But I have a buy order into my broker, just in case.
I don’t have a broker. But if this goes forward, I’ll certainly have one before mid-2026.
The way IPOs usually go is the trading houses buy huge tranches first day. Then sell those, to their favorites.
I don’t buy enough to be a favorite, and my sister is retiring from Merrill, but we’ll see. This one I want for emotional reasons, as well as financial.
Elon has always said that if SpaceX ever went public that he would find some way for the everyman investor to get in on the goodies too and not just the whales.
Upon further consideration, it occurs to me that Elon might well want to incorporate some sort of enthusiasts-and-fans-first feature into the IPO for other-than-altruistic reasons. Such people as ourselves will be much more like his extant long-term investors in that we will be buy-and-hold-ers and we will not be trying to jiggle Elon’s elbow every minute of the day about quarterly earnings and all of that tedious professional greedhead stuff. We share the dream, want to let him proceed with it unmolested, and will actively fight any attempts by “usual suspects” types to go for quick killings that compromise the long-term viability of the enterprise. After we enthusiasts and fans have bought in to the extent we collectively can, the rest of the offering – if there is any – can be sold to the whales.
Even Elon has the occasional really, really, dumb idea.
I side with this. Going public isn’t the kiss of death, but it does introduce a bunch of dumb people into critical decision-making paths. Maybe these companies have been so highly valued that he’s cashing out? Or at least funding his ambitions with these stocks as collateral.
One tiny Tesla shareholder nixed his billion dollar pay package through sketchy lawfare. Image what his enemies could do to him through SpaceX,
They won’t be able to do that. None of the Musk-iverse companies are any longer incorporated in Delaware. And there is a considerable exodus of other companies that have no connection to Musk out of the jurisdiction of the recently quite arbitrary clutches of the Delaware Chancery Court.
Yeah, calling Trump a pedo wasn’t exactly a genius move.
But this IPO isn’t remotely in that category.
What follows is some analysis I posted in a shorter version earlier today over at Behind the Black:
Eric Berger has joined the SpaceX IPO chorus on Ars Technica and Elon has tweeted that his reporting is “accurate.” So it seems we were all wrong. SpaceX is going to do an IPO next year.
Why?
I’ve been giving this some thought.
Starlink has raised SpaceX’s annual revenue to a projected $15.5 billion for 2025 with several billion of that representing net profit. For 2026, revenue should handily exceed $20 billion with perhaps 10 of those billions as net profit. So what could be the impetus to raise $25 – $30+ billion in the next six months if that amount in profit should be in-hand within the next three years anyway?
The answer appears to be AI. It is known that Musk thinks the future of AI must not be left in the hands of people like Sam Altman and Larry Page. The only way to insure that doesn’t happen is for Musk and his companies to be the dominant AI players. I find this rationale quite convincing.
Toward this end, xAI and Tesla are both building out as much terrestrial AI data center infrastructure as they can. But there are sizable obstacles on this path. Siting hassles and energy supply are the most consequential of these. And these obstacles are faced by all major AI players, not just the Musk-iverse.
These considerations have driven increasing discussion, and a few start-up efforts, aimed at siting the needed AI data center capacity in space rather than on the ground. Among other things, it was the motivation behind Sam Altman’s recent unsuccessful attempt to buy Stoke Space.
Elon and his SpaceX, xAI and Tesla folks have apparently been busily formulating scenarios and running numbers for awhile and the conclusion seems to be that there is a way for the Musk-iverse companies to secure and extend the lead they already have over AI competitors by upping the ante still further and not only basing giant AI data centers in space, but actually doing most of the fabrication required on the Moon using lunar-sourced materials.
This sort of thing, especially at the required scale, needless to say, cannot be whistled up overnight. While a massive effort to industrialize the Moon is being ginned up, SpaceX has come up with an interim path forward involving use of modified versions of its Starlink birds as nodes in a distributed space-based AI data center that can begin deployment quite soon in the form of piggy-backs on additional Starlink build-out that was always going to happen anyway.
To meet the schedule required for all efforts, it seems, a fairly prompt infusion of at least $25 billion is needed sooner than it can be accumulated via Starlink-driven SpaceX profits. Thus, the IPO, even with all of the known downstream downsides, is now deemed worth pursuing because it gets the Musk-iverse AI efforts into an even more commanding position faster than would waiting for a comparable accumulation of profits from extant operations at their projected growth rates.
