让AI进入轨道的100亿美元卫星帝国、芯片为何胜过火箭，以及中国第一开源模型 | #266

Today, Planet's a $10 billion company. You've coined the term “large Earth models.” What's that mean?

It's a bit like Google indexing the internet to make it searchable. We're indexing the Earth to make it searchable. It will finally enable us to be smart stewards of our planet.

The elephant in the room here, Will, I have to ask it: How do you compete with Elon's plans for orbital AI data centers?

Everyone apart from SpaceX has to pay the SpaceX launch tax right now. Everyone apart from NVIDIA and Google has to pay the NVIDIA tax. Which tax is more important? In the near term, it's the launch, but longer term, it's the compute.

That is brilliant.

Our next story should keep the U.S. labs up at night. It's a Chinese model called GLM-5.2 that, in some cases, matches or exceeds the top models from OpenAI and Anthropic.

This level of performance in an open-weight model is absolutely shocking. You can burn tokens to get more intelligence, and the Chinese have figured out how to do it.

The Chinese are evidently figuring out how to reason more efficiently, or at least more cheaply.

I'm here with my magnificent moonshot mates: Salim, the father of organizational singularities. Good to see you, pal.

Good to be here. We're going to be home.

Yeah, everybody's home. This is a great—this is home day. When was the last time this happened? This is like—never.

Never. Never.

A.G., our in-house super genius. Good to see you, Alex.

I will say, in my defense, coining the term “planetar”—a quadrillionaire who can own an entire planet—was not an advertisement for Will and Planet.

Uh-huh. Yes.

And Dave Blundin, our wizard of AI investing. I'm Peter Diamandis, your host and your optimism evangelist.

We have a special guest here with us today, a friend of nearly 20 years, a man who's building humanity's orbital AI and data layer. Will Marshall, the CEO of Planet.

Hey, thanks for having me, guys.

I have a question. Do people ever say you're the CEO of the planet?

I should be.

You should be.

FYI, Planet is a public company. The ticker is PL. You can track it in real time as we speak today.

As always, we've got a packed “WTF Just Happened in Tech?” episode. The singularity waits for no man and no agent. We're going to kick off a discussion on Planet's push for large Earth models and its orbital AI cloud. We'll jump into Eric Schmidt's newest launch company, Relativity Space, and its upcoming Mars mission. From there, we'll jump into the AI talent reshuffling and President Javier Milei's provocative statements on AI personhood. Of course, Alex, I expected you're the one influencing him, but we'll ask you behind the scenes.

Alex Goryachev

You think I'm pulling Javier Milei's strings for AI personhood?

I think you're influencing him.

We're going to close out with the shocking performance of China's open-weight GLM-5.2 model. ByteDance's new 4K video model will wrap up with the collapsing price and exploding capital expenditures of intelligence, fueled by nuclear and fusion power plants.

So, Will, I want to kick it off with you, buddy. I hope that, at the end of this conversation, everyone listening is going to understand what this layer of AI and data capabilities that you're building is going to mean for them.

Let me do a proper introduction for you. Will Marshall is the co-founder and CEO of Planet. It's the world's largest Earth-observing satellite fleet and will soon have orbiting AI data satellites. He's a physicist who earned his PhD at Oxford—not a bad place. Not MIT, quite, but not a bad place.

Before Planet, Will worked at NASA. He and I met back in 2008, along with Salim. Today, Planet's a $10 billion company. Following up on your point, Dave, I looked at the ticker, Will, and saw a 450% increase in the price over the last year. That's extraordinary. Pretty amazing.

Will is operating 200 satellites in Earth orbit today, generating 25 terabytes of imagery every day. We have 2 major stories. The first is about planetary intelligence. The second is about Project Suncatcher.

So, Will, let's kick it off, buddy. Let's talk about what you're building with planetary intelligence. You've coined the term “large Earth models.” What's that mean?

