Because Starship is the first ever fully reusable rocket. That's really what changes everything. We know that what we're doing works because the International Space Station has a radiator that dissipates about 70 kW. The problem with the ISS radiator is it is both expensive and heavy. And so the core challenge for us is to make this thing cheap and light. And that is a manufacturing challenge more than anything else. Our YC launch was the most viewed launch ever. I think it was like a few million views and it was universally negative comments on there. There was a guy called Scott Manley. Do you know him? He's a space influencer.
So, he made a 20-minute takedown video of us like as we were going into demo day. We had like at least 20 VCs say no. And we were like, maybe that was a bad idea to do that announcement and things. And then a couple of them started to say yes, and we picked the one we liked the most.
This week, Philip Johnston, CEO of StarCloud, visits Mike, Misha, and Shing Huay at MIT. StarCloud is building data centers in space and just raised $170 million at a $1.1 billion valuation. The idea has sparked widespread criticism online. So, we dive into the scientific and economic details to better understand whether it can actually work. Johnston walks us through the physics, opportunities, and challenges of building a startup that put the first data center class GPUs into space on a shoestring budget. Please welcome Philip Johnston.
StarCloud is a company that is building data centers in space, initially to provide cloud and edge services for other spacecraft, and then so particularly DoD and Earth observation spacecraft, and then later...
DAW is a DoD, the new DoD acronym now.
If you want to get any contracts with the DoD, you have to call it DAW. All right. Sorry for that.
Yeah, if you're looking for more money from them, probably DAW is a good idea. So, and then later on a sort of two, three year time frame as Starship ramps up production to compete with all terrestrial data centers on energy cost. So you can think on a 5 to 10 year time frame, we'll be launching hundreds of Starships per month. Each Starship is about 10 megawatts of new capacity. So we're talking about several gigawatts of new capacity per month we can launch and tens of gigawatts per year. So yeah, with that, that's the end ambition is to be essentially to move almost all compute into space for the energy. So we're very quickly running up against constraints on building new energy projects terrestrially and the reason we're doing StarCloud is that this addresses the need for new energy projects.
And the energy comes mostly from solar cells. Do we have enough solar cell supply?
Well, right now China is installing, you know, at least tens of gigawatts per year, maybe close to hundreds of gigawatts or 800 gigawatts per year of new solar capacity. So I mean solar is by far the most abundant source of new energy that we have for sure.
Everybody who's on Twitter or whatever is aware that there's this feeling that there's not enough data centers and there's some kind of collapse happening where you can't build enough. I guess if you break down that problem, maybe we can start with why isn't it possible to build more data centers on Earth, and then we can go to more like how do the economics work once we agree that we can't do it on Earth, then the question is like how do we do it in space economically.
Yeah, I mean maybe since we have time we can just start even one level higher, which is sort of why do we need more data centers. And I'm sure it's kind of obvious to you guys, but there's two reasons. One very obvious, which is if you want more chat, more video generation, more self-driving cars and all that kind of thing, you need much more new data centers and much more energy projects to power them. The slightly less obvious one is the competition from peer nation, near peer adversaries as they say in the military, which means China. And so China is now building new energy projects at 10 times the rate of the US and the central government is worried, and rightly so, that this will sort of inevitably lead to China having 10 times the AI capability of the US. If you imagine a company like Palantir in China having 10 times the AI capability, that's obviously not a great thing for the US. And so there is a national security and a sort of defensive reasons to do this as well.
So the reason we can't do more new energy projects on Earth is mainly if you want to build a new, let's say, 100 megawatt energy project on Earth, which is not like crazy big, as you know many data centers in the US are bigger than 100 megawatts, but you're looking now at like a 5 to 10 year long lead time on permitting. And the reason is not many people want... So the new types of energy project that we're looking at would be nuclear, solar, and let's say natural gas or coal. Not many people want a new 100 megawatt nuclear project next to them. And if you want that amount of solar, you're talking about covering many square kilometers of countryside essentially or desert with solar panels. And most of the desert in the US is protected either as national park or as tribal land or as somebody cares about you covering that land with solar farms essentially. And so if you want to deploy new, if you want to build new energy sources quickly, the only place you can really do that is in space.
