podcasts / This Week in Startups / episode
This Week in Startups cover art
podcast
📅 2026-11-12 ⏱ 33 min episode · ~2 min read 📝 Transcript · 6,119 words 🎙 Rama Afullo

Rama Afullo: This Week in Startups transcript and summary

This Week in Startups · · Transcript & summary by WhipScribe

A founder who has worked at Tesla, Google and SpaceX explains, with a live demo, why some engineers think satellites could soon run AI computing for the planet.

Read the transcript
In this episodeSatlytSpaceXStarlinkTeslaGoogleGemmaNASAStarshipKubernetesVint Cerf

▸Chapters

Tap a chapter to play it from that moment in the show's own player; the transcript opens there too.

Transcribe your links. Paste a YouTube, Google Drive or podcast link and WhipScribe gives you the transcript. Transcribe a link →

▸Notable quotes

“actually only 11.7% of your data was useful for you to downlink.” Rama Afullo · at 21:53 — ▶ hear it from here
“they're paying $15,000 a kilogram for rideshare” Rama Afullo · at 16:26 — ▶ hear it from here

▸What this episode covers

Rama Afullo, founder and CEO of Satlyt, argues that computing in orbit solves problems that ground data centers face: local opposition, power limits and latency. He traces his path from a family of engineers in Kenya through Tesla Africa and Google to SpaceX, where he says he left after learning that earlier orbital-compute proposals had ended careers there. Jason frames the idea as science fiction that is quickly becoming real.

The technical case rests on satellites relaying data to each other and to ground stations. Afullo describes software that packages jobs as containers, places them across satellites and keeps each module under about 15 watts. In a demo, a small Google Gemma model runs onboard to filter data before downlink, and only 11.7% of one dataset proved worth sending. Cost is the hard part: Starship is estimated near $1,000 per kilogram to orbit, and Afullo says parity with ground data centers needs roughly $100.

The near-term uses he lists are NASA work streaming 4K video from the Moon, direct-to-device phone service that needs no change to handsets, and faster emergency response. On heat, he says both sides of the debate are right: space is cold, but without air the heat must be pushed out through large radiators, so chipmakers are building hotter-running modules. The episode closes with a sponsored read for Google for Startups.

▸Rama Afullo on This Week in Startups: the answers

Who is Rama Afullo?

Rama Afullo

Rama Afullo is the founder and CEO of Satlyt, a company building software that lets AI run across satellites. He previously worked at Tesla in Africa, at Google as a cloud solutions consultant, and at SpaceX's Starlink.

What did Rama Afullo say about Google's Gemma?

Rama Afullo said Satlyt runs Google's Gemma small language model on a satellite to filter data before it is sent to Earth. In the demo he described, only 11.7% of the data was worth downlinking, and the rest was discarded in orbit.

“actually only 11.7% of your data was useful for you to downlink.”Rama Afullo · 21:53

What did Rama Afullo say about Starship?

Rama Afullo said a typical satellite operator pays about $15,000 per kilogram for rideshare launches, while Starship is estimated near $1,000 per kilogram. He said orbital compute only matches ground data centers near $100 per kilogram, which depends on Starship working.

“they're paying $15,000 a kilogram for rideshare”Rama Afullo · 16:26

What did Rama Afullo say about Direct-to-device service?

Rama Afullo said orbital computing could beam data straight to phones without changing handset hardware, so a ground telco would no longer be required. He predicted that within about ten years telcos could become a relic, much like the landline telephone.

“They become a bit of a relic like the landline telephone”Rama Afullo · 14:57

What is Rama Afullo's main argument on This Week in Startups?

Rama Afullo argues that putting computing in orbit can solve problems that ground data centers face, including local opposition, power limits and latency. He says satellites can filter data onboard, that launch costs must fall to about $100 per kilogram before orbital compute matches Earth, and that the approach will eventually reshape telecom.

