Mo Islam
Sid, thank you very much for being on the show.
Sid Dixit
Likewise, Mo, always nice to chat with you and talk about our common passions.
Mo Islam
Well, you’ve had quite the career without going too far back because we could talk about it for some time, but you’ve held senior roles at Planet, Amazon, and Maxar. Maybe just to kick things off, give us a little bit on each and how you ultimately ended up as CTO of Maxar.
Sid Dixit
Well, let me start by saying I’m a space fan and I’ve worked at the intersection of Earth observation and artificial intelligence and related technologies for last 10 or so years. I’ll start with my journey at Planet. I used to work at Microsoft where I built and launched Surface tablets and there was an opportunity to lead the launch of 150 satellites in space to build the largest constellation. And when that opportunity presented, I was like simply, who can miss this opportunity to be part of something revolutionary and game changing in space industry? That started my space journey there. I joined Planet. It was probably like 60, 70 people company. And over the course of the next few years, we ended up launching 150 or so satellites, building the world’s largest constellation, acquiring Google’s Skybox imaging and BlackBridge, the Canadian company with rapid eye constellation, and ultimately built a company which went IPO in 2021.
And it was the start of my passion with not only building and launching satellites, but building large data platforms which handle humongous amount of data to be able to process things using artificial intelligence and machine learning such as computer vision. That passion continued in my journey at Amazon, but I had an opportunity to build not only devices such as robots and Amazon Alexa and Echo shows, but build a large platform to collect millions of data points, curate them, tag them, label them for machine learning and training, and create a data pipeline to be able to build algorithms and deployed into production, building one of the largest EIML platform and operations in large tech.
And then after four and a half, five years, there was a unique opportunity where I was being hired and presented by the same search firm which placed me at Planet to go and do this for Maxar technologies. Maxar at that time was looking for somebody to build a geospatial platform which can transform the way the business is done for them and take the conventional approach of selling satellite imagery as data breaks to convert that into a platform-driven, SaaS revenue-driven platform where you take satellite imagery and sell data as a service, software as a service, APIs and UI, but not only build and also build an artificial intelligence and machine learning platform to get insights out of this data. I joined there as a VP and ended up building and launching Maxar’s geospatial platform and then finally left as the CTO for the company after private equity acquisition.
Mo Islam
So we’re definitely going to unpack a lot of the things you’ve mentioned, especially on the technology side and on the acquisition side. But actually, before we talk about Maxar and EO, I do want you to share a little bit about the robotics work that you did at Amazon, because that was particularly interesting to me when we first met. So, yeah, tell us a little bit about what you’ve built, because I actually wasn’t even aware that that product existed in market.
Sid Dixit
It’s a very fun story, Mo. I was a few months into my Amazon journey, and there was a secret project being started, and I was being recruited internally for that. Little did I realize this would turn out to be a monumental project. It was a project to build a product called Amazon Astro. And what Astro is, is a small home robot and a small business robot which does a number of things for you. So think about a use case where there’s a small robot and you went to a tire shop to get your tires changed and it comes to you, instead of you standing in a line to give your keys and get your appointment, to your appointments comes to you like, hi Mo, how are you doing? Welcome, I see you have a tire change appointment. And then you say, yes, I do. Please drop the keys at the back of my car and I’ll take it to the technician, but hey, by the way, would you like to get your brakes changed?
And that was the robot we ended up building. But the most funny story I have around that robot is as follows. Most of this development happened between 2018, 2019 to 2021. And this was time of pandemic. And I also had my youngest son born who’s now four and a half years old. So literally the robot and my son started to grow together. And my youngest son used to think like he’s a family member. He will play with it. He will dance with it. Eventually the robot became so intelligent that it will hear my wife come down the stairs, intercept her in the kitchen and say, hi, Chevy, how are you doing? I haven’t seen you for a while. But that was the journey of being part of the team which built Amazon robot astro is available on Amazon from homes and small businesses.
Mo Islam
So what ultimately led you to decide, well, building robots were fun, but I got to get myself back into Earth observation?
Sid Dixit
I think I did for robot got me even more passionate for some of my unfinished business in Earth observation. I always wanted to take the very latest and greatest of artificial intelligence techniques and platforms to manage and handle large volumes of data and modernize that space. What better opportunity to take that learnings and apply them for the highest resolution satellite imagery at Maxar. So when that opportunity presented, it was a no-brainer for me to go contribute to such a monumental project. We ended up building and launching Maxar geospatial platform as part of that journey.