The projected timing of the IPO – mid-2026 – is likely based on a desire to demonstrate that Starship has reached at least an initial state of operational readiness by that time, making SpaceX’s stock more desirable than it would be at present. That allows more money to be raised by selling fewer shares to the general public.
Raising, say, $30 billion in an IPO that would value SpaceX at a total of $1.5 trillion implies selling just 2% of company-owned outstanding shares. Given how much appreciation their shares have already seen over the years and the huge bump in value that will be occasioned by the IPO, I foresee little or no prospect of major prompt share sales by long-time SpaceX investors.
So SpaceX will be public, but with very little of its total value in public hands, rendering moot many of the more common forms of public company downside. A hostile takeover would be impossible and even the inveterate Musk-hating shorts would find very limited leverage. Political activists with single shares and large agendas would find themselves with no leverage at stockholder meetings.
For SpaceX, the challenge of relatively quickly bootstrapping up a sizable industrial infrastructure on the Moon – from absolute scratch – means, at the very least, the following:
1) Everything Moon-related anent Starship will be pursued even more aggressively than it already has been even in the wake of all of the recent argy-bargy, in Congress and elsewhere, about Starship anent Artemis and the recent test stand failure of Super Heavy B18.
2) On-orbit refilling will be a thing ASAP in 2026 as will deployment of the first Starlink V3s with the AI data center payloads. Both might well happen before the IPO as that would boost the yield from same still more.
3) There will definitely be Earth-to-lunar-orbit-and-back Starships in both crew and cargo variants developed. The first of these could well be ready to sub in for SLS-Orion on Artemis 3.
4) The scale of what SpaceX and the rest of the Musk-iverse intend to do on the Moon will dwarf the Artemis program, as currently envisioned, into comparative insignificance. If the Musk-iverse lunar efforts can be compared to WW2, then Artemis is roughly comparable, on that scale, to a gang rumble in Chicago. The notional lunar efforts of the PRC will, on this scale, be comparable to a street mugging.
5) Given that Musk is already pretty much a half-trillionaire, with his net worth actually having briefly exceeded that amount on three separate occasions in recent weeks and continuing to be just below that level, this IPO should instantly catapult him over and past the trillionaire mark on his way to being the species’s first deci-trillionaire sometime in the 2030s.
6) As SpaceX employee #1 and current CEO of Impulse Space Tom Mueller noted recently, the proposed AI-motivated Moon industrialization provides a way to keep all of the SpaceX Starship infrastructure gainfully employed between Mars armada departures.
7) A by-product of this upcoming effort will also be to provide a very clear answer to the question so frequently posed for years by skeptics of manned spaceflight – “What is there to do on the Moon that would make it worth the expense of going there?”
8) With a sizable, capable and continually growing lunar industrial base as a given, starting perhaps only a few short years hence, there will inevitably be additional economic activities initiated on the Moon and/or in cis-lunar space that would have been much tougher rows to hoe had they needed to first build out all of the industrial infrastructure as a precursor to addressing their principal intents. Tourism and the orbital condo real estate business are two that come to mind. Large-scale extraterrestrial food production is another.
It would be better if SpaceX didn’t have to go public, but tempus fugit and Elon – based on hard experience with Tesla – is going to arrange the SpaceX IPO so that as much as possible of the public company downsides do not descend upon SpaceX. I don’t think he would be planning to do this unless he thought it would provide him with advantages, overall – including accomplishing the Musk-iverse’s AI ambitions significantly sooner – that far eclipse the inevitable downsides.
If we think we’ve been living in interesting times to this point, I think we ain’t seen nothing yet.
Yeah, calling Trump a pedo wasn’t exactly a genius move.
Particularly given that there were no actual pedos involved with Epstein.
True.
“the species’s first deci-trillionaire”
That’s deca-trillionaire; he’s already a deci-trillionaire.
Oopsie. Quite right. Get my orders of magnitude mixed up sometimes.
Try a slide rule. Oh, wait!
Heh. Though I do still have the one I bought back in high school during the Late Pleistocene.
Try a slide rule. Oh, wait!
