost of ownership of a computing asset is several times greater than the cost of the actual asset.
Think of it this way: a dog can be obtained for a very nominal cost (or free) but the cost to house, feed, entertain, and provide healthcare for is non-trivial.
it's not unheard of for just the costs of deploying a new device into a large organization to be something like EIGHT TIMES what the cost is for the actual asset. That's just to get the hardware deployed and NOT the cost to keep it running.
Cutting down on TCO and streamlining the deployment of resources is a big part of the sell for cloud deployments. Particularly for computing assets that may otherwise spend a lot of their time idle.
They're great boxes for cheap on-prem etl clusters.
I prefer sell R810s over those hps because they're 2U and have better power consumption with the same specs.
Who monitors that PC, and does preventative maintenance, etc? If a part looks like its going to fail, where do you migrate your workload to, so you can take that server offline for repairs. (you need multiple servers)
Since you have multiple servers, how much does your 40GB networking cost (with 100GB uplinks) so you aren't constrained by the network? And what kind of storage network do you have, so that you can live-migrate these running machines around to different machines?
Lastly, if you co-locate the server somewhere, what does it cost for multiple redundant internet connections to the facility? And where is your failover facility, that is at least a few hundred miles away?