October 1, 2026

Karan Kunjur on Why the Future of Space Needs More Satellite Power

Four years ago, Karan Kunjur and his brother Neel started K2 Space around a contrarian idea: instead of making satellites smaller, what if you could make large, powerful satellites dramatically faster and more efficiently?

At ICONIQ Frontier, Karan joined ICONIQ’s Tengbo Li to talk about how that thesis is playing out in orbit. Their conversation spans the launch of K2’s first satellite, Gravitas, what the team has learned from operating a 20-kilowatt system in space, and why power is becoming increasingly important as communications, national security infrastructure, and compute move beyond Earth.

Karan also shares how AI is changing the pace of engineering inside K2 and why he believes space-based compute is ultimately an engineering challenge, not a new physics problem.

Tengbo Li (00:00):

We spend a lot of time here at ICONIQ trying to figure out whether a founder's big ideal will actually fly. And with K2, a few of us got the opportunity and the answer quite literally at 4:00 AM this March. So we went down to Vandenberg Space Force Base to watch the launch of Gravitas, which is K2's two-metric-ton satellite. And with a burst of energy, we saw a fantastic liftoff.

(00:27):

Once in orbit, this satellite deployed 40 meters of solar panel arrays, fired the most powerful electric thruster ever operated in space, and brought 12 commercial and national security payloads online. And so in an industry where ambitious timelines often move, Karan Kunjur and his brother Neel set out to do something exceptionally hard and delivered on their very first try.

(00:53):

So that told us a lot about the team. And today, K2's approach is attracting significant demand from both commercial as well as national security customers. I think they've announced publicly that they've secured over a billion dollars worth of contracts across commercial and national security. And soon, there's going to be another enormous need that's going to be met by K2's capabilities. The compute that many of you are building in this room and supporting in this room has to run somewhere.

(01:20):

And as you heard from Nikesh, it can't be all on Earth. And so increasingly somewhere, that somewhere might be in orbit with K2 helping to power the next frontier for AI. So let me welcome to the stage our friend, Karan Kunjur.

(01:47):

Long walk. Thanks for joining us, Karan.

Karan Kunjur (01:50):

Thanks for having me.

Tengbo Li (01:51):

So maybe just to paint the picture, you guys manufacture some of the largest satellites that have ever flown. And just to give the folks in this room a sense of proportion, the current Mega Class satellite that Karan, Neel and the team have built is about the width of this room. In fact, I don't think this room could actually contain that satellite. And their next class of satellites will be probably three to four times the length of this room.

(02:13):

So that's at a time when the satellites being sent into orbit, for the most part, are actually pretty small. They're attritable systems, larger constellations, they tend to be low Earth orbit. You guys took a very different path, a very contrarian path, almost in completely the opposite direction. Why?

Karan Kunjur (02:30):

I'll break it down a little bit with space in general. I think most people here have become much bigger space experts in the last 6 to probably 12 months than they were the 12 months before, but almost every application in space ties back to power. If you think about the applications that we currently have in space, it's communications, it's starting to be compute, it's a bit of missile warning, missile tracking, and it's a bunch of other stuff on the classified side. All of those applications tie back to power or payload mass. The size of your satellite when you're generating all power from solar arrays directly scales with that power. The bigger the satellite, the more power, the smaller the satellite, less power.

(03:09):

When my brother and I started this company about four years ago, it was viewed as conventional thought that the only way you could do space in a low-cost, efficient, industrial way was to go build small satellites. Big satellites were too expensive. They took too long to produce. And as a result, the only solution was to build small satellites, almost the size of these chairs to actually go proliferate in space. The problem is when you go that small, you give up all power and you give up all capability.

(03:43):

So Neel and I had kind of a contrarian thought. We were like, "Why has no one just tried to figure out how to go build a big satellite in a better way?" And they talk about being naive when you're a founder starting something. I'm glad we were naive at that point because I don't think we appreciated the reason why no one had done it is there's like 20 to 25 different super hard engineering challenges need to go solve to go do that. You need to build giant solar arrays, you need to build giant reaction wheels, giant batteries. You need to basically go build an entire stack, both on the hardware and the software side for operating in space completely from scratch. Which is why when we started it, we were like half the people that we talked to said, "That's going to be impossible. You're not going to be able to do it." And the other half told us, "No one's ever going to need that much power or capability."

