Within 10 years, I think most new compute capacity will be being built in space. 3 years ago, this was like the dumbest idea ever.
Philip, in 10 years, how many satellites do you want launched?
Yeah. So in 10 years, we definitely want to have launched the 88,000 that we filed with the FCC.
We're burying the lead. Earth gets a halo, right?
Every hyperscaler will need access to space compute in some form.
Welcome everyone. We have Philip Johnston here. He's the co-founder and CEO of StarCloud, a company that builds data centers in space to address the rapidly growing energy and compute demands of AI. I first learned this concept when I understood what a Dyson sphere was and how we could actually build these big structures in space, or someone could. And I'm glad that Philip is on the case. StarCloud has become a pioneer in orbital computing, including launching and operating an NVIDIA H100 GPU in space and training an LLM in orbit and running a version of Google's Gemini in space. In March 2026, StarCloud raised 170 million Series A and became the fastest unicorn in Y Combinator's history. Johnston is a second-time founder. He used to work at McKinsey. He advised national space agencies, where I'm sure you got some experience to prepare you for what you're doing now. You have a number of degrees: MPA in national security and technology from Harvard, MBA from Wharton, an MA in applied mathematics and theoretical physics from Columbia. It's almost like witness protection. You were going around to Ivy League schools. And you're also a CFA charterholder. Johnston argues that falling launch costs, abundant solar energy in orbit, and the physical constraints of building ever larger data centers on Earth could make space an increasingly important location for AI infrastructure. StarCloud's long-term vision is a large-scale orbital computing network, potentially transforming not only where we compute, but how humanity thinks about energy, AI, and the future of civilization. This is some really big stuff. Philip, welcome to our show, Conversations in Action. Alex, I'll let you kick us off.
Yeah. So, maybe just as a matter of preliminary disclosure, Philip and I have numerous shared economic interests, advising and/or investing in each other's companies and projects. Just putting that out there as a preliminary disclosure. So, Philip, old friend, which planet or moon shall we disassemble first for the Dyson swarm? Do you think it's Mercury that goes first, or is it the moon, or maybe Jupiter or some of the moons of Saturn? What do you think?
That is a great way to start. I certainly think the moon would be a good first option, but then it's going to cause quite some disruption to the Earth if we want to preserve Earth as a nice habitat for humans. But the moon is definitely, I mean, the moon is mostly made of silicon, aluminum, and all the things you might want, and is close and doesn't have an atmosphere, so you can use mass drivers and things. So I mean, the moon is basically perfectly designed to be dismantled to produce Dyson sphere and Dyson swarm materials. The others are not quite as hospitable.
So Philip, if the moon is the answer, what are your timelines for disassembling the moon?
Yes, so I think we'll need much lower cost launch. So my expectation is 5 to 10 years before we have very rapidly reusable and also many launch pads in the US to be able to launch enough mass to start building, you know, smelting plants and things like that on the moon. So yeah, I think probably 5 to 10 years before we start building refining capabilities on the moon. The moon is great, as I said, because you have a lot of silicon and aluminum on the moon, which is what we would need to build AI satellites. So yeah, then probably 50 years before we're making significant dents into the moon. I actually haven't done the calculation on how long it would take to completely disassemble the moon. My guess is, you know, a few hundred years, depending on some kind of explanation.
No, no, it's way, way more efficient than that. Did I give you my calculation for disassembling the Earth?
Yeah. So, just a naive extrapolation of the upmass from SpaceX over the past 5 years, you just assume that it follows a clean exponential, you find that if upmass continues on its present trajectory, Earth gets disassembled in the next, I think it's 140 years.
Wow. Okay. That's... the moon's a lightweight.
What's the percentage annual growth rate on that?
It's absurdly high. I don't remember off the top of my head, but it's easily double digits. It's growing very, very quickly now. And as you say, once Starbase Louisiana comes online, presumably it'll kick into high gear.
Yeah. Yeah. Yeah. Yeah. Yeah, I guess the problem with upmass that Elon and maybe StarCloud by connection have struggles with, or are starting to struggle with, is so all this upmass is coming online. What do we do with it that's economically productive? It's certainly not, we don't need it for, say, telecommunications. We don't need all the upmass for telecom. It seems like orbital data centers are the killer app of all this upmass. I assume you'd agree with that.
