They say that they store data in three separate servers in two data centers. Three HDD's must fail between two data accesses or scrubs to lose data from HDD failures.

For 99.99999% durability their way of doing things is good enough to protect from HDD failures. Of course, there are other failure scenarios.