Planetary intelligence, to me, is sort of the next era in machine intelligence, and it's all about building models of the real world. Of course, AI models are only as good as the data set they're trained on, and we have gobs of real-world data.

I think of it in 2 phases. The first phase is combining planetary sensing—which is already in space, for all the obvious reasons we've been doing planetary sensing in space—with large language models to create what I call large Earth models. That's wrapping up not just the text of the internet that's embedded in LLMs, but all of Earth's data, so that you can ask physical questions about the real planet.

The second phase, in big-arc terms, is putting the compute up next to the sensors in space, and we're doing that, too. That's a bit further out, but that leads to phase 2 of the capabilities in this world. We're mainly focusing on phase 1, which is pulling all of our Earth-imagery data into models such that you can ask questions about the physical world.

Whether you're a farmer, a journalist, or somebody interested in national security, you want to know about the real world.

Every European government. You forgot to mention billion-dollar-plus contracts with every government in Europe.

Everyone who can't afford to launch their own satellites.

Well, exactly.

Absolutely. Governments want to see around the corner. They want to see new threats. They want to respond to disasters. Everyone wants to answer questions not just about the text of the internet.

That's where all the large language model companies are going. All the AI companies—Demis has talked about this, and Dario has talked about this—the next-scale models are going to be real-world models. For real-world models, you obviously need real-world data.

Planet has 3,000 images for every point on the landmass of Earth over the last 10 years, documenting every day's changes. You basically have a huge stack—it's 150 petabytes of data—of information about the real world and how it's changing over time.

About 8 years ago, I did a TED Talk talking about how we're a bit like Google: indexing the internet to make it searchable. We're indexing the Earth to make it searchable. It's just that large language models are making it way faster to do that.

Now it's unleashing all of this potential latent in Earth-imaging data—not just ours, but the whole field.

Well, if I wanted to pay you to take 1 point on Earth—Peter's house, say—and stitch together those 3,000 images you collected over 10 years and turn it into a little movie, could I buy that from you?

Yeah, sure. It's getting easier and easier. Historically, most of our users have been big entities. NASA has used it, the National Reconnaissance Office—the intelligence community—huge agricultural companies like Bayer and Syngenta, hedge funds in New York, and so on.

Alex Goryachev

Will, I myself have purchased data from you in the past. I really wonder: Can I purchase it via API?

Yes, you can. But typically, you're unusual, Alex. Most people can't get much value out of it because processing terabytes of satellite imagery has been too hard.

Now AI comes along and just shortens that gap. You could ask Gemini or Claude, “Hey, find me images from Planet. Tell me how my farm field has changed over time. How can I improve it next year? How can I improve it today? What do I need to do?” It will go off and do that analysis and come back to you with just the answers.

Alex Goryachev

You're one of the few, if not the only—correct me if I'm wrong, Will—you're the only, or one of the only, vendors that actually offers high-quality historical imagery. If I specify lat/long coordinates and I want a bit of historical imagery of different types, do you offer that via API?

Yes, we're the only company in the world that images the whole world every day at high resolution. Think of it as the Google Maps satellite layer, but that layer is maybe 3 years old—sometimes 1 year, sometimes 10 years, but a couple of years old, let's say. We're doing that every day for the whole Earth and have a time axis.

It's like the Google Maps satellite layer, but with a time axis. Yes, we're the only ones doing that. Until someone invents a time machine, even if somebody erected a whole load of satellites, they can't go back in time and get our historical archives.

All of our clients use not just today's image. They almost always want to know, “How does this compare with normal?” Let me give you an example. In Ukraine, we're very much helping them with the defense of their country. They don't just want to know where the Russian positions are, where their military bases are, or where the industrial facilities are. They want to know how it compares with the last couple of years, so they know if it's normal or abnormal, and therefore what the threat level is.

The same is true with the U.S. intelligence community.