Mo Islam
Let’s talk real quick about the acquisition because that was a pretty significant one, and I do think it caught a good chunk of the industry by surprise. So it was May 2023, Advent and BCI, two institutional kind of PE firms, acquired Maxar for about six and change a billion. At the time, Maxar hadn’t launched its Legion constellations yet. There was a number of delays. The company had about a billion and a half in revenue in 2022. Obviously, we don’t know how the 2023 ended, but I just want your general sense. And I obviously need to be respectful of the fact that you were an exec. And I don’t want to put you in a complicated position, but why do you think Advent and BCI acquired Maxar? Why don’t we start there?
Sid Dixit
So let me, before I answer that question, I’m a former, I no longer work at Maxar and of course, bound by the confidentiality agreements and respect for the company. So I’ll answer this as a private citizen and I think something to reflect on the larger earth observation industry. I think there’s a great opportunity in earth observation industry to be able to capitalize on a number of things, basically. There are a few large players which have direct inbuilt relationship with governments across the world. They have customer intimacy. They have well-developed go-to ]-market channels, basically.
And they have technology which is great, but can be built towards and further improved upon. Then there is a lot of fragmentation in the industry. So there are a lot of small players with multiple point solutions there are number of earth observation players I think it is only bound that the industry is going to see some consolidation it is going through go through acquisitions and I think the larger strategy is to serve broadly both commercial and defense customers serve internationally and I think some of these acquisitions allow you to do all of this plus bring the very latest and greatest of technology to make grow the size of the pie basically. So I think yes, it could be surprising news as when these acquisitions happen in the industry. But I think this is the beginning. I think I would expect there could be more consolidations. Again, this is purely my personal opinion as I see in hyper fragmented industry.
Mo Islam
So, well, you answered my second question, so I’ll move on to my third, which is, what do you think Advent and BCI’s end goal is here? Do you expect a $6.5 billion acquisition? At the time, a lot of folks argued that, hey, this was a great bailout for the current shareholders, given the share price of where the company was taken out at. Do you think that the. Do you think current owners intend to bring the company back into public markets? Do you think this is a potential sale down the line? And if so, who buys a company like this? Any thoughts there?
Sid Dixit
I’d say that could be in the realm of a little bit of speculative, but let me try to answer this briefly. I think not particularly with this entity, but in general, the goal post-acquisition is to increase the pie, grow the value of the company, and then find whether it’s a public exit or a private exit. Those are the options on the table, especially when you’re talking about multi -billion dollar deals. So there aren’t any surprise options out there. And I think that in this particular case, these are some of the most esteemed companies in the business who have driven this acquisition. What they allow to bring on the table is very best in market development, very best in leadership, very best in talent development, very best practices, which have been successful industry across industries. And I mean, I’m very optimistic that this is going to drive the growth of the market, the value of the pie, the intrinsic value. I think the company, as I have been watching from outside, has been successful in launching first off the Allegiant satellites. And that’s a testament to the positive outlook which is being practiced there. So, I mean, congrats to the company. I’ll always continue to cheer from the outside, but I see great things.
Mo Islam
All right, okay, fair answer. So taking a kind of step back, looking at the industry in general, maybe describe what your thoughts are on the current state of EO. Who’s succeeding, who isn’t? Maybe let’s just, why don’t we start there?
Sid Dixit
So let’s take a very, I want to take a 30,000 foot view or maybe I should call it like a, you know, 600 kilometers view on the Earth, right? And even let’s take a step back even further and go further and see how the space industry is evolving basically, right? I think once we establish that baseline, it will be very useful in analyzing how Earth observation industry would do, because there are many parallels to it in the mainstream time. So overall, the space industry, by going by the dimension of number of spacecrafts basically in space, is going through a rapid transformation. When we built and launched like 150 plus satellites in space around 2015 2016 time frame. It was the world’s largest constellation of space like that regard is far broken in 2013 there were no more than like 1100 or so odd Space crafts in orbit. I could be off by a number but that’s pretty much close. By the time you go to 2033, 20 years thing, that’s going to be over 60 000 such objects in flight, if not more in space, which is like a humongous, somewhere between 50 to 55 fold increase basically.
