> You seem to not even understand the very basics of heat transfer.
Basic physics: The moon is very cold in surface shadows and below the surface. It is an enormous pre-chilled heat sink.
The surface is also the support structure for any scale of radiative cooling with the same heat physics as orbit, but much better for larger and enhanced radiative engineering.
For example, heat pumps can centralize waste heat energy. Higher heat density vastly increases radiative efficiency.
• Permanent shadow: 40-60 ˚K, -230 to 210 ˚C
• In polar shadow: 25-30 ˚K, -250 to -245 ˚C
• Under 1 meter of surface, equatorial: 250 ˚K, -23 ˚C
• Under 1 meter of surface, polar: 200-220 ˚K, -75 to -50 ˚C
Many advantages beyond unlimited heat sink/radiative area: all compute in one place, i.e no size limit, so low inter-center latencies, no orbit safety negotiations or periodic orbit re-lifts required, able to update entire data center in a single trip, easier maintenance and stability in gravity on a surface, solar panels can be distributed over distance limiting total space debris risk, different component lifetimes don't result in wasted components, ...
Only downsides are a higher Earth-Datacenter latency, lunar dust resistant design, and a need to be at a pole for all-month solar power.
Nuclear power, or nuclear + solar, would allow any site.
Note that shade can be created anywhere on the surface via reflective shielding, and power can be used to heat, in order to stabilize temperatures in a desired band. Buried installations can use insulation for even greater temperature control.
> The moon is very cold in surface shadows and below the surface. It is an enormous pre-chilled heat sink
Technically true, but not really. "Radiative cooling" is heat loss through thermal radiation and it's really ineffective. We use air cooling / water cooling for a reason.
Satellites and spacecraft are engineered to make sure they can shed enough heat and they use a fraction of the power a datacenter would. All that energy eventually gets turned into heat, and it has to go somewhere.
It's a ridiculous idea that's never going to make even a tiny bit of economic sense.
IMO, they're bad, but not so bad as "have their brain examined".
While they absolutely do have huge problems at current costs, and I don't trust Musk's estimates for future costs.
It's not implausible that collectively humanity (well, China: it's not like the ESA appears to value cheap launches yet) is going to get launch costs down a lot further, something that makes the question of "how
cheap is cheap enough?" worth asking.
Then you can take a look at the existing constellations and their combined power throughput, look at whatever fraction of that power budget is not radiated by RF/laser output for comms, and trivially that's the power budget with minimal redesign for compute.
IMO all of space is still not good enough to be worth caring about: the moon is about twice the difficulty of LEO, and LEO now getting to the point that we're seriously asking about Kessler cascades; but also in space the waste heat is currently only a problem with no currently-useful side-effects, whereas down here on Earth we have possibilities for using the waste heat as an industrial input, e.g. using DCs as the heat source for district heating, or combining with ocean water to become evaporative desalination (which is otherwise pointlessly energy-intensive).
That, and the arguments about space-based power is as yet still marginal given how hostile an environment space itself is to PV. And PV on the moon doesn't even get the advantages (launch cost or ~24h light) of PV in a sun-synchronous orbit around Earth.*
But it's close enough to not be insane to do a real engineering analysis. Even if the answer turns out to still be 10x more expensive than the ground, which is what I'm expecting it to be.
* Side note: for a while I've noticed that China has production and money to afford to build a global power grid on Earth with 1 Ω resistance the long way around. This would allow 24h PV everywhere from deserts on the other side of the world including across seasons. Less material would be required to do this on the Moon because it's smaller, and also you don't really need to go across the equator so it can be much shorter, but also this would need someone to put an aluminium plant onto the moon that has negligible consumables and IIRC we don't have one of those yet.
Still, if moon-base design were up to me, I'd suggest sending up 1000 km of HVDC cable on some early missions and put a ring around one of the poles, with some PV every 60° or so.
This is still not a sensible design for moon-based compute.
10x is very very optimistic. Practically it would be more.
Even if you assume launch cost = zero its most expensive and less practical.
And the moon is even worse. Still you can assume launch cost = zero. But energy is one part, to actually reliably land on the moon with your whole infrastructure. Connecting all that infrastucture up with power and everything else.
