Elon Musk quote: SpaceX will lean in big on the Moon

[meekGee]

The idea that there is any advantage in putting AI in orbit or on the moon is so stupid, I'm wondering if Elon is just trying to make up an economic justification for having a moon colony. I'd be concerned if I thought he really believed this.

Greg, since I share(d) some of this: What are the reasons it is so stupid?  I want to compare notes.

[sdsds]

It's an open question: does Musk's vision for Moonbase Alpha imply he thinks civilization/consciousness might need a backup sooner than a Mars settlement could become self-sufficient? Assuming a global collapse of civilization on Earth could a lunar AGI help restore some semblance of what we have today?

[Twark_Main]

when we collapse  the worst that can happen is we go back to the stone age. We can still make babies though.

When an artificial civilization (which, mind you, still carries the "light of consciousness") collapses, they can't even make new ones anymore - it's a much bigger problem.

New "UAP are future Earthlings" conspiracy theory unlocked. 

Yes yes, where were we?

[Twark_Main]

It's an open question: does Musk's vision for Moonbase Alpha imply he thinks civilization/consciousness might need a backup sooner than a Mars settlement could become self-sufficient? Assuming a global collapse of civilization on Earth could a lunar AGI help restore some semblance of what we have today?

I think the main overall goal is that a Moon base / Mars city will provide a forcing function to improve space travel and space habitation technology until humanity is scattered across the Solar system. So not just one backup, but (like a good sysadmin) ultimately many backups.

The Moon vs Mars is largely short-term strategizing to maximize the pace of innovation this decade, which maximizes the probability of that ultimate success.

Ultimately the reality is there's a "histogram" of possible existential threats and time horizons, and you want to reduce the "expected area under the curve" as fast as possible. So I expect there's a mix of long-term and some shorter-term risks that are meant to be mitigated. IMO.

[Cabbage123]

When an artificial civilization (which, mind you, still carries the "light of consciousness") collapses, they can't even make new ones anymore - it's a much bigger problem.

It's a fascinating point, although maybe one underpinning assumption is incorrect (i.e. that a future artificial civilization, will not find a way to make its reproduction immune to the collapse of its civilization, or at least as immune as ours is).

Maybe the first task on the list of a omnipotent silicon based AI would be to make itself an omnipotent biological AI.

[Vultur]

When an artificial civilization (which, mind you, still carries the "light of consciousness") collapses, they can't even make new ones anymore - it's a much bigger problem.

It's a fascinating point, although maybe one underpinning assumption is incorrect (i.e. that a future artificial civilization, will not find a way to make its reproduction immune to the collapse of its civilization, or at least as immune as ours is).

Maybe the first task on the list of a omnipotent silicon based AI would be to make itself an omnipotent biological AI.

No physical system can be omnipotent. It is necessarily bound by the laws of physics, and also (being limited by the speed of light) can only access finite amounts of energy and other resources in less than infinite time (even if the universe is spatially infinite & contains infinite resources - this may not be the case).

-

If the first goal of a sapient AI would be to recreate biological sapience, what would the point be?

Current (misnamed) "AI" power demands make me much more impressed by the capability of the human brain, which runs on maybe 20-25 Watts (about a quarter of basal metabolic rate).

[meekGee]

When an artificial civilization (which, mind you, still carries the "light of consciousness") collapses, they can't even make new ones anymore - it's a much bigger problem.

It's a fascinating point, although maybe one underpinning assumption is incorrect (i.e. that a future artificial civilization, will not find a way to make its reproduction immune to the collapse of its civilization, or at least as immune as ours is).

Maybe the first task on the list of a omnipotent silicon based AI would be to make itself an omnipotent biological AI.

yes - just like human civilization is built on top of a biological/chemical layer that's independent of it, an artificial life-form will be smart to figure out something similar.  A back stop.  So it's both an assumption and a call to action for our benevolent artificial successors.

But the original thought was, the AIs don't need to kill us "terminator/BSG style". It's enough that they get better and carry forward.  But that's not happening for like at least 6 years.

[novo2044]

Are we over conflating the quantum computing comments with the AI comments?

Quantum computing requires very cold temperatures, just a few degrees above absolute zero. AI data centers obviously need cooling, but nowhere near to the same degree.

The talk of such craters, is presumably for Quantum, as the baseline temperatures there are just above what is needed.

