Agreed. And there are big answers of course: data science, web front end, embedded, etc. But when you exclude those things, it does leave a huge swath of "generic back end" software where Java has a large if not dominating presence.
Even data-science is limited to just the science part. All the heavy lifting is still Java. Between Hadoop stack/Spark/Beam/Presto/Kafka/friends pretty much all data at any reasonable sized company is getting shifted around with Java.
Even more so if you are using the "hosted"/SaaS stuff as the cloud providers heavily use Java for their data intensive services (though Google has more C++ than the others).
True, I almost said "data science exploration" as a way to avoid this inevitable comment :). Java can't nearly compete with python (right now) for a certain type of data science programming, but you're right that a ton of data science infrastructure is built with Java.
The world "primarily" is important here. Sure, you _can_ use Java for embedded and even real-time stuff. Some games are written in Java, etc. But no one would say that Java is the primary language of choice in such domains.
Java has been used for soft real-time systems, such as payment processors or real-time bidding. For a GC-ed language, its latency guarantees can be pretty OK due to having good garbage collectors and runtime optimizations.
But, yes, obviously it can't be used for systems where latency is a matter of life and death, like operating the breaks on a car.
There actually are special, hard real-time JVMs that are used in missile defense systems. Do note that hard real time is not about performance (and as computers are so fast, you really don’t need much time for millions of calculations), but the guarantee of executing it in X time. But it does require very special engineering.
