Why Space and Data Centers Don't Go Together || July 15, 2026

I'm a big fan of Zeihan. Been watching his videos for a few years and bought his book. My brother is not so much. Recently he messaged me with the following points of criticism on this video. I was hoping Peter would be willing to address them (my apologies up front if this comes across a little offensive. I hope you will nonetheless take his points seriously): "First of all, the whole "data centers in space" project is predicated on Starship being available. That is the only reason it's being considered, and they won't do it until they have Starship available. So any criticism based on "not enough launch capacity" or dubiousness about Starship being available entirely misses the point. "And SpaceX is ramping up for truly astonishing launch rates, far beyond what anyone has seen to date. Musk literally wants to build a fleet of 1000 Starships, and he's creating the production capacity for it. "Secondly, his whole argument about needing habitats for construction in orbit is silly. They're not going to have humans constructing data centers in orbit. They're going to launch prefabricated data-center satellites. SpaceX is big on highly automating everything. "Thirdly, SpaceX ALREADY is dealing with the heat dissipation issues on those tens of thousands of Starlink satellites! And the Starlink V3 satellites are even bigger. They have vast experience on that particular issue. "And fourthly, there's just the arrogance of a person who is not an aerospace engineer thinking that he's uncovered an issue that the engineers who live and breathe these issues every day, people who are at the absolute top of the entire industry, haven't thought about. Zeihan apparently never bothered to dig a little bit to find out what SpaceX has said about any of the issues he raises."

The idea of data centers in space is so spectacularly absurd, I can't believe anyone takes this techbro fantasy nonsense seriously.

Although I appreciate the dialogue on this idea, I guess Elon forgot one quote he should have paid attention to: "If you want to kill a big dream of yours, tell it to small-minded people." Steve Harvey Don't you wonder why Peter Z. is so snarky and condescending about this subject? This is only an idea. It's obvious you can't process data in space the same way we do on Earth. However, if this idiot knew anything about innovation, he would know someone has to start the conversation. I imagine some 17-year-old in his garage or driveway camper will develop an idea that will make data centers as we know them now obsolete. What's remarkable is that the data centers of today will crunch the data that this imaginary 17-year-old kid will use to do it. Thank you, Steve Jobs and Palmer Luckey. Peter, you are as narcissistic as Trump.

Elon is a genius, Peter is NOT! Elon has already figured this out and plans to be building data centers in orbit within 5-10 years.

Two Points: 1. Add more power needed for constant communications equipment 2. Similar issues arise for a Mars based colony.

Two points:

Makes me wonder who is putting the cart in front of the horse. All these dreamers should to commit right now, to a system to function right here and now. But nice try, back to the drawing board.

your numbers ARE in great error, Musk is much smarter than you.

This short video is worth a view. Perhaps AI datacenters are possible, though replacing individual GPUs likely is unlikely. https://youtu.be/dv_PzR2Z7iE?si=Y5Cs-CynONFoF83a

This is EXACTLY the type of challenge that Elon excels at. Much as I am a gigantic fan of Zeihan, I believe he’s wrong on this one. Those challenges will be overcome.

Imn

i wonder what your thoughts were initially when Elon was working on catching its own rockets. i reckon you were in the camp that was stupid and impossible

i think for this one, i’ll listen to Elon and other field experts

The IEEE had a good YouTube video on the problems with the physics. https://youtu.be/_qpdUNMt2yg?si=rIUtFBVClRBS01j3

Not me opening this video just to get publicly roasted for my own question. 💀😂

Data Centers in Spaaaaace! Pick one. Scott Manley is brilliant and has the patience of a saint. https://www.youtube.com/watch?v=FlQYU3m1e80 Kyle Hill is also brilliant and very silly https://www.youtube.com/watch?v=-w6G7VEwNq0 Anastasi is brilliant and very Hot! Real sexy accent and has the most comprehensive Video. https://www.youtube.com/watch?v=osMooAvpSSs Peter is 100% correct on this topic. I'm considering changing my moniker to "Silly STEM guy."

