Cratering market share...
Though a few years old now, this is the best visual of IBM Mainframe market share I've seen:
https://itjungle.wpenginepowered.com/wp-content/uploads/2021/01/tfh012521-story01-fig01.jpg
From: https://www.itjungle.com/2021/01/25/taking-the-full-measure-of-power-servers/
It wouldn't be fair to naively project that going down to zero in a few years, but it's safe to say it's a precipitous decline.
Kyndral's finding that "Ninety-six percent (96%) of respondents are moving some workloads off the mainframe" sounds about right to me.
Being a normal Linux/Unix/Network/SAN IT guy who has been supporting mainframe systems for several years, I can easily see why...
IBM firmly embraces proprietary systems and protocols, making them mandatory wherever possible, and slowly and reluctantly including "Open Systems" (i.e. standard) support only once their solutions are woefully unable to compete. And they keep as much lock-in as they can manage... z/OS still requires expensive and proprietart FICON/ECKD SAN disk (DASD) solutions, while Linux s390 can use any normal fibre channel SAN. IBM could easily emulate ECKD (and I think they eventually will have to), but that would undermine their storage system sales.
IBM publishes a huge amount of documentation, but you need to read something like 100X as much of it to accomplish the same tasks as on an x86-64 server... IBM documentation has a certain consistent style that is both incredibly wordy with useful bits of info very sparsely dispersed. Similarly, there's quite a few documents for every piece of hardware and software, and it's not obvious which 400+ page documents you should/need-to bother to read through.
Additionally, they don't produce any introductory documentation, and it doesn't appear anyone else does, either (awful light on z/System introductory books on the market)... I occasionally stumble upon a piece of documentation with a chapter or two that would have been a good introduction, but I was never able to find it with google/keyword searches back when I needed it, and it's typically 15 years out of date, so enough has changed in the IBM world to limit its usefulness.
Similarly, they hold strong to their legacy terminology... Perhaps wanting to keep all the nearly-retired IBM sysprogs comfortable and happy, or perhaps intentionally raising a barrier to entry for beginners (and also making sysprogs unable to operate or even cooperate with the non-IBM computer world. Terms like "storage" referring to RAM, "DASD" referring to disk, "CEC being the IBM preferred term for a mainframe but IBMer's still calling it a "processor" or "CPC" in the same breath.
I'd advise almost everyone to move any processes they can off of their mainframes. (I'd similarly advise to move services off of Kyndryl, as my experience with them was both technically disastrous, and terribly over-priced for the service offered.) Besides extracting a huge toll for companies to keep their IBM systems operating, fighting futilely against market forces to stay relevant, the crunch of the needed skilled individuals, they're also an ever-present obstacle to integrating with any and all other services or modernization initiatives.