Mentioned in this episode

Episode facts

Released
November 12, 2026
Length
33 minutes

▸Key takeaways

01Satlyt CEO Rama Afullo, a former Tesla, Google and SpaceX Starlink employee, says his startup runs software that lets AI work across satellites.
02Satlyt's orchestration software places containerized jobs across satellites, like Kubernetes for orbit, and ran Google's Gemma model on a satellite.
03Afullo says only 11.7% of one demo's satellite data was useful, while ground stations charge $10 to $20 per gigabyte to downlink it.
04Afullo says a typical satellite operator pays $15,000 per kilogram for rideshare launches, far above the roughly $1,000 per kilogram Starship is estimated to cost.
05Afullo argues orbital compute only matches ground data centers near $100 per kilogram, which depends on Starship and Blue Origin's New Glenn working.

▸Questions this episode answers

Q1What did Rama Afullo say about orbital data centers?

Rama Afullo says moving compute to orbit solves three problems: local opposition to data centers, power limits on the ground, and latency for remote users and emergency response. He argues satellites in sun-synchronous orbit get near-constant sunlight, and that low Earth orbit offers far more usable space than land on the surface.

Q2How much does it cost to launch computing into space, according to Rama Afullo?

Rama Afullo says Starship is estimated at about $1,000 per kilogram to reach orbit, while typical satellite operators pay around $15,000 per kilogram for rideshare. He says orbital compute only reaches cost parity with ground data centers near $100 per kilogram, so large launch vehicles such as Starship and Blue Origin's New Glenn must reach full operation first.

Q3What did Rama Afullo say about running Gemma on a satellite?

In a demo, Rama Afullo said Satlyt runs Google's Gemma small language model on an edge computer in orbit. If a ground link fails because of cloud cover or interference, the system relays data to another satellite or ground station, diagnoses the problem and shrinks the file. In the demo only 11.7% of the data was worth downlinking.

Q4What does Rama Afullo say about NASA and streaming 4K video from the Moon?

Rama Afullo says Satlyt is working with NASA to relay 4K video from the Moon to Earth through satellites in lunar orbit and lower orbits. He says the networking protocol, built with the NASA Glenn Research Center for deep space, is designed to limit packet loss and bit flips. He thinks daisy-chaining satellites out to Mars may come later, after current technology is proven.

Q5How does Rama Afullo explain the debate over heat in space?

Rama Afullo says both sides of the heat debate are correct. Space is cold, but without air the heat generated by chips has to be pushed out through large radiators, like those on the International Space Station. He says chipmakers are building modules that run hotter and draw around 3,000 watts, and can throttle when needed.

Q6Why does Rama Afullo say orbital computing could change telecom?

Rama Afullo says orbital computing could beam data directly to phones without changing the handset's hardware, so a ground telco would no longer be needed. He predicts that within about ten years telcos could become a relic, much like the landline phone, and says the race for direct-to-device service is already underway.

▸Transcript

Our own transcript of the public episode, with a timestamp about every 90 seconds. Tap a timestamp to play the episode from there in the show's own player; the link button on a paragraph copies a link that opens this page there.

0:00All right, everybody, welcome back to AI Basics. Yes, this is the show where we go over all the basics and sometimes stuff that's even a little bit more complicated. And then sometimes we dip into science fiction and science fiction is quickly becoming reality. Today's guest is the founder and CEO of SatLit. You want to talk about science fiction? He's building software that lets AI run on satellites, right? There's no standard way for satellites from different makers to work together and do large computing jobs. And you're saying, hey, wait a second, J. Cal, how much compute is out in space? Well, this is just getting started. So our guest today, Rama Afullo, is skating, proverbially, to where the puck is going. Welcome to the show, Rama.

0:50How are you, sir? Thank you so much, Jason. I appreciate it. It's an honor. I'm great. You're good? How are you? Well, thank you for asking. I always take it literally. Have the greatest job in the world. I get to hang out with founders all day and tape podcasts and invest in 100 startups a year. Can you think of a better job? I mean, if it was a government job and they paid me minimum wage, I'd still do it. So that's a pretty good indication you're doing something well. There's a Japanese term called ikigai. Have you ever heard this term, ikigai, Rama? Yes. Yeah. So for people who don't know, put a bunch of circles together, what you're good at, what you're passionate about, what the world needs, what the world will pay for. And when you start overlapping those and you get

1:35all those different circles, sometimes it's a hobby, sometimes it's a pursuit that you do quietly, and then sometimes it's a life. It's a great life. And that's, I think, what we probably both share. Tell me a little bit about what you're building and why. There's the ikigai. Yeah, as you see there, what you love and what you're good at is your passion. What the world needs and what you love is a mission. What you can be paid for and what the world needs is a vocation. What you're good at and what you can get paid for as a profession. And that ikigai, I mean, that is just what it's all about. I think you fall in here too. Tell me a little bit about how you came up with the idea, when you started the company.