Now let’s translate that into Earth observation industry. Since then it has been much easier and cheaper to build satellites. And this trend is going to continue forward for two or three different reasons. Number one, there’s a miniaturization technology. We all know about everything is smaller and cheaper to build and from cube sets to mid-size sets to different things. But then there are two more trends which are going to rapidly accelerate and these are fairly new in the last one year. Now. the second tool is generative tools to write software basically, right? It was very complex to write software to manage satellites and spacecrafts and do things, but with Copilot and Augment Code and Devon, the new auto-encoding tools, it is going to be increasingly much, much, much more faster to be able to write the software.
Third thing which is happening in the Earth observation or even in space industry is just like we have tools, to be able to write software. Fast forward two years, three years down the line, we’ll have tools to be able to design hardware. Think about auto hardware tools which can design from reaction wheels to attitude determination system controls to design the exact size and layout of the panels to everything. So this trend is simply going to accelerate. The number of satellites are going to be much larger and the cost of launching satellites is already down with reusable rockets from SpaceX. I think a number of rocket launchers are also developing such technologies, such as Indian Space Research Organization or a number of others, you know, countries and European Space Agencies.
So what’s the net net out of this, right? Hey, we will have at least 60,000 spacecraft by 2033. That number could be 70 or 100k also, depending on how the technology increases. Coming down again from like us that view to again a 30,000 field you will have very different kinds of satellites in orbit. So you’ll have electro-optical satellites, you will have synthetic apartheid, you have thermal satellites, you’ll have hyperspectral satellites, you’ll have RF satellites, and there would be satellites with specialized sensors which could now be launched and built very cheaply. So what would it do overall to the industry?
Every time we see, let’s draw the parallels to cloud computing industry, or internet industry, the cost of web hosting was like humongously high in the late 90s right the cost of cloud storage was human vastly higher the cost of cloud compute was you make us… all these costs are continuing to bring down like you can get you know cost of Internet communication has come down similarly the cost of earth observation data is bound to come down it will create a situation where five ten years down the line you’ll have a humongous volume of data and humongous diversity of data. So Earth as our precious planet when we launch the planet’s constellation for the first time we can image it every day. Now it’s going to be imaged thousands of times a day by cluster of satellites. So I’ll pause there to a mega trends and then we’ll go to micro trends there.
Mo Islam
Sure. Well, a number of things to talk about, but I want to get one question out of the way, which is a question that I’ve heard a whole host of varying opinions on. Or I should say people kind of fall on one side or the other of the coin, which is SpaceX. I’m curious that if you agree or disagree with the idea that SpaceX will own Leo, and just curious what you think that impact is going to be on the Earth observation industry. Or, yeah, maybe let’s talk a little bit about that and Starlink and Starshield and just general, like, what SpaceX is doing with LEO kind of manufacturing.
Sid Dixit
So I would, there is a fair chance that SpaceX can, or I should rather say SpaceX would be a major player in LEO, right? They have proven the propensity to be able to build and launch thousands of satellites. I mean, we are looking at Starlink satellites, which are in multi-thousands right now and that has completely changed the dynamics of internet connectivity from space business completely like draftically change on commercial space the revenues if I read it correctly is nearing five billion dollars a year like that’s no small feat What it took for conventional pairs to do that over 20 years now they have done this in like five years, right?
I think SpaceX have made their plans clear to be able to go after defense and intelligence business and government business There’s a huge opportunity to be able to provide internet connectivity for these such players And and I think that’s something bound to happen and they will they have a very strong foot in there. But there’s one more aspect of it. The way they have designed their satellites is very modular. They have designed as a bus and a pillow and they have shown propensity to be able to manufacture a large number of satellites. There’s nothing stopping SpaceX to be able to replace their payload, which is an internet and RF connectivity payload, with optical payload, with hyperspectral payload, with SAR payload.
With SAR it’s more complex, it requires a complex array of antennas and other subsystems, but thermal payload and RF payload, and provide multi-sensory data, provide internet connectivity for both assets on the ground, but internet connectivity for assets in the space, and most importantly, reduce the latency. And I want to talk about what does reducing latency means, you know, and that changes the game of the industry, right? Say, hey, let’s imagine a defense use case, a Earth observation provider is taking mission critical images from a place of where active or situation is going on. Now when a satellite takes these pictures, it has to find the next ground station to download it. The typical orbit is 90 minutes. You may do it every 90 minutes or you may intercept either Amazon’s ground station or some of your own multiple ground stations to do less than 90 minutes. But there’s going to be a lag, but there’s 10 minutes to 90 minutes that applies for most of the players now, right?
