First, the overstatement. The internet is, in relation to this topic, not all that different from a library. Before, you could get books in the library, now you can get them on the internet. I include lectures here, because a lecture without interaction is really not very different from a book. Some people prefer to listen rather than read, so having both is good for them, but I do not see the great big leap forward here. Both are unidirectional means for transferring information.
Second, the good bits. What I think is worth thinking about is what aspects of the university experience (and I am going to focus on that here) are and are not easily duplicated over the internet. Lecturing is easily duplicated and I think that we will come to see some faculty lectures, particularly for introductory courses that are heavily textbook based and have fairly standardized material - think of introductory economics - replaced with recorded lectures given by the textbook author. This would mean that students would hear from Greg Mankiw himself rather than a live faculty member. Given the wide variation in the quality of lecturers, I think it makes good sense to have more specialization than we do, so that everyone benefits from the best lecturers. This would not mean the end of local faculty, but they would then devote all of their time to the aspects of the course other than unidirectional information transfer. This means, for example, interactive sessions in which students can ask questions and discuss issues raised in the text and the recorded lecture. In some contexts, it may also mean grading, though for many introductory courses multiple choice questions with automated grading are already used.
Many aspects of a university education besides just problem sessions cannot easily be put on the internet. First, universities select peers for the students. Student learn a lot from their peers, not only in group projects but in informal study sessions and in the broader social project of "finding oneself" that takes place during the university years. Universities also, essentially, act as dating services via their selection procedures, and having a highly selected pool of potential partners to choose from is a non-trivial part of the value of attending a top school.
Second, universities can afford to purchase the physical capital required to effectively learn many disciplines such as chemistry or medicine, in part because they can spread the cost over many students and via the joint production of teaching, research and health care. There is no substitute for actually participating in research if one wants to learn how to do research.
Third, universities can provide direct interaction with faculty. Students, particularly undergraduates, make too little use of this aspect of university. For gradual students it is absolutely critical. In learning to give seminars, you need to give seminars to audiences that understand the data and methods being used and the background literature. Similarly, to learn how to write papers you need not just readers, but expert readers who, ideally, have produced their own papers and can give informed feedback.
So, no revolution yet, but I think still some internet-based change is coming down the pike in the teaching part of academia.
Addendum: Peter Thiel simply does not know what he is talking about in terms of the numbers. Perhaps he should read the literature. It's on the internet. The issue with the numbers is more the thoughtless extrapolation of positive (and broadly compelling) estimates of the impact of "treatment on the treated" to the presently untreated. Also, the numbers are about averages, and there are surely individuals for whom higher education does not have a big financial payoff (and maybe even not a big utility payoff, which is the relevant metric for decision-making). There is likely progress to be made in figuring out more effective ways to identify such individuals ex ante.