(04:32):

In the last six months, both of those groups have been proven wrong. The second group with the rise of compute and the rise of high-throughput communications, all the stuff that's happening on the optical side and the RF side is showing why power is important. And then as Tengbo mentioned, about four and a half months ago, we launched our first satellite, which was a 20-kilowatt satellite. It was as large as the highest power, largest satellites that have ever been deployed. And it's now 135 days into operation at this point and operating well. So it's been a step change, right?

(05:07):

Basically by doing this thing that was super hard that everyone told us that nobody else could do, we are now in a position where, as a four-year-old company, we've signed a billion dollars in contracts with commercial customers, with the US government, and are in full on scale mode in our factory in LA.

Tengbo Li (05:22):

Great. There's a lot of folks who are building software in this room. What you built is largely inaccessible unless you go to the factory and see it, but otherwise the satellite's going to be flying in space. And so maybe just to help folks conceptualize what you are actually doing for your commercial customers, for national security to the extent you can talk about it, can you talk about some of the payloads and use cases that you're actually supporting?

Karan Kunjur (05:45):

Yeah. The best way to think about what's happening in space right now is there are a variety of applications that happen on the ground that are all for different reasons getting moved to space. So you think like Starlink, which is probably the most ... Everyone knows about it at this point. What do they do? They took broadband communications and they took it from a ground-based capability and put it all in space in a distributed fashion with global coverage, with massive performance improvements. What you're now starting to see is that exact thing happening in different flavors. So now, you have sovereigns who want to be able to deploy that exact same capability, but have them control their telco infrastructure.

(06:26):

So one of our first big commercial customers is a company called SES. They're a Luxembourg-based company. They're effectively building the sovereign communications infrastructure for space for Europe. It just so happens that they decided to build that entire constellation off of our K2 platform. We knocked out Boeing, we knocked out a bunch of European players just because of how much more performant, how much more powerful the satellite was and the speed at which we could do it.

(06:51):

We have US government customers that are thinking about using us for other comms applications. So in addition to broadband, you also have lower-frequency comms, which is like, think like a soldier with a handheld radio on the ground who historically could only communicate like line of sight, now being able to communicate beyond line of sight using satellites. Satellite based, basically a flavor of direct to device. That is now possible because you can deploy a lot of power in a proliferated fashion globally.

(07:19):

There's many other flavors of what I just described that go off into different places that I can't really talk about. But the whole point is, when you suddenly become capable of deploying a lot of power on orbit, you can start to do really interesting things on the comms side and start pushing a bunch of these applications that were sitting on the ground into space.

(07:38):

Now that's where we're at today. Where are we going tomorrow? Well, over time, you're going to see things like not just the comms network, the RF comms network move to space, but you're also going to see the fiber optic network. One of the most recent developments of the last three years has been our ability to do optical communications. Basically, use literally space lasers, which is always fun to say, to use space lasers to basically communicate satellite to satellite and satellite to ground. And what happens when you can do that? Well, now I've got hundreds of satellites on orbit, each with four to eight different optical nodes on them and you've deployed a mesh network that allows you to do terabit-per-second class optical comms.

(08:19):

You basically now have the chance for the next 5 to 10 years to take the entire fiber optic network on Earth and push it to space and push it to space in a way that no one can cut those lines. You have massive amounts of redundancy and you have massive amounts of data throughput in a way that's highly, highly scalable.

(08:36):

Layer on top of that, a bunch of compute nodes where you're basically taking terrestrial data centers and now putting a lot of compute up there that needs a lot of power as well and you've basically replicated the entire terrestrial system in space.

(08:47):

So that's kind of where we're going. What you need to be able to do it is you need a lot of power. You need to be able to do that in a robust, scalable way. And you need to be able to do that everywhere, not just in what's called low Earth orbit, which we've figured out really well, that's close to the Earth, but in medium Earth orbit, in geostationary orbit, in cislunar, and eventually beyond.

