However, your comment has me thinking that FB doesn't have a way to lock people to their cloud - they'd be forced to compete on price more than any other major cloud provider. Probably not a great business.
But is Apple a cloud provider? I recall they have some data centers but I thought that it was rumored they used Google Cloud or AWS (or both) too.
When Amazon got into the cloud business, they actually had most all of the non-software challenges solved already. It turns out that datacenters and distribution centers go in the same kinds of places and have very similar kinds of challenges. They benefited significantly from knock-on effects from running their existing business.
If Facebook got into the cloud game, they would have to learn two new businesses at once: cloud hosting and industrial real estate.
I don't think that they would be able to obtain good pricing on either real estate or server hardware. Their current hardware vendor can't absorb that large of a scale out.
although it's outside facebooks current core competency i think they wouldn't have too many issues getting into the cloud game if they wanted to. they could hire all the expertise they need and get 100 data centers up and running in 5 years.
but why would they? it's like google getting into the ISP market to compete with verizon et al. google runs its very profitable business on top of the commodity rails and keeps limited deployments of google fiber as a stick to threaten slow or uncooperative ISPs. facebook can run its profitable software based businesses on top of the infrastructure instead of joining the cutthroat race to the bottom of cloud computing that may or may not be semi nationalized/regulated as ISPs and utilities are today
Facebook can't say the same and getting 100 datacenters up and running would cost them 25 billion before hiring anyone -- that's more than half of their cash.
I agree with you though that it makes zero sense for Facebook to do this.
That sounds good in a blurb in CIO Magazine but is likely completely disconnected from reality.
The complexity of building a cloud provider is not primarily in the physical management of physical assets in a big box building.
Building data centers is hard because there are not a lot of places that you can build them effectively. Large scale, specialty real estate deals are the kind of thing that effect a company's financials in a big way and for a long time -- much longer than the market cycles where you determine whether or not to continue or abandon your hypothetical, nascent cloud offering service -- and are exactly the type of thing that market investors will pillory your company for if you fuck up.
If you're Amazon and you're starting off in the datacenter game and it doesn't go well, you can always transition the property into a distribution center. Those are useful to you anyway.
If you're Facebook, what are you going to do? Who are you going to sell this $800-1200/sq ft (that is the cost of building a turnkey DC and puts AMZN's largest datacenter somewhere between 172-258mil and that's just one of them) "big box" to? They should make a multi-billion dollar investment for that and hire a global enterprise sales organization on top of that to sell it?
Maybe it's not so much that what CIOs say is divorced from reality but that they see reality at a scale that you don't. Their abstractions may not make sense to you.
As an enterprise customer of all three, I will tell you that both Google and Microsoft have terrible UI, terrible permissions sets and terrible documentation. Microsoft has terrible policies and support but their interface still edged out Google. Google constantly shutting off APIs and forcing people to refactor isn't doing them any favors either.
IAM and the documentation are AWS' secret sauce. It's the thing that wins over those internal champions/coaches that an enterprise sales organization needs at their prospect companies to close large scale deals. In an environment where I have to manage security, compliance and administration on top of just using the thing, it makes my job 1000x easier.
They tried that with Parse acquisition but shut it down