A tool of the Progressive Movement if there ever was one! I remember always wanting to use the ‘I’ scale to keep to the *left* for better precision…
My skepticism here hinges on power. Power power power.
You’ll need SPS to the n’th degree if you want meaningful AI. Even if you disperse the power requirement across 1,000s of birds they will have to co-operate and co-communicate at speeds that point-to-point orbital latency will severely hinder. You’d be better off with some big SPS birds perhaps out to GEO that are firing wide enough angle lasers out to the AI birds in LEO. Then they can get some of the serious power they’ll need. I’m not even considering the cooling requirements. Or maybe you can power them from the ground firing lasers upwards. Not sure what the advantage of the latter is over a ground based setup.
I think the piggy-back-on-Starlink thing is being pursued because it can be quickly ginned up and can provide usable capability faster than ground-based centers with multi-year implementation-from-scratch timelines, not because it represents some sort of optimal architecture.
Musk has said the longer-term solution is really enormous data centers with much more compact connectivity built in lunar orbit from lunar-sourced materials. That, in turn, requires a phased program of massively industrializing the Moon, something that would have a lot of significant payoffs beyond enabling construction of orbital AI data centers.
It could also be that he’s burning through more money in Starship development test flights than he originally planned,
In a word, no. The cadence of construction and launch will continue to increase which will – as with Falcon 9s – continue to cut the cost-per-unit of building Starship stacks and the cost-per-launch of operating them. On-site propellant manufacture will also cut costs as it’s expensive and slow to bring in vast quantities of cryogens by truck. The most expensive Starships there will ever be have already been built and flown.
“So SpaceX will be public, but with very little of its total value in public hands, rendering moot many of the more common forms of public company downside. A hostile takeover would be impossible and even the inveterate Musk-hating shorts would find very limited leverage. Political activists with single shares and large agendas would find themselves with no leverage at stockholder meetings.”
BUT they’ll all be classified as “Elon’s friends” and a Delaware Judge will rule in favor of the “Political activists with single shares”.
Never count out how widespread the hate/envy/jealousy of Elon goes.
Delaware no longer has any jurisdiction over any Musk company.
I suspect making of the actual chips will likely be the last item to be off-planeted, but I’m certainly prepared to be wrong about that. There is certainly no question that Musk regards semiconductor fab as a core technology needed for any self-sufficient extraterrestrial settlement. Just when he may decide to go all arses and elbows at Moon- and/or Mars- and/or orbital-based chip fabs is very much an open question. But it is certainly on his to-do list.
Semiconductor manufacture is one of the very, very few industries where objects made in space for use on Earth would be economically viable. The products are individually so small that launch and recovery costs would be negligible – a big Intel CPU chip weighs about 0.28 grams. Crystals grown in zero G are of much higher quality than anything achievable on Earth. And you can’t get a better vacuum on Earth for such things as molecular beam epitaxy. Cost of a space based chip plant? Who knows? But a terrestrial chip foundry is a multi-billion dollar proposition, so it probably wouldn’t be an issue.
The other big bugaboo that this solves is seismic vibration. As your lithography shrinks into the x-ray realm this is a real problem. It’s the ol’ butterfly wing wind problem x one-gazillion.
Agree with all of that. A great deal of the expense of terrestrial fabs is in providing the cleanliness, ultra-high vacuum and vibration damping needed to make chips with high yields. That all comes pretty much for free in space.
If he were to spin off say the Tesla Optima bot production facilities from the Tesla Automotive I might be interested in that IPO…
There’s no obvious reason to do that given the large technical overlap between vehicles and bots. And it would not have a good effect on the value of the rest of Tesla which would piss off long-time shareholders. I think the net effect of such a move would be to diminish, not increase, overall value. Besides, Elon’s future compensation plan is based on the performance of the Tesla enterprise as currently constituted.
That technical overlap has also occurred to Honda, Mitsubishi, and others.
The Japanese have been very good at programmable robotics – Elon is a major user of such. But AI-driven autonomous general-purpose robotics is not particularly in their wheelhouse.
Japanese carmakers have also been quite reluctant to go full-electric and none have a credible autonomous driving capability or, really, the technical base to develop one. So there’s a lot less technical overlap in Japan than at Tesla. That’s true of legacy carmakers generally and not just in Japan.