Tengbo Li (09:07):

I think it's just easy to underestimate how much can be done in space and how much better it can be done in space, but maybe just to bring it to the here and now versus the long-term roadmap. So you guys have been ... Gravitas has been operating for 130 days, which is fantastic. One of the most successful first launches that we've seen in the space industry. But very tactically, what have you guys proven out over those 130 days?

Karan Kunjur (09:30):

Yeah. So we were briefing a government customer yesterday actually, and he was at the factory and he was like, "I am surprised at how much went well with this." Because we had to build so much of this from scratch, right? 85% of the satellite built completely in-house. So those are 85 percentage points of risk. Each of those is a new component that could have and probably should have failed on orbit so far.

(09:56):

So the beauty of Gravitas has been we've just had so much data collected. And we call Gravitas kind of our block one design. Block two, we're launching next year, about 10 different satellites across two different launches. The timing of it allowed us to basically take a lot of the learnings, feed into the block two design, and now rapidly iterate.

(10:15):

It certainly wasn't perfect. There's a ton of lessons learned. In fact, my brother who is also the CTO of this company, he's in the Pentagon. He was in the Pentagon all day yesterday and he's at the NRO all day today talking about all the lessons learned. No one's ever done this. No one's ever done this at scale. Even SpaceX, they have yet to launch a satellite at this power level. So we're really pushing the bounds of what's possible with the first 20-kilowatt satellite and Gravitas allowed us to kind of prove out a lot of the capability.

Tengbo Li (10:45):

You mentioned sovereigns before. We have folks in the room who represent the sovereign wealth funds and other organizations when in some of these countries outside of the US. And when you talk about sovereign capabilities and what folks are solving for, what are those? How does K2 actually address those needs? How are different sovereigns thinking about their space strategy relative to some of the vendors that are available here in the US versus elsewhere?

Karan Kunjur (11:12):

Yeah. So there's some fundamental capabilities that are absolutely happening from space. And if a sovereign is not currently thinking about how they're going to deploy those capabilities, they're probably about five years behind. So communications is the obvious one and communications has everything from broadband to direct to device to kind of the fiber optic network.

(11:32):

There should be a strategy if you're a sovereign for each of those. And then you layer on top of that things like missile warning, missile tracking, moving a bunch of the ground-based sensors to space is a much more effective way to ensure that you have the right level of coverage. And then beyond that, I think there's some of the sci-fi stuff that'll eventually happen, but those four categories I talked about, the three different flavors of comms and the missile warning, missile tracking, every single sovereign right now that I've talked to is thinking about it.

(11:58):

And what they're thinking about is like, "How do I multi-thread this?" So you should absolutely use Starlink where it makes sense for you. Buy it, use it, even use compute where it makes sense. But you probably, as a sovereign, also want to have capabilities that you own, that you have the ability to turn the switch on and off for, that you can customize for your own military use cases. And so it's very much a dual-track approach. It's buy where it's available and build where possible.

(12:26):

And there are very different flavors of build. You can build the hardware and own the hardware and own the full stack. You can have someone like us, build the hardware, but you own the waveforms. So you effectively, through the software side of it, have full autonomous capability of it, but you don't need to necessarily build the hardware or be deep there.

(12:43):

And there's finally all the way in the end, just buy data. Every single sovereign is thinking about it. Every single sovereign is thinking about it, and some more so, what are the risks to my network and my country's network if the deep-sea fiber cables are cut? And we're seeing more and more of that. What is my backup plan for the infrastructure of my nation? And space presents a very interesting kind of backup plan.

(13:09):

And so the sovereigns that I'm seeing that are furthest along are people that are like, "Yeah, we're going to buy some of this. We're going to build some of this. For this segment, we're going to build in this way. And for this segment, we're going to build in this way." And that sounds really simple, but it actually takes a lot of thought. And I've seen dramatically different levels of sophistication across the sovereigns I've talked to.

