I would argue that letting whichever AI that wishes to build data centers in space would be an excellent form of anti-trust enforcement: it's a borderline impossible idea and if xAI or whoever wants to go down trying it, well, that's capitalism!
Let's start with cost. Let's be SUPER generous and assume SpaceX can rapidly cut their cost per kg to space in HALF. Ok, then for a smallish industrial AI datacenter it would only cost $1bn to get the GPUs necessary into space.
Ok, AI companies have billions to spend so it's still doable (though orders of magnitude more expensive than it would be to do in just...some open space in the middle nowhere on earth)
Now the problem is heat. Data centers generate a TON of heat and dissipating heat is a HUGE problem in space. Out same smallish data center would generate 100 million watts of heat. The heat sinks required to dissipate this would be over 300,000 square meters! 38 times larger than the ISS! (The ISS, btw, cost $150 billion to build, in case you're curious).
And I haven't even gotten to the problem of radiation and what it does to computer parts!
I don't know if there is a role for public policy, but my reason for opposing the merger is that it seems like it will burden SpaceX with the heavy losses and debt from xAI/X in a way that will hobble itself before going public.
Not sure how much say, if any, the non-Elon Board of Space X has, but suspect to allow Elon to bail himself out from terrible leadership/moves with xAI/X to do this.
But perhaps it will make it easier for other companies to compete with SpaceX, which will address Matt's concerns here.
"Notably, “bring your own generation” mandates only kinda sorta solve this if your solution regarding generation is to burn gas because you’re going to raise the price of fuel. " ????
Gas is also a fuel; if there is a generation that have not that problem is some variant of renewable
Data centers in space, while surely possible, would be wildly expensive with today’s technology. I think Musk’s plan here is more about boosting the SpaceX IPO than the actual building of anything.
Can somebody explain to me when "Compute" became a noun? I've noticed it coming on as jargon for a few years. I'm not wild about it spreading.
But, hey, I'm not a prescriptivist. I even like the word "irregardless," I think it's aesthetically awesome, although I usually don't use it because people would make assumptions.
I don't think anti-trust action is appropriate in an industry that hasn't yet turned a profit. For a high-risk endeavor like data-centers in space there needs to be the potential for large profits to incentivise firms to try. Vertical integration of the launch company and the AI company may be necessary for the economics to pencil, common carrier rules could prevent data centers in space from happening at all.
"I think it would make more sense to target regulatory activism at evil right-wing billionaires rather than maintain an exclusive focus on antagonizing Democratic donors."
You didn't actually mean this, did you? Use the regulatory/prosecutorial powers of the state selectively against those with whom you disagree politically?
And why is "regulatory activism" even mentioned in a way that implies it is in any way appropriate or acceptable? "Activism" is EXACTLY what regulators should NOT be doing. That they have, in fact, engaged in regulatory activism is among the reasons why the Democratic Party has repelled voters and developed for itself the lowest approval rating in history.
The fact that the reprehensible Donald Trump targets political opponents is not license for the Democrats to do the same (as if they had never done so before). Most of the electorate wants sane, serious, and responsible people in office who will begin to repair the damage done by Trump AND by previous Democratic excesses. The Democratic Party is running out of time to provide them.
I like the common carrier requirement for launch capability but I really don't see space data centers ever making sense. I just don't see how they compete with underwater data centers. If you are willing to put the data center somewhere you can't easily access it then the ocean offers both better cooling, power in the form of wind and waves (you only need fiber going to shore not the masses of copper for large power transfer) and we already have old oil platforms.
But more important than a common carrier requirement on SpaceX is a common carrier and interoperability requirements for space communication like starlink.
I don't think it's useful to regulate an industry that is nowhere close to maturity or feasibility. That rings too close to how Europe regulates.
If Elon is deploying anti competitive measures to prevent other competitors from launching data centers that is a golden chance to regulate. But as written what this is doing is imagining a future monopoly and using regulation to choke capital deployment in a field that doesn't even exist yet.
I get it from a political/emotional standpoint that democrats need something to rally around to stick it to Musk but I think this will flop.
I'm generally a skeptic of anti-trust, so I'm pretty hesitant to sign on blocking of a merger for a market that doesn't exist yet.
