Do keep in mind that the fellowship targets very very young people, and it's exceptionally hard to show excellence in most fields by that age. Software is a rare exception.
Not just 'show excellence' but actually know enough to contribute to the field meaningfully. For better or worse, one can learn programming and some basic math in order to contribute to the market in the span of a few years. Traditionally to do good physics required a few years doing nothing but math, and then a few years developing a good physical intuition. There have been some great physics papers turned out by early-/mid- 20-year-olds. Modern molecular bio requires a few years chemistry, a few years of either math/stats/systems/programming, a few years of bio, and then a few more years developing an intuition for biological systems. It's just a much longer route to travel. A student - no matter how good - just isn't well suited to being anywhere near 'useful' by the age of 20.
Agreed with all of the above points. What it suggests to me is that the idea of bypassing college and going straight into entrepreneurship may only be testable in a limited number of fields, notably programming.