Starcloud is one of the most interesting companies I’ve ever come across. A defense prime quoted them $100M to build their first data center in space. They turned around and built it themselves for $2.5M.
31 months after founding, the company’s worth $2.3 billion.
I sat down with Co-founder and CEO Philip Johnston for two hours, and he walked me through the physics behind operating a data center in space, why space will be the cheapest place to run AI compute as launch costs continue declining, why Nvidia invested in Starcloud even after one of their GPUs died, and navigating the dynamic of competing with SpaceX while relying on their rockets to get up there.
We also get into the tranched Series A that came together after 40 VC’s already said no, what YC is really funding, and his $15M probe to Alpha Centauri.
Support this Episode’s Sponsors
Monaco: The revenue engine for startups.
Flex: Sign-up for Flex Elite with code TURNER, get $1,000 here.
Clerk: Everything you need for authentication.
Numeral: The end-to-end platform for sales tax and compliance.
Inquire about sponsoring future episodes here.
👉 Stream on YouTube, Spotify, and Apple
Timestamps to jump in:
0:00 Why space will be the cheapest place to run AI
1:34 Starcloud: Building data centers in space
4:41 The first H100 in orbit
11:37 Do things hit satellites in orbit?
12:35 Orbiting on the line between night and day
15:57 Why not run on nuclear for 50 years?
17:54 Why the cooling objection doesn't hold up
23:10 The $100M quote Starcloud built for $2.5M
26:15 Downlink the insight, not the raw data
27:51 Hardening an off-the-shelf H100 for space
29:50 The satellite that rebooted every 90 minutes
32:47 Why Nvidia invested even though their GPU died
36:15 Riding SpaceX rockets while competing with them
37:12 From beaming solar power to the Equinix of space compute
39:53 Without Starship: billions, not hundreds of billions
41:15 Why data centers end up in space within 20 years
46:40 Disassembling Mercury to build a Dyson swarm
50:27 From McKinsey to Starbase
53:02 Opontia, the Thrasio of the Middle East
57:06 Hiring: half the team is ex-SpaceX
58:53 YC funds the engineers, not the idea
1:02:20 Inside Starcloud's tranched Series A after 40 VC's said no
1:07:57 Fermi Explorer: a $15M probe to Alpha Centauri
1:11:18 Is there other intelligent life out there?
1:13:11 VC's use AI more than founders realize
1:16:50 Keep the talent bar high and capital will come
1:19:13 Lessons from Elon, PG, Dalton and Michael
Find Philip on X / Twitter and LinkedIn. Check out Starcloud.
Referenced
👉 Stream on YouTube, Spotify, and Apple
Transcript
Find transcripts of all prior episodes here.
Turner Novak:
Philip, welcome to the show.
Philip Johnston:
Thank you for having me.
Turner Novak:
I'm excited to do this. So you said that space is the cheapest place to run AI. Kind of a bold statement. So I wanna talk about how that actually is true.
Philip Johnston:
Let me clarify. Space will be the cheapest place to run AI.
Turner Novak:
Will be the cheapest, okay. That's true.
Philip Johnston:
Right now it is definitely not the cheapest place to run AI.
Turner Novak:
So why is it not the cheapest right now?
Philip Johnston:
So it is entirely contingent on the launch cost, and right now the launch cost with Falcon 9 is about $5,000 a kilo. For it to be the cheapest place, it would need to be under $500 a kilo. That is only possible with a launch vehicle like Starship or some of the newer vehicles that are coming online.
Turner Novak:
So then why are you building this company right now? Because if the cost doesn't come down, it doesn't work.
Philip Johnston:
Yeah. As Wayne Gretzky used to say, and you're Canadian, so...
Turner Novak:
Yeah. Famous Canadian quote.
Philip Johnston:
Yeah, skate to where the puck's going, right?
Turner Novak:
Yeah.
Philip Johnston:
And so we can very clearly see a path to $500 a kilo. There's also a business model that does make sense right now, which is providing cloud and edge services for other spacecraft, particularly US government and military satellites. So that's the business we're working on now.
It enables us to develop all the technology, the radiator cooling technology, all of the shielding from radiation, etc. And then in maybe two or three years, once the launch cost hits the place where it breaks even, we're gonna be way further ahead than anybody else.
Turner Novak:
So can you just describe what Starcloud is for someone who doesn't know?
Philip Johnston:
Yeah. Starcloud is a two-and-a-half-year-old company based in Redmond, Washington, with a team of about 30 engineers, and we're building data centers in space. We're initially doing that to provide cloud and edge services for other spacecraft, military and government. On a sort of two to three year timeframe, as much lower cost launch vehicles ramp up, we'll be competing with all terrestrial data centers on energy costs.
Turner Novak:
And what's the chain of thinking where you get to the point that it makes sense to do? Because when I think about data centers in space, I'm like, this is probably the craziest possible combination of things that I could state. It's pretty out there. So then what's your logic?
Philip Johnston:
Submarine made of cheese is my go-to.
Turner Novak:
What's your go-to?
Philip Johnston:
Submarine made of cheese.
Turner Novak:
Submarine... Oh, okay. So walk me through all the steps of why this makes sense.
Philip Johnston:
Yeah. So I think the best comparison is with an energy project on the ground. The cheapest energy we have is solar. So compare building a new solar project terrestrially to power data centers versus building a solar project in space for data centers. That's probably the best comparison you could draw.
So terrestrially you have three big costs to do that. Number one is the cost of permitted land, and actually in North America, that can be the largest. Not too many people will allow you to cover 10 square kilometers of countryside in the US with solar panels.
Turner Novak:
So it gets hard to just build it in the first place.
Philip Johnston:
Yeah. Permitted land for a new solar project is becoming harder to find. And it's actually linked to the energy price, obviously. The cost of that land is based on the discounted cash flow you're gonna produce from it, and the energy price is going up, so that means the land price is going up.
Number two is the cost of battery storage and backup power. You're only in peak sunlight eight hours a day, and that means you need to store power so you can use it at night for a data center, for example. And then the last one is the cost of the solar cells themselves.
Now compare that to space. Number one, we don't need permitted land, so your biggest cost is gone. You don't need batteries and backup power because you're 24/7 in the sun, so your second biggest cost is gone. And then you need eight times fewer solar cells, because one square meter of solar panel in space produces about eight times the energy over the course of a year compared to one terrestrially.
So the main additional cost that we have in space versus terrestrially is the launch cost. You can see there's definitely gonna be a break-even point where the launch cost comes below the cost of permitted land, batteries and eight times the solar. When you run the numbers, we see that at about $500 per kilo. So it's actually a complete no-brainer to build almost all compute in space once the launch cost hits that price.
I think the other thing to note is the lead time on permitted land for new energy projects for data centers. It can take five years, 10 years sometimes to get permitting for some of these projects. In space, the chips can come out of the factory, they can be loaded onto a rocket, and a week later be up and running.
Turner Novak:
So what is actually the thing that is up there in space? 'Cause I'm trying to picture it. When I think of it, is it like a data center building that's floating around, spinning, with solar panels on it and little jet engines adjusting the trajectory? So what is actually up there?
Philip Johnston:
Yeah. So I can tell you what we've launched, and then I'll tell you what we're going to launch in the end state. In November last year, we launched what we called Starcloud 1. That's a 60-kilogram small sat. It has five GPUs, three from Nvidia, two from Arm, and also some CPUs from AMD.
The most important one on there was the Nvidia H100. The reason it was important is it's about 100 times more powerful GPU compute than has been operated in space before. With that, we were the first to train a model in space. We were the first to run a version of Gemini in space. We were the first to do high-powered inference on satellite imagery and SAR and other data.
So that kind of proved out the case. Before that, people literally thought you couldn't run these power-dense GPUs in space, and we've proved that you can.
Then there's the constellation that we're building towards now. In January, we're launching Starcloud 2. It has 100 times the power generation of Starcloud 1. It also has by far the largest commercial deployable radiator in space. That's our core technology.
But the end state constellation, and we filed with the FCC for a constellation of 88,000 of these, will be a three ton, 200 kilowatt spacecraft. They're about six meters long, and they fit in the fairing of Starship.
Turner Novak:
So this is like a tube-type cylinder shape thing?
Philip Johnston:
It's actually a flat disc. If you've ever seen the Starlink satellites, it looks a bit like that. The reason it has to be flat is it has to slide out the side of Starship. It has a dispenser door.
Turner Novak:
You said like a PEZ dispenser?
Philip Johnston:
Have you seen the Starship deployment? You probably haven't.
Turner Novak:
No, I haven't seen that.
Philip Johnston:
Yeah, so at the bottom of Starship, there's this little slit. Normally, a rocket has a clamshell door that opens like this, and then a big payload comes out. That's quite expensive to engineer, and it's kind of annoying, and you don't really need it right now.
So the way Starship works right now is at the bottom they have this slit. The stack goes down and a disc slides out, and the stack goes down and a disc slides out, and the stack...
Turner Novak:
So all the payloads that were going up on that trip just shoot out kind of like a PEZ dispenser once it hits the correct altitude up above the Earth. Okay.
Philip Johnston:
Yeah, exactly. So that's the Starcloud 3 form factor as well, and it's about 200 kilowatts per spacecraft. The way you can think about that is a flat disc in the middle with a bunch of compute. It's about six meters by six meters square. And then 100 meter by 100 meter deployable solar panels and radiators.
Turner Novak:
So you said six in the middle and then 100 on each side?
