I probably misunderstand how the fediverse works, but my worry is that the small instances won’t be able to hold an ever-growing amount of data forever.

Let's pretend you run a small Lemmy instance (~100 users).

If you federate with a large instance, you (i.e. your instance) will only receive new posts from communities that your users subscribe to, or users that your users follow [1]. These are deduplicated, in the sense that if all 100 of your users subscribe to the same community, you only need to download and store one copy of that community's posts in your database.

[1] AFAICT. The current implementation of Lemmy seems to handle federation using the activitypub_federation crate. I skimmed the docs of that crate, but they aren't 100% clear about this.

the posts I’m posting here today might get lost in time because the instances that annex it will have shut down by then?

You have the same problem with any data you put online anywhere: The people currently keeping your stuff online might delete it anytime they decide it's not worth the trouble to keep it online.

If it's important to you that certain information stays online, keep a copy on a disk in your house; check back periodically to be sure it's still online, and if it's not, you can always use the copy in your house to put it online again somewhere else. If it's very important to you, keep multiple copies on multiple disks hosted by multiple companies on different continents.

50 years from now on

Predicting what will happen in tech in 50 years is a pretty daunting challenge.

50 years ago, in 1973, all the computers on the ARPAnet (the predecessor of the Internet) could be easily listed on a single piece of paper. The home computer was still years from birth. The Zilog Z80, Intel 8080, Motorola 6500 and the MOS Technology 6502, which would play key roles in early home computers and gaming consoles, were just beginning to enter the market.