Centralized entities emerge to absorb costs because nobody else can do it as efficiently alone.
>What happens when someone storing decentralized data decides to exit?
They exit, and they no longer store decentralized data. At the very least, IA would still have their copy(s), and that data can be spread to other decentralized nodes once it has been determined (through timeouts, etc) that the person has exited.
> Will data be copied to multiple places[...]?
Ideally, yes. It is fairly trivial to determine the reliability of each member (uptime + hash checks), and reliable members (a few nines of uptime and hash matches) can be trusted to store data with fewer copies while unreliable members can store data with more copies. Could also balance that idea with data that's in higher demand, by storing hot data lots of times on less reliable members while storing cold data on more reliable members.
> who pays for the decentralized storage long term? [...] who is going to pay for doubling, tripling or more the storage costs for backups?
This is unanswered for pretty much any decentralized storage project, and is probably the only important question left. There are people who would likely contribute to some degree without a financial incentive, but ideally there would be some sort of reward. This in theory could be a good use for crypto, but I'd be concerned about the possible perverse incentives and the general disdain the average person has for crypto these days. Funding in general could come from donations received by IA, whatever excess they have beyond their operating costs and reserve requirements - likely would be nowhere near enough to make something like this "financially viable" (i.e. profitable) but it might be enough to convince people who were on the fence to chip in few hundred GB and some bandwidth. This is an open question though, and probably the main reason no decentralized storage project has really taken off.