At those power levels they would have to use some kind of highly error-corrected modulation and coding scheme to provide enough coding gain to overcome the path loss. I agree they are pretty optimistic, but until they detail their modulation scheme, it's hard to tell.
A few years ago I was experimenting with 900 MHz LoRa for a work project -- we had need to communicate a very small data payload from inside elevator cabs, with forgiving latency requirements. So we took a LoRa board to a hotel building 2 city blocks away from our lab and cranked the coding gain up to the max, which gave us about a 1 byte payload every second. Perfectly sufficient for our application. Astoundingly, we had great copy in our lab even when the doors of the elevator cab were closed, inside a building 2 blocks away. I can't remember the power level, 500mW I think, but I may be wrong.
A few years ago I was experimenting with 900 MHz LoRa for a work project -- we had need to communicate a very small data payload from inside elevator cabs, with forgiving latency requirements. So we took a LoRa board to a hotel building 2 city blocks away from our lab and cranked the coding gain up to the max, which gave us about a 1 byte payload every second. Perfectly sufficient for our application. Astoundingly, we had great copy in our lab even when the doors of the elevator cab were closed, inside a building 2 blocks away. I can't remember the power level, 500mW I think, but I may be wrong.