Philip Johnston:
Yeah. Six meters. What's that in feet? 24 feet, and then 400 feet on each side with solar panels and radiators. And then on the corners, it has these optical terminals for communication.
It's a very simple satellite, actually. It's much more simple than the Starlink satellites or almost all communication satellites or Earth observation satellites. It's way simpler than all of them.
Turner Novak:
Why is it simpler?
Philip Johnston:
It's simpler than Starlink because the most complex piece on a Starlink satellite is the phased array antennas. These are enormous antennas to direct beams of radio frequency towards dishes and phones and things.
Turner Novak:
This is like beaming internet connectivity to the Earth, essentially.
Philip Johnston:
Exactly. Exactly.
Turner Novak:
And you guys don't need to do that?
Philip Johnston:
We don't need to do that. So we remove those RF antennas.
Turner Novak:
Yeah. So it's basically like a giant X shape that looks mostly like solar panels.
Philip Johnston:
Yeah. Most of it. It's a blob in the middle and then two big wings.
Turner Novak:
Oh, just only two.
Philip Johnston:
Yeah, exactly. Not four.
Turner Novak:
And is it in orbit around the Earth?
Philip Johnston:
Yes. So it's in low Earth orbit. Starcloud 1 is orbiting at around 550 kilometers altitude.
Turner Novak:
I've always wondered about this. Do you have to worry about traffic up there, like stuff hitting it? I feel like it's actually less of a deal than people think it is. But how do you make sure it's all safe and another satellite doesn't hit it, or a plane doesn't hit it? Is any of this stuff you have to think about, or not really?
Philip Johnston:
Yeah, no, you definitely do. Probably the best case study for this right now is the Starlink satellites. They have around 10,000 satellites in orbit, and every so often the satellites will have to do these collision avoidance maneuvers. It's where they slow themselves down a little bit or speed up a little bit.
Turner Novak:
To avoid the trajectory of someone else.
Philip Johnston:
Exactly. We're gonna have to do much less of that because of the dawn-dusk sun-synchronous orbit. Where the Starlink satellites fly, they cross each other like this. But we're actually just trailing ourselves over the Earth in this horizontal band. So it's much, much less prone to collisions in general.
Turner Novak:
So you don't need to worry about positioning in terms of constantly having a beam-type connection, like a Starlink connectivity satellite would?
Philip Johnston:
So we use lasers to connect to other satellites, and then one of them will connect to the ground. The whole constellation is internally networked. But we don't have RF antennas to connect to the ground.
Turner Novak:
And I think you guys orbit on the time between night and day.
Philip Johnston:
Yeah, they call it the terminator line.
Turner Novak:
Oh, is this what it's called? So why do you do that?
Philip Johnston:
I will show you with the help of a glass. So imagine this is the sun.
Turner Novak:
Yep.
Philip Johnston:
Just make sure you can see it on the screen.
Turner Novak:
Yep.
Philip Johnston:
This is the Earth, and the Earth's rotating on its axis. The Earth is going around the sun like this. If you orbit the Earth like this, half the time you're in front of the Earth, and half the time you're behind the Earth. And that's annoying because it means you have to have batteries.
So what you wanna do is actually orbit the Earth like this, flying over the poles. It turns out that means you're flying exactly over the shadow between day and night on the Earth.
Turner Novak:
So you make sure you're constantly flying on that line as it's changing around the Earth?
Philip Johnston:
Yeah. And so the problem you might think of is, okay, well, when you go around here, now you've got a problem. Unless your orbit was being kicked around, half the time you're now going over noon and midnight, basically, which is bad, 'cause then you have the same problem.
Turner Novak:
Yeah.
Philip Johnston:
So what you do is you fly seven degrees off this, actually. Because the Earth is slightly flat, like slightly oblate, there's this extremely nice property of nature. It means you get a one 365th kick of your orbit around, so that by the time you get around here, it's still staying exactly over the terminator line.
Turner Novak:
Okay. So you're going vertically over the Earth instead of horizontally around the equator.
Philip Johnston:
Exactly.
Turner Novak:
And you're kinda rotating the orbit as the Earth orbits around the sun, so you're constantly in that line.
Philip Johnston:
Exactly.
Turner Novak:
Okay. And so why is it so important to be on that line?
Philip Johnston:
It's very important so that you don't ever go behind the shadow of the Earth, and also so that you don't cast a shadow on the Earth. So you're always in the sun, basically. If you were not on that line, every other orbit would have at least some period where it's behind the shadow of the Earth, and that means you need batteries.
Turner Novak:
So you have no batteries on the thing?
Philip Johnston:
No batteries. The only batteries we have are to buffer from the solar panels to the chips, but you don't need like an hour of battery. You can have much less than that.
Turner Novak:
I don't even know if this is true, but if there's a solar eclipse, what happens? Do you just lose power?
Philip Johnston:
Every so often we will have... The solar eclipses that we see are when the moon passes in front of the sun. There will be periods where the moon is casting a shadow on our solar panels, and then we'll have two or three minutes of no power.
Turner Novak:
So do you need constant power coming to this thing for it to work properly?
Philip Johnston:
Yeah, you do. I mean, two or three minutes is fine. We'll probably have enough battery even for the two or three minutes. But even if you didn't wanna have battery for that, you can just distribute the workloads to the satellites that do have sun. The moon isn't big enough to cast a shadow on all of the satellites at once.
Turner Novak:
Why don't you just put batteries on it?
Philip Johnston:
Batteries are expensive and heavy.
Turner Novak:
And that just makes the launch costs more expensive?
Philip Johnston:
Exactly. It's not a deal breaker if we need to have five minutes of batteries, but you'd rather not have to.
Turner Novak:
And also, this importance of solar is just that solar is the lowest cost solution. So we can use nat gas or oil or nuclear in the short to medium term, but long term, solar is probably just everything. Everything's gonna be solar. And I don't know if that's gonna take 100 years or 10 or 1,000, but eventually it's all gonna get there.
But couldn't you just say, "All right, well, it's 50 years away. Let's just do nuclear for 50 years or something, or natural gas for 50 years"?
Philip Johnston:
Yeah, you could, but solar is by far the cheapest form of energy we have now. I think for a new fission plant you're looking at at least 6 cents per kilowatt hour. For a new solar project, depending where it is, you're looking at down to 1 cent per kilowatt hour.
And then in space, because we don't have to pay for land or batteries, we can get down to much less than half a cent per kilowatt hour, including the launch cost. So even if we were not gonna run out of oil, this would still be by far the cheapest way of running compute.
Turner Novak:
And you talked about putting a shadow on the ground. So how big does this thing get? Is it this massive thing where I look up and there's one of those alien spaceships from Independence Day, covering all directions, this orbiting data center? What does it look like eventually?
Philip Johnston:
No, it's not like that. You won't really be able to see them. When it's only 100 meters and the wingspan is only six meters, or the width of the wing is only six meters, you wouldn't actually see any shadow on the Earth.
If you wanted to train a large model, for example, you need the whole thing to be internally networked. And so you might wanna dock together a large structure. We have a concept video on our website of a four kilometer by four kilometer solar panel. So that would be-
Turner Novak:
That'd be visible from the ground, right?
Philip Johnston:
Yeah, that definitely would cast a shadow. But in the orbit we're flying, it actually would never cast a shadow.
Turner Novak:
So if I Google data centers in space, people have opinions. It works, it doesn't work. It feels like one of the pushbacks is related to this cooling concept. Can you explain what they're saying, like, "Hey, this doesn't work," and then why it's actually not true?
Philip Johnston:
Yeah, for sure. There's sort of a few layers of understanding of that, so I'll start at the bottom and work my way to the top. At the bottom end, people think space is cold. Space is about minus 270 degrees, three degrees Kelvin.
Turner Novak:
And data centers, when you're running these models, they overheat. So you're like, "Okay, cool. Put it in space. Cool it off."
Philip Johnston:
Yeah, exactly. So your first instinct might be, "Okay. Yeah, let's just dump it in space. Space is cold."
The second instinct is, okay, but space is actually a vacuum. There's a reason vacuum flasks are effective at insulating, which is that you don't have convection or conduction through a vacuum. You only have infrared radiation. That's the only way you can get rid of heat. And so people are like, "Oh, well no, you're screwed. The heat's just not gonna go anywhere. It's just gonna build up on the satellite."
Turner Novak:
So the satellite stays hot while it's extremely cold all around it.
Philip Johnston:
Yeah, because it's a vacuum, right?
And then you have the third piece. The infrared radiation that you get on a satellite is proportional to the fourth power of the temperature differential. And the way that you emit energy in infrared is you have to keep a large surface area warm, essentially, and then it will just glow in infrared.
For people who don't know, infrared's nothing fancy. Everything that is warmer than its surroundings glows in infrared. So if I had an infrared camera on my face right now, you'd see that I'm glowing in infrared.
And the same is true in space. The difference in space is that you have a temperature differential of 270 degrees Kelvin. And by the way, if you can increase the temperature of the radiator by a small amount, even from 50 degrees to 80 degrees, that means you can cut the surface area on the radiator by close to half. And so that's how you get around it.
This isn't new physics or anything else. The International Space Station has had a radiator for the last 20 years that operates perfectly fine. It's exactly the same mechanism we're using, which is a liquid loop that draws heat from the main central piece and then goes out to this large surface area to emit the infrared.