And that's because there's a permitting problem, but in space you do have the 50% over six-year rule of deployment, something like that, which could also affect your business model, I guess.
So yeah, it's a great point. So, when you file for an orbital slot with the FCC, once you've been granted that slot, you are required to fill it within a certain time. If you don't fill it in that certain time, there's basically no penalty. You just refile.
Unless somebody else has filed in the meantime, but so you don't really pay anything for those orbital slots either. It's not like you're paying the FCC for them, you submit a filing fee, but beyond that, there's no other expense to getting those orbital slots. Whereas with land, obviously...
One of the three biggest costs of terrestrial solar is number one, cost of permitted land. Number two, the cost of battery storage and backup power. And number three, the cost of the solar cells themselves. Well, in space we don't need permitted land, so your biggest cost is gone. We don't need batteries and backup power, so your second biggest cost is gone. And we need eight times less solar because one square meter of solar panel in space produces eight times the energy of one square meter of solar panel on Earth. And so there's a...
That comes from the 24-hour day and like no absorption from the atmosphere, or is there a third?
The third is sort of seasonality. If you're anywhere besides the equator...
Just the radiance changes basically.
So your capacity factor is close to maximum.
Yeah. So we're at like 90, 98% whereas terrestrially you might be at like 15% if you're lucky.
Over the course of a year.
Okay. But let's just... the idea that we don't have any space, I just want to, you know, I feel like I'm not completely satisfied. So let's say in America we're too wealthy, we don't want to give up, you know, parts of our desert or whatever to, you know, but I mean what about Kazakhstan or like Antarctica? Like I feel like...
Antarctica is a problem because you don't get much solar there anyway. You get very weak, you don't get great solar even in midsummer where you got 24/7. You're not getting very high power solar because it's going through a lot of the atmosphere.
You could do that. You could do it in Kazakhstan. You could potentially do it in the Sahara Desert too.
US companies are not comfortable or are not even allowed in some cases to store their US customers' data outside the US and to process it outside the US. So they do need it to be... they're allowed to do it on satellites, but I mean Starlink does it on satellites, but they're not allowed to do it in Kazakhstan or in the Sahara Desert.
What about waters, ocean?
International waters? It's possible. Yeah. If you want to build massive solar projects out to sea, it's possible. They're way more expensive than doing it in space. Way more expensive because like the...
It's a very harsh environment, right?
It's an extremely harsh environment. The salt and everything else.
Yeah. So in a way it's like somehow if you're doing solar, yeah, anything electronic, right?
Well, okay, what about, I mean just coming up with stupid ideas. So what about you put your compute on the border with Mexico and you put your solar on the other side of the border because the compute is small relative to the solar?
I'm sure we're going to do that, but there's only so many parts of Mexico you can cover with solar panels too.
Yeah. I mean, I'm sure the US is going to start buying power from Mexico if Mexico wants to cover their desert in solar panels. I'm sure.
So, basically, even then at the limit, it's just basically... and there's also, you know, the space feels like it's straightforward, predictable. As long as you can get there cheaply enough, you know what's going to happen. There's no human factor kind of.
Yeah. Exactly.
Okay. And I mean even if you solve it, so even if you say okay cost of permitted land is free, you still have the cost of batteries, backup power, right? Chillers, especially where you want to build solar farms is usually hot, which is why you don't want to build data centers, so transmission then becomes a problem because data centers need a lot of cooling. So yeah, in space the cost of infrastructure is essentially batteries, backup power, cooling towers, chillers, and a few other things. Depending where you are, AC to DC converters and things like that. In space, we don't need any of that. We just need these dirt cheap solar panels and dirt cheap radiators.
So, you buy solar panels off the shelf. You don't need...
Standard terrestrial silicon. Yeah.
Is the assumption that you're going DC solar panel to DC power for the GPU? There's no in between. There's no AC in between.
There's a step down but there's no...
Yeah, there's a voltage step down but no AC in between.
What's the... let's say if we can compare solar cell versus nuclear plants and how this kind of energy path can be expanded for data centers.
