All the answers so far are missing a crucial point. The “bitness” of a CPU or application refers to the width of the data bus, not the address bus. Put simply, a 64 bit CPU, on accessing memory, receives a 64 bit wide word from that memory. This is the data width and only vaguely related to the amount of addressable memory. The CPU works with data in chunks equal to its bitness. This is the main reason a 64 bit OS needs more RAM than a 32 bit OS, for the same number of fetched instructions twice as much RAM has been accessed.
To give an old example, a Z80 has a 16 bit address bus and an 8 bit data bus. Guess what its bitness is? Yep, it’s an 8 bit CPU but can address 65536 RAM locations. Pointers such as the instruction counter are held in two 8 bit registers, so the problem of addressing 16 bit wide addresses in an 8 bit CPU is solved.
We don’t need an instruction set any bigger than can be represented by a 16 bit word, so any data bus width beyond that is only useful for well, data. A wider data bus means more data can be moved between the RAM and CPU in a given number of clock cycles, and wider registers in the CPU means larger or higher precision numbers can be processed natively. But we reach a point of diminishing returns. There’s very few applications that require precision beyond 64 bits, and the requirement of more RAM to carry the redundant leading binary zeroes becomes ever more prohibitive.
The sweet spot at the moment is probably somewhere between 32 and 64 bit data widths, and it’s unlikely that in anything other than super high precision specialty applications this will ever be any different. So no, a 128 bit operating system will likely never come to the consumer market.