So possibly future "Quantum powered AI" but not your common or garden variety.

If I recall correctly, you can just dig about 1-2 meters down from the surface of the Moon and it's a fairly constant -40 degrees Celsius.  A simple cooling loop can take advantage of the fact the Moon is pretty cold unless it's directly illuminated by sunlight.

The permanently shadowed craters drop to 25-30K, which is way colder and totally unnecessary for cooling data centers, but might be useful for quantum computers.

It'll be a long time before automated factories are making PV panels, chips, and mass drivers on the moon.  But in a couple decades?  Doesn't seem crazy to me

[Vultur]

Are we over conflating the quantum computing comments with the AI comments?

Quantum computing requires very cold temperatures, just a few degrees above absolute zero. AI data centers obviously need cooling, but nowhere near to the same degree.

The talk of such craters, is presumably for Quantum, as the baseline temperatures there are just above what is needed.

So possibly future "Quantum powered AI" but not your common or garden variety.

If I recall correctly, you can just dig about 1-2 meters down from the surface of the Moon and it's a fairly constant -40 degrees Celsius.  A simple cooling loop can take advantage of the fact the Moon is pretty cold unless it's directly illuminated by sunlight.

The permanently shadowed craters drop to 25-30K, which is way colder and totally unnecessary for cooling data centers, but might be useful for quantum computers.

It'll be a long time before automated factories are making PV panels, chips, and mass drivers on the moon.  But in a couple decades?  Doesn't seem crazy to me

I am fairly skeptical of the degree of automatic hardware manufacturing that is really possible / practicable without human on site maintenance. *Maybe* on the Moon it could work, since light lag is short enough to do things by telepresence, but... Still pretty aggressive.

(Elon Musk has accomplished a lot, but he's not right about everything. The practical usefulness of AI and humanoid robots may be one of those things.)

[Greg Hullender]

The idea that there is any advantage in putting AI in orbit or on the moon is so stupid, I'm wondering if Elon is just trying to make up an economic justification for having a moon colony. I'd be concerned if I thought he really believed this.

Greg, since I share(d) some of this: What are the reasons it is so stupid?  I want to compare notes.

1) The cost of lifting equipment into space is exorbitant.
2) You have to put up not just your hardware but also power supply and cooling.
3) Assembly. A big data center is big. You'd have to make dozens or hundreds of launches to send up the pieces, and then how do all the pieces get put together?
4) Bandwidth from space is pitiful. Latency is bad too.
5) Radiation in space is going to mess with your hardware, unless you add lots and lots of shielding. MMOD shielding is another requirement.
6) If you put something on the moon, you add in super-high speed-of-light latency.
7) Even if you get it set up, how will you maintain it? You can't easily send someone up to pull out bad components and plug in new ones.

Now, what are the legitimate advantages of a data center in orbit? I cannot think of a single one.

[meekGee]

Are we over conflating the quantum computing comments with the AI comments?

Quantum computing requires very cold temperatures, just a few degrees above absolute zero. AI data centers obviously need cooling, but nowhere near to the same degree.

The talk of such craters, is presumably for Quantum, as the baseline temperatures there are just above what is needed.

So possibly future "Quantum powered AI" but not your common or garden variety.

If I recall correctly, you can just dig about 1-2 meters down from the surface of the Moon and it's a fairly constant -40 degrees Celsius.  A simple cooling loop can take advantage of the fact the Moon is pretty cold unless it's directly illuminated by sunlight.

The permanently shadowed craters drop to 25-30K, which is way colder and totally unnecessary for cooling data centers, but might be useful for quantum computers.

It'll be a long time before automated factories are making PV panels, chips, and mass drivers on the moon.  But in a couple decades?  Doesn't seem crazy to me

I am not sure the rock stays cold for long when you dump 100s of MWatts into it.
Meanwhile solar collection is rather difficult, with the variable-azimuth near-horizontal illumination.

But since I'm probably rethinking my "orbital AI is stupid" stance, I should be careful with "lunar AI is stupid".

[CoolScience]

The idea that there is any advantage in putting AI in orbit or on the moon is so stupid, I'm wondering if Elon is just trying to make up an economic justification for having a moon colony. I'd be concerned if I thought he really believed this.

Greg, since I share(d) some of this: What are the reasons it is so stupid?  I want to compare notes.

