Bitcoin nodes don't need to store or provide access to historical blocks to operate. They only do today (to the extent they do, many don't) to aid new nodes coming up securely, but in the future that will be accomplished via other means because transferring terabytes of blockchain to process and throw away whenever someone starts a new node won't be sufficiently viable.
Very interesting. Is there some technology under consideration to replace the traditional "initial blockchain download", or some concrete research on one?
There was, of course, the "Mini-Blockchain scheme" [1] (Bruce, 2014) but it had been described as "flawed" (in a discussion I don't remember); I believe due to some attack vectors [2]. Another scheme is Rollerchain [3]. Maybe also [4] (Matzutt et al., 2020) and [5] (Sforzin, Maso et al.) are relevant.
It's an extremely interesting topic as it would also probably solve the problem with the "right to be forgotten".