We've built our own dataset, largely based on the US Census Bureau's TIGER/Line data which we've converted into a useable format. This is why we are US-only -- most countries don't provide such data, and if they do they charge an arm and a leg for it (ex., the UK).
By the way, how big is your dataset: millions, billions records for the U.S.? Or how many (approximately) records are there for, let's say, Washington city or LA?