Wait till Elon bends the laws of physics to make DCs in space work! And don't forget to buy SpaceX stock, of course ;)

envidia & deeram

Glad Peter Zeihan addressed this topic. Remember space is a very hostile environment and semiconductors don't do well in hostile environments. Assuming we can get everything up into space and do the construction. If you factor in all the cost the data center would need something like $100,000/token just to recoup the cost.

What about the Starcloud satellite in late 2025 that carried an NVIDIA H100 GPU running Gemini and training an LLM in space? What about the conclusions of the ASCEND feasibility study released in 2024 by Thales Alenia Space? What about the fact that Spacex, Google and others are willing to spend billions to overcome the technological hurdles? I disagree that it is impossible. I think Peter underestimates the tech industry. I think it is another new tech frontier that will eventually be overcome.

Anything in direct sunlight in orbit will heat up to about 250* F. In the shade of earth -250*F. There is no air so no conduction of heat.

Ars Technica happened to have an article on this basic subject with some more detailed analysis. It's, of course, in contrast to what Mr. Zeihan has posted. There's some overlap, of course, because some of the basics are still basic.

God, you’re smart. Wow

The moral of the story is to get rid of the starship launches and build a space elevator. Then build data centers in space.

Idiotic to talk about putting data centers in space until some AI company makes a profit. Not income, profit.

Space elevator?

Here comes Peter Jetson

This was a waste of time! Seriously? Why don't we talk about taking out Chump and crooked cronies OUT????????

I looked at the design, and I agree that it is far-fetched. However, some parts are plausible. The design involves a 150 kW solar panel and a 110-square-meter radiator. 1) It doesn’t need to rotate; the solar panels face the sun, and the radiator surfaces stay perpendicular to the sun. 2) The 110-sq-meter radiator has to dump 150kW of heat. For black-body radiation into space, Power = Area * 5.67e-8 * T_kelvin ^4. For this to work out, the radiator has to work at 120 degrees C. Not sure if chips can run that hot. 3) I don’t think you do maintenance; you just deactivate and bypass failed chips. 4) I don’t think it needs heavy server “racks”, I imagine it would be like those rolls of LED light strips (like find along with all the other crap on Amazon Haul) unfolding automatically. 5) It would have to be self-constructing (like unrolling or opening an umbrella) or buildable by robots. No way humans can be involved. Otherwise, all the other criticisms are justified. The design has huge engineering problems. The only benefit I see over normal data centers is that you don’t need to pay for electricity. Hard to imagine it is worth it.

Peter, thanks for answering this "pesky" question. I suspected all this hyperbole about AI data centers in space sounded "too good to be true." Sure, enough your logic and counter arguments sound on point. Sadly, where are such counter arguments anywhere in the legacy media? To all the retail investors who may get hooked by Elon Musk's hype , I hope this message goes viral. Musk needs more of the public's money like a fish needs a bicycle.

For the first time in a very very very long time I agree with Peter Zeihan Space Data Centers will not work.

"So many levels of dumb to this" 🤣🤣🤣 Im going to be using that at some point lmao

The thermodynamics point about vacuum being an insulator is the part most coverage skips — glad someone’s making it. A few things I’d love your take on though: the proposals actually moving forward (Google’s Suncatcher, Starcloud, SpaceX’s FCC filing) are all constellations of autonomous satellites rather than one monolithic crewed station, so no habitats or orbital construction crews. Does the modular approach change your math at all? Also, running the Stefan-Boltzmann numbers at Starcloud’s published ~633 W/m², a gigawatt pencils out closer to 1-2 km² of radiator than tens or hundreds — still unprecedented, but a different order of problem. Seems like the whole thing really lives or dies on whether Starship gets launch costs near the ~$200/kg Google says it needs. Would enjoy a follow-up on that economic threshold specifically.

I dont think a data center in space would be anything like what is on earth.... no pumps no liquid

And this immediately turned my thoughts full-on Heinlein. There is a possibility of under the lunar surface. You can have a full (albeit small) population with the data center output as the main export. Solar power can come from arrays at the surface (and if they are at a pole, there can be constant solar power). You can have efficient waste heat radiation from deep lunar shadow. People can live underground, protected from surface radiation. OK, now I have to go back and read TMIAHM.