If they take SpaceX public, they should consider spinning off Starlink and going public with that while keeping launch private.
Major league bad idea. That would require Starlink to pay retail for launches. Pretty much a replication of the Amazon-Blue Origin situation. It’s going to be tricky enough putting together whatever joint-venture framework will be needed to enable SpaceX, Tesla and xAI to all pull in harness together on the whole lunar industrialization and orbital AI data centers thing without needing to stay at arm’s length from one another.
SpaceX has conducted over 100 Starlink launches so far this year. I’m sure an independent Starlink company would qualify for enormous volume discounts.
Launching at cost is cheaper still. Musk has quite a bit of race yet to run. It will not help him at all if he ties his shoelaces together mid-race.
Lunar resource and manufacturing to make things that use solar power in Earth orbit and export product to Earth.
Somewhere, Gerry O’Neill quietly smiles.
I don’t know that the plan is to site the data centers in Earth orbit. Until someone cleans them up, most Earth orbits are pretty messy these days with almost 70 years of accumulated space schmutz. Lunar orbits would be much more pristine and could far more easily be kept that way.
You’d want reasonably low latency, and Lunar orbit is not that.
Give the scale of budget a major orbital AI push would have, I’d think that orbit cleaning might be feasible on the side. Just part of the development process, preparing the site…
Except for lag time, one second round-trip per packet is a killer except for passive retrieval of data.
The round trip latency to the moon is closer to 3 seconds.
Most applications of AI – particularly vehicular and robotic – will involve a separation of the learning part of AI – which generates the inference models – from the actual use of the inference models by vastly more modest hardware in vehicles, robots and terrestrial data centers in the case of large language models like Grok and the rest.
With no need for any real-time connection between the learning-centric AI data centers and the terrestrial nodes of inference execution, signal latency is not a problem. The heavy-lifting AI of learning and inference model creation can be done in lunar orbit and the inference models transmitted back to Earth for local application. Inference model updates would take place at frequent intervals via a process like the over-the-air-updates long employed by Tesla.
I’m no expert, but I think extremely low latency is required for training neural nets with billions of weights. The latency and bandwidth between GPUs is a major limiting factor in the speed of updating the weights. A better use initially for individual Starlink-scale satellites might be performing inference for earthbound users, where satellites in LEO could be very responsive.
So far as I know, that is all correct.
But Elon, himself, was responsible for developing an architectural fix for an inter-processor connectivity issue that was limiting the scale of terrestrial AI data centers. There may be ways to minimize the impact of inter-satellite latencies too. My best guess, based on approaches long taken in addressing other large-scale types of parallel computations, is some form of pipelining of the learning computations.
The piggy-backing-on-Starlink approach may not be ideal, but it can be quickly implemented and incrementally expanded. Even sub-optimal computation that is available is plainly superior to more-optimal computation that takes years to come on-line. Elon seeks to steal a march on his competitors by availing himself of a good enough solution only he can reasonably have access to in the near-to-medium-term.
Sam Altman wouldn’t be waving his checkbook under the noses of rocket company CEOs if he didn’t think there was any value to Elon in going this route.
Yes, it is curious. Just based on first principles (i.e., speed of light) it takes orders of magnitude longer to propagate a signal between satellites than between circuit boards in a rack. What would be needed is an algorithm that allows localized updates to the matrix that are then somehow combined, perhaps every 10 or 100 iterations or so.
It’s amazing to see the exact same issues discussed today as were discussed on the Fountain Society Yahoo group 25 years ago.
Every interview he is asked about how people can invest in SpaceX. Aside from all the great reasons people already talked about, if Musk really believes the things he says, he may view public investment as a good deed that will help mitigate no one having jobs by being part owners of one of the companies that will take all the jobs.
I wouldn’t entirely rule that out, long-term. Musk has certainly broached the topic of what he now calls Sustainable Abundance many times recently. Exactly how that would work is still TBD in certain ways, but AI is a central enabling technology.
Meanwhile, in the years which may be remaining to capitalism and the economics of scarcity, the work needed to get to even some beta version of Sustainable Abundance will provide lots of employment of more and more exotic types.
If he’s doing it….
But I have a buy order into my broker, just in case.
I don’t have a broker. But if this goes forward, I’ll certainly have one before mid-2026.
