SpaceX IPO

[reboots]

There is no world in which AI datacentres in space make sense.

And 'they' must know that. So the question should be, what's the real game here?

The cynic in me says, SpaceX has realised that it needs more external investment in order to continue development of Starship and any future Mars missions, and an IPO has been chosen as the vehicle. So why not find a tenuous link between the company and the current AI frenzy, to pump the issue price as high as possible by association, before that particular bag of tulip bulbs hits its shelf life.

Later down the road, you can quietly 'rest' the idea, citing feasibility studies.

This seems insightful. My completely unofficial understanding is that while there is a market for Starlink, it was conceived as make-work to subsidize SpaceX, and this at least partially drives a lot of Starlink technology and deployment goals. Fulfillment of these goals is a perpetual crisis for SpaceX, since there's not enough market for their launch capacity otherwise and they depend on utilization to fund their own rather more ambitious goals.

[tom66]

There is no world in which AI datacentres in space make sense.

And 'they' must know that. So the question should be, what's the real game here?

The cynic in me says, SpaceX has realised that it needs more external investment in order to continue development of Starship and any future Mars missions, and an IPO has been chosen as the vehicle. So why not find a tenuous link between the company and the current AI frenzy, to pump the issue price as high as possible by association, before that particular bag of tulip bulbs hits its shelf life.

Later down the road, you can quietly 'rest' the idea, citing feasibility studies.

The rumour is whilst Starlink has been a hit, bringing in billions for the company (est. $5 billion annually), it hasn't been as much of a hit as the leadership wanted.  So the company is expanding further because you must feed the beast.  If you have a look at the maps of service available over, say, Africa; urban areas are full, but the rural areas are usually available.  But to serve those urban areas, you need quite a few more satellites, which then don't get particularly well utilised once they drift out of that area.  And urban areas are probably best served by terrestrial 5G or copper/fibre anyway so... maybe there's a limit to demand in those areas anyway. 

There's certainly a future for Starlink but in 10 years all of those satellites will be EOL and need replacement and whilst Falcon 9 launches are cheaper than the competition, it still costs about $20m per launch to SpaceX (including a fair estimate for engineering time and refurbishment of first stage plus new second stage on each launch).  If you need 15000 satellites to provide full coverage and they last 10 years on average, then that's 1500 satellites per year.  A Falcon 9 can launch at most 30 sats in the mini configuration, so you have about a billion dollars a year in launch costs alone to keep the network alive. And that doesn't include ground stations & equipment, customer support, random failures etc.

Leadership wants to go to Mars and that's gonna need a lot more revenue to make that feasible, more on the order of $50 bn a year.  Admittedly Starship does change the maths as it can launch a lot more sats and is in theory fully reusable so perhaps cost per launch will fall below $10m once that is in mass use.  There's also longer term risk of competition from Amazon Kuiper and OneWeb.

I am using "Leadership" for a stand in for "you know who" to avoid fouling Rule 1...  ... but I suspect the master plan is more than just the tweeter in chief as SpaceX has large private investors involved now.

[nctnico]

SpaceX IPO is tied to AI bubble - now SpaceX wants part in it before it bursts.

Don't think so. Keep in mind Starlink is part of the IPO and SpaceX seems to have a pretty good record of launching rockets reliably once they get the design right. So there are two big money markets here: high speed wireless internet (which will probably need internet traffic routing happening in space) and bringing sattelites into orbit + all kinds of flights to the moon and to Mars later on.

High speed wireless internet may seem far fetched right now but looking back, I can see 5G (and soon 6G) offer similar or even higher speeds compared to what is standard for wired connections. Especially in countries with lacking infrastructure. For example: about 15 years ago I visited a relatively poort country in Asia. The wireless internet speed over 2G/3G was much higher than that of the (not so cheap) hotel. Basically that country has skipped the whole ADSL and fiber-to-the home and went for wireless instead because the infrastructure is much cheaper. So potentially Starlink has the ability to kill off all mobile providers globally IF they can get the hardware cost and size down considerably.

