Planet Labs 工厂探访

We’ve built about 700 satellites here in the last 15 years. That’s more satellites than any other company on Earth except for SpaceX. All 700 of our satellites cost less than a typical satellite at NASA. That’s about a 1,000× reduction in cost. But this is our AI interface. We can ask it questions. We’ve tied an LLM to the back end, and it can search the entire Earth in under a second. Someone at NASA kept saying when we were there, “Why are we spending a billion dollars on a satellite when a phone like this has most of what a satellite has, but it costs 500 bucks?”

So what was the answer to that? All right, we’re here at Planet Labs. Thank you for having us.

No problem. Welcome.

So what are we going to see today?

We’re going to see some satellites.

Okay.

Also some AI stuff. You’re going to see some of our new AI interface to all of our satellite data. But let’s—we’re starting down here in the satellite lab because it’s cool.

I got a little peek, so you guys are in for a treat. There are some vibration plates and some… What is that one thing you have, a thunderstruck—

Yeah, that tests the—

…testing table.

…the shock table. Yeah.

Yeah. Okay, so as we walk through, we’ll start over here. I know you have some older satellites we can look through, but what are the different areas that we’ll see today?

Basically, this is our satellite manufacturing facility, so this is where we build loads of satellites. We just launched 20 more last week. We can pump them out. In fact, we’ve built about 700 satellites here in the last 15 years. That’s more satellites than any other company on Earth except for SpaceX. This is the manufacturing facility for satellites. We launch about four or five times a year. We’ve done it on over 40 rockets.

What has really changed in the satellite game has been the sheer size and number of satellites.

Mm-hmm.

Planet pioneered miniaturization of satellites, taking satellites that are bigger than this room and shrinking them down to a small size so that we can launch loads of them. With loads of them in space, we can do different kinds of missions. Before, each satellite was the size of a double-decker bus and cost a billion dollars minimum. Some NASA missions cost up to $10 billion. The cheapest ones cost a couple of hundred million. All 700 of our satellites cost less than a typical satellite at NASA, so it’s about a 1,000× reduction in cost. That’s all to do with the miniaturization of electronics, and that’s what we pioneered here in this lab.

Something I definitely want to talk about during our sit-down is how the use cases have changed substantially, especially with the new thing you guys just launched with Google.

Yes.

So you’re moving into the world of intelligence, which, by the way, I heard has great premiums on it. But can you talk through—

Yeah, it’s fundamentally a data business. We’re selling information services.

Yeah.

Exactly. Even though we have a satellite back end, our data business is a very high-margin business. Each sale of a picture or analytics from that picture has very high margins.

Mm.

I love that because, as a businessperson, we do have some CapEx, but it’s a very high barrier to entry. People launching satellites are few and far between, and they’re very complicated, so we have a high barrier to entry. But our margins look like a software business, and so it has the best of both worlds.

Amazing. Okay, so what do we have behind us?

Yeah—

Let’s talk through the lore of Planet Labs.

The lore. Exactly.

Because we have some right here.

Yeah, we do. This is the whiteboard from our garage, where we were playing around. This is our initial team. We were building our first satellite there in the lab. We were still working a little bit at NASA, and then we slowly left. We were doing this on our nights and weekends, and then we were like, “Okay, this is too cool.” So we started building these satellites.

You can see various bits of them: solar panels and the actual boards that we make for the satellites. I can show you the optical telescope assembly. This is the telescope and the camera system. You can look down there and see the actual telescope.

Oh, yeah. Oh.

You can see yourself in the mirror, so to speak. That is looking down, taking pictures. Now we’re 500 kilometers away, about the same distance as from here to Los Angeles. We’re taking pictures from that far away, but we can see individual trees. We can see houses. It’s amazing that we can do that from 500 kilometers away with a telescope this big.

This is what the satellite looks like all put together, except for the solar panels. This is one of our older ones, but it gives you a sense. This is the telescope. These are the motherboards. This is pointing down, right? That’s taking pictures. It also looks up into the stars. It automatically recognizes the constellations and then goes, “Okay, well, that constellation’s there, therefore I’m pointed exactly in this direction.” Then it can turn itself around using these little wheels, so it can spin itself in any direction to take a picture that we want it to take.

