why assume if they are publishing annual failure rates as well as the level of redundancy? Your assumptions are wrong.
They published that they use 17:20 erasure coding. Have they published exactly how much overhead they get from their filesystem? I don't think it's unreasonable to blindly guess 5% for that.
The published drive failure rates are about one drive per year per 60 drive vault, which is exactly what I said. And we can estimate the labor cost of replacing that drive as tiny. What we don't know is the cost of power, networking, and physical space for the server. So I picked $400 (minus the drive fraction) per month as a nice round "this is double what I can find for many colocation offers, but still avoids losing money" number.