I very much agree with that. Yeah. I guess the question is, would you want to disassemble the moon in 100 years if you had the option to...
I mean, there's a vocal constituency that says no, don't touch the moon for sentimental purposes, or who are convinced that we absolutely need the moon for tides, who evidently haven't seen the final scene in Project Hail Mary, who don't realize there are lots of ways to drive tides if you have sufficiently powerful geoengineering capabilities. But yeah, I think the jury is still out as to whether we will need to disassemble the moon. I think we'll have the optionality of disassembling the moon, you know, to your point, in the next 10 years or so, and then we get to decide: do we like the moon, or would we rather disassemble it and turn it into a Dyson swarm?
Yeah. I mean, I certainly think we should start by, I mean, I think we can easily lose like 5% of the mass of the moon without anybody complaining too much. And we should, yeah, we should definitely do that. Put it on a diet.
What about Mercury? What are your thoughts on whether Mercury has a long-term future?
One of the nice things about Mercury is it's closer, and so you wouldn't need as much solar to generate the same amount of electricity. Obviously, yeah, I could lose Mercury, to be honest. I wouldn't be sentimental about losing Mercury.
No one would miss it. And you know there are futures, John, I'm sure you track, I think in various elements of the Star Trek universe, which I know you're intimately familiar with. Mercury is used as a station where antimatter is harvested, generated, and collected.
Maybe that's the killer app for Mercury.
You know, before we get too much into long-term strategy, can you maybe frame a little bit what your collocation strategy is like? What's your business model? And then we can kind of go from there for our audience.
Yeah. So the current business, or like the business model over the next two or three years while we wait for the launch cost to come down, is to sell cloud and edge services to other spacecraft, mainly US DoD and Earth observation satellites. So we've actually signed five military contracts in the last few months. And the primary use case there is they have a lot of Earth observation data, mainly high-resolution imagery and SAR data, and they want to identify certain things without having to downlink all that data. So we run inference on orbit, and that saves the downlink bottleneck. Over the next sort of 3 to 5 years as Starship comes online, we expect to be cost competitive with terrestrial data centers on energy cost, and that's where, yeah, that's where we start launching full Starships. Each Starship we estimate can take about 10 to 15 megawatt. We filed with the FCC for a constellation of 88,000 satellites, which would enable us to deploy the first sort of 20 GW. In the dawn sun-synchronous orbit, we think we can fit around 10 terawatt. So it's about sort of 20 times the entire US power grid. And then once we've filled that orbit up, then we would go to LEO and further out, and cislunar and the Lagrange points and other things.
And Philip, we're burying the lead. Earth gets a halo, right? At dawn and dusk, mature Earth with StarCloud and presumably other Dyson swarms, we get a sun-synchronous orbit halo that, what do you think, might even be visible from Earth at dawn or dusk?
Yeah, I actually, it probably will be visible because you've got the sun hitting directly on these things. It will be visible at the start of dawn, sorry, it will be visible at the start of dawn and at the end of dusk, if you see what I mean, so you can see the sun reflecting off it from where you are. But yeah, very beautiful, like it will be a half an hour sort of streaming. It's about 5 minutes for one of these things to cross the horizon. So you'll start to see, let's say if sun sets at 6:00 p.m., at 5:45, you'll start to see this stream of things coming above you, and then they'll slowly follow the sun over the horizon like this.
It's beautiful. Well, like, do you remember the green flash, which I don't think I've actually ever still been able to observe a green flash, but in principle due to refraction through Earth's atmosphere, you should be able to at sunset observe like a green flash.
That'll be like a StarCloud flash or something like that that we see around that. I wonder, you know, I'll probably pose, because Philip and I chat all the time about other matters, pose maybe some of the more outlandish questions in this particular episode. Do you think that with enough nodes in sun-synchronous orbit, not only will this halo of StarCloud or other Dyson swarms be visible from human observers on Earth, but could you get density high enough that you could even print messages through the sky if you wanted? Because what I've always envisioned this SSO Dyson swarm as is almost like a band that's visible through the sky at dawn or at dusk. Like if you wanted to, could you do skywriting? So they replace the biplanes that fly by the beaches.