They want to know what's going on across China. They don't just want to know, “Is there something new in China?” They want to know how that compares with the normal activity levels in that place.

The same is true with the farmer. They don't just want to know their farm yield output now; they want to compare it to the past and know if their agricultural interventions would be better if they did it like the neighbor does it or someone else does it. They need the historical background to know that. So, yeah, the archive is really important. It goes back 10 years.

This is going to be really cool for, you know, Salim and Peter and me. We're looking at island real estate and mountaintop real estate. We're big on this eVTOL thing coming online very soon, so places that are normally—

Yeah, but this API—

It's got a totally—

Well, why don't we do it together? Let's get our fund together for this and use your historical data to analyze the perfect location.

The search query is: Which piece of land will generate the most profit for us when eVTOLs arrive?

I'll bet you I could do that during this podcast. I think it relates to how many wiggly roads there are: how far it is by travel time now and how short it's going to be after eVTOLs. I've bought property myself within a 100-mile radius of San Francisco, knowing that the value, I think, is going to go up.

I've got 2 questions. First, when you talk about imaging, what's the resolution at which you're imaging? And do you also do infrared and other bands?

Yeah. As a basic explanation, we have 3 fleets. The scanning fleet is 3-meter resolution. That's the one that does the entire Earth, and it does it with 8 spectral bands. We're improving that with our next generation. We're launching a first tech demo this year of Owl, which enables it to go from 3-meter resolution to 1 meter.

The latency of the present imagery is several hours, and we're reducing that 10× as well, to well under an hour. So, that's that system. A second system does high resolution, so we can go up to 40 centimeters today—40 to 50. It's going to 30 centimeters tomorrow.

We launched 9 of these satellites, and we're launching a whole bunch. We're working toward a 30-by-30-by-30 goal: 30 centimeters, 30 times a day, and 30 minutes from request to getting the image back in your hand anywhere on Earth. So, that time axis is being shrunk, and that's for 30 centimeters. Each pixel is 30 centimeters across, so about a foot.

Then we have a hyperspectral imager, which is the first and most sensitive one in orbit, according to JPL, who we built it with. It has 400 spectral bands. The human eye has 3—RGB: red, green, and blue. This has 400, and that crosses from infrared to ultraviolet.

Those extra spectral bands enable you to essentially take a signature, a fingerprint, of the planet. Where we look in each pixel—which in that one is even bigger, 30 meters—we can actually tell the species of the tree, gas emissions, or, if it's a tank—

Gas emissions?

Which tank site built that vehicle. I mean, incredible. It's like the signature, the fingerprint, on the Earth's surface.

Why hasn't anybody done this, Will? I mean, I don't think anybody's close to the fleet that you've built, and you bought part of your fleet from Google early on. I'm just curious.

Peter, are you sure you don't mean, “Why hasn't anyone in the private sector done this?”

That's what I mean. Yes, the private sector. Obviously, defense has—how many defense imaging satellites are there in orbit right now, do you think?

Well, it's technically classified for the US government and probably—

You can tell us. No one's listening.

Roughly half a dozen really high-resolution ones. They have much higher resolution than we have—more than 10 times or so higher resolution—but—

Golf-ball pixels?

But that sort of resolution has a trade-off in coverage. They have even less coverage than we do. In fact, not just a little bit less. We cover about 200 million square kilometers every day. The Earth's landmass is about 150 million square kilometers, so we cover a bit more than the Earth's landmass every day.

They probably cover less than 1% at anything like that resolution—probably much less than that.

I mean, for the entrepreneurs listening, it's important to realize Will started this with Robbie and his partners by launching a phone into orbit—PhoneSat. It was a crazy idea. You got in trouble for it. It worked. You got Steve Jurvetson's attention; he became a major investor, and it kicked off a $10 billion company. It's extraordinary.