So the lowest cost I've ever seen for a new nuclear project is around 8 cents per kilowatt hour, factoring in the cost of building it. For a depreciated nuclear plant, you're talking about four or five cents per kilowatt hour because you still have to pay for the uranium operating cost.
In space, we're targeting much less than half a cent per kilowatt hour. Like potentially 0.
That's including depreciation, launch. That's assumed launch cost with Starship.
But the assumed launch cost was something like $500 per kilogram.
The break even is 500 bucks a kilo. For that estimate we're talking about say five times the marginal launch cost of Starship for SpaceX which is around the 50... they're expecting $10 to $20 a kilo is their marginal launch cost. So if we get charged say 5x that, that would be 50 to 100 dollars per kilo.
But maybe yeah, maybe to run a comparison with say like let's say you have a way to get power for 10 cents per kilowatt hour on Earth which is like, you know, for a new gas project that's certainly the range you're looking at because if you want to build a new data center in America you're really almost all of them now build with gas turbines because it's the only one you can scale fast enough. So with gas turbines you're talking about 10 cents per kilowatt hour the energy cost alone.
Let's say for a 10 megawatt data center, and I picked 10 megawatt specifically because it's what we can fit in one full Starship payload bay very conservatively, that's with the Starlink V3 form factor, so 200 kW per flat sat spacecraft. Let's say your energy cost alone is about 50... is close to $50 million over 5 years, which is how long the chips are going to last in space.
Let's say we do have the five times or 10 times Starship marginal launch cost, so $5 million for the launch. We're targeting $4.5 million per megawatt for infrastructure cost. So $45 million for the infrastructure. And so yeah, we're saying $50 million for 5 years that gets you the same price as the electricity cost, but that means we don't have to pay the infrastructure cost.
So infrastructure costs for data centers on Earth, and that's things like chillers, cooling towers, batteries, backup power, just that alone is about 15 to 20 million per megawatt. So we're basically getting all of that for free.
But to what extent are you sensitive to or dependent on other missions because this is like economies of scale. So the cost of Starship really depends on there being a lot of demand. Does your own business already satisfy that demand? Like would StarCloud itself generate the economies of scale that would then bring the cost down or do you depend on other missions?
We can consume almost unlimited amounts of launch capacity. So...
So you could drive the launch cost down even just with...
Assuming there's a market for the compute.
Well the compute, there's a market for it so long as compute is being done anywhere, it will be being done cheaper in space than on Earth. And so...
Yeah, maybe we can take a step back and just do the very basic like let's say I purchase a GPU, right? Let's do this on a single unit basis, right? So I buy a GPU. What does that cost for how many kilowatts of power that it consumes? Like kilowatt hours per year, however you want to do the math.
So let's take one Blackwell chip which is about a kilowatt. So what you don't want to take into account is the chip cost. So it's very important that we don't have a higher failure rate on the chips than we do on Earth because the chips are actually by far the largest cost of what we're doing in all of this. So let's just say the chip cost is the same in space as it is on Earth and we're working very hard to make that the case.
Yeah, that's a whole separate can of worms like radiation and everything. But okay, let's assume you're able to take your GPU which is 1 kilowatt and what would it be? What's like the scaled cost of that device? Like if you're buying whatever the numbers are.
The chip itself is about $30,000 per chip.
But as I say, let's ignore that because it's going to be the same cost on it as it is.
Oh, no. But I actually think it's an interesting question because like what percent of the whole... if you amortize it, I don't know, over 3 years or however long it is, 5 years, if a lot of the cost is the chip then like the energy question becomes more of a bottleneck but not like an economic...
The most of the cost is the chip.
Okay. So even over... so like how much energy, let's say if you're doing this on Earth, what's your capex opex like ratio over five years?
Let's do megawatt because it's the number that I have closest to my head. So one megawatt of infrastructure cost on Earth is 15 to 20 million per megawatt.
And then your energy cost again is another... it's about a million dollars per megawatt year. So another over five years another $5 million for the energy.
Okay. If it had 10 cents per kilowatt hour.
Okay. So basically if you were getting completely free energy, it would be, you know, 30% cheaper or something over a 5-year period.
Yeah. But the key is the infrastructure is also cheaper in space.
