What about tape?
Let's discuss the flash solution. It could be made in small or large modules, each would have its own advantages and drawbacks. Even though it's frugal, flash still needs power to function. The larger the basic module, the more power it will use. Then there's the matter of reliability. Larger modules would fail more frequently per module than small ones. And ultimately, they would be more expensive per byte because they would need more complex controllers. These factors would favor smaller modules. However, smaller modules would require more complex routing, switching, finally it would need very complex controllers per each brick of modules.
Would it be cheaper than current flash technologies? Sure. Would it be cheap? Not by a long stretch. Flash is still 8-10 times more expensive than spinning drives. TLC doesn't bring the cost down far enough.
It's also not a matter of density. At the same node, I suppose flash makers could make features denser, but even if they were twice as dense (which is rather unrealistic), we're looking at only four times the raw capacity -- which is still more expensive than spinning media.
Interference would become a greater problem, and it would probably cause the usable capacity to not increase as fast as raw capacity did. Durability would suffer, of course, but as the guy said, it's not a problem for them, especially since they already don't delete the content, but keep it hidden.
Nevertheless, it's still not a solution. Perhaps Facebook will be happy with the resultant module, even if it's expensive, if they think it will save power, or if it would be less complex to build and maintain, but I don't think so.
Which brings me to tape. There are T10000C drives that offer 5 TB per tape, and T10000D on the horizon which will offer more -- that's beyond the LTO roadmap at the moment, so I'm not talking about LTO. Tape has the nice property that when it's idle, it's not using up power and when a cartridge is needed, automation takes care of picking it up and mounting on a drive.
That said, I realize that if he said that waiting times for spinning up disks are too long, waiting half a minute or so to access a tape would probably be much too long for a user to wait. Caching part of the content on disk to wait until a tape is mounted would probably alleviate some of this concern. However, the service is free of charge, so Facebook pretty much has all power to set SLAs for it.