Thank you. Yeah, well, it's been quite a ride. As you were saying, the stock right now is a rocket ship, and part of the reason is this AI piece, because the AI piece is lowering the barriers to entry. I think we're going to see this massive takeoff, and it's just at the very beginning.

I said before, space and AI are getting married. A lot of people understand how AI is affecting every discipline, and it is—it's affecting every sector. But space is one of those unique sectors that's producing gobs of data, and therefore space is actually important for AI as much as AI is important for space.

AI is not just eating space as a sector, like it is eating almost every sector. Actually, space has something to offer it, because all the AI companies, as I said, are trying to build these physical-world models. For that, they need real-world data.

Space comes along. AI is useful to space because it makes extracting value out of all this data much easier for those smaller organizations, but at the same time it gives AI something it really didn't have, which is this information about the real world. All they're trying to do is help people ask and answer questions about the real world.

Let me give you an example. The farmer going onto an LLM right now and saying, “Hey, how do I improve my crop yield?” The LLM will say, “Well, here's all the theory of agronomy.” But it doesn't know about his or her field and how it's doing today, how it compares with yesterday, how that compares with last year, how that compares with the farm next door, and therefore what they can do about it. How's the soil doing? How's the water content? How's the agriculture doing? We can tell all that and help the LLM answer that question.

Or take the journalist. They're investigating some flood. They don't want to know the theory of a flood. They want to know how that flood is doing today in that village and where the emergency response people need to go. So, basically, real-world data is going to come into these AI models, and that's going to enable them to be 10 times more powerful than they are today.

So many questions. I'll start with the simple ones. To the extent that you draw a parallel between your large Earth models and large language models, I think one of the most important questions I could possibly be asking is this: Yes, you offer historical or archival imagery via API, but I think most people would love a crystal ball that autoregressively extrapolates Earth into the future.

Sort of a Sora video model at meter or submeter spatial resolution, projecting the video into the future. Where are the future extrapolations of Earth? Where's the crystal ball powered by Planet?

I think it's coming. It's very exciting. One step at a time. We're focused first on retroactive analysis and how AI can unlock that. But already, predictive analysis is coming to the fore, and obviously AI has been very good at tokens and guessing the next token.

That's what AI is doing when it's guessing the next token, giving you text output: each time, it's guessing the next word in the sentence, and it's doing a very good job of coming up with coherent answers. So, obviously, with 3,000 images, you could easily ask it to guess the next few images—what's going to happen next.

We've already just studied this. Some people—you would be terribly surprised—are interested in us tracking data centers across China. So, we loaded into it all the data centers across the entire US, which have to be registered. Then we showed it, got it to look back through the imagery of their development, track that, and then extrapolate that model to China, and said, “Go find all these things in China and track their development.”

But it also got really good at predicting, based on the US data, when they would be complete. Within a few days, it could guess way out when they're going to complete, because it has taken into account all this construction information in the nearby region and various other things.

Now it turns out this model is pretty good at predicting when data centers will be complete. Lots of people are interested in that right now. So, that's the kind of thing—it's the first time I've seen it really work like you're suggesting, Peter. It's really the very beginning, but I think we're going to get there relatively quickly, just because of the nature of the tech: it sort of already can do that out of the box.

But surely you have enough data. Stop calling me Shirley. Surely, you have enough data to be able to take everything you already have and pre-train an autoregressive video model to extrapolate the Earth at the pixel level into the future. Do you feel like you have enough data to do that already?

I think we do. Yeah.

Have you done it?

No, but I think it’s—

Why have you not? I think you are perhaps the only entity in the world with the power to build an honest-to-goodness crystal ball.

Yeah. [laughter] I love the vision. Again, we’ve been doing this in some bespoke areas, but the main thing has been looking backwards because we believe there’s already a $100 billion market just in the retro, in the rearview mirror.

But you’re right, it’s very tempting. I think the biggest thing that we’re working on that’s super relevant to that is embedding models, because doing that for the entire Earth—each layer is around 30 terabytes of data. It’s 4 million 47-megapixel images. It’s just a huge amount of data, times 3,000 layers, right?

