They have a lot of servers and it's more cost effective to come up with their own solution for management, monitoring, maintanance, etc?
You can't exactly wait for a kernel patch or a fix for a breaking change if you're working with over a million servers.
Also to get bang for your buck You need your programs with a lot of patches to make them fit for your specific requirements and stripped of any code that goes unused or is deemed a security risk which means a whole lot of packages that are compiled in-house. hell even individual packages that are forked are posing problems for them (see https://boringssl.googlesource...)
All of this means you have to use your own package repositories which means even if they use redhat as a base it's not redhat anymore.
Did I mention ridiculous pricing?
Well, that used to be true about five years ago.
One of the laptops is a sony vaio; sdcard, bluetooth, keyboard backlight, nvidia card , screen backlight(yes, it was black), hdmi and audio jack didn't work or needed days of figuring out.
With linux > 3.4 they all work out of the box now.
Damn, someone please do this. make LFS, link/dev/null to/dev/siberia in your build environment and s/null/siberia/g on every package. PLEASE. I know there are nutjobs among you good folk who are crazy enough to do this.
In highschool I had some variation of this (from here) memorized as a number and would convert it to binary before every damn physics/math/trigonometry test.
Don't forget to 'shred -u' your files after doing that. nothing is safe.