Link to paper: http://www.nature.com/nphoton/journal/vaop/ncurrent/full/nph... -- its contents are behind a paywall, but the abstract and several illustrations are free.
For example, the City of Melbourne (Florida) uses wireless links for intra-city WAN traffic. The cost analysis versus building a fiber network was laughable. Granted, you don't get the speed and latency of fiber, but even when they built in extra software (to provide LAN nodes for latency sensitive services), wireless came out ahead.
This development won't help with latency, but more throughput is always welcome.
I get the speed but where do you lose the latency? LoS is the shortest path and fiber is slower than speed of light so I'd actually expect latency to be lower. Wall Street low latency trading is moving to microwave to exploit this:
For example, with 1GHz of spectrum (a huge amount compared to WiFi (40MHz for 802.11n) or LTE (20MHz)), with an excellent 50dB signal to noise ratio, you can never move more than 16.61 gigabits per second.
Using the same 50db SNR for WiFi at 40Mhz we get:
ans = 6.6439e+08
It doesn't seem like they are using this though in their experiment.
I imagine that in either case, they're using multiple computers to do the transmitting and receiving.