And with SpaceX, you have a web of satellites circling there is a hexagon all interconnected, you’re going to transmit that information for the speed of light. That latency going from hours to minutes to seconds completely changes the game. So they have the infrastructure. Can they match that infrastructure with different kind of Earth observation payloads? Absolutely they can. And will they? That’s to be seen, but it seems like a very real possibility. Combine it with the reusable rockets where they can build and launch their own payloads multiple times a week. You’re looking for a perfect storm to engulf the industry.
Mo Islam
So, if they do end up going after the remote sensing market, what do you think happens to all the commercial startups that are operating in Leo right now?
Sid Dixit
You know, we started by talking about the space of space is starting to get crowded and no pun intended here. You know, there are advantages and disadvantages of doing that. When SpaceX builds and launch their own satellites, let’s say for our observations. That doesn’t mean it cannibalizes all the business of all the startups, right? So let me talk about why the startups are peer side basically, right? It’s a startup funded by Techstars and that’s recently raised six, seven million store funding. But what are they doing? They’re building a vertical application to use SAR imagery to do a maritime solution basically, which combines SAR imagery, electro-optical imagery, EIS signals and be able to provide, you know, illegal phishing and other such things, right? Now, I don’t think so SpaceX is going to directly cannibalize startups which are building a fully integrated vertical application. They know their customers very well, have customer intimacy and understand their pain points. But what it may do is, in fact, assist them.
This company such as these no longer has to launch an EOS satellite, an AIS sensor satellite, and a SAR satellite all of a sudden that own, but provide them data cheaply enough to be able to build rapid solutions for their customers and do what they do best, which is focus on their core competence. So startups like these are going to survive. But if you think about companies which purely are going to be in the space of like startups or otherwise selling data as data breaks or data as a service, right? I think if you see more proliferation of Earth observation satellite. That is going to be a market where the cost of data is continuously going to come down over the next few years. And I’m drawing analogies purely from cloud compute and internet things like, you know, like the Moore’s law, which applies to semiconductor is also going to apply to space in some way, shape or form where per kilometer of satellite imagery is starting to come down as more and more satellites are available for cheaper and cheaper price points in space.
Mo Islam
So we’ve spent some time talking more about the supply side of the technology. I want to talk a little bit about demand. So the common color that we’re hearing and supported by the earnings calls, like recent earnings calls of the public companies, is the slowing pace or like the much slower pace of commercial demand. So curious, what do you think about commercial demand in the industry? Is it where it should be right now or is what you or maybe what you predicted while you were at Maxar and Planet where it would be today? And if not, sort of where are the commercial customers?
Sid Dixit
I think with the most recent downturn, there was certainly, my sense has been that when it comes to order of priorities, for Earth observation data is not on the highest order for many big tech companies. So imagine companies like Facebook, Google, Microsoft, Amazon. As you have noticed, these companies have been going on, I’ve called it like voluntary margin expansion basically, where they have improved their cost structures and reduced their cost and now the companies tend to be hugely profitable. They’ve improved their margins. Often the demand for Earth observation becomes, is one of the factors which is scrutinized. What does that mean basically? Let’s say that Google is gulping data from two or three or four major players basically. They will still buy imagery. They might just buy from one or two highest resolution imagery providers basically, but cut down multiple suppliers where they will buy every available data on the market, right? So those things are playing out.
The demand has certainly softened. But I want to drive to the other tail end of it basically, which is very interesting. It’s the rise of generative AI and these large language models, which have now become large audio models and large image models, basically. There’s an increasing trend on the commercial side to do what is being done for the rest of the imagery, to recognize human subject traffic and everything on Earth, to do the same thing for geospatial. In fact, many of the models like Chat GPT-4.0 have already done that, where they can substantially recognize things from satellite imagery. There is an increasing trend for training data, which is now how much do these two trends balance out as to be seen. I think overall it’s a very exciting juncture, you know, where finally there are more use cases which could be solved using generative AI and that will eventually drive that band back up.
Mo Islam
Well, let’s hone in on technology a little bit. So maybe talk a little bit. So, okay, let’s actually start with the current startups. Like there’s a number of startups out there, Optical, SAR, you know, hyperspectral, all sorts of different types of imagery and at different resolution qualities, right? You know, especially on the optical side, it’s like 50 centimeter, 30 centimeter. Now folks working in VLEO at 10 centimeters. So like, are the startups that you’ve seen, that you’ve worked on and that you’ve communicated with, do you see those as marginal improvements in technology or do you see them building stepwise changes? Like where do you think all of the startup, how do you sort of relate all of the startups with one another? They’re so, like, I mean, I could name 10 right now off the top of my head that have raised fairly substantial amounts of capital. Like it’s starting to actually, I don’t want to answer for you, so I’ll toss it over to you.
