They’re measuring individual photons, right? I’m old enough to remember modems slower than that. No doubt they will get faster, but what if they were used only for (say) key exchange?
"20k photons at 99% fidelity" usually means "we can detect each of these 20k photons separately with 1% error rate", which would be roughly 20kb/s. Low bandwidth, but it has little to do with how much light comes from a flash light or from the night sky, because they are using detectors that detect each photon (as opposed our eyes or typical network equipment which might need a large average number of photons per bit).
AIUI, this sort of system is likely to end up a lot like TLS negotiation and asymmetric crypto in general. Even if it never has very much bandwidth, the ability to share secrets on a channel that is not just declared secure by engineers, but declared secure by physics has some advantages. Asymmetric crypto is generally atrociously slow too, but incredible useful because we can use it to bootstrap symmetric crypto which is very fast.
Of course, if it does upgrade to the point that it can move gigabits per second, they'll use it for that. I don't know of a particular reason why this would be fundamentally impossible. I'm just saying that's not a necessary condition for the technology to be useful.