So yeah, we've designed, built, and fabricated our radiators in-house. They're about 10 times less mass per watt of dissipation than the International Space Station radiator and about 500 times less cost per watt of dissipation. And that's really the challenge. It's a manufacturing and engineering challenge. How do you build cheap radiators?
Turner Novak:
And I think you mentioned earlier that's kind of what you're doing that's so unique.
Philip Johnston:
That's our secret sauce, yeah.
Turner Novak:
So could I right now just straight up make something, book a launch, and get my own similar type of cluster up in space? It might cost me a lot of money, and it might not cool properly.
Philip Johnston:
I mean, in the same way that you could launch your own constellation of Starlink if you wanted. And yeah, you need to have a lot of money and a lot of engineering people that are good at that stuff.
Turner Novak:
Yeah. So I think I saw that when you were first doing this, you were working with a defense prime, and they quoted you something like $100 million to build this thing. And you guys launched the first one for about $2 million, $3 million, I think.
Philip Johnston:
Two and a half. Yeah.
Turner Novak:
So what was the difference in that price that made it so much more expensive, and what were you doing differently?
Philip Johnston:
I mean, you'd have to ask them.
Turner Novak:
So you don't actually know.
Philip Johnston:
We had a call, and we told them the specs, and I said, "What are we looking at?" And they said, "Well, it would probably be about $75 million to $100 million to get this program going." And as I say, we did it for two and a half.
I can tell you what the two and a half cost us, which was $300k for the launch, a million dollars for the bus, about $500k for the payload, and then some overheads. It's a 60 kilogram small sat that had the first H100 on board. It's still operating now, by the way. We're still running government workloads on it. So yeah, the system is just broken. That's why these primes are costing so much.
Turner Novak:
Isn't what's kind of happened that a lot of them have booked capacity that they then sell out to other people? So you're probably paying for a lot of overhead and bureaucracy. They're a middleman, essentially, managing some of these programs. Is that sort of part of it?
Philip Johnston:
It's sort of part of it. The launch cost is reasonably well known, though. I think more of it is coming from this cost plus on R&D. And also, people just thought it was impossible to run an H100 in space. So they were like, "The amount of R&D hours we're gonna need to make this work is like..."
And the way we made it work in the end is Adi and Ezra just built the payload by hand. They built this enclosure for the H100. It was immersion cooling, so we had the whole thing submerged in this phase change material, all of the motherboards, power supply, everything else. Two completely independent systems.
Turner Novak:
So can you describe what that was like when you guys were making that?
Philip Johnston:
Janky. If Starcloud 2 works, I really wanna show the videos of us working till 7:00 AM. We worked through the night. And then, yeah, we didn't have time to do thermal cycling, which is a pretty important test because you wanna know that the thing isn't gonna burst when the phase change material heats up.
Turner Novak:
Thermal cycling, so that's so when it hits space, it doesn't catch on fire?
Philip Johnston:
No. Thermal cycling is basically where you expose it to the thermal environment it's gonna have in space over a two or three-day period. What it means is that every 45 minutes it's gonna be very hot and then 45 minutes very cold. It'll go down to minus 50, and then it'll go up to plus 100, and then down to minus 50, and then-
Turner Novak:
So this is what happens when you're in space. The temperature around you is constantly changing in those huge-
Philip Johnston:
Yeah. With our first spacecraft, it's not on the terminator line. It's in a mid-inclination orbit, so it's going behind the Earth and it's going in front of the Earth, and that is a 45-minute cycle.
And so we have this phase change material to keep all the components cool, and it solidifies and melts during that period. And so we needed to make sure that the satellite doesn't explode when it melts and things like that.
Turner Novak:
Is that usually something you'd have to worry about if you're doing a satellite, or is it unique to-
Philip Johnston:
Yeah. Every satellite does a thermal cycling test, like Starcloud 2 is doing now.
Turner Novak:
Okay. So what are you guys doing on it right now? You said you're running some government workloads. What kind of stuff are you doing? Are you allowed to say what is going on?
Philip Johnston:
Yeah. So we did a bunch of these more experimental workloads, like training things, and now we're doing inference on SAR data, so synthetic aperture radar. That's very useful for government because they have a whole load of satellites that are collecting SAR data right now, and-
Turner Novak:
They're collecting what data?
Philip Johnston:
SAR. It stands for synthetic aperture radar. It's basically building a 3D map of the world underneath it as it flies, and it's useful for identifying tanks or ships and things like this for the military.
But the problem is it collects enormous amounts of data, like 5 to 10 gigabytes of data per second, not gigabits. And so right now they have to wait till they hit a ground station. That might take an hour and a half. Then they have to downlink as much data as they can, which might only be 1% of the data they collected, and only then can they do processing on it.
Whereas if we're in space, they can send all of that to us with an optical terminal. We run the inference on orbit, and then we downlink just the insight. The insight might be the coordinates of a tank or the coordinates of a vessel or something like this.
Turner Novak:
So you're only sending basically change data to the map versus the whole thing.
Philip Johnston:
Yeah, which is way less, which you can do in real time. You can do that with RF in real time.
Turner Novak:
Did they need to use Starcloud 1 to do that?
Philip Johnston:
Starcloud 1 is definitely a demonstrator. What we're doing right now is demonstrating that we can run high-powered inference with an H100 on SAR data in space. It's not being used in a battlefield situation right now.
Turner Novak:
And you did this with off-the-shelf stuff, not space grade. So can you explain what space grade materials are? I'm not really sure what you'd actually do differently, and then how it actually ended up working out anyways.
Philip Johnston:
Yeah, for sure. There's people that manufacture electronics components for space, and they're designed to be radiation tolerant and low mass and have nice thermal properties and all the rest of it. An H100 is not designed for any of that. And a lot of the subsystems that you need around it, like power and other things, are also not.
So there's a few different steps to it. One is with the H100, we put a bunch of chips through radiation chambers. We went to the cyclotron in Knoxville, a proton cyclotron particle accelerator. You put it in this chamber, and over a 24-hour period you expose it to five years' worth of radiation dose. And then we did the same at the Brookhaven National Lab for heavy ions. We did that four or five times, and we used that to inform the choice of components.
But once you've picked the components, you then have to cut the mass, make sure it survives thermally, and remove components you don't need. So for example, for the H100, we removed all the casing and removed the AC to DC converters. We switched out the cold plates, so instead of copper, we're using our own custom aluminum plates. There was a whole bunch of different things we had to do to make it work.
Turner Novak:
And how long has it been up there for?
Philip Johnston:
November 2nd is when it launched, so it is almost a year.
Turner Novak:
Okay. I think by the time this comes out, it'll probably be past November 2nd, I'm pretty sure. I'd have to double check.
So have there been any changes to anything? Any, "Oh shit, we lost a part," or something got damaged because of being in space?
Philip Johnston:
Well, yeah, it's working remarkably well, but there's been a whole bunch of things we had to fix. For example, for the first few weeks, the satellite kept restarting itself every hour and a half.
Turner Novak:
Do you know why?
Philip Johnston:
What we figured out was that it was a software failure. One of 20 different restart modes was triggering. They're usually safety mechanisms in case the satellite overheats or something like this. But there's 20 of these different mechanisms, and we weren't sure which one was triggering unnecessarily.
And the problem is you only get a ground station pass every 90 minutes. So what we had to do is turn them all off and turn one on, wait for a ground station pass, and see if it's triggered. And then turn that one off and turn another one on, wait for a ground station pass, and see if it's triggered. So that was three days of software debugging just to... And sometimes you don't hit a ground station, so then you'd-
Turner Novak:
So can you explain this process of getting data up and down? Why can't I just press the button and get the data? Why is that such a big process?
Philip Johnston:
Yeah, it's a nightmare right now. If your satellite's over the middle of the Pacific right now, there's no ground stations near it, and so it doesn't have connectivity. There's a few very low bandwidth satellites you can ping data to. The Iridium network, for example, you can send very small packets of data to. But in order to send down large amounts of data, like imagery, you need to be above a ground station, basically.
And there's no ground stations in China, none in Russia. Russia's the biggest country in the world by land mass. There are hardly any in the Pacific Ocean, which is the biggest place by surface area. So you've just got a lot of time where you're not connected to anything.
Turner Novak:
So you basically need a line of sight to shoot the data, and then it's received at another thing? That's essentially the point of all this?
Philip Johnston:
Yes. Let me... I'll even show you, look. I'll show you where Starcloud 1 is now.
Turner Novak:
Okay. So what happens a lot is you're pinging from satellite to satellite to satellite to then hit a ground station? That's sort of how it works right now?
Philip Johnston:
Yeah, exactly. Well, at the moment, no, you're just waiting for the ground stations. So-
Turner Novak:
For Starcloud 1, the first model you guys launched.
Philip Johnston:
Yeah. Okay. And speak of the devil, we're right now in the Pacific Ocean.
Turner Novak:
Oh, wow.
Philip Johnston:
So at any given point, the chance of you being above the Pacific Ocean is about 50%, 'cause the Pacific Ocean is just massive.
Turner Novak:
I'm taking a picture on the screen.
Philip Johnston:
Yeah, go for it.
Turner Novak:
Actually, maybe what we could do... Actually, take a screenshot of that and just send it to me, and we'll put it over top of the screen while you're showing me that.
Philip Johnston:
Yeah.
Turner Novak:
Yeah, I guess for somebody who doesn't know anything about how satellites work, I guess that makes sense. You take this stuff for granted. You just pull out your phone, and you text your friend who's in Europe, and they just get it a quarter of a second later.
