From there, I can then start digging into what the real problem is. Maybe it was a desire for greater performance, or easier deployability. Maybe his concern was about protecting intellectual property, and he thought that binaries would be harder to reverse-engineer than bytecode.
All of those could be legitimate problems, and none of them would really be solved by a Java-to-native compiler.
Quite the contrary.
Excelsion JET, IBM, PTG, Aicas have supported native Java compilers since the early days.
Oracle added AOT compilation for Linux x64 to OpenJDK on Java 9, with the remaining platforms being supported later. The Java 10 fork, already has Windows and macOS support as well.
The Graal team recently announced the Polyglot VM, which is actually another AOT compiler for Java.
The demand for native Java compilers is going quite thanks to Fintech and embedded development.
Companies selling commercial native compilers for Java are still in business since the early days.
The trouble with a native compiler is dealing with dynamic class loading.
But that's all beside the point, which is people will often tell me what they want in a product, but don't actually want it.
There are/were a couple others, but the customer base is very... exclusive
I said it was a great idea, and I had already implemented one. He could start using it today.
He promptly lost all interest in it.
(It turns out, on the complex plane the demand for native Java compilers is on the imaginary axis.)