There is market for reliability, security and scale in compact size, so they can get new customers.
Companies like Robinhood may discover that it's actually cheaper to by reliable and secure hardware and write software into it than try to write fault tolerant and secure software over COTS hardware.
Is it really that easy to write reliable software for mainframes? What sort of abstraction do you get from them that you don't get in commodity hardware?
Nice easy shot at Robinhood of course, but they're a very young company. Do the big banks really have a fundamental edge that's not just more invested hours?
I think the idea is that your locks and threads are so close latency wise that you only need a few to get the job done. Versus trying to figure out how to parallelize things, some of which are inherently non-parallelizable. But sometimes you don't know until you try! Ain't life grand?
