The point I was making is that it is an estimate of the lower bound on security. You shouldn't assume that a transaction worth $50 million is safe after 6 confirms.
Probabilistic speaking then, having the $50million split in multiples transactions sent out at different block levels would considerably lower the risk of a full transaction loss.
This entire rented hash rate would also need to be in place before that transaction is broadcasted.
Trying to re-mine, 'alone', at least 6 new blocks to build a chain long enough to take over the consensus, would bring the entire network to a long wait as new blocks are not broadcasted. This raises suspicion, and at the first complaint of a successful attack the price drops to near-zero, making the attack unprofitable in the end.
It's interesting to think that the economics required to attack the network successfully are made even worse if the attack succeeds.