2:13What's the origin story here? Because this is pretty far out there, putting compute into space and then getting them all to work together. This is science fiction that's quickly becoming a reality. Yeah, I mean, I love that. Ikigai, actually, I just got married this weekend. My wife's half Japanese and half Chinese. And we were in Japan earlier this year. And so I think that center of that Venn diagram as well for my life, outside of my professional life, that works out well. But yeah, getting into this, so it probably starts with family. I'm originally from Kenya. I'm a Kenyan American. My dad's a professor of electrical engineering. And a lot of the work that he did, his PhD was focused on SATCOM

2:58and microwave and satellite technology, right? And my mom's a computer scientist, right? And so she was uh very much like she's actually the person who taught me how to copy paste and that changed my life right copy paste very underrated quick keys oh you talk about saving an hour a day it compounds pretty nicely it's revolutionary like i i mean just that concept was was next level to me way back when, right? And I mean, I've got two brothers as well. And my older brother always was very entrepreneurial. We were the only people in the block, which is when I lived in Botswana, the only people in the block in the neighborhood who had a CD writer. Back when CDs were a thing. And he used to, like, copy CDs for folks and sell them back to people.

3:45And I just thought, wow, there's a real need for this thing that he's doing. But yeah, to fast track, I mean, my younger brother's a stuntman, he's in movies, and we've always been a very dynamic family. But to come up with this idea, I actually knew what my startup would be called about a decade ago. So just before I started working for Tesla, I was the third employee of Tesla in Africa. And just before I started that, I thought, well, I want to work in satellites. And back then, there was this sort of craze of how you just you take a common name and shorten it a little bit. And that was like the way Silicon Valley made their names for their startups. Right. And I took satellite and it's actually a backronym for space and technology analytics, right?

4:27And the idea back then was, well, I'm probably going to do something to do with satellites at some point in the future. And then I always thought, well, at the time, there was OneWeb, right? And they were more or less the original Starlink. And the work that they were doing was so interesting to me, right? And I've, of course, always been fascinated by SATCOM since I was surrounded by it at home. And that really is like, I decided to buy the domain about the satellite.com, the satellite. A few of them, a few different extensions, but the .ai like four or five years ago. But to fast forward through that, I worked at Tesla, then I worked at Google. And when I was at Google, I was really pitching for Google

5:11to push on orbital compute and orbital data centers. And this is just at the beginning of the AI boom. So, of course, their priorities were specifically terrestrially, how do we make sure that our LLMs are as powerful as what ChatGPT is doing, right? So I understood that. And I left Google and joined SpaceX. I thought, well, surely they'll take this seriously. Like orbital compute will be the next big thing, right? And I was told under no uncertain terms at the time that the last people who proposed something like this are no longer at the company, right? So I said, well, you know what? I think there's a path forward for us here. And I'm going to either stay here and become part of the furniture or leave SpaceX. I was a part manager at Starlink,

5:57leave and go and do this on my own. So I opted to leave three months into my time at SpaceX. And I started this just over two years ago and happy to deploy from there. Yeah. And so for people who don't know the why, why would we want compute in space? I think most people listening here are going to understand it, but you started this before the why existed. So I'm wondering what your why was, because there's a recent why, which has to do with, you know, a lot of societal issues, but take us through your why and society's why. Sure. So I've always felt like I know a secret within a secret. So the first secret is that we should put space, we should put compute in space for a number of reasons, right? One of