I also fully listened to the Musk and Dwarkesh podcast. I came into it thinking that putting data centers in space is just so much more expensive, it could never make sense, but I came out of it having to admit that "putting data centers in space" is less expensive than "build solar farms on earth" if the government won't permit your solar farms regardless of your willingness to pay.
My takeaway wasn't "anti-trust would work here" but rather "boy it sure would be good to have some NEPA categorical exemptions for solar farms on federal land or else it will be cheaper to literally launch solar panels into space".
What other companies are planning data centers in space?
Are we looking at LEO like StarLink supposing there is some sort of hand-off (likely more complicated than StarLink)? Or, is geo-synchronous the plan?
The life span of a GPU chip seems pretty short. Are they cheap enough to just throw away after failure?
Basically, Musk is making a bet here using money from Tesla and SpaceX, which themselves were bets. However, the risk of Tesla and SpaceX predicated on somewhat known technology compared to xAI would appear to have been less risky. In other words, Musk is betting the farm.
MY is taking a lot of heat here on antitrust grounds from a lot of people who (I have to say) do not seem all that well-versed in antitrust. But I want to point out that he also used this as an opportunity to yet again imply that Earth-based renewables and solar in particular is not "practical". Par for the course with MY I guess, but please note that we could 6x the amount of land-based utility scale solar in the world and it would still be less than the land area currently used for golf courses or barley farms. We could 10x utility scale solar in the US (from 239GW to 2,390 GW -- about 2x all current US generation capacity from all sources) and still use only a fraction of the land (6.4M acres) currently dedicated to corn production for ethanol (29.7M acres). And that's not counting the additional 1,000GW of additional rooftop solar capacity potential. The "not enough land" or "conversion of farmland" arguments are bogus. See also a recent Berkeley Haas Energy Instiute working paper estimating that data centers co-located with existing gas generation facilities that have adjacent land suitable for solar and storage could meet projected US data center electricity load, with solar and storage providing 95% of the generation hours and the gas facilities (already operating at non-critical capacity factors) providing the difference, without having to wait on costly and long lead time transmission projects.
> One is that some of the controversies that often arise with regard to internet companies just don’t apply in the common-carrier world of telephones. There’s no question of asking the phone company to censor “misinformation” that’s carried over phone lines.
If the phone had just been invented today, in our current political environment, there would absolutely be this kind of pressure.
I've genuinely been surprised over the last 10 years or so to not really see attacks on common carrier/universal service obligations.* I've only been able to conclude that's because the relevant parties are too ignorant to figure that out and I've thus been reluctant to say anything about it on Twitter or other more "public" spaces for fear of speaking it into existence.
*: The big one I've been expecting for ages (but keeps not happening) is for a red state to remove universal service obligations from utilities so that people can then pressure the companies to drop, e.g., abortion clinics, gay bars, etc. as customers for electricity, gas, water, etc.
"But actual politics is just a bunch of specific things, not a bunch of vague slogans."
I would like to believe that this is true. However, I think it would be more accurate to say that in the past, this was more true than false. But it remains to be seen just how much truth remains in that statement at this point.
For a non-antitrust lawyer, Matt’s antitrust takes are usually pretty solid. This one isn’t.
Even setting aside the speculative chain of events required to create a real market for “space compute,” it’s far from clear that space compute would even be a standalone antitrust market, much less a bottleneck controlled by a combined SpaceX/xAI. From there, you have to layer additional speculation — without evidence — that the merged firm would be reasonably likely to substantially lessen competition even within that hypothetical market.
The bigger issue is that this analysis collapses “dominance” into the actual U.S. legal standard. That’s not how U.S. antitrust works. We do not prohibit mergers or conduct simply because they could produce dominance, or even monopoly. A monopoly itself is not illegal to obtain. Section 2 requires exclusionary conduct plus anticompetitive effects (or a dangerous probability of achieving monopoly power for attempt claims). Section 7 is forward-looking and asks whether there is a reasonable probability the merger may substantially lessen competition.
In practice, that usually requires real evidence of a foreclosure mechanism, actual rival dependence, and economically grounded proof that the merger would likely produce real-world competitive harm — not just a plausible future story about how markets might evolve.
I would argue that letting whichever AI that wishes to build data centers in space would be an excellent form of anti-trust enforcement: it's a borderline impossible idea and if xAI or whoever wants to go down trying it, well, that's capitalism!
