There are still some use cases where light-based computation could work very well without miniaturization, just don't expect it to replace general-purpose chips

There are still some use cases where light-based computation could work very well without miniaturization, just don't expect it to replace general-purpose chips. One possible application is in the amplification stages of optical communication links. If you could do error correction in pure optical circuits, you could avoid the optical -> electrical -> optical conversion, and save lots of energy -- something quite relevant for submarine cables, for example.