So you can’t just throw that into a machine. No machine can just take that into RAM and do the processing, right? What do you have to do? You have to put it into an embedding space.

Alex Fielding

We’ve been working with Google Research and DeepMind models on this, as well as their important work called AlphaEarth, which you can look up, and some open-source models like CLIP—a remote-sensing CLIP model—and then fine-tuning it on our data.

What this does is it’s sort of an image- or tile-to-text conversion. You can do that for each area, say, a kilometer by a kilometer, and then you could search for arbitrary objects around the Earth.

We’ve got it to a point where we can put the entire one layer of the Earth into BigQuery in this embedding space. Now you have the potential to do what you were talking about: putting in thousands of layers and then predicting the future. So I don’t—

I got it. So the vision is you’re tokenizing the Earth.

Alex Fielding

You’re tokenizing the Earth first because it’s like a massive compression, and then you want to do the prediction.

What we could do—the middle school my kids went to commissioned this. They took collections from all the parents to buy this huge globe. It’s 6 feet in diameter, and it’s all LED. Basically, you can make it the Earth, Mars, Venus, or any other planet you want with this little console.

Alex Fielding

It’s so cool.

You could say, “Well, that’s semi-cool,” but you overlay the planet data and you can actually dial time backward and forward in the real world. If you built one of those for your lobby, you could sell those like crazy.

But, Alex, where were you going with this? Where do you see the value there? What would you say, “That’s the thing I want to then predict into the future,” apart from everything?

Alex Fielding

I want to be able to predict everything into the future, not just at the data-center level.

That’s the answer you weren’t meant to give. [laughter]

Alex Fielding

Okay, excluding that, I want to be able to—I’ll give you 1 concrete example. Other than that, I’d like a reasoning model that I could layer on top of it.

I would argue every LLM wants to be a large reasoning model, not just an autoregressive LLM. I’d like to be able to reason RL-style about what changes at the pixel level on the Earth’s surface could, say, maximize GDP.

We talk on the podcast all the time about maybe GDP triples year over year due to this singularity that some of us think we’re in. But you have the dataset at the planetary scale to actually build a reasoning model via reinforcement learning that lets us historically back-test various theories of, say, land use.

If we literally tile the Earth with compute, as some of us think we’re doing, what would be the hypothetical effect on GDP if we get rid of a few gas stations—

Or let’s—

Coffee bean futures.

Alex Fielding

Yeah, futures. Well, certainly, on futures markets, you can imagine that, but let’s get above GDP for a second and go even beyond that.

It will finally enable us to be smart stewards of our planet. We are effectively stewards of the planet, but we’re not always doing it in the smartest way—not in terms of efficiency, nor in terms of how we’re taking care of the precious ecosystems and complex environment that we have on Earth.

Now we finally have a system that understands it all, from the local level to the global level, and can integrate all of that into recommending what course of action you, as that farmer, you, as that insurance guy, or you, as that finance guy betting on markets or whatever, can take to make a smarter decision. But go ahead.

There’s a huge issue that comes with this, though. If you go back, the internet was born to operate at planetary scales, but then governments domesticated it, right? What you’re doing with orbital mapping is you’re re-globalizing that.

How do you handle the aspects of this? Governments own the map, but you own the sky above the map, right? This raises hugely unsettling questions. You’re shifting from national infrastructure to planetary infrastructure. Is there a global kill switch? How do you handle sovereignty? You mentioned Ukraine already, or China. There’s enormous geopolitical tension in this that must drive you crazy trying to navigate.

Alex Fielding

I would say it’s not what drives us crazy. It’s a founding part of our mission. We call it giving greater transparency and empowering everyone. That leads to greater security and greater sustainability.

No one can hide. No one can hide anymore.