Ah, so you're also... what's your cost?
Ours is $4.5 million per megawatt for the infrastructure. So solar panels and radiators because we don't need chillers, cooling towers, batteries, backup power.
When you say infrastructure, that doesn't include the chip.
Okay. Yeah.
Dude, how many... if you're doing a megawatt, how many chips is that?
It's about a thousand kilowatts.
So, okay. So roughly it seems like on Earth it's like roughly a 50/50 where like half of your cost, let's say you're amortizing over 5 years, half of your cost is going to be this infrastructure and half of the cost will be the chips. Something like this.
So you're no matter what the chips...
Well, so 30,000 is... yeah. $30 million worth of... yeah, exactly. So it's roughly like 25-30 type...
And then you said per year on the infrastructure it's like about 15 to 20 million.
No, no, that's over 5 years to build it.
Okay. So it's 50/50.
So if I take the whole pie, right, it's like 50% is already the chip and at best you're going to probably match that, right, because it has to be radiation hard and everything, but let's assume it's matched. So that's... you're kind of stuck with that cost and then everything else you're competing on the other half and then on the actual ability to do this at all. Right? Like a binary like can we even scale this?
Yeah. Yeah. The reason why the chip is more expensive is that the infrastructure usually can last for say 15, 20 years while the chip is only 5 years. So you have to replace that chip every 5 years.
Right. Right. Right. What's the cost of operation to do that? Like to factor that in, to replace the chip, you're talking about another $30 million every 5 years.
So basically like assuming you get the infrastructure cost down to 5 million, let's say, on this megawatt...
And the energy cost is like essentially free for you guys, right? Once it's up there. So you're essentially cutting like 75% of the... it's 75%. Okay.
But in the end that ends up being like 40% of the total because you still have to amortize the chip. Okay.
So how much is the launch like these days? Let's say per kilo.
What's the...
Right now we're paying $6,000 a kilo for launching on Falcon 9 on the rideshare program.
Yeah. And that used to be $60,000 a kilo not too long ago.
Okay. And then for that megawatt infrastructure, how many kilos do you need to send to space?
It's around one ton per megawatt. So it's...
No, wait. Sorry. No, it's around 10 tons per megawatt. So...
Yeah. 2 and a half tons for 200 kW. So 10...
$60 million for deployment.
No, but then they're expecting the cost to come down. Falcon 9 is $5,000 a kilo. The break even point is around $500 a kilo.
And the marginal launch cost with Starship is projected to be $10 to $20 a kilo. So because Starship is the first ever fully reusable rocket, that's really what changes everything at the moment. It's like flying a plane from London to LA and you have to rebuild the plane every time. That's how expensive it is.
So you need to decrease like 10 times.
The launch cost needs to come down by 10 times for this business model to make sense.
Mhm. And it gets very interesting if there's like another zero and another zero.
Yeah.
But the expected... you're saying that like the hope is that it comes down to $50. Yeah. Something like that. Okay. So that's basically a 100-fold cost reduction.
Yeah. And it could go down even further than that. I mean, but yeah, we're break even anything below about 500 bucks a kilo. And arguably, you know, given that we can deploy faster, if we hit that break even point, there'll be a lot of incentive to switch over.
And a single rocket fits, you said 10 megawatts.
Okay. So, if you're doing like a gigawatt data center, that's 100 launches.
Yeah. So, potentially, you know, once Starship is flying at rate, it'll be doing that every month.
Okay.
What's the weight ratio of solar panels, radiators, and I guess the chips are probably the lightest part.
Yeah, chips are pretty light. It's about a third, a third, a third. So, a third satellite structure and chips, a third radiators, and a third solar panels including deployment.
So, it really makes sense for you to get the weight of the radiators down or the solar panels like...
Yeah. Exactly. And that's where Stefan Boltzmann comes into play because what you want to do is run these chips as hot as possible. Basically, you want them to be as power dense as possible.
So, you want to spread the...
Yeah. So, because if you can increase the temperature differential by only 10% or just over 10%, you're close to halving the surface area of the radiator. Because it's T to the fourth for the...