Your basically doing a gigantic civil engineering project all with only roboitcs, while we can't even do a civil engineering project on earths with only robots.
And if your going moon, nuclear is clearly the better option then solar towers. And if you go nuclear anyway, just do it on earth.
I mean tbh the problem was that Nevermark was making strong claims without the actual analysis. There's a lot of analysis out there already but he wasn't even doing the simple ones. There's a lot of armchair experts on space.
Look, I'd love to do more things in space, but we'll never be able to do them if we lie our way and aren't realistic at the costs, and benefits. It just creates strawmen that are trivial to tear down. The armchair experts aren't helping, they're hurting.
> Even if the answer turns out to still be 10x more expensive than the ground
You're off by at least an order of magnitude.
Using Musk's optimistic numbers, to put things into LEO, it is >$1k/kg for single reuse, ~$100/kg with ~5 reuses, and <$50/kg with like 50 reuses. That's to LEO. Moon is way more expensive.
> I'd suggest sending up 1000 km of HVDC cable
I'm sorry, WHAT?
I'll let you do the math on that one, because that stuff is not light weight... We're talking several kg/m minimum... Then consider payload...
You're being pretty cavalier about all the hard things... You can't just hand wave away these details because these "details" are just a fraction of what makes all of this so difficult.
If you're not willing to have 8 starship landings for power infrastructure, why even bother? Even with 8 landings and a magic power system, it would only be on the scale of one of the smaller Antarctic research bases.
(100 Ω is completely arbitrary, FWIW. It's a dry vacuum, so bare metal just lying on the surface could run at 1MV. Above 1.044 MV, you actually need to care about random photo-ionised electrons turning into a cascade of positron-electron pair creation events for at least part of the line, but do also consider that this is the potential at opposite ends of a loop rather than vs. ground).
> You're being pretty cavalier about all the hard things... You can't just hand wave away these details because these "details" are just a fraction of what makes all of this so difficult.
I think you misunderstood me. I'm absolutely not saying "this would be easy" (nothing in space is), I'm saying "this is what my sales pitch would be".
Consider this as what I think is the MVP of being serious about the moon, that anything less than this scale is just rah-rah flag-waving.
As an aside, I prefer the moon to mars as a "first attempt" target for this kind of thing, precisely because I expect all kinds of disasters. Toy example: Accident, hardware failure, or meteorite impact that kills the water supply? Dehydration would kill you in 3 days. Emergency return from the moon (or resupplying the moon from Earth) is fast enough to solve that; but if it happens during all but the most survivable 0.4% of a Mars mission, everyone dies.
> If you're not willing to have 8 starship landings
Your math is WAY off.
You used LEO payload... Their GSO payload is 21tons[0] and the moon is a lot further than GEO. If you use the GSO numbers you get about 37 launches. But TLI (Trans Lunar Injection) is probably closer to 15% LEO payload capacity[0], if we estimate off of Atlas V, so let's say about 50.
Not 8, 50. You're off by 5x-10x.
> I think you misunderstood me.
Look, I don't want to call you dumb, I actually think you're pretty smart. But rocket science is famously hard. Many things are non-intuitive (true for most hard subjects).
I also think you should take a step back here and think about what you're saying. Look at your number of lander estimates here and how far off you are by a simple naive assumption. I get why you made that assumption and I understand why the error was made, but also these are not the kinds of mistakes people make when they have expertise in the domain. I knew it was more than 8 before even running any numbers, I knew it was more than a few dozen. But I also know people frequently make the claim that getting to LEO is the hardest part and that this warps people's perceptions and makes for bad assumptions. You have passion and I don't want to kill that passion, but if you are this passionate then use that passion to drive you into diving deeper into the topic. Don't be satisfied with shallow knowledge, your passion is greater than that.
So I want to address the full
> If you're not willing to have 8 starship landings for power infrastructure, why even bother?
Because 100 launches is a non-starter. There were a little over 300 for all of 2025. It's a big improvement, since 5 years back we barely broke 100, but you're talking about way more. About 100 of those were from China and SpaceX hit 170 total. That's a wildly impressive number, mind you, but you're also talking about 10xing their Starship launches. These things are hard to scale. They've been doing about +30/yr since 2020 on their Falcon 9. Impressive numbers, but not fast enough and scaling Starship will be harder.