Yeah.

Then it beams the pictures down to the ground using these radios. You might think, “What the heck are these bits of wire?” We can get two gigabits a second from our satellite down to the ground station over 1,000 kilometers away. We’re talking about two gigabits a second down to the ground over a range of 1,000 kilometers. Sometimes our internet in this building is worse than that.

I understand. I understand.

Basically, all of this electronics tells you where the satellite is in space—that’s the GPS—and where it’s pointing, with a star camera and the sun sensors.

Mm-hmm.

Then we take a series of pictures, and we embed that metadata on each picture so that we know where we’re pointing. Then we stitch that to the Earth’s surface. We have about 200 of these in orbit. They image the entire Earth every day. It’s the only system that does that, that images—

And—

—the whole Earth every day.

…and how many images do you do a day, and what is the file size of that?

It’s about 4 million 47-megapixel images, basically. So it’s on the order of 100 terapixels.

Oh my God.

Yeah. And 8 spectral bands for every pixel, so it’s actually kind of 8 pictures—

Mm-hmm.

—for every point on the Earth’s surface with different colors. But not just the 3 that we have, RGB, what our eyes see—

Mm-hmm.

—red, green, and blue. We have 8 spectral bands that cover near-infrared into the optical wavelengths.

For people who might be newer to Planet Labs, why is image data so important, maybe versus radio occultation data?

There are all sorts of data that you can get from space. Imagery is, of course, super useful because it’s the kind of data that we can visually interpret. We’re visual creatures, right? You name it, we can see it, we can understand it. Also, there’s a huge amount of work in vision-language models now.

So, graduating from the machine-learning things that we’re doing—recognizing cats and dogs in videos online—

Yeah.

—we can now recognize buildings and ships and planes and trains, or whatever it is that you’re trying to find. AI is very ready for it, just like we’re very ready to look at it. Of course, we can’t look at 4 million images per day, so we’re using machines to do most of the heavy lifting there.

Mm-hmm. Okay, so—

Do you want to hold it?

Sure. Oh my God.

What do you think?

Oh, it’s not that heavy.

No, don’t drop it.

It’s really light.

Actually, if you dropped it, it wouldn’t be so bad, because that’s more or less like dropping it from about this height—

I heard one of your jobs is to throw this against the wall.

Yeah, exactly. I’ll demonstrate in a second. But dropping it from about this height onto a marble floor is about the equivalent of the shock loads it takes when the 2 stages of a rocket separate.

Okay. Yeah.

It’s kind of—

That’s intense.

Yes, 200 Gs. It’s extremely intense. One of the hard things about designing these satellites is not just that they need to work, but that they need to work through the launch, which is 15 Gs of shaking and 200 Gs of bolt shock loads. Then there’s the thermal environment of space, where we’re going from minus 40°C to plus 20°C every 45 minutes. That’s much colder than your freezer and much hotter than the hottest day, in a sort of oscillation. This telescope must stay in focus for that whole time. Even movement of a fraction of a wavelength will put this out of focus.

So this thing has to survive those shocks, survive that thermal environment, and stay in focus. The difficulty is not so much the fundamental optical design, but how do you encase it in a system that will survive all of that load?

Mm-hmm.

Right. And so, the first version that you had—I'll put this down—you put a phone up into space?

Yeah. This was a slightly catastrophic event, as you can see. Basically, we put this phone onto a suborbital rocket. We were trying to show—well, basically, let me step back one second.

Okay.

Someone at NASA kept saying when we were there, “Why are we spending a billion dollars on a satellite when a phone like this has most of what a satellite has, but it costs 500 bucks?” It has GPS—

Right. So what was the answer to that?

It doesn't need to cost a billion dollars. Phones work just fine in space. This first test did not work, but amazingly, all of the phone electronics survived, and we managed to get all the data. This was a case in point: even though this—

Yeah.

