Comment by pveierland
7 hours ago
Beyond aggressively optimistic timelines, I find it difficult to disagree with the premise. The aggressively optimistic timelines is also what makes it feasible to even attempt these things, where e.g. the amount of iteration required for Starship would have broken most other companies.
> In the long term, space-based AI is obviously the only way to scale.
In the long term - all mass and energy available is outside of Earth - what is here is not even a rounding error. If you wish to continue scaling compute it then becomes a question of time before you'd want to go off planet. Personally I'm quite keen to see near term space based compute explored, as it could end up becoming a much better trade-off than allocating ever more ground to power and operate terrestrial compute which directly conflict with the biosphere.
SpaceX started the Starlink design phase in 2015 - started launching Starlink satellites in 2019 - and they now have the most dominant satellite constellation ever deployed by a large factor. They have their own launch systems, launch sites, satellite bus, communication stack - both in-house designed and built.
What is really going to be that difficult with space-based compute? Radiation hardening and cooling? These are clear engineering challenges that can be simulated, tested with earth analogs, and then rapidly iterated across design generations. There's napkin math all over the internet on this, but it really seems like small challenges compared to the other engineering SpaceX have already sorted.
Beyond radiation / cooling / servicing - it seems like the biggest hurdle is to crack the scaling of designing / scaling the necessary amount of compute they will need to scale space based compute according to the laid out plans.
> In the long term
In case anyone is wondering how Tesla’s stock price remain wildly detached from its business reality, keep these four words in mind. If you can convince people that anything about you and your business has to be evaluated on a literally astronomical timescale, you can justify any valuation you desire, because your believers will give you infinite time to realize their investment returns. It has nothing to do with business. They are selling you a vision — which can also come in a pill form, labeled "salvia" and sold at gas stations.
I still see people say the cybertruck is built for mars environments, conveniently ignoring the vast technological and economical barriers stopping us from driving commercially produced vehicles on mars. This space data center thing is the same deal. It doesn't matter how long it will take to solve the technical issues with cooling, radiation, maintenance. It doesn't matter if it will make economical sense or not. It doesn't matter if spacex will be the one to actually do it. You just have to believe, and give them some time — a lot of time, so much time that a monkey can type out Hamlet and type it out again backwards.
See also the buffoonery coming out of Bay Area "effective altruist" and "longtermism" communities.
I fully agree on the reality distortions and valuation chaos surrounding Tesla. This does also follow from the company being very volatile and chaotic, which becomes harder to price. How do you accurately price in e.g. Optimus - it seems really hard to tell at this point - which I guess is also one of the motivators for these strategies.
However, in this particular trajectory, SpaceX did build the rockets and did build Starlink which is now the best global-scale wireless communication network for many use-cases. Stretching this trajectory to scale up the technology to facilitate in-space computing is vastly more grounded than Shakespearean monkeys.
All share price fundamentals are based on the long term. Short term trading is why some shares are very briefly high or low.
All share price fundamentals based on the long term have this pesky thing called discount rate which means your [hypothetical] earnings from something expected to happen in 2050 get weighted a lot lower than your 2028 earnings and your 2100 earnings barely figure in it at all though.
That's the case on a pure "I could invest my money in something that makes a bigger profit now, and use that money to buy shares in the longer term bet afterwards" basis, but is even more the case when you factor in uncertainty. And "SpaceX's 2026 near monopoly of launch and the 2026 datacentre build rush will still be relevant once we're far enough into the future for inference chips to not need regular replacement and orbital megastructures to be cost competitive with ground ones due to the amount of orbital recycling going on" is pretty uncertain...
2 replies →
Well, at least that is what they teach you in college when you learn about the efficient market hypothesis. In reality, investors are getting less rational every year. The modern stock market has become pretty much decoupled from reality.
1 reply →
> believers
If it were only retail investors, your assumptions could make sense.
However plenty of the share ownership is institutional investors. Most of them care a bit more about fundamentals. (I'm ignoring passive investors just using indexes).
A lot of institutional investors get caught out all the time when they make mistakes about the fundamentals.
They also care about networking and connections.
Secondary effects can also make this a good investment decision: if you have enough other delusional people to buy into it, you still gain from it, even if you believe it's bullshit.
I love how in today’s quarter-based world a decade is now a “literally astronomical timescale”.
Chipping in here. There's a lot of speculation on this subject matter, all of which entirely wrong.
Technical concerns aside, the main risk is financial. Success is based on the premise that we need this enough that the costs are justified but the costs are going to be much higher. That is totally unproven on any financial modelling scenario I've seen. In fact there's likely no actual ROI on what has been spent so far and no qualification of demand. With geopolitical problems on the table, no one is going to fund this.
