An application (decentralized or not) needs to store data, often a lot of it. For example a social media platform needs to store the list of users, everything they’ve written, the images they upload, etc.
People who don’t have all the data on their disk need a way to be able to access the data that they’re missing when needed.
When the application is a centralized service, the company behind it just pays and manages the servers that do store the data, and people can send requests to access said data. (I’m a bit abstract here, but basically you send an HTTP request and you get an answer with what you want). If the servers go down the application entirely stops working, and the company behind it obviously doesn’t want that to happen.
In the case of a decentralized service, the people who are missing some data need to download it from people who have it (and then verify that this data is accurate, which is done with hash functions and is an already solved problem).
But this creates a sort of prisoner’s dilemma. If many people store the data, then everything works fine. If nobody stores the data, then everyone loses. If only some people store the data, the people who store the data lose (as they have to pay for the disks that store the data, and serve the download requests) and the people who don’t win.
There’s no individual incentive to store the data. It’s done on a goodwill basis.
This goodwill works up to a certain point, for example Polkadot’s data is only a few dozen gigabytes, and many people are okay renting a server for a hundred bucks per month because they want to support the project. But huge websites like Facebook generate at least terabytes of data per day, and nobody is going to store hundreds of terabytes of data purely by goodwill.
There are some pragmatic solutions, such as having a treasury system pay for people for hosting data. But the people who are getting paid can always easily cheat. For example, they can simply store nothing at all and redirect all the access requests they receive to a different hoster.