—was a failed launch, these things are incredibly robust. They're designed exactly for you to drop them on the floor and for them not to break, because people do shit like that.

Mm-hmm.

It was a surprising example of just how robust consumer electronics are. Back in the '60s, '70s, and '80s, NASA was pioneering all this technology because it had to. To go to the moon, you had to put a capsule with humans in it, and you had to have a computer. So they had to develop microprocessors to get a computer small enough, because computers were the size of a building.

They had to pioneer solar panels and various other technologies to get it done. They were used to the idea of building everything themselves because they couldn't go down to Fry's or whatever and get the components. They didn't exist. You couldn't buy the batteries they needed, the computers they needed, or solar panels, so they invented it all. NASA had a mentality of inventing everything. It did incredible things as a result, but now the consumer electronics we have in our back pockets—

Yeah.

—are far superior to what they had. The dominant R&D is being spent by Apple and Google and whoever on developing these electronics to be better and better. So we took the attitude that instead of developing all of that, let's take the best stuff that's already being developed commercially and use it for space. We learned to follow the consumer electronics industry rather than trying to pioneer everything.

The whole approach was to use cheap electronics that we can buy online and, by the way, throw up a few more than we need. If 1 or 2 of the satellites breaks, no big deal.

Mm.

That's a totally different risk attitude from what NASA had, and it enabled us to reduce the cost by about 3 orders of magnitude, which is a big deal for any sector.

A huge deal.

It's a bit like the mainframe-to-desktop computer revolution. All the applications we have with desktops now are completely different from what we did with a few mainframes, one per country or whatever.

So I have endless questions. I guess before I get to all of them, what is the biggest difference between V1, where we're putting a phone up into space, and one of these going up there—what we're about to see right now?

Yeah.

The way to think about it is how much data they're producing and getting down to the ground. When we first launched the first satellites, some of which are here in various parts, they would get data rates down to about 1 megabit per second. We'd take lots of pictures and dump them down with a radio going at about 1 megabit per second. If you're taking 47-megapixel images, you can't get very much data down at that rate.

Now our data rate is about 2 gigabits per second, so about 1,000 times better. That means we're getting roughly 1,000 times more data than the first ones.

Mm.

Per satellite, per unit time. So, huge increases in the amount of data. And that's what enabled us to get to this first mission one, we called it, of imaging the whole Earth every day.

Let's go take a look inside.

Yeah.

I know you and your team work very closely together across manufacturing, testing, production, all that. So how do they communicate with one another? What's the flow here?

First, we build all the satellites here. You'll see some of it. We're actually going to first go into the place where we build the bigger satellites.

Okay.

All of this is done on these automated test stands. This gentleman here is working on them, these guys here. They build all the subsystems. That all comes together in this main clean room area, which you see here. We can actually put our hairnets on and go in so that we can see one of the satellites.

They're a bit bigger than those first ones because, as I was mentioning, these are trying to get to a much higher resolution on the ground.

Mm-hmm.

I don't know if you were asking about how they communicate with each other, but we can talk about that. Basically, each satellite dumps its data to the ground stations, and we've got 48 of those ground stations all around the world.

You have your own ground stations?

Yes. We built them. Basically, no one had that either. You can't go to Fry's and get that either. We built about 48 of these parabolic dish antennas, doing dual S- and X-band, in a few dozen locations.

We often have 4 or 8 of these ground stations at any one location because we have so many satellites, and there are many to look at at any one time, even from the same spot. They're in Alaska, northern Canada, northern Norway, the southern tip of New Zealand, and the southern tip of Australia.

We put them at the highest-latitude locations because our satellites are mainly in polar orbits, so they get many more contact points the closer your station is to the pole. We put them as far north and south as possible with an internet connection because we have to get all that data back, obviously.

Yeah. Okay. So as we go and look through these, I want to address one thing. All of these satellites are named after birds. Why?

Yes. This is actually a space tradition that precedes us. In NASA, there was the Eagle that landed. There are obviously a lot of military missions, like Kestrel Eye. We wanted to name ours after birds that were part of our mission.