> I'm absolutely not saying "this would be easy" (nothing in space is), I'm saying "this is what my sales pitch would be".
So this is why you misunderstand me, and, I think, the conversation. Maybe the "sales pitch" works for people who don't know any better, but it isn't going to work on those with even junior level experience in the industry. The numbers are so off they will set of alarms and you get dismissed. It only makes it worse when pressed that the numbers look even worse.
Because I don't think you're suggesting it would be easy, if you did I would have laughed in your face. But I think you've underestimated how hard it is, even though I think you think it is really hard. There's no limit to how difficult something can get so it becomes easy to underestimate the difficulty. This is just like it is easier to make bad estimates of distance when looking at something very far away, it is easy to think something is 5 miles away when it is 10. This doesn't make one dumb, but rather that we need to more accurately be aware of our level of uncertainty. And in this case, it is pretty high. Why wouldn't it be? There's literally no expectation for it to be unless you're an aerospace engineer working on lunar systems.
[1] The Wiki says 100k but with in-orbit refueling and there's even a note about needing a better source. So that doesn't really count for our estimates. Saturn V and SLS has a bit better, hence the range in the next line. But also remember Saturn and SLS don't have to do returns... You'll find this helpful: https://forum.nasaspaceflight.com/index.php?topic=49117.0
> So this is why you misunderstand me, and, I think, the conversation. Maybe the "sales pitch" works for people who don't know any better, but it isn't going to work on those with even junior level experience in the industry. The numbers are so off they will set of alarms and you get dismissed. It only makes it worse when pressed that the numbers look even worse.
Yes, absolutely this. I'm not even coming at it from the side of the conversation you're arguing here.
Again, I don't actually believe Musk, and all the stuff I'm saying absolutely should *not* be treated as a complete ready-to-go mission plan; it's nowhere near that detailed, and I know it.
None of this was intended to be a "he can do it!" cheerleader, because I don't believe Musk can even get close, I'm saying "As a less bad alternative to him talking about a million people on mars by 2100…" or perhaps "As a less bad alternative to waxing lyrical about something that would be within spitting distance of fundamental thermodynamic constraints even if we start by assuming we've tiled the entire surface of the Moon with theoretically perfect PV" (which is ball-park what I get for his 1000 TW/year number).
What I suggested was only "what I'd be talking about if I had what he's promising", not what I think Musk can actually deliver. Even where Musk has beaten incumbents, it's by being less bad at price-timeline estimates rather than actually good at them.
Also:
> Look, I don't want to call you dumb, I actually think you're pretty smart.
Thanks, but do feel free to call me an idiot on this. I mean it: it took what I now regard to be an embarrassingly long time before I became skeptical of Musk's claims.
And I am not, and do not claim to be, even a junior level experience in space. Well, except for processing data from earth observation satellites, where I can claim *exactly* junior level experience and no more than that.
> You used LEO payload.
As per your own references, landings, not launches. Yes, this may be over-optimistic, but hopefully I'm saying often enough in this comment that I don't take Musk seriously any more.
I absolutely agree SpaceX have not demonstrated what they need to demonstrate to actually pull off the orbital refuelling plan, but (and as per your [1]) the target payload *if* they could was still 100 tons last I checked… well, assuming Musk doesn't randomly change everything again, which at this point I expect him to do instead of delivering any of this.
I'd like to not be skeptical of Musk, but, well, he's repeatedly demonstrated reason to be skeptical of every claim he makes in every field, and unfortunately SpaceX is merely his least-wrong domain rather than one where he's close to correct.
Still, steel-man and all that. Given what he's saying he plans to do, what would I do with that? Not Mars, not space data centres.
> Because 100 launches is a non-starter.
I don't expect Musk to actually succeed with his prices, but *hypothetically* if he did, the target price per launch is order-of $10M (and if that target price sounds stupid, I'm more inclined to believe anyone on this forum dissenting than Musk's own claims on this), *if* then it would be lower than a Falcon 9's current price to launch. Again, I don't believe him, Cybertruck's launch spec was higher price for worse everything than he initially announced and that was something he should've been able to estimate better.
