We're currently optimising towards minimising both the distance from the compute to the storage (for speed), and the egress costs of getting data from the object store to the compute. We started with S3 & Lambda because speed & cost are minimal, but BD can also lift data from any HTTP source (with the caveat that it could be slower and/or incur egress fees).

That said, we will expand to other clouds at some point, so this is a relatively short-term complication.