Peter, I'm sorry you had to waste time debunking this errant nonsense, but it's obvious that some of your subscribers actually believe random bushwah from the South African huckster, just because it's something that he babbled, even though neither he nor anyone else is taking even step one to implement this extremely stupid idea. Fractally stupid. As the saying goes, we really are in the dumbest timeline.

Peter is right in terms of overall presentation. His specifics are (typically) overstated. Space based data centers would be designed with specialized equipment that was much lower weight. They would be packaged as a lot of small, single-launch platforms that could function on deployment - not as one big honking satellite that would need to be assembled and crewed in orbit. Still, each satellite would require big solar arrays, cheap heavy LEO lift (Starship has to work), and cooling would be a real challenge. High-end GPU chips are the most sensitive to radiation, leading to need for error-correction (radiation causing bad data randomly on the fly) and loss of function over time. My back-of-the-envelope calculation is about the same: about 250-500 launches equals one medium-sized on-orbit data center. That's not going to be at all useful for commercial applications. However, it will be fundable for some applications - national security redundancy, military drone AI inference globally, deep backup of critical governmental functions, etc. Eventually each Starlink satellite could provide a few GPUs for rapid inference, networked to some bigger satellites for more compute. This could provide a little bit of space-based infrastructure for such critical use cases, at about 10-20x the cost of ground-based infrastructure.

Chip engineer here: ICs that need to go into space have to be rad hardened and tested extensively which on average would cost 5x or more than the commercial version. + chips eventually burn out (some subsystems faster than others) Due to DC current being more corrosive as compared to AC.

The whole talk of putting the data centers in space has a real we need to go and colonize Mars feel to it, and I remember when Neil deGrasse Tyson stated something to the effect of it is much easier to go to Antarctica, where you can breathe the air, survive outside in the atmosphere and technically rescue is never more than a day away and barely anyone will go there for more than a six month tour on a research facility… Maybe instead of being focused on going to Mars, we should just fix earth. We can just build the data centers on earth, it’s just don’t expect them to be this big economic slam dunk, they just need to abide by the water regulations abide by the energy regulations, and we need to build energy production at scale like China is doing. We need wind, solar and nuclear… We also need a larger more integrated grid and we probably just need to nationalize the power production company since they have proven to be unreliable in the medium to long run because they do not plan for future expansion and they do not maintain what they have. I mean how we’ve seen in Texas in California. The two states with the two largest private energy companies constantly refusing to invest… Meanwhile, you have places like Home State in Nebraska that can tank a blizzard and we rarely lose power because the Nebraska public power district trims trees, and maintains the integrity of our power grid… not to make a profit but to provide a service

So the other items listed in the Space X IPO, included asteroid mining, solar power beaming, lunar colonies, etc. The reason these things are in the IPO is that I and others at NASA worked with Congress in 2010, to shut this down as they were technically infeasible on any practical level as Musk wanted the taxpayers to flip the bill. Instead, he has found a deep well of gullible investors using the IPO. I should point out that starting around 2006 NASA was compromised by individuals wanting to bring the Chinese onboard to invest in low cost launch and spacecraft. You can read it here. https://chinaselectcommittee.house.gov/media/press-releases/select-committee-releases-report-on-nasa-s-research-security-and-enforcement-of-the-wolf-amendment. Why this is important it that the Chinese learned that instead of copying Starlink in LEO, they are investing in Laser communication from GEO. Sadly the NASA LADEE mission to the moon was the first tech demo of this technology and it wound up in the hands of our adversary's.

As a NASA Engineer with over 30 years experience, I can say Peter is mostly correct. The elephant in the room that no one mentions is atomic oxygen drag. In low earth orbit, (LEO), there are enough oxygen molecules to slowly degrade a satellites orbit. This is why constantly loses Starlinks. The size of both a solar array and thermal radiator would dramatically limit useful life of a data center in LEO.