Philip Johnston:
Yeah. It's cool. So that text most of the time is going through subsea cables. It's not even going up to satellites. If you're using Starlink, then it's going up to these low Earth orbit satellites and then getting relayed through a bunch.
Turner Novak:
So a lot of what's happening is transferring satellite, satellite, satellite, and then coming down.
Philip Johnston:
Yeah, for Starlink, yeah.
Turner Novak:
Interesting. You always need a line of sight to hit something.
Philip Johnston:
Yeah.
Turner Novak:
I think there was actually a second chip on board Starcloud 1, and it actually died. So what happened with this?
Philip Johnston:
Yeah, that chip actually died before we launched it. I'm not 100% sure what the failure was, but we presume it was a connection that broke during vibration testing.
Turner Novak:
And you didn't catch it or didn't notice it?
Philip Johnston:
No, we did. But the problem with vibe is, if you're a company like us, you're usually running right up to the deadline for when you're allowed to vibe. We basically didn't have time. We would have to open it up, fill it with everything, and then revibe it, and vibe houses get booked out weeks in advance.
Turner Novak:
What is a vibe house?
Philip Johnston:
A vibe house is a company that will have a vibe table that you can rent.
Turner Novak:
What's a vibe table?
Philip Johnston:
A vibe table is a huge block of metal that shakes. It has a solenoid inside. A solenoid's like a big magnet that vibrates at every frequency. So you basically strap the satellite to this table, and then it vibrates at every frequency and every amplitude.
It does all the, what they call, random vibe, and then it does quasi-static, which is like shock testing. And the reason you're doing this is, with SpaceX, you're launching on a stack which has a bunch of other satellites on there.
So they're looking for two things. Number one, nothing can fly off, 'cause if it does, it's gonna smash into everything else in there. And number two, you can't have resonance below a certain frequency. I think the cutoff is 40 hertz or something like that.
The problem is if you have resonance at 10 hertz, the thing will be going like that, and it'll be pulling the whole stack left and right. So you need to check every axis, every frequency, every amplitude, and then every potential type of launch vibration it's gonna experience.
And we didn't have our IM5 table back then. We do now. We just didn't have time to fix it, basically. So we were like, "Okay, we're just gonna have to launch with..." Thankfully, it wasn't the H100, it was the A6000, which is another chip from Nvidia. And so we were just like, "Yeah, screw it. We're just gonna launch."
Turner Novak:
And I know Nvidia ended up investing in Starcloud also.
Philip Johnston:
Yeah.
Turner Novak:
And I think you've said that they did way more diligence on what you're doing than any other investor in the company, and there's specific data that they saw that they were really excited about. So why did they invest? What was their diligence process like?
Philip Johnston:
Yeah. So we sent a bunch of reports on the results of Starcloud 1, which I don't think we'd even put together by the time we did the Series A. Because they're now building this Space Rubin One chip, we're the only people in the world that have any telemetry and data for how an H100, or something that powerful, operates on orbit.
So yeah, they're interested in how Starcloud 1 is operating, but they also want our input on designing this next Space Rubin One chip as well.
Turner Novak:
So what's the name of this new chip that they're making?
Philip Johnston:
The Space Rubin One chip. It's basically a Vera Rubin chip that's optimized for space, so optimized for mass, thermal, and radiation.
Turner Novak:
So they think that there's gonna be a lot more compute happening in space.
Philip Johnston:
They are very convinced of that. They're working with SpaceX as well, and SpaceX has been very vocal about their ambitions, so yeah.
Turner Novak:
I think that's the other really interesting thing about what you're doing. You're basically using SpaceX rockets to get into space, and they're kinda like a competitor really at the end of the day. How do you feel about that as a startup? You're kinda reliant on a competitor.
Philip Johnston:
Yeah. Right now, we have a very good relationship with them. We submitted our application to the FCC, and they wrote a supporting letter for that, and we wrote a supporting letter for theirs. I think our interests are pretty well aligned right now.
We've got a bunch of launches booked. We're paying for some communication stuff as well. They will have a lower cost base than us because they own the launch. But we will have a lower cost base than all of the other hyperscalers, so long as we get launched at something like less than 20 times their marginal launch cost with Starship. Right now, we're paying three times the marginal launch cost with Falcon 9. So-
Turner Novak:
Do you guys have different customer bases or use cases that you're going after?
Philip Johnston:
A little bit. Yeah, a little bit. They're mainly serving their own internal workloads and Grok, and maybe they'll offer a third-party cloud service. One of the business models we've been working on since the beginning is we're more of an energy and infrastructure play, more like an Equinix.
So for example, with Crusoe, the agreement we have is we will provide them with power, cooling, and connectivity. Then we'll work with them on whatever chip architecture they want, and they can sell to their customers at whatever price they like.
Turner Novak:
In space, using the-
Philip Johnston:
Yeah. And we're just an energy and infrastructure provider.
Turner Novak:
And that's kind of interesting, because when you first started the company, I think it was kind of an energy company originally.
Philip Johnston:
Yeah. So we're back to the...
Turner Novak:
Yeah. What was the original thesis when you first started working on this? I think it had a different name also.
Philip Johnston:
Yeah, we were called Lumen Orbit, based on that business model: collecting solar power in space and beaming it down to Earth.
Turner Novak:
Yeah. Similarly, kind of a crazy idea. How does that make sense? What's the science and physics behind that, how that works?
Philip Johnston:
So that also definitely does make sense. But instead of the $500 a kilo breakeven that we have, it requires a launch cost closer to $50 a kilo, which is within range of what Starship should be able to do. And that's why we started working on it. We were like, "Okay, we can see this is coming. Let's get going now."
Initially we were trying to sell that power. We were basically trying to sign energy off-take agreements with various people. And we realized almost all of the new energy projects being built today are being built for data centers, essentially.
So we were like, "Okay, well, either directly or indirectly, this power is gonna end up going to data centers." So maybe we can rerun the numbers, assuming that we have a cheap way to get the data center to space.
The problem with beaming power from space is you lose like 90 or 95% of the energy in transmission from space to Earth. It basically warms up the air between space and the ground, and that's how you're losing energy.
Turner Novak:
But you collect way more 'cause you're just in space.
Philip Johnston:
Yeah. So you can build tens of kilometers of solar and then beam that power down. And you'll have 24/7 sun, and you have some other nice properties. It's a concept that was written about. I think even Asimov in the '40s was writing about space-based solar as a concept.
So we were just jumping the gun a bit, thinking, okay, if Starship really does deliver this $50 a kilo, we should be able to break even with this. And then we found a much nicer business model, which is moving the data center to space instead.
Turner Novak:
Yeah. I think the interesting thing with Starship launches is, I mean, you used to be able to book them. They're very hard to get now, right? I don't know if there's very many that are... I think there's no more available past 2028.
Philip Johnston:
For Falcon 9, yeah, Falcon 9's booked out. Starship, they have not started selling Starship yet.
Turner Novak:
And you don't know when that's gonna open up, right?
Philip Johnston:
No.
Turner Novak:
So in theory, if there's no more launches, you're screwed. In theory. Or do you need to just find other launches?
Philip Johnston:
We've booked on a bunch of different... I think it's public now. We've booked on Stoke, we've booked on Firefly, we've booked on... I don't think I can say the others, but we're booked on at least three more.
Turner Novak:
So there's all these other providers that do similar... Are they smaller rockets?
Philip Johnston:
Smaller rockets, much more expensive and much less well-proven right now. We can survive on those for the next five years just selling to the government. We don't need a reduction in launch prices for that business model, for the cloud and edge services business.
Yeah, if there is never a competitor to SpaceX and they won't sell to us, then we won't be able to compete with terrestrial data centers on energy costs. We can still be a several billion dollar a year business from this government use case. But we can't be a several hundred billion dollar a year business from competing with terrestrial data centers on energy.
Turner Novak:
And there's other approaches people have taken. Some people are putting them in the ocean. I don't know if anyone's doing them underground.
Philip Johnston:
People are doing them at the bottom of the ocean. Yeah, I didn't know about underground.
Turner Novak:
Yeah. So where do you think we're gonna end up? There's been a ton of pushback, like people don't want us to build new data centers. There's a lot of people that are very against that. What do you think are some of the steps, and maybe the end state, of a lot of this pushback?
Are we gonna figure out how to make them better? Are we gonna put them all in space? Is there gonna be some in the ocean?
Philip Johnston:
They're definitely all going to space eventually. On a sort of 20-year timeframe, they're definitely all going to space.
The thing with all of these alternative energy methods is, even if you can get them to be cheaper than they are today, they're not gonna be below 5 cents a kilowatt hour. For a new natural gas generator plant, you might be looking at something in that region, probably much more.
But yeah, the lowest forecast cost I've ever seen for nuclear fusion is 20 cents per kilowatt hour. I think that's from Helion, and they're pretty incentivized to give a low forecast. We're targeting much less than half a cent per kilowatt hour, including the launch cost.
So yeah, all data centers are definitely gonna go to space. That doesn't mean you're not gonna have them on the ground. There's use cases for them on the ground: very low latency inference, training large clusters, and all these kinds of things you would need them on the ground for. But your bog standard coding agent is gonna be done in space for sure.
Turner Novak:
And then how do you make sure that they go fast enough, though? 'Cause when I'm coding, I don't want a 90-minute latency. That would suck. So do we just need a-
Philip Johnston:
Well, no. To anywhere in low Earth orbit using lasers, it's sub 50 millisecond latency. You could have a video chat with 200 millisecond latency and you wouldn't notice any difference. So it's totally fine.
