The data volume isn't very big. Let's assume Uber records locations for 20 million users and pings each user every 5 minutes. We're talking about 20 million users * 12 events per hour * 24 hours * 16 bytes (my generous assumption of how much it would cost to store a user id, timestamp, and precise long/lat data) = ~100 GB a day or 3 TB a month. And 20 million / (5 * 60) = 67k events per second.
I could probably build out the infrastructure to process and store a year's worth of user location data for 20k a year on AWS.