As someone who cares about the environment, I am not too happy about even the current rate of launches given the apparent lack of any of those Sabatier machines he kept saying would let them do ISRU for return trips from Mars. So I also kinda want him to fail here.
But if, *if* I was taking those numbers seriously, if I was worth a few hundred billion and wanted to demonstrate serious commitment to building up off-Earth infrastructure, *then*.
Things scale so differently, we don’t need a parts list to make a general tradeoff relationship.
Of the moon and orbital, orbit is much closer and will be cheaper to start with.
But a lunar site would scale to much greater mean density and unlimited total capacities. And be much cheaper for reasons I gave, at some threshold scale.
Neither is easy, and it’s not at all clear that either is actually better than down here. Especially with nuclear efforts and funding rising quickly.
Well then, use your math to address a point. Just one.
In a vacuum, radiative heat loss per time hyper scales with temperature to the 4th power.
In orbit large and complex heat transfer systems are not going to be practical. On a surface, specialized heat pumps can localize heat energy to very high intensity. With critical reliability advantages of stability, vibration control, complete sun shading, weaker size constraints, etc.
That is a tremendous advantage that will overwhelm most other details and tradeoffs, because the two main constraints, and operating costs, are energy production and heat dispersion. The latter imposing a limit on the former.
(You have no knowledge of how effectively I use my time. If you have a valid point, make it, instead of - whatever you are doing. Claiming you know things without sharing your reasoning and aspersive language are for the posers. Just communicate why you think, what you think.)
> In a vacuum, radiative heat loss per time hyper scales with temperature to the 4th power.
Correct, but you do know that the Stefan–Boltzmann Constant is 5.67e-8 W/(m^2K^4), right? And that emissivity <1 in all real world applications? It is only 1 when a blackbody is radiating in a vacuum.
This is how I know you don't understand the math. Because you didn't take the time to understand it. Plug in dummy numbers. I'll make it easy, 100^4=1e8.
You didn't think about how T works and the domain we're operating in. T's power isn't a huge factor when we're trying to dump lower levels of heat.
Let's say we're trying to discharge our power draw of 300W at 100C, that's still going to take a 0.5m^2 black body radiator sitting in perfect darkness PER CPU!!! A data center has hundreds of thousands!
You realize how much fucking surface area that is?
And this is before we consider all the other heat generated from the datacenter and the fact that you're dumping heat back to the moon's surface which will radiate it right back at your radiator making it much less efficient. Add the sun and you're fucked.
> You have no knowledge of how effectively I use my time.
Of course I do. You made a statement so preposterous I know you don't use it to do math or physics. Maybe you watch some math YouTube but that's not the same. I don't have to know everything you do do to know what you don't do.
You know how I know this stuff? It's because I've put things into space. Yet you were even too arrogant to check NASA's website
Basic physics: The moon is very cold in surface shadows and below the surface. It is an enormous pre-chilled heat sink.
The surface is also the support structure for any scale of radiative cooling with the same heat physics as orbit, but much better for larger and enhanced radiative engineering.
For example, heat pumps can centralize waste heat energy. Higher heat density vastly increases radiative efficiency.
• Permanent shadow: 40-60 ˚K, -230 to 210 ˚C
• In polar shadow: 25-30 ˚K, -250 to -245 ˚C
• Under 1 meter of surface, equatorial: 250 ˚K, -23 ˚C
• Under 1 meter of surface, polar: 200-220 ˚K, -75 to -50 ˚C
Many advantages beyond unlimited heat sink/radiative area: all compute in one place, i.e no size limit, so low inter-center latencies, no orbit safety negotiations or periodic orbit re-lifts required, able to update entire data center in a single trip, easier maintenance and stability in gravity on a surface, solar panels can be distributed over distance limiting total space debris risk, different component lifetimes don't result in wasted components, ...
Only downsides are a higher Earth-Datacenter latency, lunar dust resistant design, and a need to be at a pole for all-month solar power.
Nuclear power, or nuclear + solar, would allow any site.
Note that shade can be created anywhere on the surface via reflective shielding, and power can be used to heat, in order to stabilize temperatures in a desired band. Buried installations can use insulation for even greater temperature control.