SpaceX wants to put 1 million solar-powered data centers into orbit

[IanD]

This is why everything comes down to launch cost per kg. The additional few percentage points in capacity factor doesn't swing the pendulum much. The whole concept seems viable and competitive even at >$250/kg, Musk's target is $10-20/kg. At $10-20/kg, the economics will be entirely in favor of orbital datacenters, just brute force with more batteries and solar panels. Until Starship is perfected, all of this is all speculation, but they have a good track record with F9 and I'm pretty sure they've considered all the stuff we're discussing here with their internal planning.

The FCC application calls for orbits from 500km to 2000km, the latter of which approaches MEO. There's more concern about damage from the Van Allen belt and energetic particles higher up, so one more factor to consider when going up. As for decommissioning and deorbiting, that's something they've been spending a lot of time and focus on since it directly impacts their future. I believe they've only had one, at they put it, dead as doornail satellite to date, the rest have been deorbited or are in the process of being actively deorbited. Their hall effect thrusters use only $10 worth of argon gas per satellite, their relentless effort to reduce cost is, quite impressive.

View attachment 4186
Not sure about active use times, there's a few charts out there, but the bigger problem with AI is the short peak-to-trough intervals on the order of milliseconds. This is problematic for the terrestrial grid and cannot be resolved without massive battery storage, but becomes less of a factor if all power will be provided through a battery in orbit.

View attachment 4187
Refer to this link for more on that: https://www.nerc.com/globalassets/w...ristics-and-risks-of-emerging-large-loads.pdf

Which is exactly what I said earlier, but which was voted down, possibly by the usual Elon-haters who want SpaceX to fail... :-(

(just for the record, I too hate Elon with a vengeance -- but that doesn't change the facts about whether space datacentres will work or not, though they do need Starship to get the cost per kg down)

If the satellites can be positioned in the orbital band over the poles which always sees the sun then the need for batteries disappears -- yes they'd then have to hop via other global-coverage Starlink satellites to get to worldwide users but this is already happening. The short-term power peak-to-troughs can be managed with small/light supercapacitors if no long-term (hourly/daily) storage is needed.

The objection because of the heat radiation/cooling problem doesn't look valid if the area needed for backside cooling is similar to that for frontside solar, which seems to be the case. To avoid heavy-duty power/cooling distribution it's likely that the NPUs will be distributed across the array and linked by low-power low-latency high-speed optical links, the same as those already proposed (e.g. CPO) for replacing copper in AI datacenters -- so you could end up with ~2kW panels (one per NPU?) with solar on one face and radiator on the other, with the hot bits sandwiched in between.

Still not guaranteed to happen though, even if the technology and economics both work the idea could be scuppered by launch capacity (though 4-8 launches per 100MW datacentre doesn't seem bad) or politics/regulation... ;-)