SpaceX is amazing. As best as I (and Claude) can tell, the situation is as follows:
Competent competitors to SpaceX are developing rockets that should provide around $2000/kg cost-to-orbit. This is a big improvement over legacy space competitors like ULA, Ariane, etc. which range from $5000/kg to $50,000/kg. (It’s amazing that those competitors are still getting any business… the answer is nepotism/corruption basically afaict)
However, these competent upcoming rockets that can do around $2000/kg? They aren’t ready yet. Probably it’ll be like 5 more years before they really solidly hit that milestone at scale.
SpaceX, meanwhile, has already hit that milestone with Falcon 9 and has been there for years. It’s why they have Starlink and nobody else does.
But that’s not all. SpaceX is working on Starship, which is afaict about as close to being finished as the aforementioned competitor rockets, and when it is finished it’ll should provide somewhere between $15/kg and $150/kg. So… ONE TO TWO ORDERS OF MAGNITUDE CHEAPER STILL.
This is a really big deal. The reason space looks the way it does today (mostly empty, a few satellites and probes, two space stations) is that it’s so damn expensive to go there. SpaceX brought costs down by an OOM already and this unlocked Starlink already, and probably in the fullness of time will lead to 10x more of the other things as well (10x bigger space stations for example). Or heck, could be more than 10x; when the costs drop by 10x you often don’t merely do 10x more of the same thing, you do even more than that because now a bunch of stuff is profitable that wasn’t profitable before.
And then once Starship is online we’ll get another 1-2 OOMs on top of that. We won’t have finished adjusting to the price drop from Falcon 9 (which only happened over the course of the last decade), when we’ll be hit by another even bigger price drop!
Intuition pump: A typical house weighs something like 50,000 kg. and costs about $300,000 to build, plus whatever you pay for the land. (in e.g. Berkeley, the cost of the land may be close to a million dollars!) At $15/kg, you could launch your entire house into orbit for less than the price of a house-sized plot of land in Berkeley. (!!!)
I want to say that again for emphasis. If cost-to-orbit gets close to $15/kg, then the price to buy a house in Berkeley will be about the same as the price to buy a similarly big house in an orbiting space station. (!!!)*
It may actually be more affordable to build some kinds of high cost-per-kg structures (e.g. datacenters, high-tech factories) in space than on land.
We are in the middle of a qualitative shift in humanity’s relationship to space. And it’s basically all thanks to SpaceX.
*Yes, you probably need more expensive materials to build in space, because you need to be vacuum sealed etc. But on the other hand, if you use lighter materials (steel instead of concrete and wood for example) then you can cut the launch costs by a lot. I’m waving my hands and assuming these factors roughly cancel out.
Ex-aerospace engineer here! (I used to work at Xona Space Systems, who are working on a satellite constellation to provide a kind of next-gen GPS positioning. I’m also a longtime follower of SpaceX, fan of Kerbal Space Program, etc) Here is a rambling bunch of increasingly off-topic thoughts:
Yup, SpaceX is a big deal:
yup, Spacex is a totally off-the-charts success compared to basically any other aerospace company. (Although maybe historically comparable to the successes of early NASA?) It’s not just that their rockets are good; their Starlink satellites are also very impressive in a variety of ways—basically no other satellite company can match them on cost-vs-capability, the uniquely efficient flat-pack design, etc. And they do other stuff well too, like developing their Dragon spacecraft that certainly does a better job than Boeing’s Starliner or Sierra Nevada’s Dream Chaser.
It’s correct IMO to pay a lot of special attention to SpaceX when analyzing the aerospace industry and even perhaps the big-picture future of space exploration over the next few decades. (I presume you are thinking about SpaceX in the context of researching how space exploration might go in various “AI 2030” singularity scenarios?) Although SpaceX probably isn’t a totally unstoppable juggernaut—it’s totally plausible that Starship might continue to see troubles & delays, while Blue Origin’s “New Glenn” and RocketLab’s “Neutron” and other rockets might manage to beat expectations and scale up quickly, creating a more competitive world rather than a monopolistic Starship-fueled continuation of the famed “SpaceX steamroller”.
“SpaceX brought costs down by an OOM already and this unlocked Starlink already”—yeah, people don’t realize that Starlink constitutes 75% of all satellites in orbit (8,800 / 11,700). This is maybe not a totally fair comparison insofar as Starlink satellites are a little smaller (and in lower-energy orbits) than the big honking GEO satellites of yore, but still—in a certain sense, Starlink versus all the traditional satellite industries is a little bit like Uber versus the taxi market. It’s not just that SpaceX has captured a large percentage of the preexisting launch market; they’ve made the market way bigger.
Why are Ariane & other legacy launch companies even still alive, lol:
To your question of “why are ULA / Ariane still getting business; is this just nepotism / corruption?”—I think the situation is more accurately described in terms of national security concerns.
For Ariane, national governments want to maintain sovereign access to space—Germany, the UK, etc, don’t want to have to hand over their spy satellites to Russia or America or any other major powers for launch! But the “European military/intelligence satellite launches” industry isn’t big enough to really sustain an entire launch company like Ariane. So, Europe pressures its own commercial satellite companies (including a lot of the broadcast & communication companies operating GEO satellites, who have always launched on Ariane all through the 1980s / 90s / 00s when Ariane really was the cheapest and best option) to keep buying Ariane contracts so there’s enough European launches happening to support a European launch company. (The pressure / implied threat being that if those GEO satellite operators defect to launching on SpaceX, Europe might cut them off from contracts / subsidies / whatever other kind of industrial-policy support they’re currently providing.)
One reason why this works alright is that rocket launches are often cheaper (like $100m - $200m) than the satellites they’re launching (which can be many hundreds of millions for GEO commsats, or billions for fancy military / science missions). So the rocket launch is only a minority of the overall cost.
In ULA’s case, they are mostly propped up by the Pentagon being (reasonably IMO) concerned that they don’t want American space launch to become a monopoly because then SpaceX could charge very high prices, so they do stuff like giving 60% of their launches to SpaceX and 40% to ULA according to a big contracting process. In the future, Blue Origin might surpass ULA and mostly take over their role in the industry.
The continued existence of SLS is totally just corruption though, lol… (combined with extreme bureaucratic inertia, unwillingness to do proper decisionmaking under uncertainty / take certain perceived risks, while ignoring the risks like that you might spend tens of billions of dollars just to develop a way-more-expensive-than-the-competition rocket...)
Also, some satellite-constellation companies that think they’re “competing with SpaceX” (more like losing to SpaceX, amirite?) refuse to launch on SpaceX vehicles. Mostly I’m thinking of Amazon’s Kuiper satellite internet constellation (which wants to mostly launch on Bezos-owned Blue Origin), and the european-ish OneWeb. Also, like a more extreme version of the situation with Europe and Ariane, obviously China doesn’t let Chinese companies just buy Falcon 9 launches.
Another important factor in “why is anybody still buying these expensive-ass non-SpaceX rockets??” is that the DID prefer buying SpaceX rockets, but then SpaceX raised their prices (from $70m some years ago to about $100m today, IIRC), and then SpaceX got booked solid and ran out of rockets (despite their impressive scaling over the years), so if you’re in a big hurry to launch soon, you need to start looking at other more expensive companies (indeed, even many of these companies are booked out for many years, scaling up as fast as they can manage, etc).
Will Starship make a house in space cheaper than a house in SF? No:
House-in-Berkeley versus house-in-space is of course a weird comparison, but I very much doubt Starship could singlehandedly make it cheaper to live in space even if we used all Starship capacity for building a giant space station. An orbital space station needs a lot of complex expensive stuff to make it work (thrusters, momentum wheels, batteries, solar panels, life-support equipment for recycling water and air), plus stuff in space breaks down a lot more quickly than stuff on Earth which would increase the cost through faster depreciation. (The ISS is made of fancy aluminum pressure vessels and micrometeoroid shielding and stuff, but—despite the fact that its 7-person crew spends a huge portion of their time doing fixes & maintenance—it’s springing all kinds of weird leaks and is gonna have to be deorbited soon, even though most of the station is less than 25 years old. Contrast this with the house where I live in Colorado, built a whopping 35 years ago, which still basically does fine with just minimal home maintenance, occasional new appliances, etc.)
Plus obviously your house will need lots of supplies (food, amazon packages, but also stuff like air and thruster fuel), and transporting these supplies into orbit will be much more expensive than going to the grocery store in Berkeley.
Obviously if it was just one little house in space, then it would be SUPER expensive (since you’d need all those subsystems just for your one little house) and there would be no feasible way to do regular (like monthly) deliveries since you don’t eat a Starship full of groceries every month. But what I’m saying is that it would still be expensive even if you wanted to save money by aggregating all the houses together into one giant space station to cut down on subsystem & resupply costs.
Perhaps a more interesting point: the reason why Berkeley is expensive is because the land is expensive. But as launch gets cheaper and cheaper thanks to Starlink, the most valuable orbits will start becoming very crowded, and we’ll probably start charging for them. Right now, spots in orbit are basically given away for free (although before you launch, you’ve gotta get an FCC license to operate your satellites, which is a paperwork-intensive process, almost like the space version of getting a pharma drug approved by the FDA). But in the future, I suspect we’ll probably implement some kind of “space Georgism” to prevent kessler syndrome and properly allocate the most valuable orbital slots. (Where “we” is ideally some kind of international agreement, but in practice will probably just be, like, the USA’s department of commerce, and then China does their own similar thing, and no other country launches enough satellites to be relevant.) Under such a system, valuable spots in orbit might be auctioned off a la elaborate electromagnetic spectrum auctions. So, if you want to live in Space-Berkeley (a valuable, crowded orbit like sun-synchronous LEO), most of your cost might soon be space-land (some complicated notion of orbital crowdedness + making credible promises to maneuver around debris and de-orbit your satellite at the end of its scheduled lifetime) instead of just the construction cost. Unless you want to live in some random radiation-filled MEO orbit not really useful for anything, like Space-Rural-Oklahoma.
Will it really be cheaper to build factories in space?? Probably not pre-ASI, but possibly, idk:
You’re probably right to focus on “high cost-of-kg” operations as things that are most likely to be done in space. Lots of people talk about this dumb zombie idea of putting solar panels in space, even though it has only become less sensible over time. People are like “omg, space launch is cheaper now, maybe now it finally makes sense to implement the techno-optimist 1970s dream of solving the oil shock by putting solar panels in orbit!!” But solar panels have gotten cheaper much faster than space launch has gotten cheaper, so the trend is actually in the other direction—don’t even bother mounting the panels on a basic single-axis tracking system to follow the sun over the course of the day; just drop them directly on the freaking dirt to save on installation + mounting costs.
Obviously in an ASI-singularity scenario (or even, just, the long-term trajectory of a non-AI human civilization growing at 2% per year), we are eventually going to use up all the land, and then the natural next thing to do is to start launching lots of solar panels into space. But it doesn’t make much sense to start doing this now.
I doubt that factories is a winning idea either:
Factories are usually defined by needing lots of input material and producing lots of output material. Shipping this stuff to space and back would be expensive, so it only makes sense IMO if either the inputs are coming from space already, or the outputs are destined to stay in space.
Working in zero gravity + vacuum tends to make most things more difficult, not easier. Lots of factory processes designed on Earth will break in space. So, doing anything with moving parts in space is probably way more of a hassle than doing the same thing on earth, unless there’s some amazing special advantage to working in vacuum or zero-gravity. Some proposed special advantages I’ve heard mentioned:
People used to talk about doing pharma research in space, because proteins crystallize much more easily in zero gravity?? But I think the reason people were so hyped about crystallizing proteins is because we hadn’t solved the protein folding problem yet! (You can work out the folded structure of individual proteins by exhaustively studying protein crystals.) Now that we have AlphaFold, I think that use-case has sailed...
