1. What is the projected cost, versus terrestrial data centers in the future. The main reason we hear about putting AI in space is the load on the power grid. Take into account on earth we can build point-of-load power sources, be they nuclear or solar that would not cause greenhouse effects. IIRC, the incident solar radiation in space is only a little more than 2x what it is on the ground. Does that factor justify putting data centers in orbit?
2. The orbital altitude stated, I believe puts at least some of the satellites in the Van Allen belt, meaning more shielding would be required, presumably adding weight to the payload.
3. If the satellites are not in a near-polar orbit, a fraction of them will be eclipsed by the earth and lose power. To remain on-line some means of energy storage would be needed.
4. Why would so many satellites be needed? Why not scale up to larger satellites/solar arrays and fewer launches.
5. Does latency in data exchange over distance, either between satellites in space, or between space and ground introduce a degraded performance?
Not being negative here, considering Musk has an amazing track record of getting ideas into fruition. There are always technical problems facing anything of this magnitude, but Engineers are quite adept at finding solutions that work.
Redundancy. Plus many smaller satellites can be launched at the same weight of a large satellite.