If each article weighs in at 1mb (probably an overestimate) then 1mil papers weighs about 1Tb, so for their currently reported 47mil papers you are looking worst case at 50Tb. If we assume they have about 50% coverage (which seems reasonable base on this article [0]) then 100Tb total under the 1mb avg. I'm guessing that number is quite high because the true average will be significantly less than 1mb and compression could bring it down even more.