Nowadays people talk about doing semiconductor manufacturing in space, on the grounds that semiconductor manufacturing is extremely afraid of dust (so it might actually help to do in vacuum), and the machines are all so high-precision that they might as well be aerospace-grade anyways. Maybe there’s something to this idea?? But if you need vacuum so much, you could probably just build a vacuum-sealed assembly line, or even build an entire vacuum-sealed wing of the TSMC factory (with employees walking around in pressure suits and everything) for cheaper than building an orbital factory. (The vacuum quality of low-earth-orbit isn’t even especially great compared to what you can get pretty easily on the ground!) Semiconductor manufacturing is infamously one of the most difficult, complicated things that human civilization does; I’d be surprised if you could just move it all into space without a million little things going wrong.
Something about optical fibers, carbon nanotubes, and other advanced materials potentially being easier to manufacture in low gravity?? I don’t know much about this—mostly I’m just remembering the plot of Andy Weir’s book “Artemis” and hoping that the optical-fiber McGuffin plot-point was based on plausible background research. You could imagine an AI angle here too, if we need tons of super-high-quality optical fiber to make the interconnections between our vast datacenters full of TPUs or photonic chips or however we multiply matrices in the year 2040.
One entertaining niche application of space manufacturing is to produce “extinct polymorphs”—chemicals like the HIV drug Ritonavir, which once were easily manufactured on earth but have since become nearly-impossible to create, thanks to a bizarre ice-nine-style process where they get “infected” by misfolded versions of the same molecule! Varda Space is an aerospace startup which actually produced some Ritonavir in space precisely to make this point. But I hardly expect “bringing back extinct polymorphs” to become a major portion of GDP in the future; it seems intrinsically niche. (Barring, perhaps, some mirror-life related catastrophe such that we are only able to grow crops and preserve natural necosystems in pristine space environments, a la the sci-fi stories Interstellar, Silent Running, and Speaker for the Dead.)
If you have AGI / ASI, then maybe you can simply have the AI redesign all your manufacturing processes from first-principles to work well in the space environment. Maybe in some objective sense, space is actually a better place to do most manufacturing! But in that case you do need the AGI, and you also need some time to bootstrap the entire alternate manufacturing ecosystem. This might face some of the same pros & cons as Carl Feynman’s concept of creating an alternate manufacturing system of self-replicating automated/miniaturized machine shops (see my comments here), although of course I’d expect a true ASI to power through the various troublesome issues of transitioning over to a whole new industrial base pretty quickly.
Putting datacenters into space is a little more plausible IMO, because you don’t have to worry about tons of moving parts and manufacturing processes, and your input is just energy while your output is just heat + information.
But you do need energy, which you can either beam up from earth via some kind of microwave laser (but this hasn’t been tested IRL, has some pretty serious efficiency losses even just in theory, etc), or manufacture locally with solar panels or nuclear power (but this is gonna drag down your cost-per-kg launching random solar panels).
I will say that, compared to the idea of space solar power, space-datacenters is a big improvement because you’re increasing the cost-per-kg of what you’re launching (all those GPUs), and instead of beaming back lossy microwave radiation energy, you’re beaming back information, which seems easier.
But I’m doubtful that you could do AI training in space very easily, since you’d have to formation-fly a ton of satellites close together (thus spending a lot of fuel?? unless you want to do something ridiculous and experimental like electrostatic-based formation flying), connect them all with extremely high-bandwidth laser links (dunno how this compares to the bandwidth Starlink already achieves...), etc. My impression is that if you don’t have high-bandwidth interconnects, you’re probably limited to AI inference instead of training? (idk that much about AI training though...). I’d also be worried that both training and inference would require lots of data to be transmitted to and from the ground, except then I remembered that the whole point of Starlink is to put the whole planet’s internet infrastructure in space, and it seems to be working fine—so at least bandwidth won’t be a problem!
And you do need to get rid of all the heat, which in some ways gets a lot harder in space (there’s no air to do convection, nor a ready supply of cheap water), although in some ways it gets easier (space is really cold, so you can cool down by just blocking the sun with a big mylar mirror and radiating in every other direction).
I’m not sure, but I’d be worried that radiating heat doesn’t scale well (due to square-cube-law) while piping cold water around scales better. (Until, of course, you cover the entire planet in datacenters and solar panels and fusion reactors, melt the icecaps and boil the oceans, and you’re forced to resort to radiative cooling because you’ve run out of places to convect to!)
And this is very different from the berkeley house case, because (at least at the moment) there’s still a vast amount of basically dirt-cheap useless desert land on which to build datacenters, power infrastructure, etc. People complain about regulation and permitting, but:
Space will also feature regulation & permitting obstacles, around things like space-debris mitigation, electromagnetic spectrum for beaming vast amounts of information back to earth (unless you can figure out space-to-ground laser comms, perhaps?), and power generation. Whether you are launching nuclear reactors into space, constructing gigawatt-scale microwave lasers that hostile superpowers will perceive as anti-ballistic-missile defenses, or even just building a mass-catapult on the moon so you can hurl hundreds of tons of lunar-manufactured solar panels towards the Earth, somebody is probably going to want to submit some comments during the 30-day public notice period...
My impression is that datacenters could route around much of the most severe permitting issues (such as around transmission lines and energy interconnect queues) if they were willing to go off the grid and build all their own power + battery storage. Datacenter builders don’t want to do this because that’s more expensive. But building datacenters in space also means going off-grid, plus you have to do a ton of other stuff! (Yes, I get that space-based solar is 4x more effective and cuts down on your need for batteries, but space launch isn’t the only thing allowed to reduce in price over time—batteries and other kinds of energy-storage technology are also getting cheaper all the time.)
So, what are we gonna use all those Starships for, if not in-space manufacturing or datacenters??
Right now, the biggest and most-valuable use of space is for communications (like Starlink internet, but also military communications, television broadcasts from GEO, specialized connections to airplanes and ships, etc), navigation (like GPS), taking photos of the earth (mostly for military intel, but this also has applications in agriculture, finance, etc), and assorted military applications. So, in the immediate future, I’d expect us to just keep massively scaling those applications rather than using Starship to do wacky new stuff:
You kind of only need one navigation constellation (although it’s due for an upgrade, re: Xona’s plan), so this won’t be a huge number of satellites.
With satellite internet, more & bigger satellites = more internet bandwidth, so I’d expect this to grow a lot. Claude says that Starlink today only represents “perhaps 1–3% of international backbone capacity—and international traffic is itself only a subset of total internet traffic”. Surely it would be profitable to scale this up such that satellites are providing as much internet bandwidth as all existing ground-based infrastructure combined (ie, +100% instead of +2%), and we could still find reasonably productive ways to use that bandwidth? So that would be like 50x all the Starlink satellites that have been launched so far.
So far there have been over 350 Falcon 9 launches full of Starlinks (and US-military-branded Starshields) -- assume that you want to launch 50x that amount, but also that Starship can launch 5x as much mass (90 tons instead of 17.5 tons). That comes out to 3,500 Starship launches to achieve that goal. If you look at the 11 launches they’ve done so far, pretend they were all successful and all happened this year (they were not and did not), and assume they’ll double their number of launches each year (22 in 2026, 44 in 2027...) and launch nothing but Starlinks (NASA will want a word with them about their Artemis lunar mission contract, which involves up to 20 starship launches per moon visit...), it would take them eight years to clear that backlog.
Taking photos of the earth probably scales more than navigation but less than internet—Planet’s fleet of cubesats already take 3m-resolution photos of the whole earth every day. How much more valuable would it be to have 30cm-resolution photos every hour? Or continuous video?? idk, but maybe once adding more satellite internet capacity hits diminishing returns, this becomes the next most valuable thing to scale up.
Earth-observation is also probably the place where it makes the most sense to be putting GPUs in space right away, since the satellites are already very bottlenecked on bandwidth for beaming images down to earth. If you put a GPU on all your spy satellites, you could do image-classification analysis locally (and immediately!) and only beam down the most interesting stuff. Plus you could maybe even make dynamic decisions like “oh, that’s really interesting, let’s take some more photos of that spot”—usually these kinds of decisions are delayed by the time it takes for a satellite to pass over a ground station, download an image, get the image analyzed, and then for commands to be uploaded later, so it might be a big deal to make these decisions locally & immediately.
Military expenses are dictated by adversarial / arms-race logic, so the amount we’ll spend on military stuff in space is perhaps kind of a wildcard driven by how intense the overall military competition with China gets, multiplied by how many advantageous military-stuff-in-space ideas we can dream up.
Are there any huge economic markets (besides already-discussed manufacturing, solar power generation, and datacenters) that might open up besides orbiting satellites beaming down information?
At some point, maybe asteroid mining becomes a thing? This has all the disadvantages of “moving parts in space”, plus you’re dealing with very messy inputs, but it has the advantage of the fact that certain asteroids have very high concentrations of metals that are rare on earth.
My guess is that doing an expensive space mission to bring back very valuable, rare metals (like gold, iridum, etc) is a better business plan than the zombie idea of doing an expensive space mission just to launch solar panels and simply beam back power via lossy microwave laser—beating dirt-cheap earth-based solar panels sounds impossible, but beating earth-based mining sounds a little less impossible. The business case for “solar power in space” probably only closes once you have finished covering all of earth’s deserts in solar panels. The business case for asteroid mining probably closes earlier, though I bet you’d still need pretty immense scale (like “over the next decade we’re aiming to bring back an amount of gold equivalent to 10% of all gold ever mined in history”) to amortize the huge cost of a gigantic deep-space mission and bring the whole project below the cost of earth-based mining.
An even better business plan would be if we could manufacture something even more difficult to make on earth (maybe semiconductors, optical fibers, or some other advanced material??), which would probably also have to be very high value-per-kg (to minimize the cost of transporting the inputs and outputs). But who knows if we can actually figure out anything that fits that criteria. And whatever we figure out might not be scalable to trillions of dollars (like, Ritonavir definitely isn’t) in the way that asteroid mining clearly is.
You can also mine asteroids to melt ice and make hydrogen + oxygen rocket fuel, but of course this requires some customer who’ll buy a lot of rocket-fuel for going beyond low-earth orbit (like colonizing mars?) or who has other reasons to be maneuvering all the time (like military satellites that want to constantly change their orbit to stay unpredictable?).
Melting ice is probably a lot easier than processing ore, so probably the first demonstration asteroid-mining missions are about water. But to scale up, they’d need customers (and customers high above low-earth-orbit, since their product has to be cheaper than just launching extra rocket fuel on Starships!).
Other than satellites, I think space-based industry & exploration is going to be heavily debt-financed in a really big way for a really long time.
Satellites are profitable and normal; we are obviously gonna blot out the sky with immense numbers of very large, high-powered, super-Starlink satellites (mostly for internet, also for taking photos of the earth).
But asteroid mining maybe only pays off once you scale up to some preposterous level, like bringing back a trillion dollars’ worth of gold.
I am recalling all those debates about whether Amazon or Uber were really sustainable businesses—they’re pouring billions of dollars of investors’ cash on infrastructure build-out or user subsidies; are they REALLY gonna flip to profitability and turn this all around someday?? Or the current debates about the even vastly-larger sums being invested in datacenters for training AI models. If asteroid mining for precious metals ever happens, it is gonna be that kind of situation all over again.
What about using Starship for its intended purpose of settling Mars??? IMO, nobody has thought up any plausible reason to think that a Mars city would ever return significant capital—settling another planet would be a gigantic money-sink basically forever. Yet, in some abstract sense it seems obviously likely to be worthwhile (in terms of cultural influence on humanity’s future, if not literal investment returns) to be at the forefront of colonizing the solar system! In this respect it feels similar to some parts of European colonial history—what was the ROI of Britain starting colonies in North America? In a certain sense, somewhere between low (you spend decades building it up, finally get a little bit of stamp tax, and then they go and fight a revolution against you) and extremely negative (start a Jamestown or Roanoke, almost everybody dies, then the town straggles on pointlessly for decades, consuming resupply ships but not figuring out anything to export). But in another sense, extremely high (insofar as Britain got to put their thumbprint on what later became the mighty USA).