1) The cost of lifting equipment into space is exorbitant.
2) You have to put up not just your hardware but also power supply and cooling.
3) Assembly. A big data center is big. You'd have to make dozens or hundreds of launches to send up the pieces, and then how do all the pieces get put together?
4) Bandwidth from space is pitiful. Latency is bad too.
5) Radiation in space is going to mess with your hardware, unless you add lots and lots of shielding. MMOD shielding is another requirement.
6) If you put something on the moon, you add in super-high speed-of-light latency.
7) Even if you get it set up, how will you maintain it? You can't easily send someone up to pull out bad components and plug in new ones.

Now, what are the legitimate advantages of a data center in orbit? I cannot think of a single one.

It appears you haven't actually looked at the posts from Musk that initiated this topic, or you would realize the answers to your supposed problems exist.

Please also keep in mind Musk is talking about 100 TW level power generation, something like 30 times the entire world electricity consumption. These are not next couple year projects and break scales for the standard "don't put datacenters in space" arguments.

1. This is in the process of being reduced by order of magnitude. I think this is also part of why Musk is talking about using the moon long term, put the production there, and depending on needs it goes into orbit or stays on the surface.
2. Cooling is again a reason to use the moon, just the waste heat of that on Earth would have issues.
3. For the slightly nearer term, what assembly are you talking about? Separate satellites with laser interconnects. Longer term this involves industry located on the moon.
4. Google "Starlink" At this point it seems like you must be joking.
5. This is again, an unprecedented scale, error correction is a must regardless of radiation.
6. Only for direct queries to the full compute power, good thing AI doesn't require that.
7. This is around talk of building permanent bases on the moon, railgun launches of things manufactured on the moon, human maintenance is far from a relevant problem. Also, how do you think they maintain the thousands of Starlink satellites on orbit today?

I have no stake in what the best solution for the problem is (or even if this is a thing that should be built), but Musk has stated 100 TW of AI compute as the specific problem. The closest thing to an accurate issue with doing it in space you suggest is the scale, but this scale is also broken on the ground. If you have an alternate solution please present it.

[Greg Hullender]

Please also keep in mind Musk is talking about 100 TW level power generation, something like 30 times the entire world electricity consumption.

This is supposed to make me take this idea more seriously?

1. This is in the process of being reduced by order of magnitude. I think this is also part of why Musk is talking about using the moon long term, put the production there, and depending on needs it goes into orbit or stays on the surface.

That's not enough, though. It still adds a huge cost to just building the things here on Earth. There has to be some advantage to doing that.

2. Cooling is again a reason to use the moon, just the waste heat of that on Earth would have issues.

No, cooling is much harder on the moon. And the waste heat on Earth is not an issue. CO2 emissions are a problem--not waste heat from datacenters.

3. For the slightly nearer term, what assembly are you talking about? Separate satellites with laser interconnects. Longer term this involves industry located on the moon.

Assembling the data center. If you split it into separate satellites, the data rates are too low. You're talking Gbps but you need Tbps. And the latency is insane.

4. Google "Starlink" At this point it seems like you must be joking.

Again, the transfer rate and latency don't compare to what a data center needs. Starlink can barely support Fortnite.

5. This is again, an unprecedented scale, error correction is a must regardless of radiation.

Error correction isn't going to eliminate the need for serious shielding.

6. Only for direct queries to the full compute power, good thing AI doesn't require that.

Huh?

7. This is around talk of building permanent bases on the moon, railgun launches of things manufactured on the moon, human maintenance is far from a relevant problem. Also, how do you think they maintain the thousands of Starlink satellites on orbit today?

But here you've moved decades (or centuries) into the future. And there's a big difference between discarding a used Starlink satellite vs. discarding an entire orbital data center!

I have no stake in what the best solution for the problem is (or even if this is a thing that should be built), but Musk has stated 100 TW of AI compute as the specific problem. The closest thing to an accurate issue with doing it in space you suggest is the scale, but this scale is also broken on the ground. If you have an alternate solution please present it.

Well, if 100 TW of AI is the specific problem, then I stand by my earlier statement: It's so insane, I'm surprised anyone is taking it seriously at all. But even if you did need to build such a thing, it'd still make the most sense to keep as much of it on the ground as possible.

[Greg Hullender]

Are we over conflating the quantum computing comments with the AI comments?

Quantum computing requires very cold temperatures, just a few degrees above absolute zero. AI data centers obviously need cooling, but nowhere near to the same degree.

