p2p syncing of the headerchain using a table for the 2016block intervals is very fast anyway. and yeah its not even 80mb at the moment. The only problem is with web-apps, thats a bit more clunky, but mempool.space has an API where you can get the header you need based of the height in 2 calls (first you get the headerhash back, and with that you can get the header), so that also works; can't verify the difficulty(as in how high it is) that way, but the internal consistency atleast (PoW in the header hash etc). Needing headers its a trivial problem regardless the environment is my point