Turner Novak:
And so what does the future of this look like? Is every single piece of the sky just a data center floating around? It kind of sounds maybe a little grim, I don't know. How do we make this practical for people?
Philip Johnston:
So you won't see them during the day, for example, and you won't see them during the night. The only time you might see them would be at dawn or dusk.
Turner Novak:
Okay.
Philip Johnston:
And practically what that would look like is, let's say the sun is setting over there, right? The sun's about to set on the horizon. This is the horizon, and the sun is about to go down over there.
What you're gonna see as the sun is just setting is these satellites streaming over the horizon. It takes about five minutes for a satellite to cross the horizon. Then they'll slowly start passing above you, and it'll take about half an hour for this stream. There'll be a line of these satellites. And then-
Turner Novak:
Are they rotating, like spinning?
Philip Johnston:
No, they're not rotating. They're just going like this.
Turner Novak:
Okay.
Philip Johnston:
And then slowly, you'll see them trail off over the horizon, following the sun around, basically. And then they'll pass over the horizon and you won't see them anymore.
Turner Novak:
But we're gonna run out of space to do that eventually, right? There's not gonna be enough room up there.
Philip Johnston:
So we think you can fit about 10 terawatts in the dawn-dusk sun-synchronous orbit. The 88,000 that we filed for enables us to deploy the first 20 gigawatts. 10 terawatts is about 20 times the entire US power grid right now, so we've got some decent space.
Once we've filled that orbit, then we're looking at medium Earth orbit, which is way larger, and then beyond geostationary, and then the Lagrange points, the lunar Lagrange points, cislunar orbits, sun orbits. In the end state, we'll be putting them in a sun orbit.
Turner Novak:
That's what's called a Kardashev sphere? What's it called?
Philip Johnston:
A Kardashev type II civilization is where you surround your star.
Turner Novak:
Dyson sphere, that's what it's called.
Philip Johnston:
Dyson sphere, yeah. Dyson swarm is what people like to call it, but I still call it Dyson sphere 'cause I'm old school.
Turner Novak:
And this is basically you have a star that's emitting all the energy, and it's surrounded by energy capture devices, essentially.
Philip Johnston:
Yes.
Turner Novak:
Okay. When do you think that's actually possible? Is this something that we might get in 10 years, or is this like 100 years, 1,000 years away? How...
Philip Johnston:
You could do it now. You could put a satellite in a sun orbit which captures sunlight and processes data right now.
Turner Novak:
Wouldn't it burn up? I mean, isn't the sun like 10,000 million degrees or something?
Philip Johnston:
You would put it at the same distance from the sun as Earth. So you'd just put them in a sun orbit that's the same distance as Earth. It would be exactly the same amount of energy hitting it as if it's in low Earth orbit. Except instead of orbiting, like back to this example, if this is Earth, instead of going around like this, it's going around like this. So they're just a bit in front.
Turner Novak:
Got it. So it's not like you're right around the sun, wrapped around. It's literally the same as a planet is, just orbiting it.
Philip Johnston:
Yeah. Yeah.
Turner Novak:
Got it. So maybe a dumb question, but why don't you do that today? 'Cause you can get a shit ton up there and no one would even notice, right?
Philip Johnston:
Oh, you could. Yeah, you can put probably 10 to the 20 more compute there than you can in low Earth orbit. So we will do that. But we have tons of space in low Earth orbit now. The benefit of low Earth orbit is you have low latency, and you don't need as much fuel to get there.
Turner Novak:
So what do you think is the future, the end state of humans interacting with machines in this way? Is it basically that in every piece of the solar system there's compute orbiting around the sun? Is that ultimately what we're gonna get to? I mean, it might take a very, very long time to get there.
Philip Johnston:
Yeah, my estimation is it would probably take on the order of 5,000 to 10,000 years to build a proper Dyson swarm. You'd probably need to disassemble the Moon and maybe Mercury to get enough material to build a proper Dyson swarm.
Turner Novak:
You'd disassemble them? What does that mean?
Philip Johnston:
Well, you'd need to mine all of the materials on the Moon. Maybe we wouldn't use the Moon, though, because it would mess up the tides. But you need to mine an amount of material that's kind of similar to that. You might wanna go out to the asteroid belt 'cause there's a bunch of material there.
Turner Novak:
So you're thinking you couldn't use the Earth 'cause there wouldn't be enough. You'd have to take so much out of the Earth it would mess things up.
Philip Johnston:
Yeah. And also the Earth has quite a strong gravity well, so it's actually kind of annoying to use the Earth for Dyson swarm material. 'Cause you...
Turner Novak:
Have to launch them all off?
Philip Johnston:
Yeah, you have to launch them all off. It's kind of annoying. And also, yeah, we wanna keep the Earth...
Turner Novak:
As is.
Philip Johnston:
As is, right. Yeah.
Turner Novak:
It's kind of nice.
Philip Johnston:
I think probably Mercury's a good candidate for Dyson swarm material.
Turner Novak:
Really?
Philip Johnston:
But the Moon is by far the most perfect because it's close to Earth, and it doesn't have an atmosphere, so you can use mass drivers to launch it. It has a bunch of aluminum and silicon, which are the two main things you're gonna need for Dyson swarm material, and it has iron and other things.
Turner Novak:
But it would probably mess things up on Earth, so we probably just shouldn't...
Philip Johnston:
Shouldn't build it down there.
Turner Novak:
We don't know.
Philip Johnston:
If we didn't have tides, I think there's a way to make artificial tides as well. I can't remember. Alex Filipovich has some good opinions on this. I'm sure we could figure something out.
The Moon would be great, but if not the Moon, Mercury's also a good candidate because it's way closer to the sun, so you need less material for the same amount of solar energy. And also, we're not gonna start building huge civilizations on Mercury anytime soon.
Turner Novak:
It's just inhospitable, right? It's just way too hot there.
Philip Johnston:
Yeah, it's very hot there. You could, though. It's tidally locked, so there's only one side that faces the sun all the time. And so people have talked about the terminator line. There's a difference between the day and the night. The day is very hot, the night is very cold, and they're the same places on Mercury all the time. So maybe that line between day and night might have places which have a more stable...
Turner Novak:
Like a strip of civilization?
Philip Johnston:
Yeah. Could be.
Turner Novak:
Yeah, but it's also like, why would you choose to go there when maybe Mars is better, the Moon's better?
Philip Johnston:
Yeah, yeah, yeah. Earth is still...
Turner Novak:
Pretty good.
Philip Johnston:
That's great. Yeah. So yeah, at some point we might disassemble Mars to build Dyson swarm material.
Turner Novak:
Oh, man. But...
Philip Johnston:
No, not Mars, sorry. Mercury.
Turner Novak:
Yeah, I was gonna say Mars. It's like...
Philip Johnston:
Yeah, no. Mars is a good one. We wanna keep Mars 'cause it's a good stepping stone to the outer planets. Not too many people are gonna complain if you start using a lot of asteroid material.
Turner Novak:
Yeah, 'cause you could argue they're gonna hit Earth eventually anyways in a long enough timeframe, so let's get rid of them all.
Philip Johnston:
Most of them will never hit Earth, actually. Most of them are too far out.
Turner Novak:
What do you believe about...
Philip Johnston:
But anyway, yeah. Nobody's attached to asteroids. Whereas the moons of Jupiter, probably lots of people would be pissed off if you started mining those.
Turner Novak:
Oh, Jupiter, you think people would be pissed off?
Philip Johnston:
Well, the moons of Jupiter are like planets in that they're sort of more habitable. You can't live on the gas giants, but you could live on the moons. So maybe we would wanna use that material. But Jupiter's too far away. The energy you get from the sun at Jupiter is very low compared to closer in.
Turner Novak:
Yeah, that's fair. I think one thing that might be interesting is backing up, going back to how you actually got into space originally. I think you worked at McKinsey doing some space things.
Philip Johnston:
Yeah.
Turner Novak:
So what all did you do?
Philip Johnston:
So I studied math and physics, undergrad and masters. I wanted to be an astronaut since I was a kid and all the rest of it.
Turner Novak:
Nice. What kid did not wanna be an astronaut?
Philip Johnston:
Yeah, same as everybody.
Turner Novak:
It's like the coolest thing ever.
Philip Johnston:
Yeah, yeah, yeah.
Turner Novak:
Did you read a lot of those space books?
Philip Johnston:
I read the Asimov Foundation series and a bunch of those, yeah. Then I got sidetracked and did end up doing finance and business stuff for a bit.
Turner Novak:
I feel like everyone's like, "I need to make money. You need a job."
Philip Johnston:
Yeah, yeah. You probably had the same.
Turner Novak:
Yeah.
Philip Johnston:
And then I went back to grad school and got a job at McKinsey in the Middle East. The Middle East governments have big ambitions in space, and so they're basically setting up these space agencies from scratch. During my time there, the UAE sent a probe to Mars. They had their first astronauts and things like that. They're doing some super interesting stuff, and I ended up...
Turner Novak:
Why do they care so much about space?
Philip Johnston:
They care about big ideas. They're kind of ambitious by nature. If you've ever been to Dubai, you get a sense of it. That's it. And then they have a lot of oil money that they wanna...
Turner Novak:
They got a lot of cash flow they gotta put to use.
Philip Johnston:
Yeah. So I ended up getting staffed on a few of those kinds of projects, and realized then that the launch cost was coming down rapidly with the Falcon 9 program. That was back in like 2019, 2020, 2021.