The way IPOs usually go is the trading houses buy huge tranches first day. Then sell those, to their favorites.
I don’t buy enough to be a favorite, and my sister is retiring from Merrill, but we’ll see. This one I want for emotional reasons, as well as financial.
Elon has always said that if SpaceX ever went public that he would find some way for the everyman investor to get in on the goodies too and not just the whales.
Upon further consideration, it occurs to me that Elon might well want to incorporate some sort of enthusiasts-and-fans-first feature into the IPO for other-than-altruistic reasons. Such people as ourselves will be much more like his extant long-term investors in that we will be buy-and-hold-ers and we will not be trying to jiggle Elon’s elbow every minute of the day about quarterly earnings and all of that tedious professional greedhead stuff. We share the dream, want to let him proceed with it unmolested, and will actively fight any attempts by “usual suspects” types to go for quick killings that compromise the long-term viability of the enterprise. After we enthusiasts and fans have bought in to the extent we collectively can, the rest of the offering – if there is any – can be sold to the whales.
Even Elon has the occasional really, really, dumb idea.
I side with this. Going public isn’t the kiss of death, but it does introduce a bunch of dumb people into critical decision-making paths. Maybe these companies have been so highly valued that he’s cashing out? Or at least funding his ambitions with these stocks as collateral.
One tiny Tesla shareholder nixed his billion dollar pay package through sketchy lawfare. Image what his enemies could do to him through SpaceX,
They won’t be able to do that. None of the Musk-iverse companies are any longer incorporated in Delaware. And there is a considerable exodus of other companies that have no connection to Musk out of the jurisdiction of the recently quite arbitrary clutches of the Delaware Chancery Court.
Yeah, calling Trump a pedo wasn’t exactly a genius move.
But this IPO isn’t remotely in that category.
What follows is some analysis I posted in a shorter version earlier today over at Behind the Black:
Eric Berger has joined the SpaceX IPO chorus on Ars Technica and Elon has tweeted that his reporting is “accurate.” So it seems we were all wrong. SpaceX is going to do an IPO next year.
Why?
I’ve been giving this some thought.
Starlink has raised SpaceX’s annual revenue to a projected $15.5 billion for 2025 with several billion of that representing net profit. For 2026, revenue should handily exceed $20 billion with perhaps 10 of those billions as net profit. So what could be the impetus to raise $25 – $30+ billion in the next six months if that amount in profit should be in-hand within the next three years anyway?
The answer appears to be AI. It is known that Musk thinks the future of AI must not be left in the hands of people like Sam Altman and Larry Page. The only way to insure that doesn’t happen is for Musk and his companies to be the dominant AI players. I find this rationale quite convincing.
Toward this end, xAI and Tesla are both building out as much terrestrial AI data center infrastructure as they can. But there are sizable obstacles on this path. Siting hassles and energy supply are the most consequential of these. And these obstacles are faced by all major AI players, not just the Musk-iverse.
These considerations have driven increasing discussion, and a few start-up efforts, aimed at siting the needed AI data center capacity in space rather than on the ground. Among other things, it was the motivation behind Sam Altman’s recent unsuccessful attempt to buy Stoke Space.
Elon and his SpaceX, xAI and Tesla folks have apparently been busily formulating scenarios and running numbers for awhile and the conclusion seems to be that there is a way for the Musk-iverse companies to secure and extend the lead they already have over AI competitors by upping the ante still further and not only basing giant AI data centers in space, but actually doing most of the fabrication required on the Moon using lunar-sourced materials.
This sort of thing, especially at the required scale, needless to say, cannot be whistled up overnight. While a massive effort to industrialize the Moon is being ginned up, SpaceX has come up with an interim path forward involving use of modified versions of its Starlink birds as nodes in a distributed space-based AI data center that can begin deployment quite soon in the form of piggy-backs on additional Starlink build-out that was always going to happen anyway.
To meet the schedule required for all efforts, it seems, a fairly prompt infusion of at least $25 billion is needed sooner than it can be accumulated via Starlink-driven SpaceX profits. Thus, the IPO, even with all of the known downstream downsides, is now deemed worth pursuing because it gets the Musk-iverse AI efforts into an even more commanding position faster than would waiting for a comparable accumulation of profits from extant operations at their projected growth rates.