Tengbo Li (13:30):

Makes a ton of sense. Maybe shifting gears a little bit, you guys are squarely a physical hardware company and there's increasing talk about physical AI, certainly a lot of reflection that we have in terms of the intersection between physical systems and software or AI. And just to paint the picture of how hard it is to do what you do, there's a thing called the Van Allen belt, which exists above low Earth orbit where there's a ton of radiation that comes in from all parts of the solar system and the universe and what have you. And the satellites that K2 has to build have to be hardened against that radiation, have to be able to survive through that radiation for extended periods of time. And it seems clear that you can't vibe code that radiation hardening to sit on the K2 satellite.

Karan Kunjur (14:12):

Unfortunately no. Yeah.

Tengbo Li (14:12):

But you guys are building a lot of software internally, whether it's around avionics or ground systems or what have you. What is the interplay between the hardware that you build, the software that you build? How do you think about building that software? How does AI play in when the tolerance for faults is so incredibly low for what you're doing?

Karan Kunjur (14:31):

Yeah. So I was actually asked the question earlier like what's our split for engineering between hardware and software. We're about 75% hardware, 25% software. On the software side, it's everything from the flight software to the control software at the lower levels of the subsystem, all the way to the ground software, constellation operation software. So we've got all these different flavors of software challenges that bring the hardware together. And because we vertically integrated so much of the hardware stack in-house, we had to also build all the software to power each of those subsystems that we built ourselves.

(15:05):

So it's a massively complex hardware-software problem. And it just so happens that you need to be able to solve that problem in not just a static environment. Our whole thesis was we're going to build a platform and it's an infrastructure play. So we're going to build a platform that maxes out power, maxes out payload mass, and you could put it in any orbit, which means you need to be able to handle super high radiation environments. You need to be able to put it all over the different orbits around the Earth and then eventually all over the solar system. It's kind of our objective. That's a phenomenally ... That 10Xs the challenge from a hardware/software perspective.

(15:38):

So it's a big part of what we're doing. It's a big part of what we're trying to solve. On the AI front, what I will say is you probably have heard multiple stories of different founders who've come in and told you it is accelerating what really good engineers can do. And kind of our strength, and it's funny, Tengbo did some analysis on our team, is we probably built the best engineering team outside of SpaceX, and the data will show you that.

(16:02):

That team combined with the last 12 months, what's happened with AI, has allowed us to move at an incredible pace. You've got these N-of-1 engineers. We've got an engineer who I would say is probably the best propulsion engineer for electric propulsion in the world. You pair him with an army of engineers that he gets from using Claude every day all the time and he's doing insane things.

(16:24):

I'm a little bit biased, but my brother, who's the CTO, he's probably one of the best engineers in the world, he spent 30 days basically just working on Anthropic with Claude and built our future kind of phase two constellation that we'll start to share a bit about in 6 to 12 months. That would've been a massive team and it turned into just one phenomenal engineer just vibe coding. So, yes, radiation challenges you can't yet vibe code, but there's a massive impact that AI is having on our business and the speed at which engineers can work.

Tengbo Li (16:55):

Love it. Love it. Maybe 12 months ago, this concept of data centers in space felt very foreign. You had Gavin Baker go on podcasts and talk about it and it just seemed so far-flung. And then you fast-forward to 12 months later, it just feels a lot more credible and in part thanks to the work that K2 is doing and some others in the space, but now increasingly you have folks like SpaceX and Google and others publicly talking about the opportunity as it relates to compute in space. And so again, you had Nikesh talking earlier about how data centers on Earth might not be a great investment. Let's start with data centers in space. Is that a better investment, and how does that actually get enabled?

Karan Kunjur (17:35):

Yeah. So what you need for data centers in space is a lot of power. I mentioned the satellite we just launched, it's a 20-kilowatt system. It's the highest power satellite that's ever been launched. That's actually the smallest satellite we will ever launch. The next satellite we launch after this will be 165 kilowatts. So take these two walls, the distance between it and multiply it by four or five. That's the scale of the satellite that you actually need to be able to start to host racks.

(18:06):

And thankfully, it uses all the same hardware that we're currently using, just more of them, more thrusters, more batteries, more solar modules. But if you can get up to that size and you have launch vehicles like Starship and New Glenn, it suddenly becomes possible to actually reasonably put compute on orbit because there's some natural benefits to putting stuff on orbit.