So this is kind of like the asteroid-mining issue or datacenter-buildout issue, but on steroids—a venture so vast and so uncertain (a lot of European colonial empires, like Germany’s scramble for Africa, were bad ideas that paid off neither literally nor metaphorically!) and requiring so much debt-financing that it leaves the realm of traditional investing, or even the realm of economic booms / bubbles and instead has to be coordinated through the mechanism of national/societal greatness, competition, and prestige.
But even though doing Mars settlement is way more expensive and speculative and uncertain than even doing asteroid mining, the perceived expected-value (or perceived cost of missing out) might be higher. So I think it’s actually likely that we choose to do something closer to satellites + mars colonies, rather than satellites + in-space manufacturing/mining.
It’s also worth noting that Starship has been explicitly designed for settling Mars (the methane fuel, the Space-Shuttle-like upper stage that would be a reusable SSTO on Mars, etc). Sending many Starships to Mars and back (where they can hope to use aerobraking for landing, and locally-manufactured fuel for launch back to earth) is probably much cheaper than sending the same number to the moon and back!
And in a similar way, a LOT of potential space activities are more like Mars colonization than satellite internet—“squatting on areas that might eventually be profitable someday in the future” rather than making actual profit today.
As mentioned, asteroid mining is like this—the first company to develop the tech and visit some of the asteroids might do this in the hope of kinda claiming & squatting the opportunity, far in advance of the opportunity actually becoming profitable.
Things like putting datacenters in space or putting solar panels in space also have something like this vibe, insofar as eventually it seems we will want to do them. But what’s the scarce resource being squatted?? With Mars or asteroids, you’re hoping to cheaply stake a claim (in the sense of legal rights, precedent, etc) on scarce physical land / ore in the hopes of expensively developing / mining it later. But launching solar panels & datacenters looks more like “expensively doing something now in the hopes of profiting in the far future”. You should rather aim to be cheaply squatting something now. Maybe this would be:
Space-Berkeley slots in low-earth-orbit?? (but if LEO becomes well-governed via Fully Automated Luxury Space Georgism, this plan is not gonna work out for you...)
Being the debt-fueled leader in some industry (like how SpaceX is the leader in launch), where the industry might 100x in size (and actually become profitable) in the future. Here you’d either be hoarding a technological advantage (doing trial missions to develop your in-space manufacturing processes, but not actually launching a lot since you lose money on each mission), or (more expensively) hoarding an industrial-capacity advantage.
If you believe in AGI right around the corner, this makes trying to squat “space-based power generation / datacenters / semiconductor manufacturing” more appealing, since the applications are more concrete and if the singularity is about to happen then you don’t actually have to wait very many years paying interest on your debt.
But on the other hand, if AGI is right around the corner, surely there are tons of more-profitable things to do here on earth? Like try to invest in humanoid robots, or do normal AI investments in TSMC / NVDA, or etc?
Basically, if you are playing this game of trying to squat the opportunity to do far-future space expansion, then you wanna put yourself in the best possible position to be at the forefront of a grabby-aliens-style expansion into the solar system (from where, presumably, you can steamroll onwards to the galaxy).
But it’s unclear exactly what bundle of technologies / legal claims / industrial capacity / etc will actually be needed for this. (Will controlling a small town on Mars be relevant in any possible way if an ASI singularity occurs in 2050?? Probably not!)
And in particular I’d be very worried that I’d spend all this time going into debt trying to develop a clever portfolio of space-related industrial / technological capabilities, only to get instantly lapped by ASI right off the starting block. Such that maybe the only resource really worth scrambling for is simply “access to ASI”. (Plus possibly launch capacity itself, which is a big capital-intensive heavy-industry that seems less amenable to being lapped software-only-singularity style than something more intricate and design-intensive, like space-based manufacturing equipment or satellites.)
But most people are less ASI-pilled. So, probably we start launching colonization rockets to Mars (and funding lots of doomed little “datacenter in space” / “solar power in space” / “bitcoin in space” startups) anyways.
I really enjoyed reading this comment, and I’d be sad to see it go unnoticed in a subthread of a shortform feed. Consider making this a top-level post?
+1. In general, if an expert has already put in the time to write such a detailed comment, I strongly encourage them to turn it into a top-level post. If it sounds daunting to edit a comment to the ostensibly higher standards of a top-level post, then don’t; just add a brief disclaimer at the top à la “off-the-cuff comment turned into top-level post” or something and link to this original comment thread. And maybe add more section headings or subheadings so the post is easier to navigate and parse.
Another aerospace engineer here! I agree with most of your assessment, with some caveats about both asteroid mining and the colonization of Mars.
Asteroid Mining Quibbles
The viability of asteroid mining within the next 30 years will depend on what geologic activity occurred on large metallic asteroids such as 16 Psyche early in the object’s history. If a process created mostly pure gold or a mixture of precious metals and brought it to the surface, then the resulting material can be mined relatively cheaply and easily. Otherwise, mining will be as speculative as you have previously stated.
If surface gold deposits can form on metallic asteroids, we won’t discover evidence for such a process until the Psyche spacecraft visits 16 Psyche. At that point, viable deposits of gold could be discovered by the spacecraft’s Multispectral Imager.
Mining a high quality gold deposit will be difficult due to distance and low gravity.
The Psyche spacecraft will take 6 years to arrive at Psyche. Barring an unexpected breakthrough or the revival of Project Orion, transit times will continue to be absurd. This leads to a horrifically long mission duration (15 years or more).
Companies will need to wait at least half a decade to find out if their spacecraft can successfully extract gold from a high quality deposit. Thus, companies need to succeed on their first attempt.
The light delay of several minutes to several hours will make troubleshooting absurdly difficult, once all expected failure modes have been accounted for and all redundancies exhausted. I am of the opinion that a general intelligence needs to be on site to figure out what went wrong, design an adequate solution and apply said solution to fix the problem.
Both humans and AI could fulfill this role. However, if a true AGI was commercially available to be installed for this mission, then ASI has either been reached or will soon emerge. Hence, humans will probably fulfill the role of general intelligence.
The low gravity makes most cutting tools difficult to use. Large amounts of ballast can be used to make tools like saws work as they would on Earth. Alternatively, high powered lasers could be used to vaporize material attaching a surface gold deposit to the surrounding metal.
Despite all of these difficulties, mining a concentrated gold deposit on the surface of a large metallic asteroid will likely be profitable. Gold is currently about $135,000 per kilogram. Even if Starship brings cost to orbit down to merely $150/kg, then the cost to send a 100 metric ton mining vessel into Low Earth Orbit will be only $15,000,000. I think it’s reasonable to believe that an ion propelled mining vessel with a return payload of 10 metric tons could be built for 500 million dollars. Going by current gold prices, the return payload would be worth 1.35 billion dollars, yielding a profit of 0.85 billion.
It is thought that about 190,000 metric tons of gold have ever been mined, so the gold market should be able to withstand small to medium scale asteroid mining.
Difficulties With Mars Colonization
While colonizing Mars will not be profitable anytime soon, as long as Elon Musk is alive and in control of SpaceX, this won’t matter. I believe that one of Elon Musk’s primary objectives is to send as many people to Mars as possible as quickly as possible. Therefore, in the 2030s and 2040s most of SpaceX’s profits will likely be used to send people, supplies and industrial machinery to Mars. This state of affairs is ultimately unsustainable because Elon Musk will die.
I think there is a less than 1% chance that Elon Musk will have a similarly motivated successor take over SpaceX. In the unlikely event that we do see Elon Musk 2 take over SpaceX, the rest of these bullets will not apply since the Mars colony will still receive the support it needs to develop on a more relaxed timescale.
There is no way for a Mars colony to consistently produce and export the high value, low mass goods it requires to be profitable. As a result, by the time Elon Musk dies the SpaceX Mars colony must have embraced autarky. If it is not adequately designed, the colony will struggle and die as it cannot endure without imports from Earth.
A successful Mars colony will need all of the following factories to keep itself alive. Some of these factories can be built with existing technology, others require new innovations. All of these combined should cover Mars’s basic food, power and fabrication needs.
A solar panel factory that produces solar panels and the machinery required to make solar panels using only solar power, basalt and human/robot labor.
If a concentrated uranium deposit is found and there are nuclear physicists on the red planet, nuclear power can replace solar power and massively simplify the colony.
A factory that can produce space suits using only the resources available on Mars.
Multiple independent methods of satisfying the colony’s food needs. This should include surface greenhouse agriculture, edible algae aquaculture and synthetic food production from H2O and CO2.
A factory that can produce large steel parts using the available energy supplies on Mars.
A factory that can make tunnel boring machines using only materials that can be made from basalt.
A machine shop that can be used to make an identical machine shop to the same or superior tolerances.
A factory that can make batteries.
A factory that can make drills and fasteners.
A factory that can make welding equipment.
A production plant that can produce adhesives, sealants and lubricants from available chemical feedstocks.
Other miscellaneous assembly lines necessary to support the solar panel factory, food production, steel production, general colonial construction and tunnel boring machines.
Unfortunately, some of the most important technologies with long lead times have not been developed. For instance, of all the solar panel factories on Earth, exactly none of them can produce solar panels from basalt using only solar power and human labor. Blue Origin’s Blue Alchemist program may eventually accomplish this, but I’m not sure if their technology will be fully developed soon enough.
Meanwhile, SpaceX (the only Western entity that has the technical knowhow and financial resources necessary to build a Mars colony) has not designed and built a test colony of appreciable scale. I’m disappointed by this since test colonies on Earth and the Moon are the best way for SpaceX to demonstrate and iterate the technologies needed for Mars.
Essentially, a successful Mars colony must be engineered for immediate, total self sufficiency. Otherwise, it will be crippled when subsidized imports dry up, since Mars lacks any viable export resources. Import scarcity will then cause an unprepared Mars colony to suffer a fate similar to Norse Greenland, a highly isolated, mostly self sufficient island economy that slowly crumbled when support from Norway was cut off by the little ice age.
A Chinese Mars colony will face similar stressors. However, China is more wiling to sink resources into a prestige project with a low return on investment. Therefore, I think a Chinese Mars colony will survive until propulsion technology dramatically improves.
Don’t exactly disagree but there’s a difference between Starship landing reliably and scaling up vs truly being “finished”.
$15/kg basically requires airline-like operations (keeping total operational cost to 3-4x fuel cost) while maintaining a 4% payload fraction. I don’t think the next version of Starship is capable of this due to the sheer number of kinks to work out to get the number of maintenance items down to ~0 per launch with a reusable upper stage, so the first time it could happen is with a later version of Starship similar to what happened with the Falcon 9 Block 5, which took 8 years after the first Falcon 9 launch and had almost 2x the payload. Also possible that it requires a complete redesign (9 meter → 12 meter diameter, new engines) or future advances in e.g. TPS material, or doesn’t happen until all the maintenance is automated.
Oh, also: Apparently cost per kg of air freight across the Pacific is $5-$10/kg. So maybe by 2035 or so, after Starship is fully operational and has had time to be optimized more and deployed at massive scale, it’ll be approximately as expensive to visit an orbiting space station, or to mail something there, as it is to fly from US to China.
(All of this assumes no AGI of course. AGI changes everything and makes things even more crazy.)
ChatGPT suggests China → US: about $6.5 per kg : US → China (backhaul): about $1.2 per kg. That means a roundtrip of 1kg is a bit less than $4. Of that around a fourth is fuel prices, so you have around $1 fuel prices per kg.
Starship on the other hand needs around $1000k in fuel to transport 150kg mean $6.6 per kg. Even if you double the efficiency you still won’t reach the same numbers.
When it comes to non-fuel costs, I would not expect them to be much cheaper. Especially when we talk about prices billed to customers.
The airplane market is very competitive with low profit margins that regularly make airlines go bankrupt so that they need to be bailed out. SpaceX will likely target higher profit margins and I don’t see a market with airplane like competition in 2035.
It may actually be more affordable to build some kinds of high cost-per-kg structures (e.g. datacenters, high-tech factories) in space than on land.
This seems to be partly the reason for Google Suncatcher project. However, it’s not clear that the have a solution to the cooling problem that allows anything of the scale of datacenter to exist in space.
