Assuming parallel work, 100 1GHz circuits are going to be as fast as a single 100GHz circuit. If that 100GHz circuit is significantly more than 100x larger it is probably going to be significantly more expensive to build, which means it won't be attacking any hashes any time soon.

These optical circuits aren't drop-in replacements for transistors, let alone direct replacements at 2025 transistor sizes. I expect they will be more attractive for SERDES applications, taking a single-lane high-speed serial optical interconnect and turning it into a wide bundle of parallel low-speed optical lanes, which are in turn converted to electronic signals and fed to a regular transistor-based chip. It would allow for 100GHz IO without needing silicon running at 100GHz.