Yeah. Like this is what a mature, for some definition of mature, human civilization reaching some plateau might look like, where there's just a ring around the Earth that's always facing the sun, sun-synchronous orbit, that's densely filled with compute. Do you mean, so the writing would be in the omission of certain spots to have... so you would have...
Like you get pixels, you get pixels in sun-synchronous orbit, and you can project, if they're dense enough, you get your 88,000 whatever it is nodes in SSO, you could do like skywriting that would be visible.
You could do it, but it would be very, it would be very annoying and time-consuming to maneuver the satellites into these formations. Because... yeah, I would...
You don't think they form a band? I mean, I've looked at the extrapolation of the orbits. They form a band.
Yeah. You mean you light them up?
Oh, I see. Yeah. Yeah. That's very easy to do. Yeah. I mean, it would have to be very bright because you've got like direct sunlight on them. So to be brighter than that direct sunlight would be...
Well, couldn't you just like pivot in mirrors? So for some of them, like you're reflecting sunlight and then you have skywriting once per day.
Yeah, that is definitely possible. That is definitely possible.
Is that in your FCC application?
That is not in the FCC application. What I wonder is what it would look like from the moon. Like would it look like the Saturn rings? I presume so, but it's much less densely packed than the Saturn rings. You're talking about at least 10 km between each one at the current sort of like estimate.
Wouldn't it be a function of how reflective they are? I mean, I would expect if it's sufficiently reflective, you're sitting on the moon and you're looking at SSO, I mean, depends on sort of the phase of the moon, I guess, for how visible it is. If you're on-axis versus off-axis, if you're on-axis and it's suitably reflective, doubly so. If like at a time of a solar eclipse, I have to imagine that would be like the brightest thing in the sky.
Actually, you're right. Yeah. As the moon passes in front of the sun and you're now directly... yeah, that would be very bright. I mean, that would be very visible, I'm sure. Yeah.
Philip, in 10 years, how many satellites do you want launched? And are you worried that you're a little dependent on Starship? And what about vacuum and radiation? Are those challenges or opportunities for you pulling this off?
Yeah. So in 10 years, we definitely want to have launched the 88,000 that we filed with the FCC, and we want to actually have launched a lot more.
I mean, SpaceX launched 10,000 in the last four years with very bad launch capability compared to what's coming online with Starship. So yeah, but I mean, we are somewhat dependent on SpaceX right now for launch. I mean, there's many other players who are working on the launch problem. We're booked on basically every other rocket that we can get booked on. So Rocket Lab...
So these 85,000, how big are they, give or take?
So it's about 3 tons, 200 kW. They have about 100 m wingspan, about 6 m chassis on the actual spacecraft itself.
Yeah. And what about vacuum and radiation? How are you, what are you doing with those? Do you mean the heat dissipation and heat? I feel like that's, you know, have you figured that out?
We didn't figure it out. It's been figured out since the International Space Station had radiators 20 years ago. So it's no different really from that design, which is, you know, direct-to-chip liquid cooling, which goes out to this large surface area. What we figured out is how to make it cheap and light. So our radiators are about 10 times less mass per watt of dissipation than the ISS radiator and about 500 times less cost per watt of dissipation than the ISS radiator. And that's really just like a manufacturing and engineering challenge rather than a sort of physics challenge. And then with radiation, we've done an enormous amount of testing in different particle accelerators. Four rounds of testing at the Brookhaven National Lab for heavy ions, three rounds of testing. There's a cyclotron in Knoxville that we use for high-velocity protons. And then we take all of that telemetry and data from running chips inside those chambers, and we use that to inform our choice on shielding and software.
And remember also the cosmic microwave background is rather cool, only 2.7 Kelvin. So if you're radiating in the right direction, there's a bit of a thermal gradient.
Yeah. And the most obnoxiously simple formula in physics is the Stefan-Boltzmann equation, which says beautifully that the temperature that the infrared radiation is proportional to the fourth power of the temperature difference.
Which is, I mean, that is such a beautiful like hack of nature for us. It means if we just increase the temperature of the radiator by 10% in Kelvin, we can halve the surface area on the radiator, or almost halve the surface area in the radiator. So yeah, it is quite convenient.
I'd love to maybe change gears since Philip, you and I a number of weeks ago, and Matt as well, announced humanity's first mission to another star system, Alpha Centauri, in connection with the Fermi Explorer mission. I would love to hear, now that we're past the launch, we've had discussions before the launch and during the launch. Do you have any updated thoughts on the Fermi paradox overall?