Nobody builds datacenters that can run without water on earth because the cooling without water is too expensive. It will get more expensive in space.
The application you didn’t list are surveillance satellites. The cost of providing 24⁄7 video surveillance of the whole world is dropping.
Project Suncatcher is a moonshot exploring a new frontier: equipping solar-powered satellite constellations with TPUs and free-space optical links to one day scale machine learning compute in space.
… In the right orbit, a solar panel can be up to 8 times more productive than on earth, and produce power nearly continuously, reducing the need for batteries. In the future, space may be the best place to scale AI compute. Working backwards from there, our new research moonshot, Project Suncatcher, envisions compact constellations of solar-powered satellites, carrying Google TPUs and connected by free-space optical links. This approach would have tremendous potential for scale, and also minimizes impact on terrestrial resources.
We’re excited about this growing area of exploration, and our early research, shared today in “Towards a future space-based, highly scalable AI infrastructure system design,” a preprint paper, which describes our progress toward tackling the foundational challenges of this ambitious endeavor — including high-bandwidth communication between satellites, orbital dynamics, and radiation effects on computing. By focusing on a modular design of smaller, interconnected satellites, we are laying the groundwork for a highly scalable, future space-based AI infrastructure. …
The proposed system consists of a constellation of networked satellites, likely operating in a dawn–dusk sun-synchronous low earth orbit, where they would be exposed to near-constant sunlight. This orbital choice maximizes solar energy collection and reduces the need for heavy onboard batteries. For this system to be viable, several technical hurdles must be overcome:
1. Achieving data center-scale inter-satellite links
re: that last point, they’re banking on price to LEO falling below $200/kg by the mid-2030s (so $15/kg would be an OOM more awesomeness still) because “at that price point, the cost of launching and operating a space-based data center could become roughly comparable to the reported energy costs of an equivalent terrestrial data center on a per-kilowatt/year basis” (more in their preprint).
“It may actually be more affordable to build some kinds of high cost-per-kg structures (e.g. datacenters...”
There is a company called Starcloud attempting to do exactly this (recently they launched their first H100). A lot of critics say the heat dissipation through radiation is an issue (radiation is the least effective form of thermal transfer), so their core IP is essentially giant expandable heatsinks.
I havent looked at it in much detail, but it sounds like a bad idea. Mostly as I am not sure the ‘benefits’ listed make much sense.
‘No fresh water needed’ - if your heat disipation tech is good enough to run it without water, then why not run it on land without water? Land gets air cooling for free in addition to whatever tech you are running, space doesnt.
‘Frees up space on land’ - if you dont care about internet ping, then land has plenty of empty deserts you can build in. If every square meter of the earth is filling up then going underground or underwater are also surely cheaper than space.
‘Solar’ - This one makes sense.
Another downside to note—increased radiation exposure. That presumably cuts the lifetime of the chips.
The argument as I understand it—and to my surprise when I did some napkin math it checked out—is that the efficiency advantage of having your solar panels in space (no atmosphere/clouds, no nighttime) is huge and offsets literally all the other disadvantages, from launch costs to the need for fancy radiators.
SpaceX is working on Starship, which is afaict about as close to being finished as the aforementioned competitor rockets, and when it is finished it’ll should provide somewhere between $15/kg and $150/kg.
Does some independent analysis exist that goes through the calculations to come up with those performance numbers for the Starship design, and maybe estimate how far Starship development is from commercial viability? My impression is that at this point no claims by SpaceX/Tesla should be given any credence, given their abysmal track record with those. (Red Dragon Mars 2018? Starship Mars 2022? Tesla FSD?) On the other hand, it can be easy to overcompensate because of this, just because many of their claims have no basis in reality, does not automatically mean that their technology is bad. Hence, it would be nice to see someone do a thorough analysis.
Starship can launch something like 150 metric tons to orbit iirc. So that’s 150,000kg? So to get to $15/kg it would need to cost a bit more than $2M dollars per launch. Or to get to $150/kg, it would need to cost a bit more than $20M dollars per launch.
IIRC the cost to build an entire Starship + Superheavy is something like $100M. It’s made of steel, so the materials themselves are pretty cheap, and the few dozen Raptor 3′s are like half a million each, and the manufacturing cost will go down a lot due to returns to scale as they start pumping out a ship a day at the Starfactory. (And eventually, several ships a day; perhaps the limits on efficiency are reached when they are producing hundreds per day like a car factory assembly line)
Falcon 9′s already get reused 30+ times, and Raptor 3 is supposed to be much more reusable than the Merlin engines. Seems plausible to me that Starship, once it matures over the course of a decade or so, will be doing hundreds or even thousands of flights. (main uncertainty IMO is the heat tiles...)
So the amortized cost of building the ship won’t prevent it from hitting the $15/kg milestone, and the $150/kg milestone should be almost trivial to hit.
what about the fuel and propellant costs? LOX and liquid methane. Both are super cheap as fuels go but still IIRC the cost of a fully fueled stack is like $2M just for the fuels? Maybe that’s where the $15/kg estimate comes from, maybe it’s a lower bound set by the price of the fuel.
I wonder though if the price of the fuel can be brought lower over time… for example, by producing it on-site at massive industrial scale, and taking advantage of the falling cost of solar panels and the bountiful sun of Texas to get unusually low energy prices to power the whole thing.
How’s that? It’s possible I hallucinated some of those numbers.
Starship can launch something like 150 metric tons to orbit iirc.
Well this is one of the main assumptions I am doubting. We haven’t seen Starship carry anything close to that. AFAIK none of the flights so far were done with a mass simulator, the most it carried was a couple of starlink satellites, which I don’t think would weigh more than like 1 ton.
Also, to what orbit? Low earth orbit, geostationary orbit, or an interplanetary transfer trajectory are completely different beasts. (But I guess for most of the examples you list for economic impact you mean LEO.) And with what reuse profile? Both booster and upper stage reuse, or just booster, or nothing? That obviously factors massively into cost, for the lowest cost you want full reuse.
Upper stage reuse in particular is completely new and unproven tech, they promised that with the Falcon 9 too but never delivered.
I would be interested in e.g. seeing a calculation of a LEO launch with booster return to launch site, and with upper stage landing on a drone ship. (Idk what equations you need here, or if you need some simulator software, the extent of my knowledge is the basic rocket equation, and that I have played Kerbal Space Program. In particular aerodynamics probably complicates things a lot, both for drag on ascent, and for braking on descent.)
What is the claimed specific impulse of the raptor engines, and what might be the actual figures? (And also keep in mind that the vacuum engines of the upper stage will be less efficient at the sea level landing, though probably that does not matter much as you burn most of your velocity via aerobraking.) How much fuel are you carrying in which stage, and what reserve do you need for the landings?
At least seeing these numbers check out, without anything physics defying would already be a plus, without even getting into any of the engineering details.
main uncertainty IMO is the heat tiles...
Agree, in particular I don’t see how they will be fully reusable? (AFAIK right now they are ablative and have to be replaced.) I remember years ago there was some presentation that the ship will be “sweating” liquid methane to cool itself on reentry, this being tossed in favor of a non-reusable solution does not instill confidence in me.
what about the fuel and propellant costs?
I agree that the exact fuel price does not matter much, once you get to the point where it’s the main driver of cost you have already reached the level for transformative economic impact.
The 100 − 150 ton numbers that SpaceX has offered over the years are always referring to the fully-reusable version launching to LEO. I believe even Falcon 9 (though not Falcon Heavy) has essentially stopped offering expendable flights; the vision for Starship is for them to be flying full-reusable all the time.
That said:
I forget where I got this impression (Eric Berger reporting, possibly?), but IIRC right now they’re not on track to hit their goal numbers; the first reliably-working version of Starship might be limited to more like 50-70 tons, because the ship came in heavier than expected (all those heat tiles! plus just a lot of steel.) and the Raptor engine, while very impressive, has perhaps not fully achieved the nigh-miraculous targets they set for themselves.
if you want to take 100 tons, not to LEO, but to Mars (which is the design goal of the system) then you have to use many starships to ferry fuel to refuel other starships, gradually boosting their orbit until you have a fully-fueled ship in a highly elliptical earth orbit, and then you can finally blast off to Mars. For the moon it’s even worse, you need maybe 20 refueling flights to land 1 starship on the moon with enough fuel to come back.
Agreed with you that the heat shield (and reusable upper stage in general) seems like it could easily just never work (or work but only with expensive refurbishment, or only from returning from LEO orbits not anything higher-energy, or etc), perhaps forcing them to give up and have Starship become essentially a big scaled-up Falcon 9. This would still be cheaper per-kg than Falcon 9 (economies of scale, and the Raptor engines are better than Merlin, etc), but not as transformative. I think many people are just kind of assuming “eh, SpaceX is full of geniuses, they’ve done so many astounding things, they’ll figure out the heat shield”, but this is an infamously hard problem (see Shuttle, Orion, X-33...), so possibly they’ll fail!
Some other tidbits:
Raptor’s claimed vacuum ISP is 380; I don’t think they’re just, like, making this up (they have done lots of tests, flown it many times, etc—it’s not a hypey future projection like “Starship will cost $4m per flight”), but I also don’t know where I’d go if I wanted to prove to myself that the number is legit (wikipedia just cites an Elon tweet...).
Apparently those Starlink mass simulators actually weigh about 2 tons each?? So flight 7, which carried 10 Starlink simulators, actually put 20 tons of payload in orbit.
The first reliable version of Starship will very likely fall short of its intended 100 ton goal (i mean… unless it takes them a really long time to make Starship reliable, lol). But they also plan to stretch the rocket, refine the engine, maybe someday make the whole thing wider, etc. So I expect that they’ll eventually hit 100 tons. (The first version of Falcon 9 could only lift 10.4 tons to LEO; the current version can lift 17.5 tons AND land the first stage on a drone ship for reuse!) But of course if you make the whole ship bigger, some of your launch costs are gonna go up too.
Personally I’m doubtful that they ever hit the crazy-ambitious $20/kg mark, which (per Thomas Kwa) would require not just a reusable upper stage (very hard!) but also hyper low-cost, airline-like turnaround on every part of the operation. But $200/kg (1 OOM cheaper from where Falcon 9 is today, using the rumored internal cost of $30m/launch and 17.5 ton capacity) seems pretty doable—upper stage reuse (even if somewhat ardurous to refurbish) probably cuts your costs by like 4x, and the much greater physical size of Starship might give you another almost 2x. Cheap materials (steel and methane vs aluminum and RP1) + economies of scale in Raptor manufacturing might take you the rest of the way.
Raptor’s claimed vacuum ISP is 380 [...] I also don’t know where I’d go if I wanted to prove to myself that the number is legit (wikipedia just cites an Elon tweet...).
The Isp of a closed cycle rocket engine with a given propellant mix is largely a function of its chamber pressure and expansion ratio, so one can use a program like RPA to plug in known numbers and see what other claims are consistent with an Isp of 380. Example (for SL variant) in this tweet.
My guess is that 380 is achievable if they close the throat and use a large enough nozzle, but they’ll opt for slightly lower in order to cram 9 engines into the upper stage. With Starship staging at record low velocities, reducing gravity losses through higher thrust might matter more than a 1% efficiency gain.
Yeah good point re: heat shields & upper stage reuse being hard. They’ve experimented with reusing Dragon heat shields but still they mostly just replace them each time.
Thinking aloud… suppose they never solve Starship reuse, but the Booster basically works great (as it seems to already be working pretty great). So the booster gets reused like 100 times but the Starship has to be scrapped each launch.
...In this world, the Starship could maybe be cheaper due to using old, worn-out engines retired from the boosters (don’t need to be super reliable since it can still complete the mission if a couple blow up, none of them are coming back anyway) and not needing a heat shield or wingflaps. Apparently right now the majority of the cost is the engines, the hull itself is cheap. And again a full stack booster+starship costs $100M. So the starship (having far fewer engines) is probably like… $25M or so right now? And price could drop further due to the aforementioned effects plus simply normal returns to scale… let me see how much does a car cost per kg and how much does starship weigh? Apparently starship dry mass is 85,000 kg. Tesla model 3 is 1600kg and costs like $36,000. So $22/kg. So if they can produce starships for the same cost per kg as model 3 cars, then the cost would be less than $2M. So yeah, obviously they won’t get there immediately but that’s the price they would naturally trend towards.
