They've got a bunch of completely independent regions. So they've really scaled their Rails monolith to handle 100s of millions or less a day and get to a billion by having 10+ regions. Which is obviously still good but not quite the same.
Author of the presentation here. While you are right that sharding is an incredible tool that makes our lives much easier, not everything in our system is/can be sharded.
Also, we do handle more than a billion requests/day, that was just for giving the title a nice ring.
That level of scale always boils down to partitioning work. The fact that these partition are distributed across regions is done for risk mitigation (e.g. what happens when AWS's eu-central-1 catches in fire). Architecturally, nothing really changes if you stuffed them all in the same region.
