Storage is not very expensive. Extremely large hard drives are very cheap.
The costs are power, redundancy, HA, retrieval, bandwidth; even with a proof-of-storage challenge to nodes that claim to store certain information, paying them to solve "what is the hash of this range of bytes in this huge file?" is not equivalent to "will you please send me this huge file in its entirety at high speed with five-nines reliability guarantees?"
Exactly and for many of the reasons you enumerated. I personally believe SiaCoin is on its way to be a decent provider of this technology.
For example, they recently enabled a feature where your data may be distributed to whitelisted hosts which one may limit to a certain geographic area such as CONUS if you happen to be in the states. This should help reduce latency for retrieval.
The host algorithm penalizes for downtime so this should translate to HA as there is an economic incentive.
For redundancy, the data is distributed to up to 21 hosts and up to 7 can fail and the data still be recoverable.
There is so much more the Sia network has to offer and have had my eyes in it for Steem as a markdown blog storage solution.
I think this would be more ideal than building from the ground up personally and the benefit should be explored.
I looked recently at their whitepaper and for me the crypto-economy of Sia (which has nothing to do with the technical prowess involved in providing cheap reliable storage) sucks big time.
On the contrary, the Crypto-economy is Steem's strong suit. So in my opinion combining the Sia technical solution with the steem crypto-economics could prove a real winner
Indeed, this is one of the main challenges, storage is otherwise not very expensive ... but it has to be paid in USD nontheless ... and you get into a chicken and egg problem - if you acquire storage and nobody comes to use it, you have wasted your hard-earned USD ...