Admittedly this would have been an exotic for the Bay Area but mundane for elsewhere use-case since we just needed really fast disk and fast compute because we just ran model tuning and backtesting on the machines. 90% utilization, 100G network, 4xNVMe. We only had an 8-member team so much of the management/IT-layer replacement stuff that Oxide enables wouldn't have been as high leverage across the tens of thousands of cores. We almost certainly left a lot of stuff on the table, though.

Also, is your username something interesting? It feels familiar but a quick `echo -n '' | md5sum` didn't yield anything.