The talk of such craters, is presumably for Quantum, as the baseline temperatures there are just above what is needed.

So possibly future "Quantum powered AI" but not your common or garden variety.

If I recall correctly, you can just dig about 1-2 meters down from the surface of the Moon and it's a fairly constant -40 degrees Celsius.  A simple cooling loop can take advantage of the fact the Moon is pretty cold unless it's directly illuminated by sunlight.

The permanently shadowed craters drop to 25-30K, which is way colder and totally unnecessary for cooling data centers, but might be useful for quantum computers.

It'll be a long time before automated factories are making PV panels, chips, and mass drivers on the moon.  But in a couple decades?  Doesn't seem crazy to me

I am not sure the rock stays cold for long when you dump 100s of MWatts into it.
Meanwhile solar collection is rather difficult, with the variable-azimuth near-horizontal illumination.

But since I'm probably rethinking my "orbital AI is stupid" stance, I should be careful with "lunar AI is stupid".

The biggest challenge quantum computing has is not the inability to cool very large areas. Nor is it clear that putting something on the moon helps all that much. As you say, how long is the rock going to stay cold? How fast can it conduct heat away from your data center? It's surely going to compare very poorly with air or water!

And when people speak of Quantum AI, they're talking about using quantum computing to solve very specialized machine-learning problems. No one has concrete ideas for using quantum computing to improve large-language-model systems. You'd want such a system to be on the ground where you could easily get access to reconfigure it. At least for the foreseeable future.

[Vultur]

This may be a response to the recent questioning of the Artemis HLS contract, a public affirmation that SpaceX isn't focused on Mars to the detriment of US lunar efforts. I'm not at all sure this can be viewed as a "normal" uninfluenced SpaceX plan; the timing is very very suspicious for that.

Orbital data centers hosted on Starlink v3 satellites seems like something SpaceX is likely to actually do, given that Elon Musk is pretty optimistic about AI and heavily involved in it (personally I think that LLM style AIs have negative value, actively making the Internet worse, but I'm not the one making these decisions). Quantum computing in lunar craters though ... I'm a lot more skeptical that that plan will be seriously pursued.

--

I think 100 TW of AI is very unreasonable, but if it were to be built, it'd pretty much have to be in space (probably not the Moon, though - the Moon is IMO pretty marginal for anything but scientific activity). Sure, it's cheaper to do things on Earth than in space; but you probably can't do 100 TW of anything on Earth even if you had infinite money. Too much established land uses, regulations, etc to deal with.

However, the far more likely outcome is that nobody ever has or needs 100 TW of anything. World population will probably peak at around 10 billion ish late in this century; 100 TW is less than the total energy use of a world population of 10B living in a completely electrified civilization at current US living standards.

 (US total energy consumption - not electricity, which is a bit under 1.5 kW/person - is not too far from 10 kW/person*, but if that were all electrified the total energy use would be lower.)


 *According to this EIA page https://www.eia.gov/todayinenergy/detail.php?id=65445 the US in 2024 used about 103 quadrillion BTU ... Roughly 10^20 joules - which is a bit over 3 terawatts averaged over the whole year, pretty close to 10 kW/person

[Twark_Main]

Quantum computing is best done in the permanently shadowed craters on the Moon

A unique and irreplaceable "ice core" record of the Earth-Moon system going back billions of years??  Let's build a datacenter on it just to save a bit of cooling. 

I know this is just brainstorming, but that has got to be one of the worst ideas ever.


There are plenty of mostly shadowed sites (far less scientifically interesting!) that can host such infrastructure. Literally just hang up a suspended Mylar curtain to block the remaining sunlight direction, if you really need that extra sliver of cooling performance...

I don’t agree with Cody there. He has fallen down the “don’t touch anything” hole.

I don't think that's a fair or accurate summary.

It's more like "don't immediately jump to preferentially destroy the most rare, irreplaceable, and scientifically valuable sites on an otherwise extremely large world."

This is closer to protecting UNESCO World Heritage Sites, not just blanket NIMBY "everywhere."

[CoolScience]

Please also keep in mind Musk is talking about 100 TW level power generation, something like 30 times the entire world electricity consumption.

This is supposed to make me take this idea more seriously?

You admit then that you didn't read the idea you were responding to and then expect people to take you seriously, the rest of your reply is meaningless as much of it rejects the very premise, so it doesn't contribute to the discussion.