6:43them, like you said, said societal like folks don't actually want data centers in their backyards like that's an issue right but then and when you look at the problem a little bit closer uh there's a number of other issues there's power budgets and link budgets and uh the direction of how are we making sure that folks in the most remote areas have the right emergency response that we're getting wildfire responses, and that you can process data in space? So this is like sort of the edge compute base case here, right? How do we make sure that you can get your data as quickly as possible? Because these single metric, biggest metric in sort of telecoms and space communication in general to measure whether or not we're doing

7:27things the right way is the total latency of the system. Yes. Which Starlink solved, right, by getting closer to the Earth. But if I translate what you're saying here to a layperson, you still need to make some decisions up in space as to what to do so that even if it was just Starlink and Hughes satellite and all the other satellites, they still need to process something on the satellite. So you do need compute in space just for that one use case. Exactly. And that's the base case. Right. And from there, you propagate. Right. there you propagate, right? You start looking at, well, what are the things we do on Earth that we've decided we've got to do terrestrially on Earth, that actually there's a better use case

8:09for us putting it up into orbit, right? And it's actually more efficient in the sort of long run, as soon as you hit critical mass. But to reach that parity, there's a lot of different steps that you've got to take in terms of the launch capacity, launch capability, thermals, et cetera. And we can get into that as well. Okay. So let's get into it. The next base case became, hey, people got scared by AI. Some people in the AI business kind of terrified people with the other side of science fiction, the non-Star Trek side, but the dystopian side, hey, we're all going to lose our jobs. And oh my God, there's going to be computers everywhere. And the public said, hey, maybe I'm not so thrilled about these data centers. They generally are aesthetically not

8:54pleasing. Sometimes they burn fossil fuels. Sometimes they make noise. There's a number of reasons to maybe not like them. And a lot of the reasons have been resolved. But hey, we are where we are in terms of the public's perception of these. And there seems to be some advantage to putting them in space. You don't need, just off the top of my head, to sign a lease with somebody. There seems to be a couple of hundred miles of low Earth orbit available. Low Earth orbit is a bigger circumference than the planet Earth. And that means you have essentially hundreds of thousands, millions of times more real estate just in low Earth orbit. And then it might be an energy advantage to having access to the sun without the atmosphere.

9:46So maybe walk us through that series of conditions, and then we'll look at your software and some demos. Yeah, sure. So firstly, you're pretty well-versed, right? So that's quite impressive, right? I think on the 24-7 access to the sun, the idea is to put satellites in sun synchronous orbit right uh for them to always be facing the sun and always have access to to the sun at a much more efficient level right is is sort of solving some of the power issues that you might get at a typical data center right where power is probably the biggest limiting factor for compute in general right then there's the sort of link budgets as well that can be solved for right but you've you've got to think about, like, how are you sending the data from a satellite to satellite?

10:29Right. So is it optical intersatellite links, which is probably the best path. But you've got to also think about how like where it makes sense to have radio frequency links. But that's a whole sort of different thing to think about. Right. But now the societal side, right? So when I was at Google, actually one of my roles was as a professional cloud architect, as a solutions consultant, right? And you start to understand the impact of having multiple cloud regions, like in different parts of the world, and why that's useful, especially as a security feature. Now, think about the way these cloud regions are built is that these are data centers that are in a fixed point. You've got a point of presence you can use to propagate these points, but really they're in a fixed part of the world and you have to build more of these.

11:17Whereas with a data center in orbit, right, you actually can, like these are moving at 17 500 miles an hour in space right which is a gift and a curse all right uh but because they're moving so fast you actually have a situation where you can have this data center wherever you want at any point on earth within 90 minutes right at any given point in time pretty good commute if you can do it. Oh, yeah. And the advantage there is if it needed to send data down to the Earth, you could do it quicker. Or if you wanted to take pictures of the Earth, I guess, you could take it quick. So when we see a spy film and they say, hey, can I get a satellite above that area, they can do it quite quickly, yeah.