So that means even if they don’t reuse the upper stage at all, if they use old engines they would have retired anyway then they could get the price down to maybe $30/kg eventually. Assuming they have 150 tons to orbit payload of course, which is unproven. If they use new engines, maybe they can only get to $100/kg.
Also, as a fun bonus, if they are building a space station in LEO, then they can use the hulls of the single-use starships as material for the space station! That’s a free 85,000 kg of steel in LEO with every launch, in addition to the payload! And it already comes formed into a spacious pressure vessel!
SpaceX is amazing. As best as I (and Claude) can tell, the situation is as follows:
Competent competitors to SpaceX are developing rockets that should provide around $2000/kg cost-to-orbit. This is a big improvement over legacy space competitors like ULA, Ariane, etc. which range from $5000/kg to $50,000/kg. (It’s amazing that those competitors are still getting any business… the answer is nepotism/corruption basically afaict)
However, these competent upcoming rockets that can do around $2000/kg? They aren’t ready yet. Probably it’ll be like 5 more years before they really solidly hit that milestone at scale.
SpaceX, meanwhile, has already hit that milestone with Falcon 9 and has been there for years. It’s why they have Starlink and nobody else does.
But that’s not all. SpaceX is working on Starship, which is afaict about as close to being finished as the aforementioned competitor rockets, and when it is finished it’ll should provide somewhere between $15/kg and $150/kg. So… ONE TO TWO ORDERS OF MAGNITUDE CHEAPER STILL.
This is a really big deal. The reason space looks the way it does today (mostly empty, a few satellites and probes, two space stations) is that it’s so damn expensive to go there. SpaceX brought costs down by an OOM already and this unlocked Starlink already, and probably in the fullness of time will lead to 10x more of the other things as well (10x bigger space stations for example). Or heck, could be more than 10x; when the costs drop by 10x you often don’t merely do 10x more of the same thing, you do even more than that because now a bunch of stuff is profitable that wasn’t profitable before.
And then once Starship is online we’ll get another 1-2 OOMs on top of that. We won’t have finished adjusting to the price drop from Falcon 9 (which only happened over the course of the last decade), when we’ll be hit by another even bigger price drop!
Intuition pump: A typical house weighs something like 50,000 kg. and costs about $300,000 to build, plus whatever you pay for the land. (in e.g. Berkeley, the cost of the land may be close to a million dollars!) At $15/kg, you could launch your entire house into orbit for less than the price of a house-sized plot of land in Berkeley. (!!!)
I want to say that again for emphasis. If cost-to-orbit gets close to $15/kg, then the price to buy a house in Berkeley will be about the same as the price to buy a similarly big house in an orbiting space station. (!!!)*
It may actually be more affordable to build some kinds of high cost-per-kg structures (e.g. datacenters, high-tech factories) in space than on land.
We are in the middle of a qualitative shift in humanity’s relationship to space. And it’s basically all thanks to SpaceX.
*Yes, you probably need more expensive materials to build in space, because you need to be vacuum sealed etc. But on the other hand, if you use lighter materials (steel instead of concrete and wood for example) then you can cut the launch costs by a lot. I’m waving my hands and assuming these factors roughly cancel out.
Ex-aerospace engineer here! (I used to work at Xona Space Systems, who are working on a satellite constellation to provide a kind of next-gen GPS positioning. I’m also a longtime follower of SpaceX, fan of Kerbal Space Program, etc) Here is a rambling bunch of increasingly off-topic thoughts:
Yup, SpaceX is a big deal:
yup, Spacex is a totally off-the-charts success compared to basically any other aerospace company. (Although maybe historically comparable to the successes of early NASA?) It’s not just that their rockets are good; their Starlink satellites are also very impressive in a variety of ways—basically no other satellite company can match them on cost-vs-capability, the uniquely efficient flat-pack design, etc. And they do other stuff well too, like developing their Dragon spacecraft that certainly does a better job than Boeing’s Starliner or Sierra Nevada’s Dream Chaser.
It’s correct IMO to pay a lot of special attention to SpaceX when analyzing the aerospace industry and even perhaps the big-picture future of space exploration over the next few decades. (I presume you are thinking about SpaceX in the context of researching how space exploration might go in various “AI 2030” singularity scenarios?) Although SpaceX probably isn’t a totally unstoppable juggernaut—it’s totally plausible that Starship might continue to see troubles & delays, while Blue Origin’s “New Glenn” and RocketLab’s “Neutron” and other rockets might manage to beat expectations and scale up quickly, creating a more competitive world rather than a monopolistic Starship-fueled continuation of the famed “SpaceX steamroller”.
“SpaceX brought costs down by an OOM already and this unlocked Starlink already”—yeah, people don’t realize that Starlink constitutes 75% of all satellites in orbit (8,800 / 11,700). This is maybe not a totally fair comparison insofar as Starlink satellites are a little smaller (and in lower-energy orbits) than the big honking GEO satellites of yore, but still—in a certain sense, Starlink versus all the traditional satellite industries is a little bit like Uber versus the taxi market. It’s not just that SpaceX has captured a large percentage of the preexisting launch market; they’ve made the market way bigger.
Why are Ariane & other legacy launch companies even still alive, lol:
To your question of “why are ULA / Ariane still getting business; is this just nepotism / corruption?”—I think the situation is more accurately described in terms of national security concerns.
For Ariane, national governments want to maintain sovereign access to space—Germany, the UK, etc, don’t want to have to hand over their spy satellites to Russia or America or any other major powers for launch! But the “European military/intelligence satellite launches” industry isn’t big enough to really sustain an entire launch company like Ariane. So, Europe pressures its own commercial satellite companies (including a lot of the broadcast & communication companies operating GEO satellites, who have always launched on Ariane all through the 1980s / 90s / 00s when Ariane really was the cheapest and best option) to keep buying Ariane contracts so there’s enough European launches happening to support a European launch company. (The pressure / implied threat being that if those GEO satellite operators defect to launching on SpaceX, Europe might cut them off from contracts / subsidies / whatever other kind of industrial-policy support they’re currently providing.)
One reason why this works alright is that rocket launches are often cheaper (like $100m - $200m) than the satellites they’re launching (which can be many hundreds of millions for GEO commsats, or billions for fancy military / science missions). So the rocket launch is only a minority of the overall cost.
In ULA’s case, they are mostly propped up by the Pentagon being (reasonably IMO) concerned that they don’t want American space launch to become a monopoly because then SpaceX could charge very high prices, so they do stuff like giving 60% of their launches to SpaceX and 40% to ULA according to a big contracting process. In the future, Blue Origin might surpass ULA and mostly take over their role in the industry.
The continued existence of SLS is totally just corruption though, lol… (combined with extreme bureaucratic inertia, unwillingness to do proper decisionmaking under uncertainty / take certain perceived risks, while ignoring the risks like that you might spend tens of billions of dollars just to develop a way-more-expensive-than-the-competition rocket...)
Also, some satellite-constellation companies that think they’re “competing with SpaceX” (more like losing to SpaceX, amirite?) refuse to launch on SpaceX vehicles. Mostly I’m thinking of Amazon’s Kuiper satellite internet constellation (which wants to mostly launch on Bezos-owned Blue Origin), and the european-ish OneWeb. Also, like a more extreme version of the situation with Europe and Ariane, obviously China doesn’t let Chinese companies just buy Falcon 9 launches.
Another important factor in “why is anybody still buying these expensive-ass non-SpaceX rockets??” is that the DID prefer buying SpaceX rockets, but then SpaceX raised their prices (from $70m some years ago to about $100m today, IIRC), and then SpaceX got booked solid and ran out of rockets (despite their impressive scaling over the years), so if you’re in a big hurry to launch soon, you need to start looking at other more expensive companies (indeed, even many of these companies are booked out for many years, scaling up as fast as they can manage, etc).
Will Starship make a house in space cheaper than a house in SF? No:
House-in-Berkeley versus house-in-space is of course a weird comparison, but I very much doubt Starship could singlehandedly make it cheaper to live in space even if we used all Starship capacity for building a giant space station. An orbital space station needs a lot of complex expensive stuff to make it work (thrusters, momentum wheels, batteries, solar panels, life-support equipment for recycling water and air), plus stuff in space breaks down a lot more quickly than stuff on Earth which would increase the cost through faster depreciation. (The ISS is made of fancy aluminum pressure vessels and micrometeoroid shielding and stuff, but—despite the fact that its 7-person crew spends a huge portion of their time doing fixes & maintenance—it’s springing all kinds of weird leaks and is gonna have to be deorbited soon, even though most of the station is less than 25 years old. Contrast this with the house where I live in Colorado, built a whopping 35 years ago, which still basically does fine with just minimal home maintenance, occasional new appliances, etc.)
Plus obviously your house will need lots of supplies (food, amazon packages, but also stuff like air and thruster fuel), and transporting these supplies into orbit will be much more expensive than going to the grocery store in Berkeley.
Obviously if it was just one little house in space, then it would be SUPER expensive (since you’d need all those subsystems just for your one little house) and there would be no feasible way to do regular (like monthly) deliveries since you don’t eat a Starship full of groceries every month. But what I’m saying is that it would still be expensive even if you wanted to save money by aggregating all the houses together into one giant space station to cut down on subsystem & resupply costs.
Perhaps a more interesting point: the reason why Berkeley is expensive is because the land is expensive. But as launch gets cheaper and cheaper thanks to Starlink, the most valuable orbits will start becoming very crowded, and we’ll probably start charging for them. Right now, spots in orbit are basically given away for free (although before you launch, you’ve gotta get an FCC license to operate your satellites, which is a paperwork-intensive process, almost like the space version of getting a pharma drug approved by the FDA). But in the future, I suspect we’ll probably implement some kind of “space Georgism” to prevent kessler syndrome and properly allocate the most valuable orbital slots. (Where “we” is ideally some kind of international agreement, but in practice will probably just be, like, the USA’s department of commerce, and then China does their own similar thing, and no other country launches enough satellites to be relevant.) Under such a system, valuable spots in orbit might be auctioned off a la elaborate electromagnetic spectrum auctions. So, if you want to live in Space-Berkeley (a valuable, crowded orbit like sun-synchronous LEO), most of your cost might soon be space-land (some complicated notion of orbital crowdedness + making credible promises to maneuver around debris and de-orbit your satellite at the end of its scheduled lifetime) instead of just the construction cost. Unless you want to live in some random radiation-filled MEO orbit not really useful for anything, like Space-Rural-Oklahoma.
Will it really be cheaper to build factories in space?? Probably not pre-ASI, but possibly, idk:
You’re probably right to focus on “high cost-of-kg” operations as things that are most likely to be done in space. Lots of people talk about this dumb zombie idea of putting solar panels in space, even though it has only become less sensible over time. People are like “omg, space launch is cheaper now, maybe now it finally makes sense to implement the techno-optimist 1970s dream of solving the oil shock by putting solar panels in orbit!!” But solar panels have gotten cheaper much faster than space launch has gotten cheaper, so the trend is actually in the other direction—don’t even bother mounting the panels on a basic single-axis tracking system to follow the sun over the course of the day; just drop them directly on the freaking dirt to save on installation + mounting costs.
Obviously in an ASI-singularity scenario (or even, just, the long-term trajectory of a non-AI human civilization growing at 2% per year), we are eventually going to use up all the land, and then the natural next thing to do is to start launching lots of solar panels into space. But it doesn’t make much sense to start doing this now.
I doubt that factories is a winning idea either:
Factories are usually defined by needing lots of input material and producing lots of output material. Shipping this stuff to space and back would be expensive, so it only makes sense IMO if either the inputs are coming from space already, or the outputs are destined to stay in space.