[BitBangBro]

So potentially Starlink has the ability to kill off all mobile providers globally IF they can get the hardware cost and size down considerably.

How so ? How is launching relay station IN A FRIGGIN ORBIT and have it working off a (space rated!) batteries and solar panels  technically superior or cheaper than having  5G station on the nearby roof?
Wherever I look, Starlink can't come close to LTE or 5G speed/price/wise.
It can compete only where there are no other good options.

[UnijunctionTransistor]

From my limited, humble point of view, this venture is only a glorified Iridium project.

And the IPO? Well, a few people will make a shitload of money, while a large amount of others will see their money disappear like dew-drops in the desert sun. But isn’t this what Peak Capitalism is all about?

And SpaceX will have the required cash flow to carry the leader’s vision for several years. When the well does dry out, he will have already figured out another scam investment opportunity.

[tom66]

So potentially Starlink has the ability to kill off all mobile providers globally IF they can get the hardware cost and size down considerably.

How so ? How is launching relay station IN A FRIGGIN ORBIT and have it working off a (space rated!) batteries and solar panels  technically superior or cheaper than having  5G station on the nearby roof?
Wherever I look, Starlink can't come close to LTE or 5G speed/price/wise.
It can compete only where there are no other good options.

Quite.  The problem is not building cell towers.  The problem is that there often aren't enough to serve an urban area, they need concentrated beam forming antennas now to get enough bandwidth to all of the customers.  Noticeably people object to these when they aren't well hidden and you have the COVID-5G-virus conspiracy nutters too that burn them down and assault staff maintaining them, as has happened in the UK already.

I don't see how putting antennas in orbit helps this, there's no way an antenna tracking the ground at 300 mph will be better at targeting users or dense areas than a fixed antenna on a building.  If anything, we need to get better at hiding antennas and/or convincing people that if they object to having antennas built near them then they also have no right at all to complain about poor signal.   

Now rural service absolutely could be improved by Starlink and I think a major benefit would be that you would never be out of coverage.  For hikers/adventurers that is a major lifeline, no longer do you need a satphone or an Omega watch with the emergency call function, you could just bring a smartphone (maybe a spare too) and you'll be fine.   And maybe it's a good idea to not have data connectivity.  Apple for instance provide emergency SOS via satellite on their newer phones, which is enough for most use cases, it can communicate GPS coordinates and requirements via a series of fixed messages.  This is using the Globalstar LEO network but that only has 25 satellites total in orbit.

I don't see how this changes the game for most people, which is usually where the money is...

[brucehoult]

If it's mainly for "AI", as some have suggested, then this is absolutely nuts. This thing has got to stop. It's becoming a literal cancer that's eating everything alive. If nothing changes short term, we're going to face a major semiconductor shortage within probably just a few months from now, much worse than what we had during covid. Prices of RAM and storage are already skyrocketing, and it's just the beginning.

The price will come down even more quickly than it went up when AI companies start going bust.

I wonder what will happen to electricity prices. I don't actually care too much now, as since buying $400 of solar panels in early September I'm spending 90% less on electricity (not counting the daily connection fee .. 70% less if including it -- I'd need a far larger investment to avoid that).

[Stray Electron]

There is no world in which AI datacentres in space make sense.

And 'they' must know that. So the question should be, what's the real game here?

   I think that the real game here is to come up with a completely novel idea (and tying it to one of the latest crazies; massive Data Centers) and suck all of the potential investors dry before they realize how utterly impracticable this idea is!  Do you remember Musk's  tunnels through the middle of the earth idea?  Who in their right mind could ever imagine that that was feasible much less even possible?

[brucehoult]

Do you remember Musk's  tunnels through the middle of the earth idea?  Who in their right mind could ever imagine that that was feasible much less even possible?

No, I don't. That would indeed be crazy.  Reference?