In mid-2023, I kind of randomly decided to take a trip to Starbase, Texas, even before the first Starship launch. Nobody was really paying attention back then. It wasn't like now. I was just blown away by the scale of what they're building there in terms of the manufacturing capability. I think these two Starship gigafactories can produce something like three Starships per day, which is just mind-blowing. And because it's reusable, in a few years that means we'll have like 10,000 times the current capacity that we have to get stuff to space.
And so I was like, "Okay, this is gonna change everything." I started speaking with Ezra, my co-founder, about, "Whoa, let's just imagine a world where launch cost was 10 times cheaper and launch availability was 100 times more." And yeah, that's what got us started on this space-based solar idea, and then that pivot, that transition to what we...
Turner Novak:
Nice. And one thing I think you did, you were in the Middle East, you did this company probably around the time COVID hit.
Philip Johnston:
Yeah. Yeah.
Turner Novak:
What was the company you were working on then?
Philip Johnston:
Yeah. So in 2021, I founded a company called Opontia. It was a digital brand aggregator, a bit like Thrasio, and I think you've posted about that before in the US. So we were the Thrasio of the Middle East, basically buying and growing e-commerce brands as a group rather than as individual companies.
Turner Novak:
What was the general thinking and thesis around starting that company?
Philip Johnston:
My co-founder was the e-commerce expert in that business, and my background was more in project finance. I'd done CFA and all that kind of stuff, and we needed to raise lots of money in VC and in debt to make that work.
Turner Novak:
Yep.
Philip Johnston:
So I was the acquisitions guy buying the brands, and he was the operator running them. I think we just thought it was a good idea. We could see it work in the US and decided to go after it.
And then, I think you've mentioned this before, in Q1 2022 interest rates went from 0% to 5% basically overnight with the Ukraine war. Anybody that had been using debt to make acquisitions couldn't refinance that debt. We'd used a bit of debt, but not as much as the others. That's why Thrasio went bankrupt, and Perch and all the others. We didn't go bankrupt, but yeah, we started to take better...
Turner Novak:
Well done. Anything else you learned from that?
Philip Johnston:
Yes. I'm much more wary of venture debt now than I was. We also hired too quickly. We hired like 150 people in the first year, and at some point we had to do a cut, and that was painful. Definitely we're super slow at hiring at Starcloud.
Turner Novak:
Very deliberate.
Philip Johnston:
I agree.
Turner Novak:
How many people are at Starcloud now?
Philip Johnston:
30. We were...
Turner Novak:
That's what the LLMs told me, which is still dated. So you're still at 30 people.
Philip Johnston:
Yeah, yeah. That's pretty slow. The LLMs are probably from six months ago, and LinkedIn has all of the investors on there. So if you go on LinkedIn, it will say 45. 'Cause they're all...
Turner Novak:
I love the investors adding themselves on LinkedIn.
Philip Johnston:
Yeah, yeah. They put it on their profile.
Turner Novak:
Nice.
Philip Johnston:
I think there should be an index. You know how well you're doing as a company based on whether your investors put you on their LinkedIn.
Turner Novak:
Yeah, or the Twitter bio.
Philip Johnston:
Twitter bio, yeah.
Turner Novak:
I remember someone... I don't really know how to track this, but there's the index of companies being mentioned by investors in their Twitter bio.
Philip Johnston:
Oh, that's right. That's right.
Turner Novak:
So then, do you get removed? 'Cause on LinkedIn you can have a million companies listed, but in your Twitter bio you have a fixed number of characters.
Philip Johnston:
Yeah.
Turner Novak:
So Starcloud gets those, what is that, 10 characters with the @. That's a slot. You maybe get 10 of those. You invest in 10 Starclouds, or maybe you get eight of them, right? So are you one of the top eight for someone who's done 100 investments? That's an indicator.
Philip Johnston:
Yeah.
Turner Novak:
So has anyone ever removed you from their Twitter bio?
Philip Johnston:
I don't think I've ever been in anybody's Twitter bio, to be honest. I don't...
Turner Novak:
Oh, really?
Philip Johnston:
I don't really follow that.
Turner Novak:
So Starcloud's still emerging. You're not even getting included in a VC's Twitter bio yet.
Philip Johnston:
I don't think I've ever seen us in a Twitter bio.
Turner Novak:
Really?
Philip Johnston:
But to be honest, I've never really seen... Yeah, no. Maybe I need to look at Twitter bios more.
Turner Novak:
Well, VC's will literally do anything to get an edge on people. One of the things is to look at how people change their bio on Twitter. There are tools for this. I used to have one. I followed this one guy, an engineer at Coinbase, and then it showed me that he changed his bio, and now it's "working on something new."
Philip Johnston:
Oh, yeah.
Turner Novak:
Right? Or it's "building in stealth." And you're like, "Oh, shit, I gotta message this person. What are you working on?"
Philip Johnston:
Yeah, yeah. A lot of people do that. I've seen that for LinkedIn. I didn't know about it for Twitter. That's good.
Turner Novak:
Oh, yeah, people do it on Twitter. Okay. I actually thought that was more common. Maybe I should be doing that more, if you've never heard of that before.
Philip Johnston:
Yeah. I've heard about it on LinkedIn but never on Twitter. Yes, maybe I should do it.
Turner Novak:
It's an edge. So how do you think about hiring now? What are you specifically looking for when you're adding the 31st person to the team? What's important?
Philip Johnston:
Yeah, almost exclusively kickass engineers.
Turner Novak:
So what does a kickass engineer look like? How do you gauge if somebody's good or not?
Philip Johnston:
So half the team worked at SpaceX. That's a very good indicator, and it's really hard to hire people from SpaceX, by the way. We're one of the few companies in the world that can actually hire people from SpaceX, 'cause all of their stock peaked when it was like $200 billion, and now it's $2 trillion. So it's pretty hard to get them away.
It's pretty easy, honestly, to tell who the good engineers are. It's easy to get them through the door, and then our guys just grill them. Everyone knows, if the guy got a 5.0 in MIT undergrad in electrical engineering and then was four years at SpaceX, the guy's a good engineer.
Turner Novak:
It's kind of obvious.
Philip Johnston:
Yeah, yeah.
Turner Novak:
But everyone's trying to hire them then, right?
Philip Johnston:
Yeah. So I guess it's pretty competitive. And really, it's a bit like VC. It's winning, not selection, that's the difficult part.
And we're pretty damn good at winning, so.
Turner Novak:
What do you think people appreciate about Starcloud? What's the pitch look like? Like, I'm that 5.0 GPA at MIT. I worked at SpaceX for four years. I could do anything I want right now. How do you convince me to join?
Philip Johnston:
I think people like the mission. I think people like the fact that we're still a small team, and they have autonomy. I think people like that we've been very deliberate on hiring, so they're gonna be working with other people that are at the same caliber as them. It's like a flywheel, a really, really important flywheel that starts spinning.
Turner Novak:
Was it hard initially?
Philip Johnston:
Yeah. It took us 18 months to make our first hire.
Turner Novak:
Really? Did you... Was that-
Philip Johnston:
And the guy was a 5.0 MIT undergrad, four years at SpaceX, two years at Helion.
Turner Novak:
So you did YC back in 2024.
Philip Johnston:
Yeah.
Turner Novak:
So what was the process of going through YC? What was that like?
Philip Johnston:
I'm the biggest YC fan ever. I freaking love YC, so-
Turner Novak:
Oh, you're a YC fan, okay.
Philip Johnston:
Yeah, yeah, I absolutely love YC.
Turner Novak:
'Cause you sell 7% of your company. A lot of people argue YC's not worth it. Like, keep your company to yourself.
Philip Johnston:
I've never met a YC founder who says that, to be honest.
Turner Novak:
That's true. There's a lot of haters.
Philip Johnston:
There are a lot of YC haters. Yeah, there's a lot of YC haters. YC is great because it gave us some kind of borrowed legitimacy at a time when people were like, "This is crazy shit." So it gave us cover for doing crazy things.
Turner Novak:
Yeah. What advice do you generally have for founders when you're raising money for this sort of crazy thing? You kinda have to say, "Hey, this is the really ambitious, really big vision that we have." But then you have to make it realistic, and you gotta raise the money for it without having proof points. You probably can't just ship some code and say, "Hey, a customer paid us, the chart went up." How do you talk somebody through balancing all that?
Philip Johnston:
Yeah, so the main thing almost all investors care about is the quality of the engineers on the founding team. When you're pitching coming out of YC, that's all they really give a shit about. What is the quality of engineering on the founding team?
My co-founder, Adi, was previously a principal software engineer at SpaceX, and 20 years on the data center side of Microsoft. He was the first to install large GPU clusters in collaboration with Nvidia and then OpenAI. If you google Adi Oltean deep learning, you'll find all these papers from him back in 2019, before it was cool.
So Adi's chief engineer. Ezra is CTO. Ezra has a PhD in engineering from Imperial and spent a decade designing and building satellites. He was recently working on Lunar Pathfinder. He designed all the deployable solar panels for that.
So that's what they're giving you money for. 'Cause anybody can come up with a wacky idea. That's one thing people misunderstand about YC. They think that because YC funded us, the idea is a good idea. YC actually doesn't give a shit about the idea. YC just funded us 'cause they thought my co-founders, Adi and Ezra, were good at engineering. That's basically it.
Turner Novak:
And so then what was your role as, like, the other guy?