The projected timing of the IPO – mid-2026 – is likely based on a desire to demonstrate that Starship has reached at least an initial state of operational readiness by that time, making SpaceX’s stock more desirable than it would be at present. That allows more money to be raised by selling fewer shares to the general public.
Raising, say, $30 billion in an IPO that would value SpaceX at a total of $1.5 trillion implies selling just 2% of company-owned outstanding shares. Given how much appreciation their shares have already seen over the years and the huge bump in value that will be occasioned by the IPO, I foresee little or no prospect of major prompt share sales by long-time SpaceX investors.
So SpaceX will be public, but with very little of its total value in public hands, rendering moot many of the more common forms of public company downside. A hostile takeover would be impossible and even the inveterate Musk-hating shorts would find very limited leverage. Political activists with single shares and large agendas would find themselves with no leverage at stockholder meetings.
For SpaceX, the challenge of relatively quickly bootstrapping up a sizable industrial infrastructure on the Moon – from absolute scratch – means, at the very least, the following:
1) Everything Moon-related anent Starship will be pursued even more aggressively than it already has been even in the wake of all of the recent argy-bargy, in Congress and elsewhere, about Starship anent Artemis and the recent test stand failure of Super Heavy B18.
2) On-orbit refilling will be a thing ASAP in 2026 as will deployment of the first Starlink V3s with the AI data center payloads. Both might well happen before the IPO as that would boost the yield from same still more.
3) There will definitely be Earth-to-lunar-orbit-and-back Starships in both crew and cargo variants developed. The first of these could well be ready to sub in for SLS-Orion on Artemis 3.
4) The scale of what SpaceX and the rest of the Musk-iverse intend to do on the Moon will dwarf the Artemis program, as currently envisioned, into comparative insignificance. If the Musk-iverse lunar efforts can be compared to WW2, then Artemis is roughly comparable, on that scale, to a gang rumble in Chicago. The notional lunar efforts of the PRC will, on this scale, be comparable to a street mugging.
5) Given that Musk is already pretty much a half-trillionaire, with his net worth actually having briefly exceeded that amount on three separate occasions in recent weeks and continuing to be just below that level, this IPO should instantly catapult him over and past the trillionaire mark on his way to being the species’s first deci-trillionaire sometime in the 2030s.
6) As SpaceX employee #1 and current CEO of Impulse Space Tom Mueller noted recently, the proposed AI-motivated Moon industrialization provides a way to keep all of the SpaceX Starship infrastructure gainfully employed between Mars armada departures.
7) A by-product of this upcoming effort will also be to provide a very clear answer to the question so frequently posed for years by skeptics of manned spaceflight – “What is there to do on the Moon that would make it worth the expense of going there?”
8) With a sizable, capable and continually growing lunar industrial base as a given, starting perhaps only a few short years hence, there will inevitably be additional economic activities initiated on the Moon and/or in cis-lunar space that would have been much tougher rows to hoe had they needed to first build out all of the industrial infrastructure as a precursor to addressing their principal intents. Tourism and the orbital condo real estate business are two that come to mind. Large-scale extraterrestrial food production is another.
It would be better if SpaceX didn’t have to go public, but tempus fugit and Elon – based on hard experience with Tesla – is going to arrange the SpaceX IPO so that as much as possible of the public company downsides do not descend upon SpaceX. I don’t think he would be planning to do this unless he thought it would provide him with advantages, overall – including accomplishing the Musk-iverse’s AI ambitions significantly sooner – that far eclipse the inevitable downsides.
If we think we’ve been living in interesting times to this point, I think we ain’t seen nothing yet.
Yeah, calling Trump a pedo wasn’t exactly a genius move.
Particularly given that there were no actual pedos involved with Epstein.
True.
“the species’s first deci-trillionaire”
That’s deca-trillionaire; he’s already a deci-trillionaire.
Oopsie. Quite right. Get my orders of magnitude mixed up sometimes.
Try a slide rule. Oh, wait!
Heh. Though I do still have the one I bought back in high school during the Late Pleistocene.
Try a slide rule. Oh, wait!
A tool of the Progressive Movement if there ever was one! I remember always wanting to use the ‘I’ scale to keep to the *left* for better precision…
My skepticism here hinges on power. Power power power.
