Why is this trivial program so large when compiled?

There's a lot in that binary:

• a header to make the binary self-describing (try running file on it)
• a symbol table, which the strip tool will remove for you (or link with gcc -s)
• the names and locations of shared libraries that you never use (five of them on my box; try the ldd and strings tools)
• startup code that loads those libraries and sets up argc and argv, then calls main
• shutdown code that returns main's return value to the operating system.

For comic effect, try linking that program statically, where your binary will include the functions that would normally be dynamically linked to DLLs. (however, this option will simplify deployment)

Don't bother.

Try to make a less trivial program and you will discover the size is not that different, until your code will start to become various hundreds of kilobytes.

Briefly: There are part of the standard library that constitute the "infrastructure" between the OS modules and the C++ semantics that manage the startup and termination of the program (all that initialize and destroy the global variables, the standard input and output etc.)

Plus: everything that maps the C++ symbols towards the memory addresses, (if you didn't require to remove it - try the -O3 -s and eliminate the -g options) so that a debugger can show the proper source code references across the execution.

Also: because of the way the memory is laid out, a binary is normally made up by chunk of fixed size. Your program may even be shorter, but at least one code segment, one data segment initializer and one shared segment (for constant values) must be present.