(18:25):

One is power, effectively unlimited renewable power. You don't have the same types of maintenance costs. You have a massively scalable problem because you can basically just add number of launches and pack out the launch vehicle and you know exactly what those numbers are and it's question of when will it reach parity and what will be the cost structure and all that, but it is massively scalable over the right time horizon, and you don't have to worry about permitting and you don't have to worry about people in your districts getting really unhappy because you're taking their water. So it's actually a really interesting dynamic.

(19:02):

And what I tell people is it's not at parity today, but it's not a new physics challenge. It is an engineering challenge and it is not an if, it's a when. So you can debate like, "Okay, is the cost parity going to happen 5 years from now, or is it going to have an 8 or 10 or 15?" That's totally fine. Those are valid arguments to have that are tied to a super heavy launch vehicle pricing and scaling and all that kind of stuff. But I think it is inevitable for space-based compute to happen, and I intend to be in the middle of it when that does.

Tengbo Li (19:36):

Yep. Thank you, Karan.

Karan Kunjur (19:38):

Yeah.

Tengbo Li (19:38):

Thank you for doing this.

Karan Kunjur (19:39):

Thanks.

Disclaimer:

This presentation does not constitute investment advice or an offer to sell or a solicitation of an offer to buy any securities (including, without limitation, any investment fund or investment product referenced herein) which will only be made pursuant to definitive offering documents and subscription agreements.

Certain statements provided in the presentation are made by Karan Kunjur, CEO and Co-Founder, K2 Space. Neither K2 Space nor Mr. Kunjur’s views are necessarily indicative of the views of other ICONIQ portfolio companies or their executives, whose experiences may differ from those expressed in this presentation.

ICONIQ did not compensate K2 Space or Mr. Kunjur directly or indirectly for participation in this presentation, but given his role as a founder of an ICONIQ portfolio company, Mr. Kunjur may be subject to certain conflicts of interest. ICONIQ holds an ownership interest in K2 Space and maintains an ongoing relationship with the company. As a portfolio company, K2 Space may benefit directly and indirectly from its relationship with ICONIQ, including through access to capital, strategic guidance, operational support, networking opportunities, and other resources provided by ICONIQ. Accordingly, K2 Space and Mr. Kunjur may have an incentive to make statements that are favorable to ICONIQ in order to maintain or enhance that relationship.

This presentation may contain forward-looking statements based on current plans, estimates and projections. You are cautioned that a number of important factors could cause actual results and outcomes to differ materially from those expressed in, or implied by, the forward-looking statements. The numbers, figures, and case studies contained in this presentation have been included for purposes of illustration only, and no assurance can be given that the actual results of any ICONIQ portfolio company will correspond with the information contained in this presentation. No information is contained herein with respect to conflicts of interest, which may be significant. The portfolio companies and executives highlighted in this presentation are not representative of the entire portfolio or investing history of ICONIQ Venture and Growth. The types, level, and terms of services and resources actually provided by ICONIQ Venture and Growth to portfolio companies may vary. For a complete list of ICONIQ Venture and Growth’s investments please contact us or visit our website.

Certain of the economic and market information contained herein may have been obtained from published sources and/or prepared by other parties. While such sources are believed to be reliable, none of ICONIQ or any of its affiliates, partners, employees, and representatives assume any responsibility for the accuracy of such information.

All of the information in this presentation is presented as of September 1, 2026 (the “Production Date”) (except as otherwise specified), is subject to change without notice, and may not be current or may have changed (possibly materially) between the Production Date and the date actually viewed by you. ICONIQ assumes no obligation to update or otherwise revise any information, projections, forecasts or estimates contained in this presentation, including any revisions to reflect changes in economic or market conditions or other circumstances arising after the date the items were made available to you or to reflect the occurrence of unanticipated events.

ICONIQ is not acting as an adviser or fiduciary in any respect in connection with providing this presentation and no relationship shall arise between you and ICONIQ as a result of this video being viewed by you.

These materials are provided for general information and discussion purposes only and may not be relied upon.

Explore more founder stories