Let's start with cost. Let's be SUPER generous and assume SpaceX can rapidly cut their cost per kg to space in HALF. Ok, then for a smallish industrial AI datacenter it would only cost $1bn to get the GPUs necessary into space.
Ok, AI companies have billions to spend so it's still doable (though orders of magnitude more expensive than it would be to do in just...some open space in the middle nowhere on earth)
Now the problem is heat. Data centers generate a TON of heat and dissipating heat is a HUGE problem in space. Out same smallish data center would generate 100 million watts of heat. The heat sinks required to dissipate this would be over 300,000 square meters! 38 times larger than the ISS! (The ISS, btw, cost $150 billion to build, in case you're curious).
And I haven't even gotten to the problem of radiation and what it does to computer parts!
I don't know if there is a role for public policy, but my reason for opposing the merger is that it seems like it will burden SpaceX with the heavy losses and debt from xAI/X in a way that will hobble itself before going public.
Not sure how much say, if any, the non-Elon Board of Space X has, but suspect to allow Elon to bail himself out from terrible leadership/moves with xAI/X to do this.
But perhaps it will make it easier for other companies to compete with SpaceX, which will address Matt's concerns here.
"Notably, “bring your own generation” mandates only kinda sorta solve this if your solution regarding generation is to burn gas because you’re going to raise the price of fuel. " ????
Gas is also a fuel; if there is a generation that have not that problem is some variant of renewable
https://youtu.be/O36NT5eY9lE?si=BE34xHT_7c9z1x3E
Data centers in space, while surely possible, would be wildly expensive with today’s technology. I think Musk’s plan here is more about boosting the SpaceX IPO than the actual building of anything.
Can somebody explain to me when "Compute" became a noun? I've noticed it coming on as jargon for a few years. I'm not wild about it spreading.
But, hey, I'm not a prescriptivist. I even like the word "irregardless," I think it's aesthetically awesome, although I usually don't use it because people would make assumptions.
I don't think anti-trust action is appropriate in an industry that hasn't yet turned a profit. For a high-risk endeavor like data-centers in space there needs to be the potential for large profits to incentivise firms to try. Vertical integration of the launch company and the AI company may be necessary for the economics to pencil, common carrier rules could prevent data centers in space from happening at all.
Data centers in space is just a ploy to raise the question of SpaceX for the public offering.
It's not happening for at least 50 years.
There are massive problems for heat dissipation and maintenance.
But beyond that, the launch cost (currently around 3000/kg) would have to improve by two orders magnitude to even start to make sense.
"I think it would make more sense to target regulatory activism at evil right-wing billionaires rather than maintain an exclusive focus on antagonizing Democratic donors."
You didn't actually mean this, did you? Use the regulatory/prosecutorial powers of the state selectively against those with whom you disagree politically?
And why is "regulatory activism" even mentioned in a way that implies it is in any way appropriate or acceptable? "Activism" is EXACTLY what regulators should NOT be doing. That they have, in fact, engaged in regulatory activism is among the reasons why the Democratic Party has repelled voters and developed for itself the lowest approval rating in history.
The fact that the reprehensible Donald Trump targets political opponents is not license for the Democrats to do the same (as if they had never done so before). Most of the electorate wants sane, serious, and responsible people in office who will begin to repair the damage done by Trump AND by previous Democratic excesses. The Democratic Party is running out of time to provide them.
I like the common carrier requirement for launch capability but I really don't see space data centers ever making sense. I just don't see how they compete with underwater data centers. If you are willing to put the data center somewhere you can't easily access it then the ocean offers both better cooling, power in the form of wind and waves (you only need fiber going to shore not the masses of copper for large power transfer) and we already have old oil platforms.
But more important than a common carrier requirement on SpaceX is a common carrier and interoperability requirements for space communication like starlink.
I don't think it's useful to regulate an industry that is nowhere close to maturity or feasibility. That rings too close to how Europe regulates.
If Elon is deploying anti competitive measures to prevent other competitors from launching data centers that is a golden chance to regulate. But as written what this is doing is imagining a future monopoly and using regulation to choke capital deployment in a field that doesn't even exist yet.
I get it from a political/emotional standpoint that democrats need something to rally around to stick it to Musk but I think this will flop.