Sid Dixit
Yeah, I think Mo I would like to answer this question at a little bit of a uber trend and drill down to the startup. Would it be fair? I give you a little bit of my thought. I think when it comes to Earth observation, we have to start looking this analysis of satellite imagery and other things as how we as humans analyze things, right? And, you know, we have language interfaces, we have visual interfaces, we have audio and perceptory interfaces, and then our brain has lobes of brains, which is like for math, which has for images, which is for text, which is for logic and all of these things. I think the way artificial intelligence is converging, we are going into the world of multimodal analysis and eventually a multimodal brain for analyzing satellite or earth observation in nature, right? And we have all fidelity of resolutions. You’re talking about low resolution, high resolution. I mean, we have peripheral vision and which is low resolution and we have audio at different resolution to our humans. With the technologies such as generative AI, knowledge graphs and racks learning and transformer models, we are looking at perfect strong where multiple modalities are starting to fuse and we’ll have a sense of fusion. So what does it means for earth observation on these startups?
Number one, it’s no longer okay or even viable to just use one kind of satellite imagery, whether it’s hyperspectral or thermal or SAR, to be able to analyze this data. So you have to analyze from multiple sources.
Number two, each source and each thing provides you some unique thing. You have to build some custom brain lobe equivalent, your secret sauce, to be analyzed that well enough. But the challenge is, You have to work in unison with other data sources many times to build the best of the customer use cases. So if you’re a defense customer, you want like every single data source fuse basically. Now there is a challenge and an opportunity for these startups, right? If startups are building marginally improving algorithms, let’s say on SAR imagery, right? Or some startup is building a computer vision algorithms on optical imagery. Those could be wiped out overnight because large language models are building these capabilities inherent and not just large. We are in the world of ImageNet and large image models, which are now part of chat GPT-4.0 already by design, right? So their secret sauce of being extracting information can go away, right? But I think what is still valuable is understanding a customer pain point and building a vertical solution. Where does this lead to?
Like these startups which have raised a lot of money to build simple things are going to find themselves challenging situation also if they don’t have that customer intimacy built in a vertically integrated customer solution built in. The other side of the coin is as the need for these multi-sensory models and multi-sensory technology goes up. I mean, we talked about consolidation in the industry and acquisitions. I mean, this is the right time for it. Like,there could be large players which have customer contracts, they have primes, they have inbuilt customer intimacy, they have large sales teams which can sell it to multiple. I mean, they can acquire the small companies which actually is a good way to get an exit. It is very, very difficult to build a point solution in a space-based startup and take it to an IPO.
Increasingly, the acquisition by a public or private companies, a great exit path basically. So it has to be balanced, but startups have to adopt and they have to rapidly adopt. They have to run super lane basically. And they have to use as many building blocks as they can do. They have to work out partnership with the data provider so they can, they don’t drown in, you know, the cost of, you know, getting the data. I mean, I’ve seen so many startups spend millions of dollars of their funding and acquiring data. And I mean, we can wise ways around it, like partner with the companies, give them equity stake, equity for imagery and stuff like that. Then you have much better chances of being successful. I love advising startups and happy to provide any insights if anybody ever needs.
Mo Islam
So with regards to generative AI, large language models, I mean, we’re talking a little bit more on the analysis side, right? Not on the data acquisition side. So maybe talk a little from a practical perspective, like how is that going to ultimately impact the user? And does that open up the, I mean, everyone’s sort of asking the question when, you know, we’ve been talking, as you look at a lot of the presentations from a couple of years ago of most of the EOs companies and startups, there’s a, you know, everyone loves to love to break down the government versus commercial revenue and you and everyone loved to show how much bigger that commercial pie was going to get over the next five to 10 years versus government. But ultimately it looks like government is the lone sort of buyer right now, imagery data. So maybe talk a little bit about the tools that you’re referring to. Is that going to help open up the market, especially the commercial market? Does it make things easier? Maybe talk, yeah, talk about that.
Sid Dixit
So two parts of the question. First is the tools such as the chat GPTs or perplexity or myths trial. How are they going to affect, you know, earth observation, imagery data analysis, number one. And how does this plays out in commercial as well as government space, right? Is that correct, Mo?