Dove is about inspiring how we're trying to help peace and security around the Earth. Tanager is our hyperspectral satellite, and that's because these are looking out for biodiverse areas. Tanagers are some endangered birds with a lot of different colors in the tropics, so we can track ecosystems.

Pelicans are the highest-resolution systems. That's because they fly really low. Pelicans get a bit of lift off the water when they fly really low. They get 10 or 20% more lift. We can do the same thing across the top of the atmosphere. We fly these satellites really low so that we get better resolution on the ground, so we named them after pelicans.

We'll continue to name them after cool birds. Our next one is going to be called Owl.

Oh, yeah. Mm.

We're launching all these owls because they've got bigger eyes, and they're wise because they have a huge amount of compute onboard. So we'll keep on doing that just for fun.

Amazing. Okay. Let's get—

Okay. We'll put some hairnets on.

Suited up.

Yeah. All right, we’re going into the clean room to see some of the bigger satellites. You’re all hair-netted up. You look cool. We look cool. Let’s go.

In here, we build the bigger satellites. That’s a stand for one of them. You can see the solar panels that will be attached to these bigger satellites. This is a bus for one of them, which contains the core spacecraft components. It handles attitude determination and control, propulsion, and radio communications. I think the propulsion is on the bottom. Basically, all of that holds the payload, which is a huge telescope that looks down.

Yeah.

Most of the volume is taken up by that. Somewhere around here, we have one of the actual satellites. This is a bit of fun: you can see the primary optics there.

Oh, yeah.

That’s the big telescope looking down.

Yeah.

That gives you a sense of the scale. Here’s the spacecraft we’re going to look at. This is Pelican-10, one of our highest-resolution satellites. It’s now integrated, and it has the telescope inside. Here you can see the back. This is the propulsion system. This is the feed system, and these are the 2 main tanks. This is the pressurant tank for the propulsion system, so that’s the back end.

The connection to the rocket is here, and these are star cameras, enclosed so we don’t get dust in them. The telescope is inside here. You can’t really see it; it’s all wrapped up in this main section.

Wow.

Any questions?

Yeah, a lot.

Go on, then.

Planet Labs is vertically integrated, so how many of these components are you building in-house, versus which parts are not?

Yeah.

Kind of—

It’s a little nuanced because, at the chip level, we buy almost everything externally.

Yeah.

But at the system level, we build almost everything internally. We get the telescope blanks from a company, but we integrate the telescope, as you saw over there. We build up the framing and the structure. It’s similar for almost everything else. We don’t build our own radio chips; we get the same chips from other people. We use WiMAX components and other things, but we integrate them into special circuits with special antennas that enable us to get the radio-frequency output we need. It depends. A simple way to think about it is: we buy the chips, and we build the systems.

Mm-hmm. How many Pelicans do you produce a year? How many Doves? How many Tanagers? How do you think through all these?

Great question.

How do you think through all these?

We’re ramping up on the Pelicans. We build hundreds of Doves. We’ve built over 650 of them here in this lab. We can build up to 40 a week. We don’t do that every week, but we can build really fast. I think each satellite takes about 8 hours to build.

Okay.

To give you a sense—

Yeah.

When I was at NASA, it was 8 years, not 8 days. It was a complete change. The bigger ones take a bit longer, and we’re ramping up production so we can build many dozens of those. We just announced a new facility in Berlin that will double our manufacturing space again, and it will be able to produce up to 60 of those a year.

Mm.

That’ll give you a sense of scale. The Tanagers are lower-volume because they’re specialized instruments we’re working on with NASA GPL. We launched our second one literally last week, and I think we have another one behind there. We can’t go in because that one is more export-controlled.

Super secret.

Yeah, we can’t go in there. You can see it’s black. Top secret. That payload is a bit more specialized and takes a little more work, so we’re launching a couple more of those per year.

Is there a goal for how many you’re going to build in a constellation, or is some of this replenishing? How do you think about that?

It is replenishing now. In fact, it’s more about what capability you have on each satellite than the number of satellites.

Mm.

