Everything going to "the Cloud" is just a backslide into the mainframe era, but with shinier graphics. I feel like the age of micros has ended, killed off by clueless users with flashy phones.
But people will be reminded (lots are already beginning to see) and the pendulum will swing back away from using someone else's server. Then after a couple decades, the pendulum will swing back again. Rinse and repeat.
Also, communications can and do suffer interruptions, and if you're using a remote server then such interruptions can be problematic.
My point isn't that "the cloud" is a bad model, though. It's that it involves a different set of risks and tradeoffs. For a given use case, "cloud" may be the best solution. For others, it won't be.
Perhaps slightly dated, but still relevant and hilarious: The Bastard Operator from Hell
The defining characteristic of the mainframe experience for me is extreme rigidity. Upgrades are practically impossible, scheduling time on the mainframe requires lots of forward planning, notice or having a close relationship with a sysop that you can buy drinks for to get jobs to jump the queue.
Cloud on the other hand brings scaling on demand scale up/scale down for bursty workloads, very quick to experiment with different configurations etc. The exact opposite - flexibility is very much the name of the game.
Shared unix servers (not mainframes) were also very different from cloud in this way but also had the difference that you were always in very noticeable contention for CPU and io with other users, whereas on cloud you are almost always on dedicated resources for your use case. I never really felt contention on mainframe workloads because they are so goddamn slow all the time it's pretty much impossible to tell if you're contended or not.