I have no stake in what the best solution for the problem is (or even if this is a thing that should be built), but Musk has stated 100 TW of AI compute as the specific problem. The closest thing to an accurate issue with doing it in space you suggest is the scale, but this scale is also broken on the ground. If you have an alternate solution please present it.

Well, if 100 TW of AI is the specific problem, then I stand by my earlier statement: It's so insane, I'm surprised anyone is taking it seriously at all. But even if you did need to build such a thing, it'd still make the most sense to keep as much of it on the ground as possible.

When really smart people are taking something seriously that you don't, it is a good idea to go revisit your assumptions. Maybe you are right, but your categorical dismissals harm the credibility of any argument you make. Again, I am not arguing either way, I am just pointing out the major flaw in what you are doing.

[meekGee]

The idea that there is any advantage in putting AI in orbit or on the moon is so stupid, I'm wondering if Elon is just trying to make up an economic justification for having a moon colony. I'd be concerned if I thought he really believed this.

Greg, since I share(d) some of this: What are the reasons it is so stupid?  I want to compare notes.

1) The cost of lifting equipment into space is exorbitant.
2) You have to put up not just your hardware but also power supply and cooling.
3) Assembly. A big data center is big. You'd have to make dozens or hundreds of launches to send up the pieces, and then how do all the pieces get put together?
4) Bandwidth from space is pitiful. Latency is bad too.
5) Radiation in space is going to mess with your hardware, unless you add lots and lots of shielding. MMOD shielding is another requirement.
6) If you put something on the moon, you add in super-high speed-of-light latency.
7) Even if you get it set up, how will you maintain it? You can't easily send someone up to pull out bad components and plug in new ones.

Now, what are the legitimate advantages of a data center in orbit? I cannot think of a single one.

The pro is the ease of deployment.

A fully functional Starship pad can launch 100 or even 1000 MWatt of satellites in a month.  How long doea it take to deploy that even in China?

--

I covered all of your con points before (from your side) and they are correct. Power in space will be more expensive. (Radiators etc)

Taking a terrestrial data center and putting it in orbit won't work. (Maintenance etc)  But limited-life (5 years?) hardware doesn't need maintenance.

Bandwidth and latency are a non-issue in comparison to the compute.

--

Basically, the issue is that he sees AI as taking more power than you can build on Earth, and he actually thinks he can make a lot of money with it.

It's not as dumb as it first looks.

[JH]

It has been a while since I heard a reason for why we should go to the Moon that was so stupid that it actually stunned me. But building quantum computers in the permanently shadowed polar craters did the trick. Wow.

[JayWee]

It's an open question: does Musk's vision for Moonbase Alpha imply he thinks civilization/consciousness might need a backup sooner than a Mars settlement could become self-sufficient? Assuming a global collapse of civilization on Earth could a lunar AGI help restore some semblance of what we have today?

I think the main overall goal is that a Moon base / Mars city will provide a forcing function to improve space travel and space habitation technology until humanity is scattered across the Solar system. So not just one backup, but (like a good sysadmin) ultimately many backups.

The Moon vs Mars is largely short-term strategizing to maximize the pace of innovation this decade, which maximizes the probability of that ultimate success.

Ultimately the reality is there's a "histogram" of possible existential threats and time horizons, and you want to reduce the "expected area under the curve" as fast as possible. So I expect there's a mix of long-term and some shorter-term risks that are meant to be mitigated. IMO.

I agree. More in-space commerce is essential for Mars development. One day, with sufficient traffic, we might even get "port cities" like Singapore/HK at Lagrange points.

Fundamentally, when people argue "Moon vs Mars", it is (subconsciously) debated under the old Apollo paradigm - flags&footprints mission, extremely expensive. Think the NASA Mars Reference Architecture (DRM), a ~$500B affair. "You have $500B, pick one".

But with Starship being so cheap and versatile, you can do both. Mars window is only every 28 months, you have a 100t fully reusable rocket capable of launching every day, you can refuel, have depots - why would you stay idle? Let's fly to the Moon.

I think Elon/SpaceX runs this sort of a loop:
  - build capability for something
  - think what else you can do with it.

F9 - kilotons to LEO/year ⇒ LEO constellations
Starship - megatons to LEO ⇒ BLEO, Moon, Mars