There is of course a well-know mathematical/physics proof that IF you could dig a straight line tunnel between two points A and B on the surface of the Earth, and they were in vacuum with some kind of frictionless movement, then if you dropped something in at A it would appear at B after 42 minutes, regardless of the distance between A and B -- it doesn't matter whether they are on opposite sides of the Earth, or 10 km apart, or anything in between -- it's always 42 minutes.

[EEVblog]

Is there even any evidence for this? What I saw was posted by a parody account.
The company says they provide liquidity for employees and investors by having share "buy backs" twice a year.

Yes, Elon replied "as usual Eric is accurate" to his post on X and his article:
https://arstechnica.com/space/2025/12/after-years-of-resisting-it-spacex-now-plans-to-go-public-why/
https://x.com/SciGuySpace/status/1998895789528264934

[EEVblog]

Let's not forget the added risk of cosmic rays damaging non radhard GPUs (can you imagine trying to radhard qualify a GPU like the Nvidia A100?) and the issues with memory/flash reliability.  You'd have to ECC everything beyond the normal level that a terrestrial datacentre runs at.  Component lifespans will also be far worse if they're in orbit, just because the environment is that much more hostile, so you'll have yet more e-waste and need to carry hot spares on board the spacecraft because presumably swapping a server over isn't something that is trivially done any more.

Starlink and other new space based mass production ideas like this (my friend is the founder of one Startup in this space), is the reliance on commercial off-the-shelf parts. Who cares if a few of them fail because they aren't space hardened, it's more than made up for the fact that the mass cheap scale production of these satellites more than makes up for it. That's the whole game. Starlink for example uses cheap commercial silicon solar panels, not space rated at all.

There is no world in which AI datacentres in space make sense.

I agree, but that won't stop them trying and getting something working.
They can probably kick this investment can down the road for a decade or more.

[mawyatt]

Starlink and other new space based mass production ideas like this (my friend is the founder of one Startup in this space), is the reliance on commercial off-the-shelf parts. Who cares if a few of them fail because they aren't space hardened, it's more than made up for the fact that the mass cheap scale production of these satellites more than makes up for it. That's the whole game. Starlink for example uses cheap commercial silicon solar panels, not space rated at all.

There is no world in which AI datacentres in space make sense.

I agree, but that won't stop them trying and getting something working.
They can probably kick this investment can down the road for a decade or more.

Iridium started this concept with Motorola building the sats using their automotive production line in Phoenix with automotive grade components. The main processor was a Power PC, same as used by Apple, but don't know if this was a Motorola or IBM chip (IBM was on SOI CMOS, Motorola's wasn't). Each sat cost ~$10M back then.

The modern nanometer feature size chips today may be more susceptible to SEU than the older larger feature size chips, but even if not, the massive aggregate area of all the large chips seems will see a constant SEU flux, followed by often large SEU "bursts", and an occasional complete SEU "blast". Making sure all these SEUs don't crash the "AI system" often , seems the overall "AI system" must take a hit in performance to survive these events.

Also, how do they plan to get the data to and from the orbiting AI system platforms? That alone seems like a huge problem!!

Seems like an opportunity for some gamblers to "short" the stock if this ever happens.

Best

[EEVblog]

Iridium started this concept with Motorola building the sats using their automotive production line in Phoenix with automotive grade components. The main processor was a Power PC, same as used by Apple, but don't know if this was a Motorola or IBM chip (IBM was on SOI CMOS, Motorola's wasn't). Each sat cost ~$10M back then.

The new gen Starlinks cost about $800k each. Earlier models were way cheaper but less capable.

[Simmed]

hmmm
i would speculate
since musk had seen what problems he faced with tesla
i dont think he will approach spacex in same way
i think spacex will remain un-IPO

* and he has $1trillion bucks in his pocket, why IPO ?