12:02Yeah, so that's part of it. And if you think about having multiple satellites within a field of view of each other where you can actually relay this data over to them, right? As they're moving, so that's when you hit this critical mass, you actually are able to relay data and send them to the quickest point to send it down to Earth, right? But you can also use them as nodes to process your data because it's this aggregated compute module, right? But you can also use them as nodes to process your data because it's this aggregated compute module, right, that you end up having in space. And what we are attempting to do is to do this across heterogeneous systems, so systems with dissimilar architectures. Ah, so what do you think

12:39the jobs people will send to space will be versus, you know, jobs we'll do on Earth? Is it going to be, hey, I've got a long-term job. I'm making the next front-term model. I'll send it up there and let it run. It'll self-organize across disparate systems, different clusters, and intelligently sort everything out and send the result back down? Or is it going to be like kind of backburner stuff? How should we start to think about when we get, I don't know, 100,000 of these satellites out there? It seems like a pretty good critical mass. I don't know, you tell me what critical mass is, 1,000, 10,000? When would this actually be able to do real work? And then what work will it do? Yeah. So I guess the number of satellites is good.

13:28We've got 20,000 satellites in space right now. It'll be at least at 100,000 in the next three to five years. But what's more important is the amount of compute, right, that we have and the ability to relay that data across different satellites and down to Earth as quickly as possible. In fact, part of the work we're doing with NASA is relaying data, streaming 4K video from the moon down to earth. That's part of what we're doing at satellite. And so part of the jobs that folks will be setting up there, you can think of different experiments that folks want to perform, right? Being able to do that in real time without having to send it via launch capabilities up to the international space station and paying for an astronaut to do

14:11your experiment for you right um or it can be uh for example um being able to have direct to cell uh and direct to device uh services right, without having to change the architecture of your phone or your handset, right? That's something that we've been thinking about quite a lot, because if you have much more capability in space, and you're able to beam this data directly to a handset, regardless of the power efficiencies of the handset as well, you can do a lot more. So you no longer actually have to have a telco on the ground, right? Which is part of why there's currently a race for the direct-to-sell side of things as well. So it essentially eradicates the typical telco, right? In the next

14:5710 years or so, right? They become a bit of a relic like the landline telephone, right? So we're seeing a lot of these different services, emergency response being a lot quicker because right now, if you think about the Palisade fires in LA, right? Insurance companies were able to get access to the data about what was going to happen in LA a lot quicker than the typical human was, typical sort of lay person was, right? And a few of them changed their policies sort of lay person was, right? And a few of them changed their policies just before those fires hit, right? So you can, it's a little bit dark, right? But so you can have some control in that respect as well. This is absolutely fascinating. And then for people who don't know the cost of doing this

15:40versus keeping, you know, a cluster on the ground right now, I think people think it's sort of 10 to 1, like maybe 10 times more expensive to put it in space. How does it get down to 1 to 1? And it becomes not just technically viable. I think we all believe you'll be able to do it. Elon, Google is working on it. Everybody's kind of working on this vision at the same time. StarCloud, a bunch of folks. When does it actually become cost effective? Is there a curve that you guys are watching for this? I think for us, for satellites, it's really cost effective because we don't build our own hardware. Sure. Right? So we've got that advantage. But if I think about the macro scale and the rest of the environment,

16:26there's a few things. You are correct, right? So right now, I guess with Starship, the estimation is that they're down to about $1,000 a kilogram, right? To put the payloads up into orbit. And in order to get parity with terrestrial data centers, you've got to get about $100 a kilogram. So you're right about on the money there, right? But for any sort of typical satellite operator who's trying to put their stuff into orbit right now, they're paying $15,000 a kilogram for rideshare, right? So we are taking the best base case, but eventually we need to bring everyone along for the ride now in order to do that there's a few different things that have to happen uh you need to have significant efficiency in terms of your your um uh launch capacity right um so starship needs to work

17:19blue origins new glenn needs to work right anyone who's building a massive vehicle to transport more satellites into orbit, those folks who can put up much larger payloads, they need to be in full operation. Yeah, and they're close. They just need to figure out how to catch it right now. So I think Elon's doing Starship 15 in the next week or two. And if he catches it, then we're off to the races. Yeah, with the chopsticks, right? Yeah, chopsticks. Yeah, pretty easy to do. It's like catching a fly with chopsticks. It's possible I may need to be a Kung Fu master to do it right now. But I think he kind of is best described as one. All right, so let's take a look at the demo you have of how this might work.

