Musk’s plan for the merged companies is predicated on several assumptions, including that AI is not a bubble, but rather a technology that will be fully embraced in the future; that orbital data centers are cost-competitive compared to ground-based data centers; and that compute is the essential roadblock that must be solved for widespread adoption of AI by society.
I guess I just disagree with the CEO on all of those.
that orbital data centers are cost-competitive compared to ground-based data centers
Might as well work off of the assumptions that magical math-faeries will do all the computing in orbit, in exchange for candy canes. It’s about as realistic.
Solar irradiance is only about 25-30% more efficient in orbit, so with 40% efficiency (VERY optimistic, ISS does about 14%) a 1GW orbital datacenter would require a mere 0.4 x 10^8 / 1300 m2 worth of solar panels, or a square 550m to a side.
If we use ISS-style solar arrays, which generate ~7.5w/kg, and double the efficiency, it would weigh 66000 tons, which translates to 3800 Falcon 9 launches. Just to power it. That includes zero frames, zero GPU’s, and most importantly, zero cooling.
This paragraph pretty much sums up where I’m at:
I guess I just disagree with the CEO on all of those.
Might as well work off of the assumptions that magical math-faeries will do all the computing in orbit, in exchange for candy canes. It’s about as realistic.
Solar irradiance is only about 25-30% more efficient in orbit, so with 40% efficiency (VERY optimistic, ISS does about 14%) a 1GW orbital datacenter would require a mere 0.4 x 10^8 / 1300 m2 worth of solar panels, or a square 550m to a side.
If we use ISS-style solar arrays, which generate ~7.5w/kg, and double the efficiency, it would weigh 66000 tons, which translates to 3800 Falcon 9 launches. Just to power it. That includes zero frames, zero GPU’s, and most importantly, zero cooling.
Press 𝕏 to doubt. Or maybe stay off of 𝕏 entirely :)
Even if I hate Elon and all he stands for, I have to disagree on second part.
Scott Manley did a good video on the subject, and he is rarely wrong on the engineering side of this kind of subjects :
https://www.youtube.com/watch?v=DCto6UkBJoI