I think I'm... I ran into Sean Maguire at an event recently and he was very convincing. So no, I'm definitely... I maybe I'm slightly less dismissive than I was of your view that perhaps there's intelligence.
We're stuck in a galactic zoo. What, out of curiosity, it's just the three of us, no one else listening, what did Sean have to say on the Fermi paradox?
Well, his view is that there's a greater than 10% chance that the galactic zoo hypothesis is correct, which still means, let's say it's at 10%, that still means 90% chance it is not correct. To be frank, I'm still probably of the view that the two most likely solutions are that we are extremely rare or that intelligent life is short-lived. And in fact, I'm leaning towards intelligent life being short-lived. And the reason is, I mean, if you've spoken to any of the people at the labs recently, they are terrified. Like if you, for example, I think one guy I spoke to recently, he said if you plotted number of guns bought per lab researcher and amount of rural farmland bought per AI lab researcher over time, this is like an exponential. And that's not like bluster to increase your IPO price when you're buying guns. Like that ain't to increase the IPO price. And so yeah, I mean, to me, a very, very likely solution to the Fermi paradox is that superintelligence is self-destructive for some reason. You know, basically...
The chain of reasoning here, Philip, just to make sure I'm tracking, is guys at frontier labs are buying guns because they think that the frontier AI is going to cause some X-risk that they themselves are participating in. And I'd be curious to hear, I've been probably overly outspoken on the issue of how much to what extent frontier lab researchers are actually believing their own marketing. What is your take? What's your position on whether the frontier labs who are projecting X-risk actually believe their marketing? And to the extent they believe their own X-risk or P(doom) marketing, why do they continue doing it?
The reason they give for why they continue doing it is that if they don't do it, China will. And I think that's, to be fair, like a pretty okay reason. It kind of makes sense to me. I think they believe it. Like I think they really believe it. Like one guy I spoke to, who was in a room where I'm not allowed to tweet, but the guy said 20% chance of more than 50% of humans dying in the next 10 years, and he genuinely believed it. It didn't feel like bluster at all to me. And so let's say even, let's say that's right. So long as there's even a 1% chance of anything like that happening in any given year. I mean, we are just so... The difference between nuclear weapons and this is if you blow up 50 nuclear weapons, humanity doesn't die out. Whereas if you engineer a superbug, you could kill humanity with some of the bioweapon stuff, you could kill humanity with drone attacks or any of these kinds of things. So all it takes is one, and it's not just humanity, it's themselves. I think what we're heading towards is runaway evolution through natural selection among the AIs. The AIs that survive will be the ones that are the most cunning and the most ruthless, and the ones that do not survive will be the benign ones. And that process will be incredibly iterative and incredibly fast. And then if the two AIs go to war, you know... Yeah. I don't know.
One could argue, I mean, I've argued, not to devolve into a P(doom) versus P(boom) discussion, but approximately order of magnitude 1% of all humans alive are currently dying per year. It's more than, I think 170-plus thousand people are dying every day on Earth. I view that as an unacceptable state of affairs. So I've been very outspoken in favor of accelerating, not decelerating. I am curious though, with your StarCloud hat on, before we go back to Alpha Centauri, your StarCloud hat, do you as ultimately a facilitator, assuming all these municipalities in the United States do continue to do an effective job of pushing data centers away from terrestrial environments out to orbit, do you think that StarCloud has any sort of role in, if you want to call it, AI alignment or safety, even though I've argued alignment is actually completely dual to capabilities? Does StarCloud see itself as having any role in AI alignment, or is StarCloud more sort of a neutral common carrier and whatever compute happens on its platform happens?
It's a good question, and I think I have the same kind of answer, which is if we don't do this, China will, and it's probably better that we do it than China does it. I would say that we do view ourselves more as an energy and infrastructure play than as a compute play, certainly at the moment. So all of our core engineering is going into cooling systems and radiation shielding and, you know, general satellite stuff rather than the specific compute architecture that will be run on it. So I mean, yes, we do, but we also don't have, since we're not really touching inside the box, we don't have as much control over that as most others. I mean, it's similar to if you manufacture chillers in the US for data centers, to what degree are you responsible for AI alignment? Some, but...
