The thing is: Most searches aren't even that complicated. Whatever "magic" (read: extra processing power) Google uses is mostly helping for extreme niche cases. The rest they seem to do, nowadays, is "editorializing" results, pushing popular websites before more relevant ones, displaying results from some internal database, etc.
As far as I can tell, 6 years ago the reason was that AWS didn't support IPv6[0]. It has for 2.5 years by now though.
[0] https://web.archive.org/web/20180909174057/https://duck.co/f...
I doubt they have any IP-dependent code, since their main selling point is not tracking anything.
I've actually run into Gabriel a few times online and had a bit of conversation with him. Nice guy. Can't really say I've chatted it up with Eric Schmidt...
Sometimes Google seems to be better at guessing the context especially if you search for C and some other term that has a bunch of different meanings.
I find it pretty funny that I automatically add !g when the Google results suck as well.
Especially on mobile, making it challenging to edit URLs on their latest chrome browser. UX consistency has gone out the window for weird things like like swapping around the tab ordering for searches (check out chicago, chicagos and chicago's here: https://pbs.twimg.com/media/Cm0C1o8VYAgLv2O?format=png)
Other things like not providing custom date ranges on mobile search are just utterly baffling. Booleans are gone, ranges are gone, and it ignores most of the words I put in. I honestly think alta vista results circa 1997 were better than what I'm getting these days. In fact, they most certainly were.
They've also abandoned their namesake "googol" ... they don't seem to care about http, newsgroups or many other things. They should probably rename themselves to "Around300orso"
Google ceased providing functional tools for technical people a while ago. There's some decent open space for another firm to come along (like DDG or perhaps the Microsoft Renaissance) and just snatch the technical user, create compelling products for just that, and own it.
I'd even pay probably $250/yr or so for such a no-bullshit resource stack - a high quality searchable answers-oriented quick-to-use, quick-to-read technical reference without a bunch of wrong information or concrete answers buried in pages of theory. Heck, I'd probably consider dropping $2,000, that sounds amazing.