The idea is completely dead before the first node leaves the planet.
Not only that! First mover advantage is mostly bunk. Even if none of what you say was true, valuing this on a horizon far enough out to solve all the technical problems gives a lot of space for competitors to emerge. So even if the idea was both technically and financially workable, there is no guarantee that an investment in SpaceX specifically would be the right move over the same evaluation term as the tech will take to play out. It only makes sense from a speculative angle because you know that if they present any more concrete excuse to believe that the tech works, the price will temporarily go through the roof.
Yes, beyond the three things that are the hard parts it's easy.
As problems go, radiation and cooling seem to have relatively low dimensionality compared to the other problems. It seems to be mostly a question of optimizing within the dimensions of dissipation / structure / deployment / service / cost / weight. When all is said and done, the cooling solution will end up being a module that can deal with some power dissipation, cost X amount, weight Y amount, have structural interface Z. This seems like something a relatively low number of engineers can iterate on largely isolated from other concerns. SpaceX does have 5000+ of them.
Comparing this to scaling the production of compute where they try to work outside the bounds of ASML (~40k employees) and TSMC (~80k+ employees), and where there is a huge number of degrees of freedom in many, many layers of the stack that have complicated interactions.
With radiation and cooling, SpaceX also has plenty of experience with both already given that they've had to solve this on existing satellites. Overall, Terafab just seems like a far harder challenge, and where I'd be more wary on timelines.
Radiators are raised because it's a known constraint and we know that Stefan Boltzmann implies a lot of radiator mass to be launched even at 100% cooling efficiency and there are also theoretical limits to launch efficiency which Starship is rapidly approaching.
Nobody is saying orbital datacentres can't be cooled, they're saying people arguing launching the mass of the required radiators into space is a better, more cost-effective cooling solution than pumping local water because "space is cold" are talking nonsense. Potential solutions don't look like trying to get 5000 engineers to invent radiators which defy the laws of physics, they probably look like amortising the costs over multiple decades of operation and ideally assembling the radiator portion of the datacentre from mass that's already in orbit, but that's not a near term profit pitch....
7 replies →
Isn't the question more an economic one: Is it cheaper to put some solar cells into the desert and to buy some batteries, or to launch things into space (plus the premium for radiation hardening and ensuring it survives long enough because you cannot service it).
Given the current trajectory of battery and solar prices I just don't that space-based systems are cheaper in any way.
Of course there is a long-term aspect should we climb the ladder in the Kardashev scale: Once we used all solar radiation reaching earth we must move to space to grow. But that is decades if not centuries away.
1 reply →
Just go to Google Maps. Just do it. Type in Arizona, or new mexico, texas whatever. Do you see all this brownish stuff? Yeah thats just empty land with A LOT of solar.
So in the long term, what do you think is cheaper and easier to maintain, upgrade, handle etc.?
A Space operation on which you need to send compute hardware constantly upwoards or a fiber connection to some more 'remote/dessert' like area which has a lot of energy available?
Starlink is not a game changer at all. It has 8-10 Million customers, from which plenty of peopple just use it for holidays, or upping there already existing internet line or because its faster to deploy than a cable.
Our planet is already very well connected. Putting lines in the ground is necessary anyway because you still need energy / powerlines.
Of course this can be done, thats NOT the question. The only question is, if its worth it and its not.
Sending some servers up in space is margins more expensive than sending some servers on trucks (you need anyway) to another earth location.
> Just go to Google Maps. Just do it. Type in Arizona, or new mexico, texas whatever. Do you see all this brownish stuff? Yeah thats just empty land with A LOT of solar.
'Brownish stuff', known more generally as natural ecosystems.
> So in the long term, what do you think is cheaper and easier to maintain, upgrade, handle etc.?
How long a term does your imagination stretch to? Are you really arguing that once provisioning, cooling, automated scaling in space, and off-planet mining are all solved problems, that shitting on our planet will still be the cheapest most maintainable option?
Whats your critisism with 'brownish stuff'? That space is free? Space doesn't disrupt anything?
Like sending up a lot of satelites doesn't hurt/poisens our atmosphere? That space debris doesn't matter? Disruption to astrophotography doesn't matter? Building a spaceship, the fuel for it and everything is ecofriendly?
But the natural ecosystem thats your issue?
Its 2026. This google maps brown areas are VERY VERY BIG. I would say we have enough space on our planet for a few hundred more years. Especially as we as a society are struggling anyway to expand as we are not even remotely able or capable of educating and handling enough people properly anyway.