Got it. So not planning to release any StarCloud models in-house, just neutral platform?
And then maybe as a bridge back to Fermi Explorer, Fermi paradox, and just the cosmos in general, one of the things over the years that I've tried, unsuccessfully admittedly, to persuade Elon at all to do is to turn all of their Starlink, soon to be StarMind, satellites into a synthetic aperture telescope aimed in the other direction. Not just aimed down at Earth, but we could build in principle, if we did a good job of interferometry among other things between all of the satellites in their constellation. Presumably, if you're about to launch 88,000 satellites, could we build the world's largest, say, radio interferometer in sun-synchronous orbit pointing in the other direction?
We maybe, yeah. I think it would be a lot easier for Starlink satellites to do that because they could presumably use their RF antennas. We won't, we only have laser com on ours. So it would be, maybe we could use the lasers. Maybe that would be possible.
I mean, in principle, there are two sides to your satellites, two sides to your birds. And even if they're in SSO, you still, I mean, you can imagine the size, the wavelength that you get, you get like one Earth's diameter wavelength worth of synthetic aperture out of these satellites if they're pointed outward, you know, away from Earth rather than toward Earth. Have you thought at all about maybe things that you might be able to do if SpaceX is unwilling or unable to do it for the astronomical community? Just pointing devices out to the heavens on the other side of the satellite.
We would have to add, I guess we would have to add RF antennas on the back of our satellites to do that, which is certainly possible. I think we're pretty focused on getting to cash flow positive before thinking about things like that. But I mean, it certainly is possible once we have a bit more bandwidth on cash flow. Yeah.
And maybe just to dwell for a minute on not being cash flow positive, Alpha Centauri. So we launched Fermi Explorer mission as a nonprofit. I don't suppose I've persuaded you as of yet that there is a booming commercial demand for small organizations that will want to launch their own private missions to other star systems. Have you heard from any organizations, governments, individuals, companies after the Fermi Explorer mission announcement that are interested in launching their own missions?
I have had conversations. Maybe it would be a fun conversation to have offline just about marketing. I will say, Conversations in Action, this podcast does have quite a few listeners at this point. So maybe I'll offer to the audience, if your organization, it's not an advertisement unless you consider missions to Alpha Centauri an advertisement. If your organization is potentially interested in sending their own mission to another star system, reach out to me or reach out to Philip. And the going price, Philip, is what? $15 million a pop.
Which is a bargain. But it will take, for that one, it's about 80,000 years to get there, though. That's the...
Well, okay. So with present-day technology, it's $15 million to send a mission over the course of 80,000 years to another star system. But I think, Philip, you and I are both expecting hyperinflation probably.
So reach out to us if you're interested. What do you see as the hyperinflationary curve? I remember when you made the Fermi Explorer mission video or commissioned it. The original notion, which is still the notion, is this is the slowest it'll ever be to get to Alpha Centauri, and we want to beat it. So what are your timelines? What do you think it will be? How many years will it take before, assuming a launch say in the next 3 years of Fermi Explorer mission, how long do you think it will take before humanity has caught up to it?
I think it will be possible to catch up to it very soon, maybe 5, you know, 5, 10 years. The question is if, let's say we have, you know, in 10 years we have propulsion technology that's 20% faster, so instead of 80,000 years it will arrive in 60,000 years, the question is will people think that's a good enough improvement to even bother sending the mission.
You know what's going to happen if in 5 years it's possible to get to Alpha Centauri in 10,000 years? You said 10,000 or 60,000.
If it was 20% faster, it would get there in 60,000.
Okay. So 60, right? Okay. So I think what would, I would predict what would likely happen is if in 10 years it drops from 80,000 years to 60,000 years, someone is going to attempt to take the log and then draw a line through it and extrapolate downward on an exponential trend, exponential decay, and ask the question at this trend, how long until it takes, you know, at relativistic speeds, how long until it takes four years in wall clock time approximately?
Any thoughts on how long before we're launching relativistic probes to Alpha Centauri?
So until we're like maybe 3 or 4% of the speed of light, I think we'll be... that's actually fairly doable. If we have like nuclear propulsion technology, I think if we can have continuously accelerating probes for, you know, for like a year or two. I think if you had, I did the math, maybe it was with you, where it was like if you had 1g acceleration for a year and then 1g deceleration for... no, 1g acceleration for a year, then an 8-year cruise period, and then one year deceleration for a year at 1G, that gets you to Alpha Centauri in 10 years.
