It’s a niche market, but banks and transaction processing are likely to continue using mainframes for the foreseeable future. One of the few remaining advantages of mainframes is the ability to run ACID compliant relational databases at immense scale, where distributed systems generally use noSQL databases and eventual consistency (which isn’t good enough for financial use cases).

What I envision happening is that IBM and the System/Z platform will hang on milking their old customers for the foreseeable future. If they can develop and adapt the platform to new use cases, they *might* be able to get some new customers, but more likely I expect their mainframes to just hold onto existing markets or shrink as companies find alternative ways to get what they need from other platforms. That last point may take some years/decades though, prior recorded attempts at transitioning from the mainframe run the gamut from highly successful to expensive and bitter failures before transitioning back.

Think of all the big web-based systems like search engines, social media services and the like — they all put low latency as number one. So the massively parallel load-balanced and fault-tolerant computing facilities that they run on are as different from “mainframes” as can be. They not only need to respond quickly, they also need to adapt quickly to shifting loads — again, something that was never envisaged in the design of mainframes.

Are there any significant workloads left where mainframes have an useful advantage? Somehow, I don’t think so…

More interestingly, Linux has allowed Amazon, Oracle and others to switch to mainframe sized ARM servers in their clouds.

I see two possible futures: either yield and performance problems will be so bad with the 7nm Power10 that they instantly go bankrupt or they could hire Timothy to turn things around.