For text parsing, it'll depend on your buffer size. If you're able to use a buffer smaller than a few GB, then you'll be fine for RAM. For parsing code or human language (or any file smaller than that), you won't need anything even close to that.

For neural nets, it depends heavily and how the code is optimized and what your inputs are. I know that isn't very helpful.

In general, the people I know who have trouble with high-end, consumer-grade hardware (~$1,000 in parts for a single machine) are usually researchers. Then you're getting to the point where you'd want to look into distributed processing.

Thanks. that clarifies a lot.