The comparison I like best is that if you waited until it was completely built, then switched on the entire array - in the first _hour_ of operation it would collect more data than every other radio telescope in the world, combined, has managed to accumulate in the entire 90 year history of radio astronomy...

That's why there's going to be ~10 MegaWatt's worth of computing power to do as much analysis and reduction of that data in realtime as possible, so you don't need to store it all.