Working in zero gravity + vacuum tends to make most things more difficult, not easier. Lots of factory processes designed on Earth will break in space. So, doing anything with moving parts in space is probably way more of a hassle than doing the same thing on earth, unless there’s some amazing special advantage to working in vacuum or zero-gravity. Some proposed special advantages I’ve heard mentioned:
People used to talk about doing pharma research in space, because proteins crystallize much more easily in zero gravity?? But I think the reason people were so hyped about crystallizing proteins is because we hadn’t solved the protein folding problem yet! (You can work out the folded structure of individual proteins by exhaustively studying protein crystals.) Now that we have AlphaFold, I think that use-case has sailed...
Nowadays people talk about doing semiconductor manufacturing in space, on the grounds that semiconductor manufacturing is extremely afraid of dust (so it might actually help to do in vacuum), and the machines are all so high-precision that they might as well be aerospace-grade anyways. Maybe there’s something to this idea?? But if you need vacuum so much, you could probably just build a vacuum-sealed assembly line, or even build an entire vacuum-sealed wing of the TSMC factory (with employees walking around in pressure suits and everything) for cheaper than building an orbital factory. (The vacuum quality of low-earth-orbit isn’t even especially great compared to what you can get pretty easily on the ground!) Semiconductor manufacturing is infamously one of the most difficult, complicated things that human civilization does; I’d be surprised if you could just move it all into space without a million little things going wrong.
Something about optical fibers, carbon nanotubes, and other advanced materials potentially being easier to manufacture in low gravity?? I don’t know much about this—mostly I’m just remembering the plot of Andy Weir’s book “Artemis” and hoping that the optical-fiber McGuffin plot-point was based on plausible background research. You could imagine an AI angle here too, if we need tons of super-high-quality optical fiber to make the interconnections between our vast datacenters full of TPUs or photonic chips or however we multiply matrices in the year 2040.
One entertaining niche application of space manufacturing is to produce “extinct polymorphs”—chemicals like the HIV drug Ritonavir, which once were easily manufactured on earth but have since become nearly-impossible to create, thanks to a bizarre ice-nine-style process where they get “infected” by misfolded versions of the same molecule! Varda Space is an aerospace startup which actually produced some Ritonavir in space precisely to make this point. But I hardly expect “bringing back extinct polymorphs” to become a major portion of GDP in the future; it seems intrinsically niche. (Barring, perhaps, some mirror-life related catastrophe such that we are only able to grow crops and preserve natural necosystems in pristine space environments, a la the sci-fi stories Interstellar, Silent Running, and Speaker for the Dead.)
If you have AGI / ASI, then maybe you can simply have the AI redesign all your manufacturing processes from first-principles to work well in the space environment. Maybe in some objective sense, space is actually a better place to do most manufacturing! But in that case you do need the AGI, and you also need some time to bootstrap the entire alternate manufacturing ecosystem. This might face some of the same pros & cons as Carl Feynman’s concept of creating an alternate manufacturing system of self-replicating automated/miniaturized machine shops (see my comments here), although of course I’d expect a true ASI to power through the various troublesome issues of transitioning over to a whole new industrial base pretty quickly.
Putting datacenters into space is a little more plausible IMO, because you don’t have to worry about tons of moving parts and manufacturing processes, and your input is just energy while your output is just heat + information.
But you do need energy, which you can either beam up from earth via some kind of microwave laser (but this hasn’t been tested IRL, has some pretty serious efficiency losses even just in theory, etc), or manufacture locally with solar panels or nuclear power (but this is gonna drag down your cost-per-kg launching random solar panels).
I will say that, compared to the idea of space solar power, space-datacenters is a big improvement because you’re increasing the cost-per-kg of what you’re launching (all those GPUs), and instead of beaming back lossy microwave radiation energy, you’re beaming back information, which seems easier.
But I’m doubtful that you could do AI training in space very easily, since you’d have to formation-fly a ton of satellites close together (thus spending a lot of fuel?? unless you want to do something ridiculous and experimental like electrostatic-based formation flying), connect them all with extremely high-bandwidth laser links (dunno how this compares to the bandwidth Starlink already achieves...), etc. My impression is that if you don’t have high-bandwidth interconnects, you’re probably limited to AI inference instead of training? (idk that much about AI training though...). I’d also be worried that both training and inference would require lots of data to be transmitted to and from the ground, except then I remembered that the whole point of Starlink is to put the whole planet’s internet infrastructure in space, and it seems to be working fine—so at least bandwidth won’t be a problem!
And you do need to get rid of all the heat, which in some ways gets a lot harder in space (there’s no air to do convection, nor a ready supply of cheap water), although in some ways it gets easier (space is really cold, so you can cool down by just blocking the sun with a big mylar mirror and radiating in every other direction).
I’m not sure, but I’d be worried that radiating heat doesn’t scale well (due to square-cube-law) while piping cold water around scales better. (Until, of course, you cover the entire planet in datacenters and solar panels and fusion reactors, melt the icecaps and boil the oceans, and you’re forced to resort to radiative cooling because you’ve run out of places to convect to!)
And this is very different from the berkeley house case, because (at least at the moment) there’s still a vast amount of basically dirt-cheap useless desert land on which to build datacenters, power infrastructure, etc. People complain about regulation and permitting, but:
Space will also feature regulation & permitting obstacles, around things like space-debris mitigation, electromagnetic spectrum for beaming vast amounts of information back to earth (unless you can figure out space-to-ground laser comms, perhaps?), and power generation. Whether you are launching nuclear reactors into space, constructing gigawatt-scale microwave lasers that hostile superpowers will perceive as anti-ballistic-missile defenses, or even just building a mass-catapult on the moon so you can hurl hundreds of tons of lunar-manufactured solar panels towards the Earth, somebody is probably going to want to submit some comments during the 30-day public notice period...
My impression is that datacenters could route around much of the most severe permitting issues (such as around transmission lines and energy interconnect queues) if they were willing to go off the grid and build all their own power + battery storage. Datacenter builders don’t want to do this because that’s more expensive. But building datacenters in space also means going off-grid, plus you have to do a ton of other stuff! (Yes, I get that space-based solar is 4x more effective and cuts down on your need for batteries, but space launch isn’t the only thing allowed to reduce in price over time—batteries and other kinds of energy-storage technology are also getting cheaper all the time.)
So, what are we gonna use all those Starships for, if not in-space manufacturing or datacenters??
Right now, the biggest and most-valuable use of space is for communications (like Starlink internet, but also military communications, television broadcasts from GEO, specialized connections to airplanes and ships, etc), navigation (like GPS), taking photos of the earth (mostly for military intel, but this also has applications in agriculture, finance, etc), and assorted military applications. So, in the immediate future, I’d expect us to just keep massively scaling those applications rather than using Starship to do wacky new stuff:
You kind of only need one navigation constellation (although it’s due for an upgrade, re: Xona’s plan), so this won’t be a huge number of satellites.
With satellite internet, more & bigger satellites = more internet bandwidth, so I’d expect this to grow a lot. Claude says that Starlink today only represents “perhaps 1–3% of international backbone capacity—and international traffic is itself only a subset of total internet traffic”. Surely it would be profitable to scale this up such that satellites are providing as much internet bandwidth as all existing ground-based infrastructure combined (ie, +100% instead of +2%), and we could still find reasonably productive ways to use that bandwidth? So that would be like 50x all the Starlink satellites that have been launched so far.
So far there have been over 350 Falcon 9 launches full of Starlinks (and US-military-branded Starshields) -- assume that you want to launch 50x that amount, but also that Starship can launch 5x as much mass (90 tons instead of 17.5 tons). That comes out to 3,500 Starship launches to achieve that goal. If you look at the 11 launches they’ve done so far, pretend they were all successful and all happened this year (they were not and did not), and assume they’ll double their number of launches each year (22 in 2026, 44 in 2027...) and launch nothing but Starlinks (NASA will want a word with them about their Artemis lunar mission contract, which involves up to 20 starship launches per moon visit...), it would take them eight years to clear that backlog.
Taking photos of the earth probably scales more than navigation but less than internet—Planet’s fleet of cubesats already take 3m-resolution photos of the whole earth every day. How much more valuable would it be to have 30cm-resolution photos every hour? Or continuous video?? idk, but maybe once adding more satellite internet capacity hits diminishing returns, this becomes the next most valuable thing to scale up.
Earth-observation is also probably the place where it makes the most sense to be putting GPUs in space right away, since the satellites are already very bottlenecked on bandwidth for beaming images down to earth. If you put a GPU on all your spy satellites, you could do image-classification analysis locally (and immediately!) and only beam down the most interesting stuff. Plus you could maybe even make dynamic decisions like “oh, that’s really interesting, let’s take some more photos of that spot”—usually these kinds of decisions are delayed by the time it takes for a satellite to pass over a ground station, download an image, get the image analyzed, and then for commands to be uploaded later, so it might be a big deal to make these decisions locally & immediately.
Military expenses are dictated by adversarial / arms-race logic, so the amount we’ll spend on military stuff in space is perhaps kind of a wildcard driven by how intense the overall military competition with China gets, multiplied by how many advantageous military-stuff-in-space ideas we can dream up.
Are there any huge economic markets (besides already-discussed manufacturing, solar power generation, and datacenters) that might open up besides orbiting satellites beaming down information?
At some point, maybe asteroid mining becomes a thing? This has all the disadvantages of “moving parts in space”, plus you’re dealing with very messy inputs, but it has the advantage of the fact that certain asteroids have very high concentrations of metals that are rare on earth.
My guess is that doing an expensive space mission to bring back very valuable, rare metals (like gold, iridum, etc) is a better business plan than the zombie idea of doing an expensive space mission just to launch solar panels and simply beam back power via lossy microwave laser—beating dirt-cheap earth-based solar panels sounds impossible, but beating earth-based mining sounds a little less impossible. The business case for “solar power in space” probably only closes once you have finished covering all of earth’s deserts in solar panels. The business case for asteroid mining probably closes earlier, though I bet you’d still need pretty immense scale (like “over the next decade we’re aiming to bring back an amount of gold equivalent to 10% of all gold ever mined in history”) to amortize the huge cost of a gigantic deep-space mission and bring the whole project below the cost of earth-based mining.
An even better business plan would be if we could manufacture something even more difficult to make on earth (maybe semiconductors, optical fibers, or some other advanced material??), which would probably also have to be very high value-per-kg (to minimize the cost of transporting the inputs and outputs). But who knows if we can actually figure out anything that fits that criteria. And whatever we figure out might not be scalable to trillions of dollars (like, Ritonavir definitely isn’t) in the way that asteroid mining clearly is.
You can also mine asteroids to melt ice and make hydrogen + oxygen rocket fuel, but of course this requires some customer who’ll buy a lot of rocket-fuel for going beyond low-earth orbit (like colonizing mars?) or who has other reasons to be maneuvering all the time (like military satellites that want to constantly change their orbit to stay unpredictable?).
Melting ice is probably a lot easier than processing ore, so probably the first demonstration asteroid-mining missions are about water. But to scale up, they’d need customers (and customers high above low-earth-orbit, since their product has to be cheaper than just launching extra rocket fuel on Starships!).
Other than satellites, I think space-based industry & exploration is going to be heavily debt-financed in a really big way for a really long time.
Satellites are profitable and normal; we are obviously gonna blot out the sky with immense numbers of very large, high-powered, super-Starlink satellites (mostly for internet, also for taking photos of the earth).
But asteroid mining maybe only pays off once you scale up to some preposterous level, like bringing back a trillion dollars’ worth of gold.
I am recalling all those debates about whether Amazon or Uber were really sustainable businesses—they’re pouring billions of dollars of investors’ cash on infrastructure build-out or user subsidies; are they REALLY gonna flip to profitability and turn this all around someday?? Or the current debates about the even vastly-larger sums being invested in datacenters for training AI models. If asteroid mining for precious metals ever happens, it is gonna be that kind of situation all over again.
