If you really needed it in memory you’d use one of the file APIs that will map it and present a direct byte buffer view over that memory.

Those APIs use long as their offset unlike the 32 ints used by arrays, and would avoid having to copy the data into some other object.

There has been some discussion over the years about how arrays could be changed in the JVM to support longer lengths, but doing so without breaking existing code and while providing truly useful functionality without providing obvious footguns isn’t as easy as you might think.