A SpaceX space-AI or an AI-RAN alone cannot function optimally on its own; they each have their own roles, much like a nuclear power plant and a solar panel do.
At night, there isn’t much data traffic, and a radio could function as a mini-data center to perform lower-priority computations instead of just generating losses by idling, especially when factoring in inventory value. Perhaps in the future, data will be priced like electricity—based on the time of day and season—to balance consumption. The nighttime would be a good time to render images and videos, or generate them with AI, or do whatever else AI can do now—heat a house(?), since during the day, data, AI, and mobile networks have higher demand while people are awake. I don’t know if that non-urgent computing would be cheaper to do in space due to energy costs, but if electricity and data are cheap at night, that’s good for radios.
What I’m trying to say is that, according to the belief of Hotard and Jensen, AI will be like water, food, electricity, or even data today. If you look at standard logistics, for example, you have a large banana ship that drops cargo at a few ports, from where it is distributed to terminals, then to warehouses, maybe back to terminals, and finally via small vehicles to stores, where people buy 1–20 pieces at a time and carry them home in bags. Electricity grids, water distribution, and data transmission work on the same principle as AI might in the future, if AI takes off. Sometimes it’s cheaper and better to have the banana tree in your own yard, as you alluded to: “It’s probably smarter to decentralize computing than to build vulnerable data centers.”
I hope that made some sense. I see both the data and AI network as a nervous system similar to all other logistics. 6G+AI will perhaps eventually replace surveillance cameras as well. 