Yeah. Yeah. So let's say instead of running your chips at 30 degrees, you can run them at 60 degrees which is like a 10% increase in Kelvin. Then you're halving your surface area in your radiator.
What's the prospect of doing that? Like does that reduce the lifetime of the chip or like basically does this... for this to really work at scale, do you need very bespoke... like you basically need a new chip essentially? Like how different do you expect it to be from whatever's coming out of the existing supply chain?
We've baselined all our calculations with just using off-the-shelf H100 or B200 or any of those kind of chips. I don't know how much I can say about this but one of the largest AI chip manufacturers in the world is very seriously looking at doing space inference chips.
And we're working with them quite closely because we're the only people that have data on how those chips perform in, you know...
So you mentioned that you don't need chillers for example in space.
We need radiators and radiators are way cheaper.
So but I guess my question is if you can cool the satellite in space in a cheaper way than a chiller on Earth, like why can't you use that on Earth?
If you aim it at the sky it's going to hit the atmosphere first. And the atmosphere... what you need is a gradient. So you need...
The differential between...
Where you're aiming your infrared at and if you're aiming at the sky, the air in the sky is giving your infrared back.
Yeah. So usually... so I guess a lot of these things maybe aren't like as... they're certainly not going to be as efficient as just plainly being in space. But there are all these efforts to make things that radiate in frequencies where the atmosphere is transparent. But I guess in terms of that's more of a science challenge whereas what you're talking about is like an engineering... Can I put a radiator in space and just aim it at the sky?
Okay.
Yeah. What we're really talking about is a manufacturing challenge actually. It's not even so... like we know that what we're doing works because the International Space Station has a radiator that dissipates about 70 kW. The problem with the ISS radiator is it is both expensive and heavy. And so the core challenge for us is to make this thing cheap and light. And that is a manufacturing challenge more than anything else.
So we've got a radiator design and we've fabricated it. We're going through TVAC testing now and it's 10x less mass per watt of dissipation than the ISS radiator and 100x less cost per watt of dissipation. And the reason is the ISS uses pretty... like they're mainly doing additive manufacturing, like this 3D printing technology, to have these very fine channels in their radiator. We can do it... like we've got a technique that's just so much simpler and easier than that.
Can you talk about that?
How is it? Maybe you can talk.
How does the Starlink system do cooling?
So Starlink actually doesn't need as much cooling. Almost all of the energy...
It's going into the actual internet data.
Exactly. It's going through the antennas.
So almost all of the energy they collect on the solar panels comes out through the antennas.
Do you know what's the percentage?
They do run them pretty hot. So I mean there is definitely a lot of waste heat that goes into the bus, but the bus is pretty large and so they don't have... as I understand they don't have active cooling like fluid loops, but the bus itself is pretty large and dissipates the heat.
So what's the thermal rejection rate then in the end that you end up with with your radiator?
So yeah, it really depends as you mentioned with the Stefan Boltzmann, it really depends on the temperature, but if we keep it at 50 degrees C, we're talking about 6 to 700 watts per square meter on the radiator. So we're talking about just essentially aluminum panels.
And then how large are those? Because then you have to balance...
The solar generates about 200 watts per square meter. And so we're talking about a third again in surface area of solar with radiator, which is really not that bad. Like some people think, oh, you're going to need a radiator that's like 10 times the size of the satellite. It's like, no, actually, we just need to add another third on the existing surface area that we have with the solar panel.
Mhm.
Oh, and unfortunately, it can't be like on the back of the solar panel because it's going to point at Earth. So you have to have it... or is it kind of... are you able to point it in a way where one side is pointed at the sun, the other side is pointed at space or...
But you're also losing half of your...
So this is a bit of a misconception. So people think, and I also thought this, that if your radiator is directly in the sun then it's going to be absorbing infrared. That's actually not true. You can dissipate 80% towards the sun as you are doing away from the sun. And that's... you just need a very custom coating that's very reflective but also because most of the sky is not sun, most of where you're dissipating the infrared is not sun.
So and to your point like if this is the Earth, we don't fly the satellites like this over, we fly them like this. Like that's let's say that's the sun, this is the Earth, they're not coming over like this, they're going around like this.