18:04I think people are interested in how software engineers will actually build software to do this. Reminds me in some ways of when I first saw ARPANET and BitNet, the precursors to the internet itself, there were routing issues. Just routing data around was the problem that was being worked on. But here we go. And remember, some people will be listening, so describe what we're seeing. So this is a fairly basic demo of what it is we actually do, right? And so the idea is, so this is our mission operation center. You've got a typical onboard computer that's in front of you right now, right? With sort of the GPUs, and we've got different mission services, right? So we can compress data, we can have a registry of your containers,

18:47so containerization is super important in this case. What that is, is how do you port software, right, in as efficient a manner as possible without having to build in the code that the software was built in order for the sort of like the resource profiler or whatever to be able to read the software, right? And it's security and encryption, storage services, et cetera. So Google's sort of Kubernetes work was really the onset of a lot of the work that we actually end up seeing here. So the idea is you have these different- Kubernetes people use for web services to build small bundles that are highly efficient for doing something like serving a web page or showing your inbox. These are little jobs, right,

19:31that can run independently or in little chunks is a good way to describe it. Yeah. Exactly. Right. That's a great way to describe it. Right. And how do you then orchestrate those? Like how do you say, well, all these different different little chunks how do we make them sort of do the thing we want to do and that's part of what kubernetes does right so you can imagine that um but like a space version of that is part of what we do because we orchestrate across these different satellites right so in this in this case so let's say we've got five satellites in orbit and we're running an nvidia just an orin nano in this case, so let's say we've got five satellites in orbit, and we're running an NVIDIA Jason Orin nano in this case, right?

20:07It's eight gigabytes, six cores. And we've got sort of our modular, we've got our sort of power budgets across these satellites where we don't want to go above 15 watts, right? So we've set up these different parameters. We know that we want to feed data from at least one of these satellites down to a ground station, in this case, a Sat-1. So as we move forward, we say, okay, well, what are we going to be running on this mission? So we're going to have a container registry, so that we can prepare different images and we can register them across these different satellites. We can analyze telemetry in space. We can compress data and serialize data. We've got security and encryption. So these are just standard.

20:53These are like the most basic managed services that we would be running on the satellite. And we can place services across different nodes. And you can also say, well, we want to reduce the amount of, reduce the power budgets as well, because once again, in compute in general, power is one of the biggest limiting factors, right? So what we have here, right, is the ability to initially cross-link data and have onboard AI running directly on a satellite. In this case, we've chosen to use Google Gemma, right, as a small language model, so running on the edge in space, right? And we've done this ground station. There might've been cloud cover or electromagnetic interference or something, right? So instead you pass and relay that data over to the ground station,

21:53diagnose the issue, pre-process the data and make sure that the data is small enough to be sent down, right? And then you find in this case, we said, well, actually only 11.7% of your data was useful for you to downlink. The rest of it was nonsense. And you pay like $10 to $20 a gigabyte to the ground station operator to send this data down. So if you can send only what's useful, it's cheaper for you, much more efficient. And you can use different resources in orbit. And then the final bit I can show you here is like, well, then what is like the next use case for us? And part of what NASA is tasking us to do here is to say, well, we've got a lunar base, right? Then using the exact same technology and using something we call delayed torrents networking right you can actually stop

22:46uh sort of packet loss in space and bit flips right and you can relay data 4k video from the moon to different uh to sort of satellites on uh in the lunar orbit down to lower orbit satellites and then straight to your satellite in really, really low latency. And that's essentially what we end up doing. I mean, 4K from the moon is going to be pretty crazy to go see an IMAX movie from the moon live. It is pretty incredible. So much work goes into this, and the payoff is going to be pretty obvious when we get to some basic concentration. Plus, you must be thinking about what happens when we go to Mars and daisy-chaining these clusters. What is the concept of doing what we're talking here from the moon and getting 4K from the