I so appreciate this commentary! Next, computers at the outset were HUGE and now I have a phone doing the same work. Why can’t data centers be small?

Build them by the ocean for cooling without damaging the already low levels of our rivers and lakes...🥰

SPCX launching (no pun intended) at circa 75x future earnings based on robots on the moon creating data centers....exactly the reason I steered away from this as an investment. Even NASA doesn't land on the moon without a series of build ups/pilot studies to fine tune the approach, spacex is no where near that stage.

Did you forget to include the need for the vast solar array in orbit to supply the energy to power any data center? That too would take several hundred lanches of Starships to get it there, not to mention the problem of space debris strikes on the whole facility.

Peter, I love you, I think you should stick with commenting on geopolitical items and leave the technical to the professionals who have constantly changed the way we live and the way technology is evolving. I'll put my money on Elon on this one.

What about energy efficiency in AI? I’m hearing that the inefficiency will create so much heat that we might be facing boiling oceans in 60 years unless we figure it out. Never mind the other reasons for warming. The data collected imo is human mental junk. How much of that really has any value? Plus how many ads do you actually watch from the data collection? All if that is garbage stored.

Fast forward four years: “if you said we would have space based data centers, I would have laughed at you”

A few comments on this one because it's actually something I know quite a lot about. What Peter gets correct: - Space is cold BUT its a terrible thermal conductor. Heat dissipation is a real challenge. - Radiation is also a very real challenge unlike anywhere on earth. The smaller the manufacturing process (think 10nm vs 20nm chips) the MORE susceptible to radiation damage the part is. This is LITERALLY advanced tech working against itself when you take it to space. - Parts are going to break and replacing individual GPUs is probably not realistic for cost. If these are installed on satellites, the whole installation performance will degrade until its not operational. Some things he gets wrong: - We CAN do radiant cooling in space and have decades of experience with this. It's a problem on the ISS and we know how to deal with it. A satellite can point cooling arrays the same way it points solar arrays to to collect energy. This is an engineering challenge but its not an unknown one, its certainly not insurmountable. - A polar orbit solve the problem of 24/7 energy as well as cooling. I wish he had mentioned it. Instead of building a 1GW datacenter, build a 20KW satellite and then launch 50 of those. The engineering challenges around mitigating the radiation and inevitable loss of GPU compute are real but not impossible to solve. Adding lead to for radiation adds weight but it also doubles as a heat sink. Liquid cooling is not required for modern AI compute and could be engineered away. I'm not saying this is a slam dunk easy project but I think Peter is dismissing it a little too quickly. We have some very smart people who solve problems like this all the time. Let's give them a chance.

Duh.

Elon Musk is a genius. All we need is Hollywood physics.

Another big one not mentioned: GPUs go obsolete fast.

Another issue that always bothered me, Starlink satellites are in low earth orbit for faster speed, but this causes them to continually de-orbit. How would a space data center efficiently transfer massive amounts of data back and forth with Earth?

I imagine that it will happen at some point at a smaller scale if the need for fast compute power where transmission times between terrestrial based data centers is too slow. I also think it will be far more likely to occur on a moon base or asteroid where manufacturing in situ will be more feasible. But we're so far away from that being a reality that I question whether the data centers as they currently exist will even be necessary.

Do people honestly don't know how Thermos bottles work ? Outstanding

Thanks for the reality check.

The only reason why space data center is so appealing to Musk, is lack of regulations, so he can produce CP and other illegal stuff without worrying about any government interfering. So he might still be willing to run it even if it is impractical economically.

Holy audio quality Batman

P.S. if there’s more than bir levels of dumb involved in a problem, I will fall for it, write a check and go back to watching Love Island.

Outstanding commentary, with all the bits unraveled. The challenge is, how do we do this on Earth more effectively?

Glad I didn’t jump on the IPO

This seems to involve a lot of numbers and I am a millenial and I only know idk maybe uç numbers and meanwhile, the people with all the money assure me that they need this and that I will make money if I agree with them so logic and reasoning like this doesn’t matter.