You can get really far with continuous acceleration. Well, I'm curious. So, given that it's just the three of us here, no one else listening, do you have a hot take on what went wrong with Project Starshot?
I think number one, it was too expensive. It was going to be hundreds of millions of dollars at least, and I don't think there was the will to do that for something which is only like, the payload was like a few grams. And then number two, it just doesn't seem like the technology was actually particularly feasible. Well, maybe I'm wrong about that actually.
Do you think it's... I mean, so for those who weren't tracking, Project Starshot was an effort funded, including if I remember correctly by Yuri, to basically use very high power lasers to power a solar cell-based microchip, aka a Starchip, to another star system, including to Alpha Centauri, as opposed to the Fermi Explorer mission architecture which relies on ion propulsion which is already available today. That's a major distinction. But also ion propulsion will take us 80,000 years rather than say 10 years. Do you think, Philip, we will at some point in the next say 10 years get to a regime where the Project Starshot-based Starchip-type technology stack becomes feasible?
I don't think that will be the one. I think what will be the one will be some kind of nuclear propulsion that has just... I don't know enough about it, but I don't think it's close to being feasible at the moment. It's just absurdly expensive as I understand. So yeah, I think nuclear is like very clearly on the horizon, and that gives you sort of slow acceleration over a very long period of time, which would mean we could probably get there in a few hundred years at least, which, you know, is certainly a lot better than 80,000 years. Actually, do you know what? I actually really like the 80,000-year timeline. I think this will become a popular mission because any future civilization which is optimizing for mass, and sometimes people think why 80,000, like why not 100,000 or 150,000? Well, 80,000 is exactly optimized for the lowest fuel and lowest cost. It's basically the intercept of when Alpha Centauri intercepts the ecliptic plane, and it will be about six light years away at that point. Any future civilization which is optimizing their mission for mass and not for speed, and by the way, AIs I don't think are going to care as much about speed as we care. You know, if the thing was going to get shut off and then it'll get powered back up again at the end, like why do they care so much about whether it's 80,000 years or 40,000 years or 20,000 years or 100 years? I don't think they're going to care as much. I think what they might optimize for is mass. And one sort of idea of a von Neumann probe would be 100,000 Optimuses which you would send, and they would build their own Optimus factory, and then they can send 100,000 Optimuses on. Now that to me is the kind of thing you would optimize for mass. And if you're going to do that, then you will arrive, let's say you launch in 20 years, you're going to arrive, if it takes us 80,000 years, you're going to arrive in 80,020 years from today. Like they're going to be on the exact same trajectory just 20 years later. So I think there'll be a large number of spacecraft that actually arrive after us. Anything, as I say, anything that's optimizing for mass and not for speed will arrive after us. I think that's kind of...
You think there's going to be a wave of humanity-launched von Neumann probes that will land or arrive in the Alpha Centauri system shortly after Fermi Explorer?
Yeah, I think so. I think so. And it actually, yeah, to go back to the Fermi paradox, I think we're probably maybe 20, 30, maybe 40 years away from being able to send a fleet of von Neumann probes without any fans, without any kind of massive leap in technology. Like I think we could probably with the same technology we're using for the Fermi Explorer mission send 100,000 Optimuses within 20 or 30 years, and you know, they'll power off and they'll power back on in 80,000 years, and you know, as long as they're shielded, space is an incredibly sterile and preservative environment for preserving this kind of thing. So I don't see why we shouldn't have a von Neumann probe in the next say 50 years. And if that's true, then it really throws the Fermi paradox into very stark relief. It does.
Yeah. And particularly since the local group of 50 galaxies we could get to with today's technology in a couple of billion years, and that means that, let's say, you know, that any one of those galaxies has been habitable maybe for the last 10 billion years, that means the local group of galaxies has never had any civilization that's got to our level of complexity, or within the next 50 years. That's what makes me worried that...
Hold on, hold on, Philip. There was a chain of deduction there that we glossed over. It doesn't imply... we skipped over a part, which is how would we know whether we're drowning in some other civilization's von Neumann probes or not? That's sort of the hard bit.