Alex Fielding

Exactly. Putin thought he could get away with people turning up on the border and then no one would notice. We put that to bed, you know? It didn’t deter him from attacking, clearly, but the potential in the future is that everyone would know they would be seen at every step.

Now everyone knows that they’re seen at every step. If you hit a school, we’re going to see the school. If you hit a bridge, we’re going to see the bridge. The accountability is going to be there for the whole world to see, no matter what.

And I think the world acts as a deterrent. Throughout history, wars have happened mainly when there’s been misinformation or a lack of information, and people have had to guess or have made mistakes based on misinformation.

Here we have more people understanding what’s going on, who’s got what equipment, where that equipment is, and can monitor peace accords and all that. I think transparency drives accountability and reduces the probability of war.

Meanwhile, on that very narrow question, on that exact topic, you assume that the US and China see everything via satellite at all times. But if you look at the 220 countries across the world, what fraction of all governments actually have satellite coverage data?

Like you said, misinformation leads to confusion, which leads to war. In Yemen right now, do they look at data or not?

Alex Fielding

Not much. But I think that’s going to change. Again, the challenge has been that digesting 40 terabytes of data every day is too much for most organizations.

NASA has teams of people doing satellite-imagery processing. They know how to deal with this. Bring AI along, and suddenly you eliminate all of that. You can actually get most of the answers very quickly, so that an NGO, the Red Cross operating in Yemen or whatever, can actually benefit from this right now and be enabled to make smarter decisions.

It changes it from a world where it’s just the big entities to one where a lot of other entities can get value.

What percentage of your revenue is government versus corporate versus individuals?

Alex Fielding

It’s about 60% defense and intelligence, about 25% civil government, and about 15% commercial. So, basically, mainly government. It’s been growing there, but commercial is now really starting to take off. Again, it’s because of all those reasons: AI is lowering the barriers to entry.

How are you going to price the AI training-data use case? I mean, that’s the big up-and-comer, obviously.

Alex Fielding

Well, frankly—

They’re going to own it, Alex. They’re going to keep it and sell the knowledge and information, not—

Alex Fielding

Build your own. So we’ll just continue to sell the data.

So if OpenAI calls and says, “We want to use it,” you’re going to say no, or are you going to say it’s $1 million?

Alex Fielding

OpenAI can call our MCP server, and off you go.

Okay, so it’s just a fixed price.

Alex Fielding

Absolutely. Train all you want, and then—

And every time you need planet data, you’re going to need to do an API call or an—

Alex Fielding

Update.

I’m curious—maybe to steelman Salim’s earlier question about information asymmetries. Correct me if I’m wrong, but am I correct in assuming that your datasets go through some sort of US government NRO filter regarding what can be made publicly available?

Alex Fielding

Not exactly. It’s a bit more nuanced than that. As a US remote-sensing company, we register under NOAA’s remote-sensing act, which means we have to register the satellites, but we can sell the data to anyone except for a blacklist—a blacklist that includes Iran, North Korea, what have you, and various terrorist organizations.

Other than that, they’re not checking every player we provide data to. We do check that, we think, and there are many people we don’t work with if we think that they would do some harm with this, but essentially they’re relatively hands-off with that.

So that's a great line of questions. That's a U.S. blacklist, right? But you have a $1 billion-plus deal with Sweden. Do they also have a blacklist that you honor separately, or do you just deal with it?

Alex Fielding

They tend to be almost the same. We respect the EU one as well, which Sweden is part of.

I'm Canadian. Are we on the blacklist? Because last time I checked, there was a problem.

Alex Fielding

Interestingly enough, they differ only a tiny bit. We have a load of ground-station infrastructure up in Canada, and we downlink there and then sell the data to certain countries that we couldn't sell it to if we downlinked it in the U.S. So we add these things up and say, "Well, let's make the master list with all the bad guys according to all the people," and then we take that off.

You must have a dedicated AI just dealing with the complexities of who gets what, where.