What about using Starship for its intended purpose of settling Mars??? IMO, nobody has thought up any plausible reason to think that a Mars city would ever return significant capital—settling another planet would be a gigantic money-sink basically forever. Yet, in some abstract sense it seems obviously likely to be worthwhile (in terms of cultural influence on humanity’s future, if not literal investment returns) to be at the forefront of colonizing the solar system! In this respect it feels similar to some parts of European colonial history—what was the ROI of Britain starting colonies in North America? In a certain sense, somewhere between low (you spend decades building it up, finally get a little bit of stamp tax, and then they go and fight a revolution against you) and extremely negative (start a Jamestown or Roanoke, almost everybody dies, then the town straggles on pointlessly for decades, consuming resupply ships but not figuring out anything to export). But in another sense, extremely high (insofar as Britain got to put their thumbprint on what later became the mighty USA).
So this is kind of like the asteroid-mining issue or datacenter-buildout issue, but on steroids—a venture so vast and so uncertain (a lot of European colonial empires, like Germany’s scramble for Africa, were bad ideas that paid off neither literally nor metaphorically!) and requiring so much debt-financing that it leaves the realm of traditional investing, or even the realm of economic booms / bubbles and instead has to be coordinated through the mechanism of national/societal greatness, competition, and prestige.
But even though doing Mars settlement is way more expensive and speculative and uncertain than even doing asteroid mining, the perceived expected-value (or perceived cost of missing out) might be higher. So I think it’s actually likely that we choose to do something closer to satellites + mars colonies, rather than satellites + in-space manufacturing/mining.
It’s also worth noting that Starship has been explicitly designed for settling Mars (the methane fuel, the Space-Shuttle-like upper stage that would be a reusable SSTO on Mars, etc). Sending many Starships to Mars and back (where they can hope to use aerobraking for landing, and locally-manufactured fuel for launch back to earth) is probably much cheaper than sending the same number to the moon and back!
And in a similar way, a LOT of potential space activities are more like Mars colonization than satellite internet—“squatting on areas that might eventually be profitable someday in the future” rather than making actual profit today.
As mentioned, asteroid mining is like this—the first company to develop the tech and visit some of the asteroids might do this in the hope of kinda claiming & squatting the opportunity, far in advance of the opportunity actually becoming profitable.
Things like putting datacenters in space or putting solar panels in space also have something like this vibe, insofar as eventually it seems we will want to do them. But what’s the scarce resource being squatted?? With Mars or asteroids, you’re hoping to cheaply stake a claim (in the sense of legal rights, precedent, etc) on scarce physical land / ore in the hopes of expensively developing / mining it later. But launching solar panels & datacenters looks more like “expensively doing something now in the hopes of profiting in the far future”. You should rather aim to be cheaply squatting something now. Maybe this would be:
Space-Berkeley slots in low-earth-orbit?? (but if LEO becomes well-governed via Fully Automated Luxury Space Georgism, this plan is not gonna work out for you...)
Being the debt-fueled leader in some industry (like how SpaceX is the leader in launch), where the industry might 100x in size (and actually become profitable) in the future. Here you’d either be hoarding a technological advantage (doing trial missions to develop your in-space manufacturing processes, but not actually launching a lot since you lose money on each mission), or (more expensively) hoarding an industrial-capacity advantage.
If you believe in AGI right around the corner, this makes trying to squat “space-based power generation / datacenters / semiconductor manufacturing” more appealing, since the applications are more concrete and if the singularity is about to happen then you don’t actually have to wait very many years paying interest on your debt.
But on the other hand, if AGI is right around the corner, surely there are tons of more-profitable things to do here on earth? Like try to invest in humanoid robots, or do normal AI investments in TSMC / NVDA, or etc?
Basically, if you are playing this game of trying to squat the opportunity to do far-future space expansion, then you wanna put yourself in the best possible position to be at the forefront of a grabby-aliens-style expansion into the solar system (from where, presumably, you can steamroll onwards to the galaxy).
But it’s unclear exactly what bundle of technologies / legal claims / industrial capacity / etc will actually be needed for this. (Will controlling a small town on Mars be relevant in any possible way if an ASI singularity occurs in 2050?? Probably not!)
And in particular I’d be very worried that I’d spend all this time going into debt trying to develop a clever portfolio of space-related industrial / technological capabilities, only to get instantly lapped by ASI right off the starting block. Such that maybe the only resource really worth scrambling for is simply “access to ASI”. (Plus possibly launch capacity itself, which is a big capital-intensive heavy-industry that seems less amenable to being lapped software-only-singularity style than something more intricate and design-intensive, like space-based manufacturing equipment or satellites.)
But most people are less ASI-pilled. So, probably we start launching colonization rockets to Mars (and funding lots of doomed little “datacenter in space” / “solar power in space” / “bitcoin in space” startups) anyways.
I really enjoyed reading this comment, and I’d be sad to see it go unnoticed in a subthread of a shortform feed. Consider making this a top-level post?
+1. In general, if an expert has already put in the time to write such a detailed comment, I strongly encourage them to turn it into a top-level post. If it sounds daunting to edit a comment to the ostensibly higher standards of a top-level post, then don’t; just add a brief disclaimer at the top à la “off-the-cuff comment turned into top-level post” or something and link to this original comment thread. And maybe add more section headings or subheadings so the post is easier to navigate and parse.
Another aerospace engineer here! I agree with most of your assessment, with some caveats about both asteroid mining and the colonization of Mars.
Asteroid Mining Quibbles
The viability of asteroid mining within the next 30 years will depend on what geologic activity occurred on large metallic asteroids such as 16 Psyche early in the object’s history. If a process created mostly pure gold or a mixture of precious metals and brought it to the surface, then the resulting material can be mined relatively cheaply and easily. Otherwise, mining will be as speculative as you have previously stated.
If surface gold deposits can form on metallic asteroids, we won’t discover evidence for such a process until the Psyche spacecraft visits 16 Psyche. At that point, viable deposits of gold could be discovered by the spacecraft’s Multispectral Imager.
Mining a high quality gold deposit will be difficult due to distance and low gravity.
The Psyche spacecraft will take 6 years to arrive at Psyche. Barring an unexpected breakthrough or the revival of Project Orion, transit times will continue to be absurd. This leads to a horrifically long mission duration (15 years or more).
Companies will need to wait at least half a decade to find out if their spacecraft can successfully extract gold from a high quality deposit. Thus, companies need to succeed on their first attempt.
The light delay of several minutes to several hours will make troubleshooting absurdly difficult, once all expected failure modes have been accounted for and all redundancies exhausted. I am of the opinion that a general intelligence needs to be on site to figure out what went wrong, design an adequate solution and apply said solution to fix the problem.
Both humans and AI could fulfill this role. However, if a true AGI was commercially available to be installed for this mission, then ASI has either been reached or will soon emerge. Hence, humans will probably fulfill the role of general intelligence.
The low gravity makes most cutting tools difficult to use. Large amounts of ballast can be used to make tools like saws work as they would on Earth. Alternatively, high powered lasers could be used to vaporize material attaching a surface gold deposit to the surrounding metal.
Despite all of these difficulties, mining a concentrated gold deposit on the surface of a large metallic asteroid will likely be profitable. Gold is currently about $135,000 per kilogram. Even if Starship brings cost to orbit down to merely $150/kg, then the cost to send a 100 metric ton mining vessel into Low Earth Orbit will be only $15,000,000. I think it’s reasonable to believe that an ion propelled mining vessel with a return payload of 10 metric tons could be built for 500 million dollars. Going by current gold prices, the return payload would be worth 1.35 billion dollars, yielding a profit of 0.85 billion.
It is thought that about 190,000 metric tons of gold have ever been mined, so the gold market should be able to withstand small to medium scale asteroid mining.
Difficulties With Mars Colonization
While colonizing Mars will not be profitable anytime soon, as long as Elon Musk is alive and in control of SpaceX, this won’t matter. I believe that one of Elon Musk’s primary objectives is to send as many people to Mars as possible as quickly as possible. Therefore, in the 2030s and 2040s most of SpaceX’s profits will likely be used to send people, supplies and industrial machinery to Mars. This state of affairs is ultimately unsustainable because Elon Musk will die.
I think there is a less than 1% chance that Elon Musk will have a similarly motivated successor take over SpaceX. In the unlikely event that we do see Elon Musk 2 take over SpaceX, the rest of these bullets will not apply since the Mars colony will still receive the support it needs to develop on a more relaxed timescale.
There is no way for a Mars colony to consistently produce and export the high value, low mass goods it requires to be profitable. As a result, by the time Elon Musk dies the SpaceX Mars colony must have embraced autarky. If it is not adequately designed, the colony will struggle and die as it cannot endure without imports from Earth.
A successful Mars colony will need all of the following factories to keep itself alive. Some of these factories can be built with existing technology, others require new innovations. All of these combined should cover Mars’s basic food, power and fabrication needs.
A solar panel factory that produces solar panels and the machinery required to make solar panels using only solar power, basalt and human/robot labor.
If a concentrated uranium deposit is found and there are nuclear physicists on the red planet, nuclear power can replace solar power and massively simplify the colony.
A factory that can produce space suits using only the resources available on Mars.
Multiple independent methods of satisfying the colony’s food needs. This should include surface greenhouse agriculture, edible algae aquaculture and synthetic food production from H2O and CO2.
A factory that can produce large steel parts using the available energy supplies on Mars.
A factory that can make tunnel boring machines using only materials that can be made from basalt.
A machine shop that can be used to make an identical machine shop to the same or superior tolerances.
A factory that can make batteries.
A factory that can make drills and fasteners.
A factory that can make welding equipment.
A production plant that can produce adhesives, sealants and lubricants from available chemical feedstocks.
Other miscellaneous assembly lines necessary to support the solar panel factory, food production, steel production, general colonial construction and tunnel boring machines.
Unfortunately, some of the most important technologies with long lead times have not been developed. For instance, of all the solar panel factories on Earth, exactly none of them can produce solar panels from basalt using only solar power and human labor. Blue Origin’s Blue Alchemist program may eventually accomplish this, but I’m not sure if their technology will be fully developed soon enough.
Meanwhile, SpaceX (the only Western entity that has the technical knowhow and financial resources necessary to build a Mars colony) has not designed and built a test colony of appreciable scale. I’m disappointed by this since test colonies on Earth and the Moon are the best way for SpaceX to demonstrate and iterate the technologies needed for Mars.
Essentially, a successful Mars colony must be engineered for immediate, total self sufficiency. Otherwise, it will be crippled when subsidized imports dry up, since Mars lacks any viable export resources. Import scarcity will then cause an unprepared Mars colony to suffer a fate similar to Norse Greenland, a highly isolated, mostly self sufficient island economy that slowly crumbled when support from Norway was cut off by the little ice age.
A Chinese Mars colony will face similar stressors. However, China is more wiling to sink resources into a prestige project with a low return on investment. Therefore, I think a Chinese Mars colony will survive until propulsion technology dramatically improves.
Don’t exactly disagree but there’s a difference between Starship landing reliably and scaling up vs truly being “finished”.
$15/kg basically requires airline-like operations (keeping total operational cost to 3-4x fuel cost) while maintaining a 4% payload fraction. I don’t think the next version of Starship is capable of this due to the sheer number of kinks to work out to get the number of maintenance items down to ~0 per launch with a reusable upper stage, so the first time it could happen is with a later version of Starship similar to what happened with the Falcon 9 Block 5, which took 8 years after the first Falcon 9 launch and had almost 2x the payload. Also possible that it requires a complete redesign (9 meter → 12 meter diameter, new engines) or future advances in e.g. TPS material, or doesn’t happen until all the maintenance is automated.
Oh, also: Apparently cost per kg of air freight across the Pacific is $5-$10/kg. So maybe by 2035 or so, after Starship is fully operational and has had time to be optimized more and deployed at massive scale, it’ll be approximately as expensive to visit an orbiting space station, or to mail something there, as it is to fly from US to China.
(All of this assumes no AGI of course. AGI changes everything and makes things even more crazy.)
ChatGPT suggests China → US: about $6.5 per kg : US → China (backhaul): about $1.2 per kg. That means a roundtrip of 1kg is a bit less than $4. Of that around a fourth is fuel prices, so you have around $1 fuel prices per kg.