So they are always in the sun. The way we've designed it is we just have the first segment of the deployable is radiator and the second segment is solar panel.
Okay. So they're actually completely... they're essentially separate.
Completely separate. Yeah.
Okay. Yeah, maybe you can talk about that. So you guys, when you came out of YC, you launched a GPU into space and trained an LLM. What's the story behind that?
Yeah. Well we just did that last November actually. So when we came out of YC summer '24 and then a year later we launched our first spacecraft. Yeah, it took us 21 months.
When you exited YC, what was the physical state of the hardware? Like did it exist at all? Did you just have like a cardboard box you would show people? Like where were you on that scale when you were doing demo day?
So I'll give you the timeline. So we did demo day September 20, 2024. We founded the company January 1st, 2024. January 2nd, we went on spacex.com and we booked whatever the first launch we could book was, which was like at that point I think it was like 18 months away. And then we were like, 'Okay, we're going to have something on that. I don't know what it's going to be, but there will be something flying on that launch because it only cost 300 grand to book a launch for a 60 kg spacecraft, only 300 grand.'
Then we contracted a company called Astro Digital to manufacture a very bespoke bus which allows us to fly lower for radiation and things like that. And then we manufactured... we had not manufactured any of the payload but we had all of the tools. So the hack for putting up like an off-the-shelf GPU was actually flying lower where you have some protection from the atmosphere.
That was one of the hacks, but we actually didn't do it in the end. We've got these thrusters on board which will allow us to do that. But we've been running the GPU totally fine where we got dropped off, which is like 550 km altitude. We've got the capacity to lower it to 380. That's what we plan to do, but since it works fine at 550, we're just like, okay, kind of leave it there.
So yeah, then when we came to demo day, we had all of the CAD models of what the spacecraft was going to look like and we knew roughly what the payload was going to look like and then we manufactured the payload. So that's all of the compute module. We submerged all of the electronics including all of the power and motherboards and everything else. We actually had five Nvidia GPUs on there. The H100 is one we talk about, but we had a bunch of other stuff. Submerged it all in this phase change material like a wax that kind of melts, paraffin wax, and then shipped that down to be integrated with the bus. We shipped it about 3 months after Y Combinator demo day.
That's impressive.
Yeah. I mean the fastest any company's gone from day one founding to having something on orbit before us was four years.
And we got a quote from one of the primes for what a spacecraft of the specs we had would be and they said 75 to 100 million dollars and we did it for like $2.5 million.
Because the launch was 300 grand, bus was a million dollars and then Adi and Ezra, my co-founders, basically built the payload by hand.
So...
What's the reason it's in wax? Is that radiation or cooling or just... but once it melts, it melts. So what's...
Yeah. So I mean it's not a very sustainable solution.
Is it still... do you train... like do you still run the computations on the orbit?
Yeah, we still run it. So...
It's just like for spikes you're basically trying to avoid spikes.
Yeah. So it also is flying in this... it's not in a dawn-dusk sun. So every 45 minutes it goes behind the Earth and then 45 minutes later it comes back up.
So it gets very cold when it's behind the Earth's shadow. So we can run it... well we don't really run it for more than an hour and then... and we can't even run it for more than an hour because we don't have enough power. The solar panel is too small. And then when it goes behind the Earth, all that wax solidifies and melts.
Oh, okay. So you kind of get... you have to cycle.
Yeah. Yeah. So what we didn't want to... the main reason to do the wax is all of those tiny little transistors and resistors on a motherboard, if they don't have an oxygen atmosphere, they'll just get way too hot and burn out. So we did that so that all of those tiny components won't burn out. For the second spacecraft we're in...
What kind of testing did you do on Earth? Like what was your feeling about whether it would run when it...
Do you want to see a video of like the testing you did? I'm sure I'm not going to show this on camera, but you guys can see how ratchet the testing on the ground is.
It was wild. Like I'm... yeah, my co... I get in trouble when I say this, but I am kind of amazed that it works as well as it does.
How many people were you when it launched? Like how many people in this...
Like well, three, including me. So...
Oh, really? Okay. So you outsourced a lot of the build or did you...
