The .NET CLR (VM) has had proper generics and value types since 2002. Java did not implement generics until 2004 and to this day does not have user-defined value types. There was an opportunity to do it right in 2004 before this became "baked in".
There is really no good excuse. Many developers have abandoned java due to the glacial progress with both the language and the VM. It is legacy now.
The main benefit (IMO) of Vahala would be for other JVM languages (such as Kotlin) to make the implementation of value types more efficient.
(I use both CLR and JVM in my work, but is clear to me that the CLR is superior on many fronts).
Well, some companies still write many of their new projects in the Java language (not to mention use the Java platform), like Apple, Amazon, Alibaba, Google, Netflix, and, of course, banks, governments, airports, militaries, hospitals, utility companies, factories, robotic warehouses [1], and most Fortune 500 companies.
To make your comment more explicit: The parent states that Java is legacy now. But given your examples above that would mean that java and many other languages are legacy now.
Of course any language could be legacy in the future but historically the probability that one of those future legacies is a hot new language now is at least equal with that of an established language currently still widely used to write significant production software.
Assuming he can uniformly invest across startups (i.e. invest in quality in addition to lower quality), and adds to winners, such an approach is akin to "ensemble-learning" approach. Works well for portfolios if the measure of prior performance & risks used to determine reweighting is proper and there is a high positive auto-correlation between the performance measure and future performance.
There is really no good excuse. Many developers have abandoned java due to the glacial progress with both the language and the VM. It is legacy now.
The main benefit (IMO) of Vahala would be for other JVM languages (such as Kotlin) to make the implementation of value types more efficient.
(I use both CLR and JVM in my work, but is clear to me that the CLR is superior on many fronts).