Exactly, I think they figure that if they have to harvest data themselves and try to beat Google at its own game, they have a larger chance of losing than if they concede on quality by not harvesting data (as much), but try to offset that by being privacy-conscious. And yeah, perhaps they'll still be able to do a pretty decent job that might not give them too much of a disadvantage compared to Google. You often see, I think, that with a few years delay many machine learning advancements can be reproduced offline.