> At this point no one is buying into mainframes are they?

If you want to be a high-volume consumer bank (insurance is similar - it's about processing a high volume of transactions), you have few alternatives. You can get a couple mainframes (you should place them in different datacenters), with their extreme reliability and stability, or you can build your infrastructure with commodity distributed systems, but then you'll need to build the same level of reliability out of your unreliable components. Current solutions are getting really good at that, but it's your job to integrate them. In the end, it's not really a matter of how much money you spend, but how you spend it - the numbers will closely match.

Regulators don't really care how you do it, as long as you don't fail to provide adequate guarantees to your customers (banking is extremely regulated, and customers are well protected).

Someone has to be buying mainframes. Otherwise, there's no point in continuing to develop them. Shareholders would revolt.

They are mostly leased, I think.