You’ll need SPS to the n’th degree if you want meaningful AI. Even if you disperse the power requirement across 1,000s of birds they will have to co-operate and co-communicate at speeds that point-to-point orbital latency will severely hinder. You’d be better off with some big SPS birds perhaps out to GEO that are firing wide enough angle lasers out to the AI birds in LEO. Then they can get some of the serious power they’ll need. I’m not even considering the cooling requirements. Or maybe you can power them from the ground firing lasers upwards. Not sure what the advantage of the latter is over a ground based setup.
I think the piggy-back-on-Starlink thing is being pursued because it can be quickly ginned up and can provide usable capability faster than ground-based centers with multi-year implementation-from-scratch timelines, not because it represents some sort of optimal architecture.
Musk has said the longer-term solution is really enormous data centers with much more compact connectivity built in lunar orbit from lunar-sourced materials. That, in turn, requires a phased program of massively industrializing the Moon, something that would have a lot of significant payoffs beyond enabling construction of orbital AI data centers.
It could also be that he’s burning through more money in Starship development test flights than he originally planned,
In a word, no. The cadence of construction and launch will continue to increase which will – as with Falcon 9s – continue to cut the cost-per-unit of building Starship stacks and the cost-per-launch of operating them. On-site propellant manufacture will also cut costs as it’s expensive and slow to bring in vast quantities of cryogens by truck. The most expensive Starships there will ever be have already been built and flown.
“So SpaceX will be public, but with very little of its total value in public hands, rendering moot many of the more common forms of public company downside. A hostile takeover would be impossible and even the inveterate Musk-hating shorts would find very limited leverage. Political activists with single shares and large agendas would find themselves with no leverage at stockholder meetings.”
BUT they’ll all be classified as “Elon’s friends” and a Delaware Judge will rule in favor of the “Political activists with single shares”.
Never count out how widespread the hate/envy/jealousy of Elon goes.
Delaware no longer has any jurisdiction over any Musk company.
Possible that on orbit manufacturing of wafers could reduce deficits so that https://www.cerebras.ai/blog/100x-defect-tolerance-how-cerebras-solved-the-yield-problem could use larger cores, giving him another advantage.
I suspect making of the actual chips will likely be the last item to be off-planeted, but I’m certainly prepared to be wrong about that. There is certainly no question that Musk regards semiconductor fab as a core technology needed for any self-sufficient extraterrestrial settlement. Just when he may decide to go all arses and elbows at Moon- and/or Mars- and/or orbital-based chip fabs is very much an open question. But it is certainly on his to-do list.
Semiconductor manufacture is one of the very, very few industries where objects made in space for use on Earth would be economically viable. The products are individually so small that launch and recovery costs would be negligible – a big Intel CPU chip weighs about 0.28 grams. Crystals grown in zero G are of much higher quality than anything achievable on Earth. And you can’t get a better vacuum on Earth for such things as molecular beam epitaxy. Cost of a space based chip plant? Who knows? But a terrestrial chip foundry is a multi-billion dollar proposition, so it probably wouldn’t be an issue.
The other big bugaboo that this solves is seismic vibration. As your lithography shrinks into the x-ray realm this is a real problem. It’s the ol’ butterfly wing wind problem x one-gazillion.
Agree with all of that. A great deal of the expense of terrestrial fabs is in providing the cleanliness, ultra-high vacuum and vibration damping needed to make chips with high yields. That all comes pretty much for free in space.
If he were to spin off say the Tesla Optima bot production facilities from the Tesla Automotive I might be interested in that IPO…
There’s no obvious reason to do that given the large technical overlap between vehicles and bots. And it would not have a good effect on the value of the rest of Tesla which would piss off long-time shareholders. I think the net effect of such a move would be to diminish, not increase, overall value. Besides, Elon’s future compensation plan is based on the performance of the Tesla enterprise as currently constituted.
That technical overlap has also occurred to Honda, Mitsubishi, and others.
The Japanese have been very good at programmable robotics – Elon is a major user of such. But AI-driven autonomous general-purpose robotics is not particularly in their wheelhouse.