We have the largest Earth-imaging fleet. It’s about 10 times larger than anyone else, and, most importantly, it images about 100 times more area than any other commercial actor, at least among Western ones. The Chinese are trying to build similar systems. They even copied our names. They called theirs Pigeon or something, close to Dove. It looked eerily similar, and they even named it eerily similarly.

Roughly speaking, we’re the only ones that cover the whole world every day. Other players might task satellites to a few specific locations, but they cover less than 1% of the Earth’s land mass per day, or maybe a couple of percent at most, whereas we cover all of it. So we collect about 100 times more area. That’s what we really care about.

The Owls are the upgrade: they’ll collect the same area, but 10 times more data per unit area, because we’re going from 3-meter resolution to 1-meter resolution in a daily scan of the planet. That’s roughly 3 by 3 times more data, so almost 10 times more. We’re also getting it back 10 times faster because we’re adding optical links. You can get it back within a few minutes. So we’re getting 10 times faster delivery and 10 times more data. But the number of satellites will stay about the same, if not go down a little, because of how we’re building them.

Mm.

It’s not how many satellites you have; it’s what you do with them, or something.

Okay. Is there anything else in here that we should see? Anything technical we should know?

We can see a few other things here.

Okay.

This is a propulsion system laid out on a bench. That’s kind of fun. This is our telescope assembly. It’s about 50 centimeters in size, and that determines how small an object we can see on the ground, together with the altitude of the satellite. That’s it.

What kind of fuel are you guys using?

In this, it’s xenon. It’s an ion engine, so it’s very efficient in space. The gas comes out the back extremely fast—much faster, about 10 times faster than from a classical rocket engine.

Wow.

It doesn’t have as much thrust, so you can’t take off from the planet using these engines. There’s nowhere near enough thrust. But once you’re in space, they’re much more efficient—almost 10 times more efficient than those big rocket engines.

I think it was SpaceX’s Starlink that just put out that they want to try to open-source the data for anti-collision avoidance.

Yes. Satellite—

So—

Space domain awareness data.

Yes.

SDA, yeah. I think that’s really cool. Early on, we had a lot more objects in space.

Yeah.

We want to stop collisions. We work very collaboratively with the other players. The biggest challenge, though, is less the satellites and more the space debris. We’ve got on the order of 10,000 satellites operating in space, but on the order of 100 million space-debris objects. These are objects from former satellites that blew apart, satellites that crashed, and satellites that were deliberately taken out by anti-satellite weapons from the ground. Those things can cause one satellite to pop into tens of thousands of bits. Okay, let’s carry on.

What a wild, wild problem.

It is. It’s wild, wild space.

Wild, wild space.

Great documentary by Ashley Vance, which you were in—

I know.

—by the way.

Yeah, it is.

Small plug.

It’s cool, right?

It’s so good.

Have you seen it?

Yes, I have. When we went to Rocket Lab, I watched that. Their little fight between him and—who was it?

Chris Kemp from Airbus, yeah.

Chris Kemp. It was just hilarious.

Yeah. They’re quite some characters, those guys.

Okay. We don’t need this for anything else.

We do not need this for anything else. By the way, this is a plot of methane-tracking activities of the Tanager spacecraft.

Oh, wow.

This is a plume of gas. By the way, when we enable them to be turned off, we just tell the gas company, “Hey, you’ve got a big freaking great leak on your pipeline.” Turning this one off is the equivalent of taking a couple hundred thousand cars off the road from a pollution standpoint. So, one leak.

When we go to the AI demo later, is it possible to see the difference between COVID and non-COVID, when everything was off the rails?

Oh, yeah. We can have a look. We can ask the AIs.

Let’s ask the AIs. Is that named after a bird, too?

No. It should be. That’s a good idea. I’m going to steal that.

Oh, yeah.

Here we have some of our testing facilities. This is a system that tests the shock when 2 stages of the rocket separate. We put our satellite here, bring this up, and let this hammer smack down. It gives a shock load of around 200 Gs.

Why isn’t this in a sound-barrier room?

I guess people put earplugs in. They’re the earplugs. You haven’t got any, but they’re up there.

