Space Observation in the Age of AI

Great to see you, Jeanne.

Absolutely. Well, look, with Ukraine, we are imaging all of Russia, all of Belarus, all of Ukraine every day. It's over 18 million square kilometers, so twice the area of the United States roughly. So it's a huge area. And then using AI on top of that to find threats. They do things like building damage assessment, plan strategic strikes, they do threat assessments. It's really early warning not just of what's going on in military installations in Russia, but also industrial activity and supply chains that might lead to that. They're trying to get the warning time not from hours or days, but maybe to weeks or months so they can plan accordingly. This information advantage has been critical to the balance of power. Drones, satellites, and AI have really been critical in Ukraine, and we've been proud to help support them.

Well, absolutely. As I was just saying, it enables the peripheral vision. Drones can't cover all of Russia for multiple reasons. Firstly, it would take millions of them to do that on a daily basis to give you that peripheral vision. But secondly, they don't have the ability to go there with air because they don't have air superiority. Another way of thinking about it is actually: where do you fly those drones? Where do you send them? You should send them to places where you know there are threats. Actually, in Ukraine they quickly learned that when they were using ISR drones—so drones with cameras—they were sending them up to the front lines, but one in two of those drones weren't coming back. Instead, now they just look in our imagery, and if there's no changes from yesterday to today, then they don't need to send the drone. So it's a smarter direction. So it's the tipping system for drones. The foundation of that in Ukraine has been satellite imagery that then tasks all of the higher resolution and other uses of drones.

We have learned a tremendous amount. As you said, it's an incubation, a test bed, and necessity is the mother of invention. Ukraine has been definitely leveraging these new tools, all of them together. And I would say what we have been learning is speed really matters. So latency of getting these answers is really important. The peripheral vision is really important, and the only way we can tap the peripheral vision—that scan of all of Russia every day—is if you have AI on top of that. So leveraging AI is critical to doing the kind of analysis that you need at speed and at scale.

Well, it's a great question. Let me give you a couple of examples of how we already are. I mean, we're just here in Ukraine, but we also work, for example, with the US Navy, and they're monitoring all of the South China Sea. Imagine you're an analyst in the Pacific trying to monitor the whole of the South China Sea. It's too big, right? So we help them by imaging all 13 million square kilometers, automatically finding ships. They can set thresholds for: tell me if there's this kind of activity; tell me if there's ship bunkering, for example, when ships get together and do some illicit activity; alert me across this vast area of those sort of challenges. So we can do things like we work with NATO on doing similar things to what we provided for Ukraine, but in the Baltic Sea and across the whole eastern flank of NATO. But there's so much more. So that's the sort of defense application. But in the intelligence world, whilst we've seen Ukraine innovate there, the US, you'll be surprised to hear, actually hasn't leveraged this data yet. The US intelligence community uses, you'll be surprised to hear, less than 1% of our data today.

Well, because their geospatial process is dependent on a few analysts. And if you just use analysts, they were used to dealing with thousands of images per day. But Planet alone has millions of images every day, so it's about a thousand times more data than they were used to. They couldn't deal with it through analysts. You've got to use AI, and what it can do is not replace the analysts; it can empower them so that they can cover a much greater area. But they haven't done this shift. It's a cultural shift; it's a change in mindset. Also, frankly, the requirements process is broken, right? So what happens is they don't even know they can ask for this capability, and the process for procurement is slow and very specialized, and this capability is fast and very general, and so it doesn't match together. We've got to have new kinds of funds that are fast and flexible. They don't need to be vast. They don't need to be in the billions. They actually can be relatively small; they can be tens or hundreds of millions. They need to be able to change when there's a new AI technique, they apply it. And I would call it moving from a sort of requirements-driven process to a more capabilities-driven process.

That's what we do in Silicon Valley. We look out at the principal needs of users, not just what they're asking for. Not just, "I want a slightly better computer like this." We think: what might they want to do? What's the power if we could do everything? If we could answer all these things? That's what the AI companies and the other tech companies around here are doing. And if you take that lens and you go, "What are the principal needs of the IC and the DoD?" Well, it's find all the threats around the world and give you real-time information about that. There's no specific requirement for that because no specific person in some specific place asked for that, but yet it's empowering across the board. So the procurement system is broken and they don't have a direct need. But I would say adapting to leverage this tool is no longer just an efficiency matter. It is faster and more efficient, and we've seen in Ukraine how important it is. It is an existential thing if your adversary is using that and you are not.

Let me be clear. The US government is using it. The NRO is using it. The NGA is using it. But they're only really using the tasking system that we have, not this scan. And so the benefit of that scan—I think we need to change that, and we need to go faster to adopt this technology. I think this administration is leaning in to adopting commercial technology and breaking some of those barriers that have existed that are getting in the way to help leverage this sort of new technology.

Not many. I think there's cultural differences and speed and so on, but not really. I think that we're seeing companies like Planet are proud to work with the government and we want to try to support them, and I think the government can so benefit from these sorts of technologies.

Yeah, this is a fun one. It's really quite an amazing project. Google and others have been looking to see where can they expand compute. They just can't go build compute fast enough. And Planet and Google are working together on a joint mission to see if it's possible to put data centers into space. Now, that might seem crazy and far out, but as launch costs come down and satellite costs come down, it's only a matter of time where the benefits of space—which include eight times more solar power per panel, continuous power, continuous cooling, plenty of space, no issues of land, and so on—can both be more efficient and eventually more sustainable as well because we don't need that land and continuous resources. Space geeks have been thinking about this for decades. We talk about the fact that we can take energy-intensive infrastructure from Earth, like compute, like power, into space and then get the benefits just beamed down. Finally, it's not crazy to think about that as launch costs and satellite costs are coming down. So we are embarking on this. It is very experimental at this stage; it's early days. But if it is proven feasible technically, it's an incredibly big opportunity. It sounds in the realm of science fiction, but it's becoming in the realm of science fact. We're putting up a first demo next year with Google. We're very proud that Google selected us, and it's because we've launched hundreds of satellites before and done that sort of scale. We've worked already with them a lot on AI, and so it was a natural fit. We were very proud they selected us for this moonshot project. It's exciting times.

Well, that's exactly some of the things we're testing on the first missions. One of the toughest things is the thermal environment. It's how to get rid of the heat that you generate. On the ground, we would use water cooling. Obviously, we're not going to do that in space. We have free cooling to the cold background of space, but it's radiative cooling, so you have to build lots of radiators. Then there's upsets from radiation that could happen. These are all the reasons we're putting a few TPUs—Google's advanced processing systems—into space and seeing how that works and learning these lessons so we can decide how and what it would take to scale a project like that.

Not too big. Google envisions clusters of spacecraft—they published a paper on this—that would be quite close together. It would be optical links between them. All of that sort of thing has been solved; there's no fundamental things standing in the way. It is a big engineering challenge, and so we're very excited to be working on it.

Yeah, I mean you have to do formation flying, which is a real technical issue, but there's nothing that we haven't solved before in doing that. It's just the numbers of satellites that you have to keep flying next to one another in proximity, but there's no outstanding fundamentals to doing that. We have to fly it low to make sure it doesn't create long-term orbital space debris, just like other big constellations. This one uniquely will fly in a sun-synchronous dawn-dusk orbit, so the solar panels always face the sun. That's one of the amazing things we can do in a dawn-dusk sun-sync orbit. So we're thinking about all the challenges of that, and it's exciting.