Japanese carmakers have also been quite reluctant to go full-electric and none have a credible autonomous driving capability or, really, the technical base to develop one. So there’s a lot less technical overlap in Japan than at Tesla. That’s true of legacy carmakers generally and not just in Japan.
If they take SpaceX public, they should consider spinning off Starlink and going public with that while keeping launch private.
Major league bad idea. That would require Starlink to pay retail for launches. Pretty much a replication of the Amazon-Blue Origin situation. It’s going to be tricky enough putting together whatever joint-venture framework will be needed to enable SpaceX, Tesla and xAI to all pull in harness together on the whole lunar industrialization and orbital AI data centers thing without needing to stay at arm’s length from one another.
SpaceX has conducted over 100 Starlink launches so far this year. I’m sure an independent Starlink company would qualify for enormous volume discounts.
Launching at cost is cheaper still. Musk has quite a bit of race yet to run. It will not help him at all if he ties his shoelaces together mid-race.
Lunar resource and manufacturing to make things that use solar power in Earth orbit and export product to Earth.
Somewhere, Gerry O’Neill quietly smiles.
I don’t know that the plan is to site the data centers in Earth orbit. Until someone cleans them up, most Earth orbits are pretty messy these days with almost 70 years of accumulated space schmutz. Lunar orbits would be much more pristine and could far more easily be kept that way.
You’d want reasonably low latency, and Lunar orbit is not that.
Give the scale of budget a major orbital AI push would have, I’d think that orbit cleaning might be feasible on the side. Just part of the development process, preparing the site…
Except for lag time, one second round-trip per packet is a killer except for passive retrieval of data.
The round trip latency to the moon is closer to 3 seconds.
Most applications of AI – particularly vehicular and robotic – will involve a separation of the learning part of AI – which generates the inference models – from the actual use of the inference models by vastly more modest hardware in vehicles, robots and terrestrial data centers in the case of large language models like Grok and the rest.
With no need for any real-time connection between the learning-centric AI data centers and the terrestrial nodes of inference execution, signal latency is not a problem. The heavy-lifting AI of learning and inference model creation can be done in lunar orbit and the inference models transmitted back to Earth for local application. Inference model updates would take place at frequent intervals via a process like the over-the-air-updates long employed by Tesla.
I’m no expert, but I think extremely low latency is required for training neural nets with billions of weights. The latency and bandwidth between GPUs is a major limiting factor in the speed of updating the weights. A better use initially for individual Starlink-scale satellites might be performing inference for earthbound users, where satellites in LEO could be very responsive.
So far as I know, that is all correct.
But Elon, himself, was responsible for developing an architectural fix for an inter-processor connectivity issue that was limiting the scale of terrestrial AI data centers. There may be ways to minimize the impact of inter-satellite latencies too. My best guess, based on approaches long taken in addressing other large-scale types of parallel computations, is some form of pipelining of the learning computations.
The piggy-backing-on-Starlink approach may not be ideal, but it can be quickly implemented and incrementally expanded. Even sub-optimal computation that is available is plainly superior to more-optimal computation that takes years to come on-line. Elon seeks to steal a march on his competitors by availing himself of a good enough solution only he can reasonably have access to in the near-to-medium-term.
Sam Altman wouldn’t be waving his checkbook under the noses of rocket company CEOs if he didn’t think there was any value to Elon in going this route.
Yes, it is curious. Just based on first principles (i.e., speed of light) it takes orders of magnitude longer to propagate a signal between satellites than between circuit boards in a rack. What would be needed is an algorithm that allows localized updates to the matrix that are then somehow combined, perhaps every 10 or 100 iterations or so.
It’s amazing to see the exact same issues discussed today as were discussed on the Fountain Society Yahoo group 25 years ago.
Every interview he is asked about how people can invest in SpaceX. Aside from all the great reasons people already talked about, if Musk really believes the things he says, he may view public investment as a good deed that will help mitigate no one having jobs by being part owners of one of the companies that will take all the jobs.
I wouldn’t entirely rule that out, long-term. Musk has certainly broached the topic of what he now calls Sustainable Abundance many times recently. Exactly how that would work is still TBD in certain ways, but AI is a central enabling technology.
Meanwhile, in the years which may be remaining to capitalism and the economics of scarcity, the work needed to get to even some beta version of Sustainable Abundance will provide lots of employment of more and more exotic types.