I used to run regional ISP's for a living (~150k users in multiple states). As an ISP, we had "cloud" infrastructure before it was cool. Among other things we had high speed internet connections, PRI connections, and vendor outsourced dial-up pools. As the head of technology for these companies, I was unable to see anything past my router interface (except incoming traffic, of course). When the T-1 to a customer crashed, there was absolutely nothing I could do except make good and damn well sure it wasn't a hardware issue. If my dial-up lines were down, all I could do was, well, make good and damn well sure it wasn't my hardware at fault.
This problem is nothing new. What's new is scale, and in a way that is unprecedented. Back in the day we had T-3's and multiple PRI's, and, at the end, an OC-3, but that was about the extent of it. However, when managing "the cloud", you're talking about the network backbone, along with servers, storage, and who all knows what else. In other words, at the hardware level, you're damn near blind.
What could possibly go wrong? Oh, wait...
Even back then it was complex. My survival tactic was to learn how to make the guys at the other end of the telephone lines believe I worked for them when I called. That, and be really nice to the ones who could actually solve a problem and get their cell phone numbers (and Norm, if you're out there, thank you for your home phone number!).
As someone stated earlier, the "cloud" is that part of the Visio network diagram where you have no idea what the #@$& is going on or what the @#$* it is running on, and you have zero level of visibility to it. Hence, the term "cloud".
I spent many sleepless nights talking to Cisco's help desk in Australia and waiting in a queue at the RBOC's office to solve a problem I knew without a shadow of a doubt was not mine. (Except the one time it was, but that guy was pistol-whipped, sock-partied, and sacked. And then re-sacked to make sure.) I can't imagine how anyone over a decade later figures outsourcing *MORE* could be a fantastic career move, especially when the technology behind it isn't really all that old. At least OC-3's, PRI's, and T-3's had at least a decade of real-world use when I was doing it. Doing it on unknown code with vendors who can't possibly have more than a few years of experience? I don't think so.