'once provisioning' -> Until then lets provision on earth
cooling -> yeah lets just leverage the heat produced by these data centers as an affordable distant heating for housing first? What do you think how much people would enjoy a DC close by if they would get very cheap heating?
automated scaling in space -> how about we start automating earth?
off-planet mining -> you watched to much scifi at this point. Do you even understand how big the machines on earth are for mining? How much we have to transport them away? If you mine anything with a little bit of gravity, the more you mine, the more energy you need to move it around.
Do you even know how to refine minerals in space?
Yeah i think 'shitting' on our planet will be the most maintanbale and cheapest option as long as Musk is alive. Easily.
1 reply →
Compute in space is doable, we already send plenty of computers up there, technologically, it is not even a challenge. It just doesn't make sense economically, even with Starship, it is making things harder for no good reason.
Starlink is different, it makes sense. Covering the entire Earth, including the oceans with cell towers for global internet connectivity is harder than having a satellite constellation. The opposite situation from datacenters.
It’s trading political difficulty for engineering difficulty.
There are now quite a few politicians running on a platform of banning data centre construction projects.
It is another common argument, but again trading a problem for a much harder one.
If politicians ban datacenter construction projects, do you think they will take kindly of the process of building them in space? Rockets are really bad from an environment perspective. We tolerate them because we don't do that many launches and the negative effects are small on a global scale.
how does the economics compare at scale building a persistent 24/7 supply of 1,000 gw isnt easy or cheap
It is certainly not easy or cheap, but doing it in space is harder and more expensive.
> In the long term - all mass and energy available is outside of Earth - what is here is not even a rounding error.
Define “long term”. Nuclear energy is practically unlimited, plus fusion (if it ever works).
I still don't get why so many people who watch/read scifi conclude "yes this is a legitimate and realistic vision of the future, and not Wizards and Dragons but with Technology"
Going to mars or staying on the moon will be a Darwin Award-level adventure.
It’s like saying within 2-3 years the sun will go out.
Almost correct, yes.
> What is really going to be that difficult with space-based compute?
Stopping some random rogue nation blowing it up.
Honestly, I don't see (sea?) it. Every advantage of space are found in oceans/seas, especially if we use dead zones where aquatic life is already dead. The cooling is cheaper, tide+wind+solar is cheaper than space solar (I know someone who worked on a lens to observe the sun, the satellite was launched but due to being cheap on the solar panels, the sunlight and radiation chipped away the coating that found itself attracted to the most massive object in the area, the lens). Anti-corrosion is cheaper than light radiation protection, and servicing is way easier and cheaper.
In the long term, the biggest problem is that space data centers are very hard to defend against missiles.
> all mass and energy available is outside of Earth
Manufacturing capabilities are quite lacking, though, in the short and medium terms, so this doesn't seem all that relevant.
Maybe a self-contained, modular solar panel / radiator / compute unit could be built, but it will be manufactured on Earth. (Where the fabs are.)
And it still seems easier to put solar panels and batteries near the data centers that SpaceX is already building on Earth.
But if solar panel is significantly cheaper and latency doesn't matter you can have servers in any part of the world. Even if they are not up 50% of the time due to limited battery it would still be cheaper.
whose wasting 50% of their silicon production capacity to save on power. doesnt make alot of sense even with money silicon is hard to buy at scale
The thesis of the space datacenter is that energy is the expensive thing not silicon. If it's silicon it doesn't matter anyways if you run it on ground or space. And energy might be cheaper on space.
biosphere interference from ground infrastructure? any idea the ground infrastructure it requires to support space based compute operations? i have a feeling that is comparable if not more impactful
you also shrug off cooling. this is not a solved problem in any way. its not even approachable as of yet. the vast size of the radiators will be hilarious regardless.
you ignore power generation. solar is not an option. so we also need nuclear reactors for these orbital data centers. thats cool spacex can just branch out into nuclear too! love the idea of unmanned nuclear orbiting behemoths.
speaking of orbital.. what is their orbit? do they go out to Lagrange points? hilariously far? or do they stay close? hilariously fuel intensive to stay out of the atmosphere for such massive structures?
but hey, maybe we distribute spaceX-AI gpu's across starlinks. a couple solar panels and a tesla battery per gpu. all launched there by spacex
'all mass and energy available is outside of earth' Yeah, and out of range for compute data connections too.
I don't agree with the feasibility or ANY sort of practicality to this whatsoever. Im all for going for it, but I wish everyone could just admit that we're doing it because it's cool, not because it's useful. I get why Elon wont say that, but not us.
Your feelings are obviously your own, but a Starlink terminal isn't that big and can transfer quite a lot of tokens.