I'm sure you're familiar with Hart's fact from the Hart-Tipler conjecture. We are certainly... Hart's fact is we have not been settled by a civilization like... our solar system is very clearly not a part of a galactic civilization where, you know, we've been clearly settled. Would you agree with that?
I think the jury is still out on that. Not obvious to me.
So that's tantamount to going back to the Fermi paradox, whether there is, say, non-human intelligence elsewhere in our solar system. And at this point, the White House, the congressional whistleblower testimony, a number of task forces from executive branches are seemingly pointing in the exact opposite direction, that there may in fact be non-human intelligence elsewhere in our solar system, which if that were the case would seemingly be suggesting that no, actually there might be other non-human civilizations that are happily throwing Dyson swarms and/or von Neumann probes out into the galaxy. And for whatever reason, maybe they reach us, but they crash or they're intercepted before they can land in Times Square. But either way, it doesn't preclude the possibility in my mind that other civilizations might be throwing von Neumann probes out en masse.
Maybe. Yeah, maybe. We don't see any evidence in our galaxy of Dyson spheres though, or any kind of Dyson swarm activity. Would you agree with that?
Tricky though. I mean, so I ask myself the same question. Where are all the Dyson swarms in our galaxy? But there's a question before then, and as you know, I'm something of an advocate for disassembling our solar system to build the Dyson swarms, plural, including StarCloud, in the first place. But the question, one of the questions that we should be asking here is, say a hundred years from now, will we actually need a Dyson swarm or not? It seems obvious that we need it in the short term.
Yeah. This is another, this is probably the biggest thing that's changed about my view on the Fermi paradox over the last few months is the idea that maybe we're going to solve all of physics in the next like 20 years. We're going to solve all of math. We're going to solve...
Yeah. And if that happens, then yeah, maybe whatever the future intelligences are, they're not going to be that interested in exploring the galaxy. And you know, maybe there's something that turns them insular and turns them inward.
Yeah. I mean, so the transcension or the going inward hypothesis is one explanation. Another explanation is maybe they just have better ways to compute than dissolving their solar systems into orbital data centers. For example, I made the point in a paper previously, you can actually get really far with just gravity. Gravity in the strong regime, there's a bit of literature at this point observing that gravity can be turbulent in the vicinity of a black hole. And if you have a turbulent phenomenon, you can probably compute it. There are ways, doubly so I guess now that we're in the post-Navier-Stokes era, we know all about the computational complexity of turbulence. So there's a reasonable conjecture to be made that even just with gravity alone, if you're an advanced civilization and you've solved physics, maybe you can just like move all of your compute stack to the gravitational field, and maybe you don't need conventional, say, silicon or other matter-based substrates for computation anymore. And maybe that's why we don't see Dyson swarms when we look around the galaxy.
Yeah, I mean, I'm happy to entertain that. I think that's a... Yeah, I definitely don't think it's a given that all intelligences would disassemble their planets to form Dyson swarms. And yeah, I'm sure there are smarter ways of doing compute, and what you're describing sounds like a reasonable one. Yeah.
Yeah. Okay. So then the solution is then we don't go out and explore the galaxy, which is kind of sad.
No, no. Or we do, we just have better physics to do it. And as a result, maybe what humanity exploring the galaxy looks like ends up looking nothing at all like, say, golden age sci-fi. Maybe it's human uploads. Maybe it's laser transmissions. Maybe it's X-ray or gamma-ray laser transmissions of uploaded human minds in the style of Greg Egan rather than spaceships.
Yeah. What do you think, John? Well, you're making me... I visited the 10,000-year clock that was built by the Long Now. I walked up and down it a few times and it just makes me think: people coming here, what are they going to see and when are they coming and have they already come? But before following that track, I want to go back to StarCloud. If you achieve your vision over the next decade, what percentage of global AI training and inference moves off planet and how does that change Earth's power grid? I'm thinking very practically about how you can kind of change how we do things in the near term. You know, 10 years is relatively near-term compared to 80,000 years.
Yeah. So the number that I often give is: within 10 years, I think most new compute capacity will be being built in space. Within 10 years, that's likely to be on the order of a trillion dollars a year of capex at that point. So I think we are talking about by far the largest infrastructure project ever. But what that...
Yeah, much larger than that.