Mo Islam
Sure, sure.
Sid Dixit
Up until now, let’s say you’re building data to analyze account ships basically on a port, right? And you can spend a few millions of dollars to train the model and deploy it and build a full product out of it. Now there are large image models, large language models, large multimodal models which are starting to come up, which actually require billions to train basically. But then for users to use, they are like super cheap basically. So you could be paying a few cents per image recognition or even lesser on bulk partnerships. What it does do is it allows rise of rapid startups. It allows formation of new use cases and things like that. So what does that mean for example? Okay, if I’m building a new startup, which does insurance underwriting basically for insurance providers to determine the risk of my house using satellite imagery, right? I would have to spend millions to acquire satellite imagery, strain my model, deployment model, and build a vertical website or apps to sell this thing. Now, this experience is simplified, I no longer need to acquire millions of dollars of satellite training data. But I can simply use existing models, maybe fine tune them or use racks or other technologies to augment it. But I can simply build a commercial web application with pricing plans and everything. I can hook up my startups to all these satellite imagery providers which are selling data to retail basically.
So you have up 42 as one of the examples, you have Skywatch and Skyfire, the other two basically. And you can hook up, I can buy a location for a few hundreds of dollars for the houses I’m wanting to underwrite. And then use that pipeline of a chat GPT for service to analyze whatever is there and deliver the result for my insurance underwriter in a simple text like, hey, this house is near a brush fire, it’s on the edge. the canyon, it looks like there could be some rest to it, there’s a pool in the house, some of the ties look missing or whatever the use things might be. What it does is it suddenly made the cost of building the startup from needing like $10, $15 million funding to be able to down under $1 million or even not even require VC funding to be raised. That does open up the market, that does open up the commercial use cases. I’ll pause there before I reflect on the government side of the business, how does it change?
Mo Islam
Well, before we go to the government, have you tried any of these tools?
Sid Dixit
Absolutely. So yes, I was playing with chat GPT-4.0 using satellite imagery and I gave it a random imagery of a military base, you know, a few military bases, you know, and see, let’s see what happens. I have no expectations what is going to happen. And the output of my results was it analyzed it. Some of his comments was, hey, it looks like a satellite imagery of a military base. Based on the architecture of building, it looks like Russian architecture. So it could be in Russia or it could be in the countries where Russia built such a thing. Based on these topography, it looks like Arid and it looks like this part of the country, right? In the image, I can see these following aircrafts being posted. They look like a bomber of a particular class. It looks like a MiG being there. And it also looks like, you know, there’s an SU-32, one of these aircrafts being done. And that’s a very precise identification.You know, it’s a very.
Mo Islam
Wait, Sid, hold on. I have to clarify. So you’re saying that you threw a random satellite image of a military base into GPT-4.0, and that is sort of the output that you got. You know, I have to actually just stop feeling surprised these days with all of these examples, because I feel like I hear one every day where I’m like, wow. But wait, so you’re saying that that was the output, and it must have been this past week that you tried this.
Sid Dixit
I, exactly, this is in the past week after the release of 4.0 model, basically. And the model was like really astute in identifying the scene, basically. I threw it in another imagery of a military base again, simply found by using Google Image Search, basically. So nothing proprietary, nothing this. And I asked it to say, analyze this second imagery, and it was like, hey, I see some trainer jets, I see some military base. I see a bomber part being there. Actually cross-referenced with the information available on the web, and it looks like this particular region in Russia basically. So it was very very precise in correlating and it used the archive of web of all the available imageries to pinpoint with the image VR. Now these are the native capabilities which are not even fine-tuned and this is like a relatively cut copy paste prompt engineering. Imagine what could happen when you are able to integrate these models into software in a much more systematic way.
Mo Islam
Right. So who becomes a winner here? Is it the software provider? Is it the acquirer of the data? What part of the stack, I know, across Earth observation is the big winner in your mind, given what you’re describing?