Philip Johnston:
Their role was why investors funded us. I don't think any investor was funding me to do a space company on my own without them, right?
My role, once we were off to the races, is everything else that's not building satellites, which is customers. And also giving some input, because I have a bit of an engineering background, but not enough to understand everything that's going on. So, customers, and then you obviously have to raise more money after that and all the rest of it.
Turner Novak:
Yeah. What advice do you have for founders when you're trying to raise money for something like that?
Philip Johnston:
Get the best engineering team in the world. That's it.
That's it, that's it. That's the only thing people care about. And if you can show proof points, that helps. We obviously booked a launch, we had the satellite under construction in Santa Clara, and we could show some proof points. But really, they're going on the quality of the engineering team.
Turner Novak:
Yeah. And I think it was Starcloud where there were people commenting on these tranched rounds. Do you know what I'm saying?
Philip Johnston:
Yeah, yeah.
Turner Novak:
You guys did a tranched round. So how does that play out from the founder's perspective? How did that come to be? 'Cause I don't know if people wake up and say, "I'm gonna raise a bunch of tranched rounds." I remember people being really up in arms about it. I don't remember what the specific things were, but-
Philip Johnston:
Yeah, yeah. So here's the way those tranched rounds come together, and the reason people should be much less pissed off than they are. We did a very structured process to raise a Series A. In the beginning of January, I went out to like 50 VC funds. Almost all of them said no. Well, at least 40 VC funds told me no, right?
Turner Novak:
They said no, like they met you and said no?
Philip Johnston:
I'd pitched them, they were in the data room. I'd done a walkthrough of the data room with them, and then they sent me a rejection. Probably at least 40 VC funds did that.
Turner Novak:
What was the most common reason that people didn't wanna do it?
Philip Johnston:
There's a few. One of the ones that was a bit outside our control is that halfway through the raise, Elon came out and said he's gonna do it. And any of the funds that were already invested in Elon, which is like Sequoia and a16z and everybody else, were already invested in SpaceX, which was a private company then. Some of them then had concerns about it being conflicted somehow. So that was one of them.
There were concerns where people didn't understand the technology, like the thermal side of things. They couldn't really be bothered to figure out if the technology worked. Yeah, there's a bunch of reasons why people rejected it.
Turner Novak:
You still had 10 left, right? Roughly?
Philip Johnston:
Yeah, I think it was like five term sheets or something. We would've probably got more if... But anyway, Benchmark gave us a term sheet and we said yes. And then once Benchmark gives you a term sheet, everybody wants to invest, basically. But it's very conditional on having a term sheet from Benchmark.
Turner Novak:
Yeah. It's kinda funny. It's like nothing changed really, except Chetan said, "I'll give you money."
Philip Johnston:
Yeah. And Chetan did a lot more work than almost anybody else. He actually came and met all the engineers. Him and Eric both came up. They saw the manufacturing process, and they had conviction. They did the work, and they understood what was going on. And so they deserve to lead the round. The problem with that is then the round's full.
So then if people wanna invest more, I tell them, "No, sorry, 'cause the round's full." And then they say, "Okay, what about this price?" And I say, "No." And they say, "What about this price?" And I say, "No." And they say, "What about this price?" And I say, "Okay."
Turner Novak:
Yeah, it kinda gets to that point where you're kinda dumb if you don't take the money. And so people were upset that, you know, those investors-
Philip Johnston:
And I never-
Turner Novak:
Oh, sorry. Go ahead.
Philip Johnston:
Yeah. By the way, I was never not upfront about that with any new investors. I told everybody, "By the way, Benchmark's invested at this price. We're doing a new round at this price. And that's how it is. Totally understand if you don't wanna invest. That's totally fine."
Turner Novak:
Yeah, that's how I always see it as an investor when there's the two tranches, or sometimes it's three or four. I mean, it's gotten a little bit crazy with the way some of this is going up. But it's never like they're lying to you or something. You always know what's going on, and it feels good when you're in one of the lower tranches and there's upper tranches or whatever. But it's never like, as an investor, I don't know that that's happening.
Like, you always kinda know, and you can make the choice. And there's a lot of times where I'm like, honestly, I don't like that price. Again, that's fine. You're allowed to make that decision.
And so I think one of the other ones was that employees got screwed or something. This was part of the pushback people gave, right?
Philip Johnston:
So I don't really get that. What they're saying with that argument, I think, is that the stock is pegged at this high valuation if I joined after the second tranche, but it's not a very credible investor who's priced it. Which is bullshit, because we had EQT basically co-leading the second round with Benchmark.
In the new round we just did, we had Nvidia investing. We had Cisco investing. We had a whole bunch of funds, like John Curtius' fund from Tiger, and Alexis Ohanian's fund. These are not unsophisticated investors. So that is the price, you know? You might not like that that's the price, but the sophisticated investors have agreed that that is the price.
So it's total bullshit.
Turner Novak:
I mean, it's just like joining, like when you're an employee, you get-
Philip Johnston:
And also, what most of the employees are thinking is, "Holy fuck, I joined this company and now my stock's worth eight times what it was like two months ago." It's great for them, you know?
And now we're hiring more people, but the people we're hiring now know what price they're being hired at, you know?
Turner Novak:
And then also, if you joined two months earlier, with the 8x change in price, you're giving each employee a certain percentage of the company. If the company is worth more, you're able to give them more dollars anyways.
So I think as long as there's transparency around what's going on, you can't really get mad about it.
Philip Johnston:
Yeah, yeah. We're very open. We tell them exactly what the funding history has been.
Turner Novak:
And so now you're doing this thing called Fermi Explora, or Fermi Explorer.
Philip Johnston:
Yes.
Turner Novak:
It's a nonprofit. Did I say this right?
Philip Johnston:
Yeah, yeah. Fermi Explorer.
Turner Novak:
Fermi Explorer. So what is this? It's kind of another one of those very big, ambitious Philip Johnston projects. So what is it?
Philip Johnston:
So just to signpost the conversation, Fermi Explorer has got nothing to do with Starcloud. It's not, like, a part of Starcloud or anything.
Turner Novak:
Okay. It's you personally.
Philip Johnston:
Yeah. Me, Adi, and Ezra, my co-founders, have created a nonprofit with the goal of sending the first spacecraft to another star, and we wanna do that within three years. We'll launch it in three years, and it will cost less than $15 million. And the catch there is it's gonna take 79 and a half thousand years to get to Alpha Centauri.
And the reason it's that exact time is we've picked the exact optimum for a low-cost mission. We're hoping to be both the first to leave and the last to arrive at Alpha Centauri. So maybe in 50 years there'll be better propulsion technology, and stuff will come past us.
And if in 100 years propulsion technology is 20% faster, by the time we arrive at Alpha Centauri, there will already have been a settlement there for 15,000 years. There'll have been Dyson spheres and O'Neill rings and all the rest of it.
Turner Novak:
So you'll be showing up with the equivalent of a canoe when everyone else has gotten battleships.
Philip Johnston:
Yeah, actually. Yeah, yeah. So my hope is that we sort of limp across the finish line. There's a welcome party and they sort of slow clap us across. They capture the spacecraft and they put it in some kind of museum or something.
It just bugs me that we have the capability to send a spacecraft to Alpha Centauri and we've never done it yet. And I think the reason we've never done it is people think it costs a billion dollars, and we figured out a way to do it for $15 million. I think we'll do it for $10 million, actually.
Turner Novak:
So why do people think it costs a billion dollars to do this?
Philip Johnston:
So the Voyager probes are heading out into deep space. They're one of the NASA missions. They're like $10 billion for those missions. And-
Turner Novak:
This is each, or for the collective of all the probes? There were two of them.
Philip Johnston:
Okay. It's tens of billions in today's terms for both of them.
And they're actually not even going fast enough. They're going about half the speed they need to be to ever intersect with Alpha Centauri. So ours will be twice the speed, and we're gonna do it for about 1,000 times cheaper.
And the reason we can do that is we're not trying to invent new technology. We had this physical AI lab at MIT come up with a trajectory where we actually spiral in towards the sun. We can use solar electric ion propulsion, which is what Starcloud 1 uses, and we can just send this thing. So it's kinda like a time capsule, although we'll have some scientific payloads on there and some artistic payloads. But yeah, I'm extremely excited about it.
One of the purposes is to draw attention to the Fermi paradox, the idea of why we don't see more intelligent life in our galaxy. And by doing this mission, I hope we can knock at least two possibilities off. One of the possibilities is that it's too hard to get to the nearest star. Another possibility is that intelligent life, for some reason, decides not to go to its nearest star. So for both of those, we'll know that there's at least one intelligent species in our galaxy that both can and thinks it's a good idea to send a spacecraft to its nearest star.
Turner Novak:
Do you think there's other intelligent life out there? Like, in the universe somewhere?
Philip Johnston:
We have no evidence of it. So I would actually be pretty surprised. What would not surprise me is if there's many sort of long-dead civilizations. There's two alternatives. Either we are the first to reach this level of complexity, or intelligent life is quite short-lived for some reason.
Turner Novak:
Like, is it that we get so smart that we use up all our resources and we just die as a species?
Philip Johnston:
That's one possibility, yeah. Maybe superintelligence is self-destructive for some reason as well. So by doing this mission, I wanna show, you know, if we are the first to reach this level of complexity and we're now about to settle the galaxy, what an unbelievable privilege that is.
