You don't merely need a lot of data, you need the right kind of data: if e.g. You have lots of users but only a few transactions per user to base your predictions on you won't get very far. There are a lot more likes in facebook's system than transactions in netflix's
It's more like the combo that's powerful: huge data, close tracking of effectiveness of recommendations and product quality, end result being open-ended subscription by user/consumers.
So he doesn't mean powerful in the sense that it's the best recommendation algorithm, just that it has a lot of data. If having data in your domain is of course a necessary condition in developing a good recommendation engine. But if it were sufficient Netflix wouldn't have to shell out $1M to outsource the task.