23:41moon down to Earth in a somewhat live fashion? There's got to be some delay, obviously. But right now, it takes a long time to get images back from Mars. I'm not sure exactly how far. But what is the theory there? Are we going to daisy chain a series of satellites that far out and they're going to be, I don't know, 10,000 miles apart? How is that going to work? I mean, I have my theories, right? I do feel that, I mean, daisy-chaining them might not make as much sense, although the high-rate delayed torrents networking protocols, so that's part of what we're using here with the NASA Glenn Research Center, was specifically built for you to be able to communicate into deep space. It was built by some of the founders, some of the creators of the internet, like Vint Cerf and the

24:29likes, right? And so the idea, I think, would be how do we send this data down with as low latency as possible, with as much, as little packet loss as possible, and bit flips. Bit flips is when the ones and zeros swap out. And everything we do in computers in general, they're just ones and zeros. So if those flip- Yeah, kind of showstopper. Exactly. That ends the conversation right there. It hangs up the phone call. Exactly. So yeah, so there could be some daisy-chaining. But if you think about the overall trajectory and how you've got to send this data over and the fact that you've got different planets spinning on the axes, you've got satellites also moving, it's quite a complex problem, right,

25:18to be able to get that to happen. But that could be a solution. I suspect, though, that before we get to that, we're probably just going to stick to the technology we've got right now to avoid some of the potential issues and figure out ways. We've got to first put some satellites in all of these different orbits, right? Because Mars is like, what, two planets away? You've got to add some satellites to those different orbits for you to be able to daisy chain that. Yeah. And if you can do this under these conditions, we didn't even talk about heat and how that's going to work, but these are extreme conditions. And like you said, double-edged sword each time. That's going to inform, I guess, how we build software and

26:02infrastructure on the planet as well. Like you're probably figuring some things out in terms of resiliency of these systems and making them not brittle. Yeah. Yes. I mean, it's the same like adage, like when folks say, why are you doing things in space? Well, without space, we really wouldn't have had the advent of microwaves, wouldn't have had solar panels. There's so many things that were built for space that we then brought a sort of a derivative of it down to earth that turned out to be incredibly useful for us here on earth. And we suspect that a similar thing will happen in this case. In fact, probably the smartest space players will make sure that they have a terrestrial sort of counterweight that they're

26:46working with, like someone who's doing something similar to what they're doing, but here on Earth, for them to feed into because they will be getting so much data and so much useful output and inputs from orbit that they can definitely transfer here. They'll make people's lives easier. As we wrap here, I just want to understand this debate. I keep hearing people talk through each other. Okay. Space is really cold and there's no atmosphere. Therefore, you don't have to chill the chips. Then I hear another group of people say it's because there's no atmosphere, these things overheat. What's the truth here? Why do we keep hearing people talk through each other? And it's seemingly smart people.

27:28I keep hearing people say like, oh, it's a solvable issue, or you're going to need really big radiators to dissipate the heat. What's the truth about heat in space and batteries in space? I mean, technically speaking, they're both correct. Maybe this is why people are in this debate. Yeah, they're both correct, right? Because you're in space, you're in a vacuum, you know, so it's technically it's quite cold, but the heat needs to be dissipated, needs to be pushed out, right? So the thermals, you're saying you need massive radiators, which is what the International Space Station has, like at the back, if you look at where their solar panels are, the back of these massive radiators, right, that essentially push this heat out.

28:11So I think there's a bit of a balance there because you do need to solve for this thermal issue at scale because you haven't had this much compute, right? In fact, like some of the computers being set up now is like 10x the total compute ever put across these 20,000 satellites that are in space right now. So this will go up in the next year. So it's a significant shift. But that means that you've got to solve for the heat. folks are doing like what uh intel starfire and nvidia's space one vera rubin module what these guys are doing is building specific compute modules built for space that can run at a much hotter temperature and that have much higher wattage right it's like three thousand watts

28:58right that they're dissipating at any given point in time and if you can run hotter uh so they're essentially just skipping the entire debate and you're saying well a runner chip's harder it's okay right and we can find ways to to throttle it right um and so yeah it has to be solved for right in in the end uh but uh both sides of the equation are correct uh that you've got you've got to get this heat out, but space is cold. Yeah. So you need some surface area to spread the heat out, and then it won't melt the processor or overheat. Or you could just make these things more industrial strength. Final, final, final question. Do these things need to be built in a more rugged fashion or a more solid state fashion with less parts since they're whipping around at such high speeds.