Or if it's true that... Because it would only take, you know, like 5 million years to settle the galaxy, only a billion years to get to Andromeda with our current technology, only 5 billion years to settle the local group of galaxies, and the universe has been going for 13.8 billion years. So if there had been life a bit like ours in any of the nearest galaxies, they would already be here by now. We would already see this big galactic empire.
There's also this zoo hypothesis, which I don't personally buy. I don't think we have enough evidence for that one.
Turner Novak:
The zoo hypothesis is that they're just kind of watching us, and they're just keeping an eye.
Philip Johnston:
Yeah.
Turner Novak:
Yeah. Really, I guess in the span of the galaxy, we haven't really been a lot. Like, we're a couple seconds if you think about it.
Philip Johnston:
Blink of an eye.
Turner Novak:
Yeah. Four billion years. All of a sudden we just have industrial power and solar energy, and we have AI. It's literally been, like, a second if you think about the time on Earth as a day or even a year. Right? It took four billion years, or five, six billion, to just get life.
And then one thing I wanted to ask you about too, I probably should've asked you earlier. You've posted about this a couple times, about VC's using AI to make investment decisions. It seems like you have a pretty strong opinion on this, as a founder, because you keep posting about it. So what's going on with how VC's are using AI to make decisions?
Philip Johnston:
By the way, VC's should absolutely use AI.
Turner Novak:
I use it, yeah.
Philip Johnston:
Yeah. My opinion is that it will very soon be much better than humans at making investment decisions. The reason I post about it is because I don't think most founders realize the extent to which VC's are using AI right now.
Turner Novak:
So what do you think VC's are doing?
Philip Johnston:
Well, I know what they're doing. They're basically throwing an enormous token budget at scouring the internet for any signs of new and breakout companies. If they need a data room and they get the data room, they'll use AI to assess the data room. And then they can benchmark all of the companies against each other based on publicly available information.
And then the remaining job of the VC is convincing the founders to take their money. But it's actually not that difficult to know which are the breakout companies just by using whatever your favorite LLM is to scour the internet for signs of it. It's not like I don't think there's a job for VC's. You know, the founders need to trust the VC, they need to build a relationship, all this kind of stuff.
But finding out which are the good companies, I don't think, is gonna be that difficult very soon. It's actually why the concentration of capital is accelerating. The good companies get all of the cash, and everyone else gets screwed. If I ask any LLM which is the best...
Turner Novak:
Market, company, founder, etc.?
Philip Johnston:
Yeah.
Turner Novak:
They're not gonna give me the same answer?
Philip Johnston:
Yeah, and also, if I do it for Starcloud... If I do it for data centers in space, since it's the one I have the most insight on, we are definitely the best, and it is very right about that.
Turner Novak:
Okay. So if I were to say, "Hey, I wanna invest in data centers in space, what should I invest in?" It'll say, "Oh, you should invest in Starcloud because of all these things." And I probably should say, "Okay, cool."
Philip Johnston:
It would be like, "I've been through every employee on every one of their LinkedIns. Starcloud has by far the best team. They've already launched something in space. They have their first Nvidia H100 in space. They are backed by Nvidia. They're blah, blah, blah." All of the reasons why we actually are the best team for this. It knows already. So it's not rocket science.
Turner Novak:
So as a VC, I see this. I'm like, "Okay, how do I get money into this thing? How do I convince them? How do I get in front of them? How do I get an intro?" That's the job of a VC.
So do a lot of founders not realize this is happening right now, or...?
Philip Johnston:
Founders that are struggling to raise should stick their company into an LLM and ask it to honestly assess it, and they'll figure out very quickly why they're not raising.
Turner Novak:
Did you do that with Starcloud? Or have you done it?
Philip Johnston:
Yeah, of course.
Turner Novak:
What is usually the biggest thing they realize? Do the LLMs say their market sucks or something?
Philip Johnston:
No. The LLM usually says there's another company in your space which is much better than you because they have a better team, better product, better traction, better investors.
I mean, sometimes there's not a lot you can do about that except carry on, you know? It's not like there's an easy fix to a lot of those problems.
But if you're starting from scratch, you need to go for gold-plated at the beginning. Don't compromise on an early hire that's not gold standard. Don't compromise on doing a demo that isn't really pushing boundaries forward, in our case. Or you really need proper, decent, growing traction if you're more like a B2B business.
Turner Novak:
It's interesting when you're talking about the concentration of capital. Maybe 10 years ago, if I'm a capital allocator, I don't know about this other really good company, so I'm giving it to a second or third tier one. But now I know about that, so I'm just gonna try to give it to the better one.
I hadn't thought about that too much, but that does make sense. It sort of explains maybe some of the capital accumulating, the whole kingmaking concept that everyone's kind of talking about.
Philip Johnston:
Yeah. I don't know if I'm phrasing it right, but it means if you can keep the bar for quality in all aspects, you know, employees, customers, whatever, traction, if you can keep that bar exceptionally high, capital will congregate around you more than it ever has before.
Turner Novak:
Yeah. And it's probably like a flywheel you really have to build, and it might take a while to start to kick in.
Philip Johnston:
Yeah, yeah. But as you say, it is a flywheel once you've got it going. And that's what I'm saying. It's helpful if you are at the looking-for-a-co-founder stage, where you haven't even started yet. You've gotta keep that in mind. Don't compromise on anything, because if you do, you're gonna get fucked.
Turner Novak:
Yeah, I definitely see some of those. I always try to meet all the co-founders when I'm investing, and some of them I won't like as much as the other ones. And I mean, that could just be my subjective opinion, obviously, right?
But I feel like there are definitely cases where you're like, this kind of felt a little bit too much like a shotgun wedding type of thing. "We wanna start a company, and we did a co-founder matching, and we're just starting a company together." That could be totally fine. That could work.
But there are definitely cases where I'm like, I don't know, it feels like not the best co-founder fit. Or, to your point, the quality bar is not as high. You could meet a higher bar maybe.
Philip Johnston:
Yeah. For example, when we got into YC, all of my mates from grad school, I did an MBA at Wharton and a bit at Harvard, tons of them were messaging me, asking, "Can I be chief of staff? Can I be head of operations? Can I be the finance guy?" And they're nice people. I like them, and they're smart, and they're capable and whatever.
Yeah, we didn't hire for 18 months because we wanted the best engineer to be the first hire. And that's what you gotta do. So many people screw that bit up.
Turner Novak:
Yeah. Do you have a favorite founder, CEO, business, historical figure or athlete that you've gotten a lot of inspiration from?
Philip Johnston:
I have an extremely cliche response.
Turner Novak:
You can give me the cliche response, but also if you have any that are, I don't know, not as cliche, I'd love to hear both.
Philip Johnston:
At the end of 2023, I read Elon's biography, the Walter Isaacson one, and it had a big impact on me.
Turner Novak:
I mean, you're right, that is probably the most cliche.
Philip Johnston:
Yeah.
Turner Novak:
Do you have any others?
Philip Johnston:
PG from YC. And recently, Dalton and Michael I love.
Turner Novak:
Oh, interesting. Okay.
Philip Johnston:
I listen to every podcast they ever do. Everything they say is gold. It's amazing. So maybe some of it's quite obvious, but...
Turner Novak:
Well, that's the thing I love about YC. It's kind of the most obvious advice, but it also is really helpful. It's always good framing.
Philip Johnston:
Yeah, yeah. It's incredibly obvious, but it's also the exact opposite of what they teach in most business schools. And so having done an MBA, it's so refreshing.
For example, in business schools they say, "Yeah, you need a really good board of advisors." Most of the stuff I'm posting is just trashing business school advice. Because YC would tell a board of advisors to go fuck themselves. YC doesn't give a shit about a board of advisors.
Turner Novak:
Yeah. Well, it's also like, you know, talk to your customers. Of course you should talk to your customers. But then when you think about how most companies actually work and run... I had so many jobs where I was just low man on the totem pole. I don't think I ever actually talked to a customer of our company.
My job was to make a spreadsheet, or I would just sit in meetings and take notes, because I was like an intern. And, you know, maybe it doesn't matter that much. But even people on the management teams of companies don't actually talk to the end customer. Your customer becomes a different department or team internally, and you don't actually know the end user or consumer of the product that you're making.
And so you just lose that empathy and that understanding of the customer. So it's the most generic, simple advice, but it's pretty helpful. Just talk to the person who uses the thing. What are they saying? Listen to them, solve problems for them, or do things that don't scale.
Right? It's another one of those classics, the most simple statement, but it's...
Philip Johnston:
Yeah. And again, it's the opposite of what they might teach in business school. In business school they might be like...
Turner Novak:
Yeah, systematize, automate everything.
I've used that anytime I do anything with AI. Because there's always this... You wanna immediately automate things with AI, right? But I think I got this from Garry Tan, the way he does it, you know? It's like meta prompting. You do it all manually, and then you go back and say, "What can I do to automate this process?" So it's kind of this self-reinforcing loop.
And I feel like a lot of people kind of do this, but I've learned that way of doing a lot of things manually with AI. And then at the end I'm like, what did we do here that I can kind of codify to then speed up next time? But you go through it manually the first time.
Right? And it would be awesome to say, "Hey, I'm doing this podcast with Philip, just make it all and do it." But you have to go through the steps of, oh, what are some different things we can talk about? What are different angles to approach this thing? You can't just one-shot this stuff. You've gotta... you know, it can't scale, essentially.
Well, it's been a lot of fun. Thanks for coming on the show.
Philip Johnston:
Yeah, thanks so much, man.
Find transcripts of all other episodes here.