Sid Dixit
Well, it’s a complex question, right? But firstly, there are new players in the stack, right? The stack is opening up beyond the space industry. So you’re going to find players such as NVIDIA, Amazon, Microsoft, Google Cloud providers, as well as large image models such as OpenAI or Anthropic or other players starting to get the pie of it, right? But then if you think about, hey, you can build all these applications, who’s going to take it to the market basically, right? And these are going to be the large players such as Airbus or my former employers or Black Sky or some of these other players which have customer intimacy, right? Which have customer established contracts, they are going to be winners, right? You will also see people like Arthur Gurman, Lockheed Martin and others which have other parts of the stack and they are prime providers starting to win. This is on the defense side basically, right? I think it’s increasingly very, very difficult for small startups to crack into defense and intelligence wars. Like they will often bring customer applications which now could be built by even larger players for a very little cost. So, it’s less attentive to partner sometimes. But on the commercial side, the biggest winner is, and even on defense side, the biggest winner is the customer basically.
I will give you two commercial and commercial and commercial and what it what it called…defense use cases. Like the first commercial use case, imagine, hey, I’m a CEO of a solar system installation company, Tesla Solar Division. And I want to know what are the good markets to go after to send my salespeople or send a direct communication to the customers to install solar. What are those good markets? Like where there’s plenty of sunlight, where there’s other people have installed solar, but the neighbors have not installed solar. So there’s penetration, those markets are fluent. You can do these identification immediately using satellite imagery and the generative AI models. Basically, you can get exact answer. Hey, go and target these zip codes because we see there’s already a movement where people are installing solars, but the penetration is still low, right? On the defense side, imagine you’re an analyst or a mid-level military officer in Japanese Navy and you go in front of your computer or software and say, hey, please provide me the situation report from last 24 hours. And the system says, I have been analyzing satellite imagery from a multitude of sources. I see a movement of ships from a country X going from port A to port B. I saw five new kind of ships in the caravan, which is unusual. They look like military ships. There’s a military formation. There’s a supply vessel. I have increased the frequency of tasking of satellites in that area from every week to every two days. Please approve this request. And I’ll continue to monitor the situation and I’ll give you alerts as soon as new data is available.
Now, that’s the kind of use case which is suddenly very big. So I’m big like, we are going towards a world where geospatial is going to become just like programming and just like other languages which we use. I think it’s going to be taught in high schools very soon. It is going to be just whether data will continue to use this other modus apprendi to know more about our earth and make decisions. End of the day, customer is the winner, whether it’s a commercial or a defense customer.
Mo Islam
Do you think satellite imagery will become commoditized?
Sid Dixit
It is bound to become commoditized to a certain extent, but I do want to call a differentiation there. Satellite imagery comes in many resolutions and many sizes, so think about… Planet was 3 to 5 meters resolution in past and then there’s sub one meter resolution between 100 centimeter and 50 centimeter. Then there’s higher resolution which is 50 centimeter to 10 centimeter. The highest resolution or ultra-high resolution, call it 50 centimeter, 10 centimeter, those satellites still going to cost tens and tens or hundreds of millions to launch. So it would not be hyper commoditized, but there are more players like Alvido’s launch with 10 centimeter satellite imagery, right? So that will be. The middle band, which is 50 centimeter to hundreds of meter imagery, that’s tons of satellite launches. And guess what? When you bring these millions of data points of satellite imagery. They’re not for human consumptions.
They’re good for machines and it is based on experiments, you know machines are really good up until like 85 centimeters of resolution, right? So you don’t need to be ultra high resolution sometimes so that sig middle segment is going to be hypercommodized and then there is a segment beyond that which is not finding as much value at least in my opinion, so You know, that’s probably going to be not too many players there, which is low resolution. Yeah
Mo Islam
Right. At what point do you think imagery quality improvements start to not make sense versus market need? Where, because image quality will start to get improved. People will figure out better ways to do it, you know, better sort of hardware flying closer to earth, you know, lots of different things that will do an attempt to improve imagery quality. But at what point do you think, hey, where it’s not necessary anymore. We don’t need to see, you know, the guy’s bald spot from space.
Sid Dixit
So Mo hidden in your question basically is also the volume of volume question basically, right? We did an experiment where we calculated f score is a terminology of how good machine learning algorithms recognize things out of satellite imagery and the resolution of satellite imagery. And we plotted them, you know, we found out that somewhere between 70 to 80 centimeters resolution, that’s a sweet spot, right? You could have 70, 80 centimeter resolution imagery and this is incredibly good enough to be analyzed by machines basically. So you sometimes don’t even need to go lower than that for I would say 90% of the use cases basically.
