>...the bottom line is that you still have to incentivize all of the nodes and peers to keep the data and you need to do so in a decentralized manner that will always be orders of magnitude less efficient than hosting on a standard cloud server. That inefficiency is expensive.

I'm not sure I agree that it will always be orders of magnitude less efficient. I think if you have enough decentralized nodes you can get closer to the ideal painted by "edge computing". I also think people wouldn't mind storing information in a decentralized way as long as it's cheap enough to do so. It kind of reminds me of the transition of most companies to cloud computing -- "why pay a subscription for a server when I can pay a one time cost for my own?"

Although people slowly began to see the benefits of paying proportionate amounts of money for arbitrary usage patterns, having a dedicated team for hardware availability, and the ease of scaling out, it did take a while to get there. Just like those who have come to understand the benefits of cloud computing, I believe people will come around to favoring decentralized models for some use cases. I also believe the amount of data most people generate and use on the web is so minuscule that the storage costs will be so cheap to pay for that people will worry about it as much as they do a purchase from the McDonalds dollar menu.