[David Hess]

Let's not forget the added risk of cosmic rays damaging non radhard GPUs (can you imagine trying to radhard qualify a GPU like the Nvidia A100?) and the issues with memory/flash reliability.  You'd have to ECC everything beyond the normal level that a terrestrial datacentre runs at.  Component lifespans will also be far worse if they're in orbit, just because the environment is that much more hostile, so you'll have yet more e-waste and need to carry hot spares on board the spacecraft because presumably swapping a server over isn't something that is trivially done any more.

It is not quite so bad as that.  Radiation induced soft error rates in DRAM leveled off, and even improved starting a few generations ago.  The same applies to the highest performance logic, except for some cache SRAM.

The soft errors are caused by charged particles injecting charge into circuit nodes, but the charge particle streak through the silicon produces an effect through a volume of material, while denser circuits have higher charge densities, so at some point, the distributed charge over more nodes does not have a significant effect on any individual node.  The exception is cache SRAM when it is optimized for highest speed at lowest power, which is done by minimizing the amount of charge necessary for writing, which is why cache SRAM is traditionally protected by ECC or parity.  Newer fast cache SRAM somehow solves this and does not require protection, but I do not know how that works.

The environment of low Earth orbit also benefits from some shielding from the atmosphere and Earth's magnetic field compared to higher orbits, and low Earth orbit does not pass through the Van Allen radiation belts.

[mawyatt]

Iridium started this concept with Motorola building the sats using their automotive production line in Phoenix with automotive grade components. The main processor was a Power PC, same as used by Apple, but don't know if this was a Motorola or IBM chip (IBM was on SOI CMOS, Motorola's wasn't). Each sat cost ~$10M back then.

The new gen Starlinks cost about $800k each. Earlier models were way cheaper but less capable.

Iridium only used 66 active sats. You can see a prototype and the engineering notebooks showing the concepts at the Smithsonian Air and Space Museum.

Best

[Stray Electron]

Do you remember Musk's  tunnels through the middle of the earth idea?  Who in their right mind could ever imagine that that was feasible much less even possible?

No, I don't. That would indeed be crazy.  Reference?

   Just Google Musk and Hyperloop. The plan wasn't for tunnels that went through the center or the earth and out to opposite side of the planet ala Jules Verne but it was sort of inferred and a tunnel or tunnels that run directly from San Fransisco to Los Angeles (~340 miles apart) were openly claimed as being the first objective of the project. And there was speculation about tunnels from LA to Los Vegas and LA to New York in the foreseeable future.

[Stray Electron]

hmmm
i would speculate
since musk had seen what problems he faced with tesla
i dont think he will approach spacex in same way
i think spacex will remain un-IPO

* and he has $1trillion bucks in his pocket, why IPO ?

  Exactly, without an IPO he can keep most of the profits for himself and he doesn't have to answer to a board or have to comply with the numerous GOV requirements that public companies have to comply with and have to report on to the SCC and other agencies. An IPO only makes sense if someone needs more cash to invest in the company and Musk seems to have more than enough to finance his ventures himself.

[EEVblog]

  Exactly, without an IPO he can keep most of the profits for himself and he doesn't have to answer to a board or have to comply with the numerous GOV requirements that public companies have to comply with and have to report on to the SCC and other agencies. An IPO only makes sense if someone needs more cash to invest in the company and Musk seems to have more than enough to finance his ventures himself.

IPO's are also for the employees who often hold private equity stock. Not easy to sell unless they IPO.
That being said, the company can always just offer to purchase them back to keep employees happy. But many see an IPO as the big pay day that makes working there worthwhile.

[BitBangBro]

The environment of low Earth orbit also benefits from some shielding from the atmosphere and Earth's magnetic field compared to higher orbits, and low Earth orbit does not pass through the Van Allen radiation belts.

LEO is out of the question for this. No one will be willing to have bazzillion of MASSIVE DATACENTERS along with ultzramassive panel arrays buzzing so low, constantly fighting against atmospheric drag and reboosting.

[brucehoult]

Do you remember Musk's  tunnels through the middle of the earth idea?  Who in their right mind could ever imagine that that was feasible much less even possible?