And that’s good enough. But then there’s no end to human curiosity, both for commercial needs as well as defense and intelligence needs to be able to look deeper into what exactly is going on. And at that point, there’s always going to be a need for 30 centimeter resolution, 10 centimeter resolution imagery, five-centimeter resolution imagery to go for these very penetrated, very deep things, you know. End of the day, we should not forget space is kind of like international waters. Anybody can image almost anything, you know, and that provides an interesting opportunity if you can do that at the highest resolution. So it’s not going away. There’s a need for it, but it’s not 90 % of the market. It’s probably 10% of specialized use cases, which could be most valuable, by the way.
Mo Islam
So a couple final questions. One, which is if you had a million, let’s just say, you know, you’re managing a million dollars to invest. Invest specifically in the space industry. How much of that million is going into EO? EO and EO-related technologies today, starting today. And Sid, I know you’re not a professional investor per se, but I’m trying to get a sense of how you view the industry versus, or how you view EO within the confines of the space industry at large in terms of growth and opportunity.
Sid Dixit
Well, I do actually invest and have been actively investing in a number of startups and as an angel investor and seed investor. And actually we have put together a fund called 5U Ventures, which I do participate in it. But to give it a specific answer, Earth observation industry is not the easiest industry to penetrate. If I’m deploying a million dollars, I’ll probably deploy 20, 25% of the earth observation, but also deploy that remaining 75% in the diversified strategy in other parts of the space industry. And let me call out one specific area, which is quite valuable. When you’re in the industry, which is, well, you know, let’s think AI race, right? Why is Nvidia winning?
NVIDIA is building shovels and pickles to dig the golden AI industry. And I think the same approach does apply to Earth observation and the industry. There’s a race, and who are the players who are going to win in this race, basically? Of course, there’s vertical apps and Earth observation players, but there are also players who are going to build buses to build these all satellites, right? People who will build common technologies, such as maybe large platforms to process all this data or some of these things. I will certainly invest another 25% in this to 30% in the areas where people are building the shovels and pickles to help accelerate this industry.
Mo Islam
Are there any Earth observation startups that you want to highlight as, hey, these are the ones I’m really paying attention to or spending time on?
Sid Dixit
You know, there are two which are very interesting to me. I think I’ll fall out of this. One is Apex basically, they’re building the bus basically, and they’re trying to commoditize the bus so you can put any kind of payload on that bus basically. And I think Apex is something to look after. They are trying to build a commoditized bus to the market and this will help multitude of startups and even big players as they want to launch satellites. Another one is XDLINX Labs. It’s an India slash US based company is in the business of building sovereign constellations for different governments. Now, because the cost of building constellation is coming down, a lot of countries are thinking, why not just launch our own sub-meter constellation?
Why we can own it, we can point it wherever we want, and we have access to it no matter what rest of the geopolitical situation and geopolitical situation of our situation in the world. I think that trend is going to come up and companies like XDLINX, which will give constant installation as a service to different countries are going to be valuable. So I do plan to invest part of my imaginary $1 million fund or I should say real $1 million plus into these companies.
Mo Islam
Yeah. Well, Sid, it’s been a pleasure. Really appreciate you being on the show and super insightful and also exciting, at least for the customer. Very exciting future. I mean, I know today you can go and spend less than a thousand dollars pretty much to a task a satellite and take a picture of virtually any place on earth except maybe say missile silos. But I actually did this recently for an event. I tasked a satellite to take a picture of Santa Monica. And I presented and said, look, I acquired this data in like two days, 48 hours. And what does that future look like in five years, especially with the tools that you’re describing? It’s pretty exciting.
Sid Dixit
I agree with you, Mo. I’ll very recent observation, there’s a startup called Atlas Pro AI, which I’ve advised that they’re building a system for a telecommunication industry where these providers can actually task a satellite and see what’s going on with their telecom projects in commercial space with like less than $1 ,000, right? Anytime they want to see. So whether it’s commercial or defense, this is an exciting world. The pie is going to grow for mankind. And I think Earth observation is going to become a crucial part of it basically in fact not the Earth observation but the space industry is going to grow I’m hyper excited about the time there you have thousands of satellite giving you real-time information almost like you know internet of space But you also have capabilities such as generative AI to write code to build new satellites or software and even generative Hardware tools to come which will build these building blocks to be able to build large satellites. I think we are at a phenomenal point in human race where the space is going to go on an exponential trajectory and we can build more spacecraft, we can launch spacecraft for cheaper and you know I would say space or galaxies are the limit. We are in the accelerated zone in the final frontier.
Mo Islam
Very well said, Sid. Until next time, thanks for being on the show.
Sid Dixit
Until next time, always a pleasure.