Yeah, we did. We did. Well, the whole payload... we outsourced the bus manufacturing, but all of the payload and everything. So like this is... let me... sorry people on the camera, but I'm not going to show people on the camera this. So this is us. So this was like the payload inside. This is one H100. This is an A6000. Then we've got like a bunch of Jetsons and other things. Each one has its own motherboard and power supply.
And then so we sealed this. We couldn't get it shut in the end. So we just had to like jam it shut because we 3D printed this aluminum box. And we'd not basically given it enough space because we were also constrained on the satellite bus. But let me go to the bit where we... So yeah, in the end we just had to literally jam it shut because like by the way, this is a membrane so that the wax that we pour in expands and that membrane when...
I was just going to ask that because otherwise it would just squish the component.
Yeah. And then directly behind that membrane, we have a sheet of aluminum with little holes so that on launch you don't have a mode, what they call a mode, like a resonance at a certain frequency.
And then, right, we had to put it through... we had to ship this thing. So what time is it now on here? It's 3:00 a.m. now. We had to ship it at 7:00 a.m. So we still hadn't done TVAC, we hadn't done thermal cycling on this yet.
So this is the wax that we're pouring in. Right. I don't know if you can see this.
That's a phase change material.
That's the phase change material.
It's very hot here, but so it's like a... Yeah. So then we got like a bucket of ice.
Oh my god. Oh my god. This is really...
This is do-it-ourselves. So this is a bucket of ice that we're dumping this thing in to cool it down. Yeah. I was like, are you sure electronics and ice and water? So it's like submerged in water and ice.
No, I don't know. Like...
Well, it's sealed, right? So it should...
Yeah, it is. It's hermetically sealed.
So if it wasn't hermetically sealed, then we had to heat it up. So we're using these like hair dryers to heat it up to like melt it.
So what time is this? This is like 6:00 a.m.
Oh my god.
And then we have to check in for this flight.
Look, we... And then Addi put it in his... Yeah. My co-founder Addi put this in his hand luggage and flew it from Seattle down to Santa Clara, and then we integrated it with the payload down in Santa Clara.
So that's why it cost $2.5 million and not $75 million.
So when you were going through YC, where were you in terms of people being interested? You know, were you like in the middle of the distribution, on the low end, on the high end? Like, what was it like before you guys had the launch?
Once we got into YC, I'd say we were on the higher end of interest. We had $3 million uncapped before we got to demo day. We were very much on the low end before we did YC. So when we raised the 2 million at Tempest, it was like an absolute struggle. Like we were going for months and we scraped it together. Once we got to YC, things got much more interesting. Then just before demo day, we released the white paper and we released this concept video of this huge 4 km solar panel and things, and all of space Twitter decided they hated us. And like there's a guy...
Oh, they absolutely panned us. Like if you go back to the initial... Oh, like YC launch was their most viewed launch ever. I think it was like a few million views and it was universally negative comments on there.
Is it something like Silicon Valley bro wet dream? It's like YC is at it again. Like fake, fake, fake space stuff out of YC. Like you guys are just burning, setting cash on fire now. You like him? And there was a guy called Scott Manley. Do you know him? He's a space influencer.
So he made a 20-minute takedown video of us like as we were going into demo day, and basically all of the VCs, we had like at least 20 VCs say no, and we were like, um, like maybe that was a bad idea to do that announcement and things. Um, and then a couple of them started to say yes. Uh, and we picked the one we liked the most.
And now there's a lot of people who like, including SpaceX, who say that this is the future for data centers. How do you feel about that? The tides turning.
Yeah. And then since Elon came out, like half of the VCs now say they're conflicted because they've got like massive SpaceX positions and they don't want to do anything that they think could jeopardize that before the IPO.
So, but the other half of the VCs are like, you know, very, very interested in...
Did you meet with Elon?
No, I've never met him. I'd love to. He like retweets my stuff and like likes... he likes my post every so often, but...
Have you slipped into his DMs?
No, maybe I should. No, I think my DM... I don't think he would get my DM if I messaged him.
Do you feel like, you know, taking on more capital like in the extension, do you feel like there's a risk of being overfunded? Like there's this kind of like funding curse that sometimes happens... too much money.