I'm generally a skeptic of anti-trust, so I'm pretty hesitant to sign on blocking of a merger for a market that doesn't exist yet.
I also fully listened to the Musk and Dwarkesh podcast. I came into it thinking that putting data centers in space is just so much more expensive, it could never make sense, but I came out of it having to admit that "putting data centers in space" is less expensive than "build solar farms on earth" if the government won't permit your solar farms regardless of your willingness to pay.
My takeaway wasn't "anti-trust would work here" but rather "boy it sure would be good to have some NEPA categorical exemptions for solar farms on federal land or else it will be cheaper to literally launch solar panels into space".
Maybe we’re a little ahead of the curve here?
What other companies are planning data centers in space?
Are we looking at LEO like StarLink supposing there is some sort of hand-off (likely more complicated than StarLink)? Or, is geo-synchronous the plan?
The life span of a GPU chip seems pretty short. Are they cheap enough to just throw away after failure?
Basically, Musk is making a bet here using money from Tesla and SpaceX, which themselves were bets. However, the risk of Tesla and SpaceX predicated on somewhat known technology compared to xAI would appear to have been less risky. In other words, Musk is betting the farm.
MY is taking a lot of heat here on antitrust grounds from a lot of people who (I have to say) do not seem all that well-versed in antitrust. But I want to point out that he also used this as an opportunity to yet again imply that Earth-based renewables and solar in particular is not "practical". Par for the course with MY I guess, but please note that we could 6x the amount of land-based utility scale solar in the world and it would still be less than the land area currently used for golf courses or barley farms. We could 10x utility scale solar in the US (from 239GW to 2,390 GW -- about 2x all current US generation capacity from all sources) and still use only a fraction of the land (6.4M acres) currently dedicated to corn production for ethanol (29.7M acres). And that's not counting the additional 1,000GW of additional rooftop solar capacity potential. The "not enough land" or "conversion of farmland" arguments are bogus. See also a recent Berkeley Haas Energy Instiute working paper estimating that data centers co-located with existing gas generation facilities that have adjacent land suitable for solar and storage could meet projected US data center electricity load, with solar and storage providing 95% of the generation hours and the gas facilities (already operating at non-critical capacity factors) providing the difference, without having to wait on costly and long lead time transmission projects.
> One is that some of the controversies that often arise with regard to internet companies just don’t apply in the common-carrier world of telephones. There’s no question of asking the phone company to censor “misinformation” that’s carried over phone lines.
If the phone had just been invented today, in our current political environment, there would absolutely be this kind of pressure.
I've genuinely been surprised over the last 10 years or so to not really see attacks on common carrier/universal service obligations.* I've only been able to conclude that's because the relevant parties are too ignorant to figure that out and I've thus been reluctant to say anything about it on Twitter or other more "public" spaces for fear of speaking it into existence.
*: The big one I've been expecting for ages (but keeps not happening) is for a red state to remove universal service obligations from utilities so that people can then pressure the companies to drop, e.g., abortion clinics, gay bars, etc. as customers for electricity, gas, water, etc.
"But actual politics is just a bunch of specific things, not a bunch of vague slogans."
I would like to believe that this is true. However, I think it would be more accurate to say that in the past, this was more true than false. But it remains to be seen just how much truth remains in that statement at this point.
For a non-antitrust lawyer, Matt’s antitrust takes are usually pretty solid. This one isn’t.
Even setting aside the speculative chain of events required to create a real market for “space compute,” it’s far from clear that space compute would even be a standalone antitrust market, much less a bottleneck controlled by a combined SpaceX/xAI. From there, you have to layer additional speculation — without evidence — that the merged firm would be reasonably likely to substantially lessen competition even within that hypothetical market.
The bigger issue is that this analysis collapses “dominance” into the actual U.S. legal standard. That’s not how U.S. antitrust works. We do not prohibit mergers or conduct simply because they could produce dominance, or even monopoly. A monopoly itself is not illegal to obtain. Section 2 requires exclusionary conduct plus anticompetitive effects (or a dangerous probability of achieving monopoly power for attempt claims). Section 7 is forward-looking and asks whether there is a reasonable probability the merger may substantially lessen competition.
In practice, that usually requires real evidence of a foreclosure mechanism, actual rival dependence, and economically grounded proof that the merger would likely produce real-world competitive harm — not just a plausible future story about how markets might evolve.