Every single satellite has sufficient cooling for its power production, otherwise they would be frying. Waste heat from a GPU is not materially different from waste heat from an amplifier. That's not cooling entire racks, but I don't think anybody talks about putting entire racks in space anymore.
I'm very much pro nuclear, but a solar cell in a sun synchronous orbit is pretty great too and eliminates most battery requirements
I very much doubt the economics of this makes sense, but I don't think a lot of your criticism is valid.
> Every single satellite has sufficient cooling for its power production
But here we're talking about putting data centers in space. It means stuffing as many gpus as possible into each satellite and running them at constant max power.
5 replies →
You can put computers underground. Cheaper than launching into space.
Why put them in space? Power? We have that on earth.
There are literally enormous problems powering AI data centers on the Earth right now. No, we don't have the power on Earth.
In terms of launch cost, Starship makes launch cost negligible. Some estimates are that it will cost less to launch a tonne to orbit, than to ship across the US by train.
Even if this figure is slightly low, that has nothing compared to the cost of real estate, construction costs, all of the building codes required to build a data center on Earth. These things all still apply underground, and underground is going to require additional shoring and structural engineering, to ensure that the structure is not crushed, damaged, and so forth.
>In terms of launch cost, Starship makes launch cost negligible. Some estimates are that it will cost less to launch a tonne to orbit, than to ship across the US by train.
So in this world vision obviously companies will start shipping iron ore and coal by starship from one coast to the other because it will be cheaper than trains. In fact all trucking worldwide would be replaced by space ships because they would be cheaper than trucks by far. I can't see how it will ever be cheaper to build a literal space ship and launch it than to put stuff on a train. This all reads like some super optimistic early 50's scifi.
2 replies →
> There are literally enormous problems powering AI data centers on the Earth right now.
Political, not technical.
Going to space replaces a domestic problem of angry locals with an international problem of angry governments.
> No, we don't have the power on Earth.
The power problem isn't meaningfully improved by going to space.
For every GW you put in a sun synchronous orbit to get permanent light, you need around 6 GW in the major world deserts given their cloud cover. But! The ones on the ground last 30-40 years, while the satellites are currently expected to get replaced every 5 years, so the quantity which need to be manufactured each year to maintain fixed useful output is actually about the same.
For scale:
The world installed 445 GW in 2024, and this number has a long term growth trend in the range of 25-35% per year.
If SpaceX's proposed million satellite constellation are each 25 kW modules, the total they need to launch is 25 GW, the ground equivalent is 25*6 GW = 150 GW, so we could deploy something of this scale on the ground three times over in 2024, and probably around 11-18 times in 2030 if trends continue.
And to pre-empt someone what-abouting night, between cars and PowerWall Tesla supplies about 150 GWh of batteries each year, so provided they didn't need replacing more often than every four years on average this would be enough to supply a data centre that size for 24 hours, long enough to wait for the sun to return and supply enough to be charging rather than draining batteries.
Of course, America only controls one such desert. China has another, makes most of the PV and far more batteries, but America wants to treat this situation as a race against China.
2 replies →
All of this may be true but the scale that Musk is talking about would require an immense amount of solar panels -- and if he has the means to produce so many solar panels why not use them to solve our climate and energy crisis on Earth?
Seems more like a grift to me, after the car grift and the Mars grift didn't pan out.
I’m not saying the math checks out, but the argument is that you get full sun with no atmospheric losses 24/7, so you produce way more energy per panel, and you don’t need batteries, because the power production is consistent and predictable.
Problem wont be energy input it'll be heat dumping. You can't transfer heat in a vacuum effectively -- just go google how large the International Space Station's radiators are just to ensure its electrical systems are cooled adequately.
Unless someone figures out how to break the laws of thermodynamics there's never going to be a cost effective DC in space.
2 replies →
>These are clear engineering challenges that can be simulated, tested with earth analogs, and then rapidly iterated across design generations.
They can. But in Elons case, its going to be his style of sending failure after failure up in the space, getting something working part time, lying about it and exaggerating how good it is, and then making fun of others for not using his inferior product.
Pretty much like everything else he has done.
> Beyond aggressively optimistic timelines, I find it difficult to disagree with the premise. The aggressively optimistic timelines is also what makes it feasible to even attempt these things, where e.g. the amount of iteration required for Starship would have broken most other companies.
Instead of wasting huge amounts of land to farming, restaurants and transportation of food it would be so much better if everyone just had a Star-Trek style food replicator in their house.
None of the tech exists but fuck it. Why bother with realities of life?
I am raising 200 Trillion Dollars for AI Space FoodX. Who is in?
"aggressively optimistic timelines is also what makes it feasible to even attempt these things"
yawn, people keep making this excuse on behalf of the South African investor with poor technical expertise.