Starship on the other hand needs around $1000k in fuel to transport 150kg mean $6.6 per kg. Even if you double the efficiency you still won’t reach the same numbers.
When it comes to non-fuel costs, I would not expect them to be much cheaper. Especially when we talk about prices billed to customers.
The airplane market is very competitive with low profit margins that regularly make airlines go bankrupt so that they need to be bailed out. SpaceX will likely target higher profit margins and I don’t see a market with airplane like competition in 2035.
This seems to be partly the reason for Google Suncatcher project. However, it’s not clear that the have a solution to the cooling problem that allows anything of the scale of datacenter to exist in space.
Nobody builds datacenters that can run without water on earth because the cooling without water is too expensive. It will get more expensive in space.
The application you didn’t list are surveillance satellites. The cost of providing 24⁄7 video surveillance of the whole world is dropping.
Made me think of Google’s moonshot Project Suncatcher:
re: that last point, they’re banking on price to LEO falling below $200/kg by the mid-2030s (so $15/kg would be an OOM more awesomeness still) because “at that price point, the cost of launching and operating a space-based data center could become roughly comparable to the reported energy costs of an equivalent terrestrial data center on a per-kilowatt/year basis” (more in their preprint).
“It may actually be more affordable to build some kinds of high cost-per-kg structures (e.g. datacenters...”
There is a company called Starcloud attempting to do exactly this (recently they launched their first H100). A lot of critics say the heat dissipation through radiation is an issue (radiation is the least effective form of thermal transfer), so their core IP is essentially giant expandable heatsinks.
Benefits:
No fresh water needed
Frees up space on land
Solar powered
Cons:
Latency
Maintenance
Harder to stop the AIs by force
I havent looked at it in much detail, but it sounds like a bad idea. Mostly as I am not sure the ‘benefits’ listed make much sense.
‘No fresh water needed’ - if your heat disipation tech is good enough to run it without water, then why not run it on land without water? Land gets air cooling for free in addition to whatever tech you are running, space doesnt.
‘Frees up space on land’ - if you dont care about internet ping, then land has plenty of empty deserts you can build in. If every square meter of the earth is filling up then going underground or underwater are also surely cheaper than space.
‘Solar’ - This one makes sense.
Another downside to note—increased radiation exposure. That presumably cuts the lifetime of the chips.
The argument as I understand it—and to my surprise when I did some napkin math it checked out—is that the efficiency advantage of having your solar panels in space (no atmosphere/clouds, no nighttime) is huge and offsets literally all the other disadvantages, from launch costs to the need for fancy radiators.
I investigated this a while back, Starship completely breaks the trendline. I arrived at $120/kg with an estimated 5 reuses.
Interactive version here.
$120/kg with 5 reuses? what about the more plausible 50 reuses?
Does some independent analysis exist that goes through the calculations to come up with those performance numbers for the Starship design, and maybe estimate how far Starship development is from commercial viability? My impression is that at this point no claims by SpaceX/Tesla should be given any credence, given their abysmal track record with those. (Red Dragon Mars 2018? Starship Mars 2022? Tesla FSD?) On the other hand, it can be easy to overcompensate because of this, just because many of their claims have no basis in reality, does not automatically mean that their technology is bad. Hence, it would be nice to see someone do a thorough analysis.
From memory, real quick:
Starship can launch something like 150 metric tons to orbit iirc. So that’s 150,000kg? So to get to $15/kg it would need to cost a bit more than $2M dollars per launch. Or to get to $150/kg, it would need to cost a bit more than $20M dollars per launch.
IIRC the cost to build an entire Starship + Superheavy is something like $100M. It’s made of steel, so the materials themselves are pretty cheap, and the few dozen Raptor 3′s are like half a million each, and the manufacturing cost will go down a lot due to returns to scale as they start pumping out a ship a day at the Starfactory. (And eventually, several ships a day; perhaps the limits on efficiency are reached when they are producing hundreds per day like a car factory assembly line)
Falcon 9′s already get reused 30+ times, and Raptor 3 is supposed to be much more reusable than the Merlin engines. Seems plausible to me that Starship, once it matures over the course of a decade or so, will be doing hundreds or even thousands of flights. (main uncertainty IMO is the heat tiles...)
So the amortized cost of building the ship won’t prevent it from hitting the $15/kg milestone, and the $150/kg milestone should be almost trivial to hit.
what about the fuel and propellant costs? LOX and liquid methane. Both are super cheap as fuels go but still IIRC the cost of a fully fueled stack is like $2M just for the fuels? Maybe that’s where the $15/kg estimate comes from, maybe it’s a lower bound set by the price of the fuel.
I wonder though if the price of the fuel can be brought lower over time… for example, by producing it on-site at massive industrial scale, and taking advantage of the falling cost of solar panels and the bountiful sun of Texas to get unusually low energy prices to power the whole thing.
How’s that? It’s possible I hallucinated some of those numbers.
Well this is one of the main assumptions I am doubting. We haven’t seen Starship carry anything close to that. AFAIK none of the flights so far were done with a mass simulator, the most it carried was a couple of starlink satellites, which I don’t think would weigh more than like 1 ton.
Also, to what orbit? Low earth orbit, geostationary orbit, or an interplanetary transfer trajectory are completely different beasts. (But I guess for most of the examples you list for economic impact you mean LEO.) And with what reuse profile? Both booster and upper stage reuse, or just booster, or nothing? That obviously factors massively into cost, for the lowest cost you want full reuse.
Upper stage reuse in particular is completely new and unproven tech, they promised that with the Falcon 9 too but never delivered.
I would be interested in e.g. seeing a calculation of a LEO launch with booster return to launch site, and with upper stage landing on a drone ship. (Idk what equations you need here, or if you need some simulator software, the extent of my knowledge is the basic rocket equation, and that I have played Kerbal Space Program. In particular aerodynamics probably complicates things a lot, both for drag on ascent, and for braking on descent.)
What is the claimed specific impulse of the raptor engines, and what might be the actual figures? (And also keep in mind that the vacuum engines of the upper stage will be less efficient at the sea level landing, though probably that does not matter much as you burn most of your velocity via aerobraking.) How much fuel are you carrying in which stage, and what reserve do you need for the landings?
At least seeing these numbers check out, without anything physics defying would already be a plus, without even getting into any of the engineering details.
Agree, in particular I don’t see how they will be fully reusable? (AFAIK right now they are ablative and have to be replaced.) I remember years ago there was some presentation that the ship will be “sweating” liquid methane to cool itself on reentry, this being tossed in favor of a non-reusable solution does not instill confidence in me.
I agree that the exact fuel price does not matter much, once you get to the point where it’s the main driver of cost you have already reached the level for transformative economic impact.
The 100 − 150 ton numbers that SpaceX has offered over the years are always referring to the fully-reusable version launching to LEO. I believe even Falcon 9 (though not Falcon Heavy) has essentially stopped offering expendable flights; the vision for Starship is for them to be flying full-reusable all the time.
That said:
I forget where I got this impression (Eric Berger reporting, possibly?), but IIRC right now they’re not on track to hit their goal numbers; the first reliably-working version of Starship might be limited to more like 50-70 tons, because the ship came in heavier than expected (all those heat tiles! plus just a lot of steel.) and the Raptor engine, while very impressive, has perhaps not fully achieved the nigh-miraculous targets they set for themselves.
if you want to take 100 tons, not to LEO, but to Mars (which is the design goal of the system) then you have to use many starships to ferry fuel to refuel other starships, gradually boosting their orbit until you have a fully-fueled ship in a highly elliptical earth orbit, and then you can finally blast off to Mars. For the moon it’s even worse, you need maybe 20 refueling flights to land 1 starship on the moon with enough fuel to come back.
Agreed with you that the heat shield (and reusable upper stage in general) seems like it could easily just never work (or work but only with expensive refurbishment, or only from returning from LEO orbits not anything higher-energy, or etc), perhaps forcing them to give up and have Starship become essentially a big scaled-up Falcon 9. This would still be cheaper per-kg than Falcon 9 (economies of scale, and the Raptor engines are better than Merlin, etc), but not as transformative. I think many people are just kind of assuming “eh, SpaceX is full of geniuses, they’ve done so many astounding things, they’ll figure out the heat shield”, but this is an infamously hard problem (see Shuttle, Orion, X-33...), so possibly they’ll fail!
Some other tidbits:
Raptor’s claimed vacuum ISP is 380; I don’t think they’re just, like, making this up (they have done lots of tests, flown it many times, etc—it’s not a hypey future projection like “Starship will cost $4m per flight”), but I also don’t know where I’d go if I wanted to prove to myself that the number is legit (wikipedia just cites an Elon tweet...).
Apparently those Starlink mass simulators actually weigh about 2 tons each?? So flight 7, which carried 10 Starlink simulators, actually put 20 tons of payload in orbit.
The first reliable version of Starship will very likely fall short of its intended 100 ton goal (i mean… unless it takes them a really long time to make Starship reliable, lol). But they also plan to stretch the rocket, refine the engine, maybe someday make the whole thing wider, etc. So I expect that they’ll eventually hit 100 tons. (The first version of Falcon 9 could only lift 10.4 tons to LEO; the current version can lift 17.5 tons AND land the first stage on a drone ship for reuse!) But of course if you make the whole ship bigger, some of your launch costs are gonna go up too.
Personally I’m doubtful that they ever hit the crazy-ambitious $20/kg mark, which (per Thomas Kwa) would require not just a reusable upper stage (very hard!) but also hyper low-cost, airline-like turnaround on every part of the operation. But $200/kg (1 OOM cheaper from where Falcon 9 is today, using the rumored internal cost of $30m/launch and 17.5 ton capacity) seems pretty doable—upper stage reuse (even if somewhat ardurous to refurbish) probably cuts your costs by like 4x, and the much greater physical size of Starship might give you another almost 2x. Cheap materials (steel and methane vs aluminum and RP1) + economies of scale in Raptor manufacturing might take you the rest of the way.
The Isp of a closed cycle rocket engine with a given propellant mix is largely a function of its chamber pressure and expansion ratio, so one can use a program like RPA to plug in known numbers and see what other claims are consistent with an Isp of 380. Example (for SL variant) in this tweet.
My guess is that 380 is achievable if they close the throat and use a large enough nozzle, but they’ll opt for slightly lower in order to cram 9 engines into the upper stage. With Starship staging at record low velocities, reducing gravity losses through higher thrust might matter more than a 1% efficiency gain.
Yeah good point re: heat shields & upper stage reuse being hard. They’ve experimented with reusing Dragon heat shields but still they mostly just replace them each time.
Thinking aloud… suppose they never solve Starship reuse, but the Booster basically works great (as it seems to already be working pretty great). So the booster gets reused like 100 times but the Starship has to be scrapped each launch.
...In this world, the Starship could maybe be cheaper due to using old, worn-out engines retired from the boosters (don’t need to be super reliable since it can still complete the mission if a couple blow up, none of them are coming back anyway) and not needing a heat shield or wingflaps. Apparently right now the majority of the cost is the engines, the hull itself is cheap. And again a full stack booster+starship costs $100M. So the starship (having far fewer engines) is probably like… $25M or so right now? And price could drop further due to the aforementioned effects plus simply normal returns to scale… let me see how much does a car cost per kg and how much does starship weigh? Apparently starship dry mass is 85,000 kg. Tesla model 3 is 1600kg and costs like $36,000. So $22/kg. So if they can produce starships for the same cost per kg as model 3 cars, then the cost would be less than $2M. So yeah, obviously they won’t get there immediately but that’s the price they would naturally trend towards.
So that means even if they don’t reuse the upper stage at all, if they use old engines they would have retired anyway then they could get the price down to maybe $30/kg eventually. Assuming they have 150 tons to orbit payload of course, which is unproven. If they use new engines, maybe they can only get to $100/kg.
Also, as a fun bonus, if they are building a space station in LEO, then they can use the hulls of the single-use starships as material for the space station! That’s a free 85,000 kg of steel in LEO with every launch, in addition to the payload! And it already comes formed into a spacious pressure vessel!
Than in Berkeley, you mean.