Great.

And then this is a vibration table. It vibrates to simulate the launch environment. We put our satellite on there, vibrate the shit out of it, and hopefully nothing falls apart.

What have been the biggest failures?

We’ve had telescopes break.

Okay.

We have to redesign it.

What’s the most delicate part of your satellite?

It’s probably the telescope.

Yeah.

Exactly.

Okay.

Think about this: the optics mustn’t move by more than a fraction of a wavelength. The wavelength is less than a micron, so a thousandth of a millimeter. The telescope’s 2 main optical components mustn’t move by more than a thousandth of a millimeter, otherwise it’s out of focus. That’s really annoying, because it’s a very small distance.

Mm-hmm.

Even when it’s in space, if you made the telescope out of aluminum like this, it would expand and contract so much that it wouldn’t work when it’s too hot or too cold, because it would be totally out of focus. We have to design it out of special materials that compensate for that expansion and contraction because of the thermal loads. That’s the kind of thing that’s really difficult. Now I can show you where we test those thermal loads.

How many satellites end up having errors? What’s been the—

Error rate?

Yeah.

First of all, out of all 700 satellites, if the launch worked and delivered us into orbit, our satellites have turned on and worked.

Okay.

We’ve had some that were out of focus. We’ve had a few problems where a solar panel didn’t deploy, a reaction wheel stopped working early, things like this. But generally, we’ve gotten utility out of 100% of our satellites, which I actually think is a problem. I keep telling folks that.

Why?

Because it means we’re not pushing the envelope enough.

Ah.

If we have 100% success—

Yeah.

—it means we haven’t pushed hard enough.

Mm.

Maybe we should have gotten the latest sensor or something that would improve it.

Yeah.

We want to push the barriers. Thanks for your help.

All right. What’s this? We’ve got a little vacuum chamber here. What are you putting in there?

That’s the crockpot.

It’s the what?

The crockpot. It’s good.

Oh, what is that? Is that for Pelican?

Yeah, for the reaction wheels.

Oh, it’s for the reaction wheels. Cool. So you’re testing that subsystem in it. Cool. Yeah, what’s going on?

We have a bunch of 3D printers.

Yeah, 3D printers. Fine.

Plastic.

Normal stuff. Those are the 3 main tests: the shock, the vibration, and the thermal vacuum. Let’s go here. By the way, I like this sign.

Oh, you actually have this.

Yes.

You know how many people will talk about this, but they don’t actually have it?

Well, these guys have it. Anyway, let’s carry on.

So let’s go upstairs. We’re going to go up to the third floor. I guess I could take this at this point.

Cool.

Come on. Here are some of our Doves. There’s a SkySat. This is our new Pelican spacecraft that you were just seeing a model of downstairs.

It’s smaller.

Well, that’s scaled down.

Oh, okay.

This is our AI interface. Basically, this is our imagery of the whole world.

Uh-huh.

You can zoom in wherever you want. You can go through time and see how it changes. If we go further north and down to December, it’s going to be snowier, so you can see all the snow cover. The coolest thing is that we have 3,000 layers of images going back 10 years—pretty much every day for 10 years.

Zoom in as far as you can. Let’s see what we’ve got here.

Where do you want to have a look? But the coolest thing, before we do that, is that not only can we zoom in, we can ask it questions.

Mm-hmm.

We’ve tied an LLM to the backend. LLMs were really only trained on the text of the internet, so they didn’t know much about the real world.

Yeah.

We have all the real-world data, and we bring them together. Then you have a system that can understand all of human knowledge and tap into real-world stuff. What does that mean? If you’re a farmer and you want to understand your crop yields, a normal LLM chat is going to tell you about the theory of agriculture, but it doesn’t know shit about your field, right?

Yeah.

But the farmer wants to know about their field.

Mm-hmm.

So this can tell you about that and help improve crop yields. Or, if you’re a journalist trying to understand a report on some physical event, like a flood or a fire, it can tell you about that flood or fire. It can search the entire Earth in under a second, see changes over time, and plot graphs. This is an example we were looking at earlier. Somebody was asking, but you can actually ask it yourself. This is a visual it gave of some deforestation.

