Y Combinator Startup Podcast
00:16:44 · S1, Philip Johnston, S4, S2
[00:16:44] S1: which was... How far out was the launch? [00:16:46] S0: It was eighteen months out, and then it got pushed to twenty one months out. And we were like, okay. Something is gonna be on that rocket. I'm not sure, 100% sure what it's gonna be at this point, but something is gonna be on that. Yeah. [00:16:55] S4: Oh, that's interesting. [00:16:56] S2: Sounds like the same advice for all the software companies. You just gotta keep launching and launching. Same thing. [00:17:01] S0: Yeah. Yeah. And what what we initially planned to launch was very unimpressive, actually. Was like we were gonna put a Jetson chip, which has been flown on many satellites before, and we were gonna use it for for kind of edge processing and things. We actually had a cofounder change quite early on when we got Adi from SpaceX. Adi was like, oh, that's that's lame. Like, let's let's put an H100 on there. We're like, Adi, you can't put H100 on a on a satellite. He's like, sure you can. The amount of people who said it was physically impossible to run a pack... Like, chip that's that power dense in orbit. Like, people just thought it was, like, literally impossible to run data center grade, you know, GPUs on orbit. [00:17:40] S1: What would have caused it to be impossible?
00:17:40 · S1, Philip Johnston
[00:17:40] S1: What would have caused it to be impossible? [00:17:42] S0: There's two things. One is the thermal side of things. They're they're powered in, so they produce a lot of heat. And the second is they've never been tested in the radiation environment in space. And both of those... So for the first one, like, it's kind of a wacky solution. Like, we submerged the entire thing in this phase change material. So everything on StarCloud one is is submerged in this, like, phase change material. So it's like immersion calling for all of the components, all of the power delivery memory, everything is submerged. Nobody's ever considered doing anything like that in space before. It's not a particularly scalable solution. It's not like you have quite a low duty cycle. You have to wait for it to melt, then then then it solidifies every so often. But it enables us to prove that you can run the H100 in space. The the next one is this director chip liquid cooling architecture that we can run continuously. But, yeah, I mean, I I give full credit to my cofounder Ali for coming out with some pretty wacky and crazy ideas. [00:18:37] S1: To get to a data center that's, like, commercial scale, where you can actually, like, make money on it, it's actually, like, profitable, it's gonna require many steps. How did you, like, think about, like, sequencing out, like, the journey to get there?
00:18:37 · S1, Philip Johnston, S2
[00:18:37] S1: To get to a data center that's, like, commercial scale, where you can actually, like, make money on it, it's actually, like, profitable, it's gonna require many steps. How did you, like, think about, like, sequencing out, like, the journey to get there? [00:18:49] S0: So as I say, first step was book whatever the, like, first launched lowest MVP we can have available. Second step is producing something which is a com... Like a product we can sell to customers. So that's StarCloud two. That's a 10 kilowatt spacecraft that we can sell compute to government and military satellites. And then the third step is the product that we're gonna be able to sell to hyperscale data centers. So that's the StarCloud three. We can... It's 200 kilowatt, three ton spacecraft. That's six meters long. With that, we can fit 50 of them per Starship, so about 10 megawatts of new compute capacity per Starship. And we've just filed with the FCC for a constellation of 88,000 of those, so that means on the order of 20 gigawatts of new compute capacity. Wow. And we can fit up to 10 terawatts, in the Dawn Dust Sun Synchronous Orbit. So that's, like, 20 times the entire US power grid. [00:19:39] S2: For context, what's the largest data center deployment on Earth? Because you're talking about 22 gigawatts. Oh my god. What's what's the largest one here on Earth today? [00:19:48] S0: They're about a gigawatt is the is the largest. Yeah. [00:19:51] S1: What's the timetable for all of this to become real?
00:19:51 · S1, Philip Johnston, S3
[00:19:51] S1: What's the timetable for all of this to become real? [00:19:53] S0: Yeah. It's very dependent on the launch cost. And so, you know, this this estimates for Starship, if you believe Elon, it'll be end of this year, but let's say, it could be end of next year, early twenty twenty eight for Starship, deploying, you know, ramping up the launch cadence. They're actually phasing out Falcon nine, so, they're gonna be moving over commercial customers to Starship. There's also a whole bunch of other launch vehicles coming online. Stoke Space has the Nova rocket. Rocket Lab has Neutron, New Glenn. Blue Origin has New Glenn. So end of twenty twenty eight, I'd say, is when we're starting to ramp up for the terrestrial business, competing with terrestrial data centers. [00:20:31] S3: So your customers are actually anyone who wants compute. And then initially... [00:20:35] S0: Initially, it's government and military, but then as Starship ramps up. So for the next two, three years, we're... We've just won four contracts with various DOW and government entities. [00:20:45] S3: And then initially also, it's people who want that compute actually, like, for other purposes in space then. Yes. Is that right?
00:20:45 · S3, Philip Johnston
[00:20:45] S3: And then initially also, it's people who want that compute actually, like, for other purposes in space then. Yes. Is that right? [00:20:52] S0: Yeah. Exactly. So right now, many satellites are very constrained with the amount of data they can downlink. And so using optical terminals, we'll have three optical terminals on our second satellite. They can ship enormous amounts of raw satellite imagery, and SAR data, synthetic aperture radar data to us. We then process that on orbit, and then we can just downlink very quickly the insight from that. And the insight might be the coordinates of a vessel or something like that. [00:21:16] S3: In terms of NVIDIA, I mean, are you... How are you partnering with them around chip design? And, you know, are there changes that you have to make to the design to... I mean, as you rescale, like, being... Making it more radiation resistant, or, you know, what does that look like?
00:21:16 · S3, Philip Johnston
[00:21:16] S3: In terms of NVIDIA, I mean, are you... How are you partnering with them around chip design? And, you know, are there changes that you have to make to the design to... I mean, as you rescale, like, being... Making it more radiation resistant, or, you know, what does that look like? [00:21:31] S0: Yeah. So we're we're working very closely with NVIDIA. They... We recently announced this Rubin... Space Rubin one chip that we're working with them on. So that's designed for the space environment. We very heavily modified the H100 to make it work in space. So on the mass side of things, we stripped out a lot of things like AC to DC converters, the, the direct... The, cold plates and everything else we sort of stripped out. And then we stiffened the board and made it radiation tolerant. So because of that, were they... They're very interested in, you know, with any people, basically, that have any data on how one of these high powered GPUs operates in space. So, yeah, we're working with them very closely on, designing this new space chip. Because you you can actually do very simple things to make these way more effectively... Effective to run-in space, and so that's what we're doing. [00:22:19] S3: YC's next batch is now taking applications. Got a startup in you? Apply at ycombinator.com/apply. It's never too early, and filling out the app will level up your idea. Okay. Back to the video.