In my mind, mainframes are associated with either:<p>1. Old software<p>2. Large not-tech companies who outsource their IT department.<p>Do any modern tech companies actually migrate to mainframes?
I've never heard of one doing so. I'd attribute it to a few things: 1) modern tech companies already have a workforce that's familiar with the cloud and modern technologies. They're unlikely to have many employees with Cobol experience (both coding experience and general Cobol interaction) 2) few of these companies want to be beholden to IBM and the like for their mainframes. 3) Many of these tech companies have requirements for near real-time streaming whereas Cobol's wheelhouse is batch processing 4) Reputation that it's difficult AND expensive to keep cobol/mainframe divisions well-staffed as there aren't many new developers coming onboard and many of the existing developers are retiring 5) Relative uncoolness of Cobol
Only when someone is extremely stupid (or has a stake in the outcome) and falls for the marketing bullshit like the following: <a href="https://www.ibm.com/it-infrastructure/z" rel="nofollow">https://www.ibm.com/it-infrastructure/z</a> (warning - graphic content)
Once you had database clusters, it became possible to run a bunch of machines rather than one giant machine. That turned out to be less expensive for most people. The machines were cheaper, because they could be commodity machines, because each individual machine didn't have the insane reliability requirements that mainframes have.
Yes, MPI and HDF5 are still a thing and processing on a supercomputer is still orders of magnitude faster than the usual distributed approaches.<p>However applications would likely be science, engineering and finance.<p>All you have to do is go to a vendors page and check out the flagship customers.
For me, the mainframe is a big expensive box. Vertical scaling is still a thing. Most RDBMS work best on big boxes. Some Java applications would require hundreds of GB of ram. There are use-cases for mainframes but people prefer new shiny Kube clusters.