29:47And they have quite a lot of G-forces going up, I understand. So are these things going to just break on the way up? Or can they come apart? Like, you can't just send existing stuff up there, can you? You sort of can, right? And this is part of what planet were really good at solving for and they're the folks working with google and the planet labs yes they were really good at like saying well you know every single misconception of like anything someone has told us about what we can and can't send in space uh we're just going to ignore all of that advice right want to throw up some some uh smartphones put some solar panels tacked to them and say, does it work? I want to throw up some stuff with moving parts.

30:32Like right now, we've got a lot of NVIDIA just in or in nanos in orbit, and they have a spinning fan and, you know, they've got all sorts of moving parts that survive the launch, survive everything, and they're operating in orbit, right? So you can, but you do, in order for you to have more longevity, you do need to radiation harden, or at least shield, right? Some of the electronic surfaces in order for them to operate a lot better. You've got to go through a lot of some expensive testing as well. So that is- That's the ruggedness. If you can build it to survive in space, you're going to have to have material science combined with, you know, the engineering to make it very rugged and be much more efficient up there because every kilogram

31:18matters, every bit matters. So if you make it that efficient, now I kind of understand your earlier answer. When we build something on earth, you can bring those efficiencies at a minimum to the Earth products, make them last longer, make them less brittle, have to be replaced less often, et cetera. It's just an amazing vision. Rama, where can people learn more? I'm sure I understand you got a little bit of funding there. And you got a couple of contracts, you're off to the races. Where can people who want to build science fiction go to work every day at your company? And where can they learn more? Yeah, sure. So our website is satellite.ai. So S-A-T-L-Y-T dot A-I. Go to our careers page. You can, if you're interested in putting your software in orbit you can uh go

32:05up there to obviously explore packages you can book uh book uh uh your spot on book some time yeah some time as well and we've got a bunch of open positions as well and yeah we're doing some cool things i'm gonna i'll send uh my grok bot up there to uh do some work in space i'm looking for some real estate in space i'll just send my grok bot up there to do some work in space. I'm looking for some real estate in space. I'll just send my Grok bot to look through whatever the equivalent of Zillow is on the moon. Amazing, amazing job. What an incredible vision. And hey, speaking of vision, let me take a moment to shout out my friends at Google. I've been friends and worked with Google since the beginning. AI Basics is brought to you specifically with partnership with Google for startups. They

32:52are amazing. They are the reason we do these episodes. They help us find the most interesting entrepreneurs in the world. And you can find all of these episodes, all these AI Basics at thisweekinstartups.com slash basics, thisweekinstartups.com slash basics. And they have a startup program. They're incredibly helpful. I have sent a half dozen founders to them, but basically every cohort. And not only are they very generous with startups and supporting them in those early years, they're generous with their time and helping them leverage all the amazing offerings over at Google. Great prices as well. It's not always about price, it's about capability, but they do have a heck of a deal. Rama, how is that for an ad read?

33:34Pretty good, huh? Pretty good partnership read? It's pretty good. I'm pretty professional. I'll be doing those for your software at some point. Hey, go get sat lit and yeah, use the promo code JCal to get 10% off when you're putting jobs in space. We're like three or four years away from this. It's so crazy. I'll see you all next time.

▸More episode summaries

This Week in Startups cover art
This Week in StartupsJason: Jensen Huang will be back at Nvidia within 12 months | E23632026-11-13 · 61 min
This Week in Startups cover art
This Week in StartupsFounders Are Now Sending VCs a File Written for AI2026-11-12 · 81 min
Transcribe your links. Paste a YouTube, Google Drive or podcast link and WhipScribe gives you the transcript. Transcribe a link →
New This Week in Startups transcripts by email
One email per new episode, unsubscribe in every message. Not shared with the publisher.