Not disagreeing with you, I was just trying to connect that the index size is a function of the breadth of information.
For example, I have digitized roughly 300 volumes (books) which collectively represent about 100,000 pages of information. That collection which represents a big chunk of my originally print reference library creates a relatively small n-gram index of about 22 GB. But it is a small sample of generally available reference material and doesn't include the 22 years of digitized Scientific American articles (much harder to parse out for indexing when starting with the PDF form). But it still answers a lot of reference queries quickly and accurately for things I am interested in. For things that I become interested in and have yet to have started curating a set of references for, its worthless.
As a result my experience is that the closer I get to my long term interests the more likely I am to find something in my library to answer the question, things that are more temporal (news, new research) are not there at all generally, and things that are only now of interest are similarly not represented. The thing that Search engines do so remarkably well is that they cover a very wide swath of interesting material preemptively.
To host that locally would be a more significant effort for me.