Mm-hmm.

I asked it your question earlier: “What happened in the last 24 hours? Show me some changes.” I haven’t got that here; I’ve got it on my computer. It found a fire in Yosemite. It found a flood that caused a load of crap to fall into a river. It found all sorts of things around the world. It was like a daily planet report, if you like. Anyway, we can ask it any question. What do you want to ask about?

Let’s go to Area 51.

Let’s go to Area 51. Is there anything in particular you want to know about it?

Let’s see the changes over the last 5 years.

Show us the changes there over the last 5 years. It’s going to zoom in on Area 51, and now it’s going to start comparing 2021 to 2026, as you asked. This is a comparison. Obviously, it’s waiting for the data to load here. It’s showing some changes over time if we zoom in. We can actually zoom in on that image and try to see changes. We can’t zoom in like that, but I see. Basically, there’s a structural change in that—

Uh-oh.

—I don’t know what that is. Look, there’s a black—

Oh my gosh.

I actually don’t know what that is. You’d see plane movements and stuff. You can see there were 2 planes on there, and they’re gone. This is at a particular zoom level, so you can get a sense of what it can do.

Mm-hmm.

It’s still early phase, but our point is to open up geospatial data to more than just the satellite imagery experts.

Yeah.

It used to be that NASA could get value out of this, and the NRO could get value out of this. But a farmer, a journalist, or what have you couldn’t get value. Now anyone can go and play with this, because all they need to do is be able to speak. And, by the way, you can speak any language into it. It’s just an LLM; it takes anything. That’s the power of AI and satellite data put together.

Amazing. So you built this internally. Did you have to bring in a separate team? How did you—

No, we just have a—

—create all this?

We have a couple of engineers, an AI team specialized in doing this.

Amazing.

Yeah.

Cool. All right. Anything else we should see while we’re up here?

Great, I’m done now.

All right.

Yep.

Well, where do we go from here?

And here are some of the AI engineers.

Oh, we have an AI engineer with us right now.

In the house.

Was that demo sufficient?

I don’t know why it wasn’t doing the search tool. That’s what you’re trying to get it to do, right? Yeah. It doesn’t do a perfect job, but it’s getting better and better.

Can anybody access that? How do you get access?

It’s on a beta launch right now, so anyone can access it, but it’s just a testing version.

Mm.

But soon we hope to launch it.

Exciting. Cool. All right. Well, we’re going to go to the sit-down interview now.

Let’s do it. There’s another floor, so—

This is the first time anybody has seen this floor.

It’s true. You’re the first one to come in. All right. I can show you just a couple of things before we sit down. This is the so-called Dove Nest.

Aw.

So this is where we’ve built all the Doves—

Mm-hmm.

—and they’re sitting in their nest waiting for the launch. You see they’re 2 by 2 in there.

Quite the difference from the garage you started in.

Yeah. It’s a little bit more substantive. That’s true.

Yeah.

Oh, and there’s one more fun thing. Thank you. You can put it back. One more fun thing for you to see in here.

Okay.

So, here. You have to go in there—

Oh.

—both of you.

Do you want to go first?

Yeah—

Is this where you scream?

Yeah, exactly. I’m going to shut the door—

Okay.

—and then you can just talk a little bit.

Okay. So we are here at Planet Labs in San Francisco, and now we are trapped inside.

Catch that?

Yeah.

It’s cool, right?

It’s really cool.

So this is where we test the radios to go to space.

Is it a Faraday cage, or is it soundproof? What is it?

No, it’s an RF chamber, so all the radio frequencies get absorbed by this. We can put our satellite here, point it there, and then reduce the power so that we can check that the right power is getting to the antenna without reflected waves spoiling our ability to tell whether it’s actually emitting at the right power. So basically, it’s a way of testing that the radios are going to work in space.

Yeah.

One final cool thing for you to see. Sorry, am I getting in your way?

No. Yes.

We were just setting it up for you. It’s your chamber. All right, let’s go.