No, I don't. That would indeed be crazy.  Reference?

   Just Google Musk and Hyperloop. The plan wasn't for tunnels that went through the center or the earth and out to opposite side of the planet ala Jules Verne but it was sort of inferred and a tunnel or tunnels that run directly from San Fransisco to Los Angeles (~340 miles apart) were openly claimed as being the first objective of the project. And there was speculation about tunnels from LA to Los Vegas and LA to New York in the foreseeable future.

That's not even slightly the same thing.

People have been running trains through tunnels under cities since the first 5km long line of the London tube opened in 1863 (shallow, using steam engines) and then deep tunnelling and electric engines in 1890. Today London and New York each have around 400km of lines and are 13th and 14th in the world behind Moscow and a bunch of Chinese cities.

There were some impractical aspects of the Hyperloop white paper, but a 340 mile long tunnel between LA and SF would be entirely doable if someone wanted to pay for it. It's certainly no more silly an idea than running trains in a tunnel from France to England -- longer, but that's only money. Based on actual costs of other projects around the world, a full-sized 10m-12m diameter tunnel for normal traffic lanes would cost $150-$200 billion given US and California practices.

If you invited the Chinese in and got out of their way (or a similar length tunnel in China) they could do it for $5-$15 billion. That's pocket change for a number of people, including He Who Shall Not Be Named

But it's not even necessary. Most of that route could be done far more cheaply than deep tunnels.

Your "remember" was just completely wrong and misleading.

[brucehoult]

  Exactly, without an IPO he can keep most of the profits for himself and he doesn't have to answer to a board or have to comply with the numerous GOV requirements that public companies have to comply with and have to report on to the SCC and other agencies. An IPO only makes sense if someone needs more cash to invest in the company and Musk seems to have more than enough to finance his ventures himself.

IPO's are also for the employees who often hold private equity stock. Not easy to sell unless they IPO.
That being said, the company can always just offer to purchase them back to keep employees happy. But many see an IPO as the big pay day that makes working there worthwhile.

SpaceX does twice-yearly stock buy-backs, to provide liquidity to staff and investors.

[Psi]

The environment of low Earth orbit also benefits from some shielding from the atmosphere and Earth's magnetic field compared to higher orbits, and low Earth orbit does not pass through the Van Allen radiation belts.

LEO is out of the question for this. No one will be willing to have bazzillion of MASSIVE DATACENTERS along with ultzramassive panel arrays buzzing so low, constantly fighting against atmospheric drag and reboosting.

Yep, astronomers would lose their shit at 1-2km wide solar panels in LEO.  GEO makes more sense if you want to stream data to it directly and makes the size less of an issue if it's that far away.
I guess if you already have crazy amounts of solar power you might be able to give it lots of hall thrusters and move it out into a really high orbit after launch. But perhaps that power is always better used to run GPUs so the cost to put it directly into a high orbit is acceptable.

(Not suggesting it's actually going to happen, just theorizing)

[tom66]

GEO would have too much latency though, for agentic AI you need something that can query Earth-bound datacentres and do something then iterate until a solution converges.  Usually such solution is wrong, but it will be slow and wrong if you add 700ms of latency to each query.

[BitBangBro]

I guess if you already have crazy amounts of solar power you might be able to give it lots of hall thrusters and move it out into a really high orbit after launch. But perhaps that power is always better used to run GPUs so the cost to put it directly into a high orbit is acceptable.

(Not suggesting it's actually going to happen, just theorizing)

I don't see how that could happen. Each additional solar panel introduces more drag.
But there is a problem with high orbits also - much greater distance mean much greater signal loss (about 5.000x) . And greater latency.
Also, these things tend to need massive bandwidth.
So I guess usual Starlink stuff ain+t gonna cut it here.
Even for ONE gigagantic datacenter, much less all of them.

And one can't really rely on using laser for communications on the Earth<-->orbit path...