I think this article underestimates the inefficiencies in large, centralized systems where efficiency is not a goal, and incentives are weighted against it. Famously, Facebook's iOS app is over 100 MB[0] and for Android, their engineers hacked the Dalvik VM because they had so many class definitions they were overflowing a 5MB buffer just for the definitions - and then, staggeringly, they bragged about this obscenity…
From what I've seen their datancenter stuff is way more competent >I can only imagine their backend systems are a similar nightmare, that leads to nonsense like FB building its own data centers to run their photos-and-ads crud app. Uh, of course they are running their own datacenters, they are one of biggest sites on net, outsourcing infrastructure to cloud makes no goddamn sense to them
> outsourcing infrastructure to cloud
Paying for AWS or building a data center are not the only two options. There are many co-location providers, there are dedicated firms with the expertise to run a data center for you...