Talk about Starcloud, what the plans are, and yeah, we'll go from there.
I also want to know how many media appearances you've had in the last like six months, you're counting.
Yeah. Too many. Too many. I need to like spend more time.
Which one is the most important one? This one.
This one. Clearly this one. This is definitely the coolest one.
If people think this is like a green screen, this is like real life stuff going on.
So I'm Philip Johnston. I'm the co-founder and CEO of Starcloud and we're building data centers in space to address the AI energy bottleneck. We just launched our first — or six months ago we launched our first spacecraft, Starcloud One, which has the first in-orbit H100 on board. It's about 100 times more powerful GPU compute than has been in space before, and with that we were the first to do a bunch of firsts. So we trained the first model in space. We ran the first version of Gemini. We're now doing high-power inference on SAR data from Capella. And yeah, we're now ramping up for our second launch. In eight months we're launching Starcloud Two, which is about an eight-kilowatt satellite, about 100 times the power generation of the first one, and we'll have by far the largest deployable radiator in space. And yeah, that's basically us.
Awesome. Well, we've been impressed to see the growth in recent months. We just had Ian on and he is manufacturing satellites at scale. He has a five-kilowatt satellite. He's going to ramp up to 20 and 100 kilowatts, but he was just discussing how difficult it is to go from a smaller, maybe eight-kilowatt, to your eventual goal of a 200-kilowatt satellite. So if you can kind of discuss what that looks like, what the cost of building a 200-kilowatt satellite would look like, and how do you assemble something that's the size of a football field? Yeah. So you recently put out an incredible report. So I think you understand better than most people the complexities of driving this BOM cost to as low as possible.
So in order to make data centers in space work, you really need the BOM cost for a 200-kilowatt satellite to be on the order of like a million dollars. Like not too much more than that. And the way that you get there really is by bringing in automotive manufacturing techniques. So, long production lines, really driving the cost down as far as you can. In principle, there's nothing to stop you doing that, you know. It's really just solar panels, radiators, a box with, you know, maybe a thruster, one or two lasers. There's not much more to it than that. Which none of those components, in first principles, need to be that expensive. But until now, SpaceX is really the only one that's achieved that. So yeah, companies like Apex are doing incredible work. We're hoping to follow in their footsteps.
We can get back to those kind of specifics on satellite manufacturing, but I want to get your take on the terrestrial issue. We're hearing talks on a moratorium of data centers in the US. Can you just outline what you're seeing and how that plays into your thesis on putting data centers in space?
Yes, I mean, Maine just banned all data center construction. Not even a moratorium, just a ban, an outright ban. And we're going to see — I think there's now eight states which are very close to a ban or have already passed. And then there's a whole bunch more which are considering moratoriums. You know, it wouldn't surprise me if it becomes close to impossible to build new data center capacity in the US over the next five to ten years. And if that's the case, then space is definitely a much more viable option for them.
So on Earth you have a terrestrial data center that can last 20 years. This is a common question that people have, and I think because I've done a lot of research on it, obviously you have too, on the annual cost of terrestrial data centers. So some of the questions around it say, 'Hey, a terrestrial data center might be hard to put up, but you can have it for 20 years. A data center in space lasts five years, has to deorbit, new data center constellation goes up.' Can you just talk about the dynamic of that, of the costs that go into a terrestrial data center and how it can kind of conform a little bit to your five-year cycles of satellite manufacturing?
Yeah. So for a terrestrial data center now, just for the infrastructure, you're looking at around 15 to 20 million dollars per megawatt. We're targeting an all-in infrastructure cost of less than 5 million dollars per megawatt. And then we're also looking at — I mean, that doesn't include the fact that we then save a million dollars per year, or close to a million dollars per year, on energy cost. So actually, as long as we're within 25% — yeah, if we're 75% cheaper and it's 20 years versus five years, then that kind of breaks even. I would say though, many terrestrial data centers are actually doing CapEx upgrades way more frequently than every 20 years. And the reason is every time a new chip comes out, you know, if you have the Rubin chip, it's way more power-dense than any chip before it. You're basically ripping out all of that cooling infrastructure and replacing it. So I think 20 years is a bit optimistic, I think, on replacement of CapEx for terrestrial data centers. And then beyond that, we actually can probably run the chips for longer in space once the satellite is large enough that the shielding is not a huge percentage of the total mass. There's a case to be made that given that our marginal cost of energy after launch is zero, rather than stopping at five years — you know, terrestrial you stop at five years because the chips don't produce more than say 10 cents per hour for 1 kW of compute. Whereas in space, because we have near-zero marginal energy cost, you can potentially run it for longer.
Is there a world — so obviously SpaceX's big advantage outside of their manufacturing capability, which is obviously a big advantage, is launch. Would you ever consider partnering with another launch company, or have you considered it, or would you consider an acquisition, M&A, to kind of own that and start to vertically integrate that stack?
It's definitely something we would consider.
Which one would you consider?
But I mean — He wants to start his own. One thing I can categorically confirm is that we are not starting a rocket company.
I mean, Stoke Space is just down the road. So, you know, if there was to be a partnership there with those guys, I think that'd be great. We also hope to continue to be huge customers of SpaceX going forward. They're going to be first to the reusable upper stage launch vehicle form factor probably by five years before anybody else. So certainly we're keen to have a great relationship with those guys. The reason SpaceX has been successful is because they have both Starlink and the launch — Yeah, same entity. So maybe it's a JV, maybe it's a merger, but I think certainly there's a world where it makes sense for people with launch and people with need for launch to team up.
