LoRa with SAFE would be for devices and for direct device-to-device communication -- i.e. the Internet of Things. Agreed, LoRa would be impractical as a platform for web browsing and video streaming. But for micro-controller devices, a 1MB chunk of data is substantial enough to update an internal data table in a reasonable time. Downloading 1MB at let's say half the speed range of 25kbps would take 40 sec. Using caches and good design, it's very usable for micro-controllers.
Alternatively, I can build my own LoRa device that acts like a SAFE data cache (server) for 10 other LoRa devices (clients). The dedicated cache device server sends and receives SAFE data in 1MB chunks on behalf of the clients' requests. The server relays the info in tiny chunks to the appropriate client devices.
The dedicated cache device server is analogous to a water tower. Water (SAFE 1MB chunks) gets pumped up to the water tower during off-peak hours and then the water pressure is distributed to individual households during peak hours. An old IBM mainframe works in a similar way by running batch jobs overnight to update the databases. It may sound antiquated today but this is the way many large systems ran in the 1970's, 80's, and through the 90's.
With hundreds of dedicated cache device servers within a local LoRa network, privacy and anonymity can be further preserved -- analogous to TOR.