Alex Fielding

It's not as hard as you think.

That's just blocklisting users or customers. What about downsampling or lowering the resolution in sensitive areas? Do you do any of that?

Alex Fielding

No, we don't. But remember, we're really a long way away—400 to 500 kilometers. It's like the distance from Los Angeles to San Francisco, pointing our telescopes from one city to the other at that distance, right? Obviously, the details you can see with that aren't the same as those you can see if you're flying a drone. If you're flying a drone, you can see people and recognize their faces. You get all the personal privacy issues. We're 400 to 500 kilometers away. We're not getting into that.

It's kind of amazing what we can see, but it's not like that. The reason that matters is that a lot of the most sensitive stuff doesn't come into the fray because of that. Countries agreed very early in the space era that they won't let each other fly planes over each other's territory without permission, or drones, but you can fly in space because they consider that so far away. You can get some data, but it's just enough for transparency, not enough to get into the details they would care about. That was basically the agreed-upon definition.

The backstory there is fascinating, right? Back when Sputnik was launched in 1957, the U.S. had to make a decision: Do we disallow that from happening because it's flying over us? But if we did that, we wouldn't be able to fly our U.S. satellites over Russian territory. We said, "Okay, everybody can fly satellites." You just ran experiments.

The physics of it dictate that, right? Whereas a plane, you can go up to Russia and turn left. You can't go up to Russia with a satellite and turn left. You're in an orbit. You're going to go over Russia. So what are you going to do? Say that you're not going to turn on the camera? That's silly. That's obviously not going to happen, so people aren't going to respect that.

There was some physics that went into that and the fact that it's far away, but that became the international norm. Famously, Gary Powers was shot down in the U-2 spy plane, and then the U.S. said, "Well, we're going to put most of the sensitive stuff in orbit. That's going to be our domain for enabling us to monitor what's going on with nuclear weapons, arms control, and all this sort of stuff." But now it's just proliferated, and far more people can get through Planet what took the entire CIA and NRO infrastructure a couple of decades ago. They can get it for a tiny fraction of the cost.

In some senses, we can do things that no one has been able to do, like the daily scan. They've never had that many satellites, so they've never had that sort of global coverage. We can do things that they can't even do. The fact that that's now possible for a private enterprise has completely changed the game.

Historically, you've always brought the data back to Earth and done the crunching at data centers on Earth. You ran some experiments in April where you put some NVIDIA chips on one of your satellites and did the processing up there. What's the significance of that?

It's basically enabling us to do the processing at the edge. It speeds up the time. This is really processing at the edge—in space, 500 kilometers up. We put some NVIDIA GPUs on our satellites, and all of our Pelicans going up now have them. The Owls will as well.

That, combined with satellite-to-satellite communication, means we're putting links in so that we can go up to other satellites and then back down. We don't have to wait until the satellite goes around to the ground station we've erected. We have ground stations all around the world, but it still takes some time. Instead, it can just send back the answer.

In the example we did in April, we took a picture of an airfield in Australia, in this case in Alice Springs. The computer automatically recognized the planes on that airfield, and then it sent us back the locations and types of planes. That was done in seconds, and then we can send it back via RF, satellite to satellite. Suddenly, you have things in seconds.

And here's a photo of that, by the way.

That image, yeah.

Oh, cool. What a flashback. That looks exactly like what I used to work on at MIT—literally exactly.

Time really matters for a number of applications. Just think of the fires in Los Angeles—the Palisades and other fires. We gave images within a couple of hours of those fires, and then we did analysis building by building: which buildings were affected and where the relief operators should go—the American Red Cross, Cal Fire—

Where the water was located.

And had we been able to get that in a few minutes rather than a few hours, could that have saved lives? Could that have saved properties, potentially? Time really matters. Processing at the edge is all about time. It's going from hours to minutes.

Can you give us some geeky numbers on that? In the Pelicans, you have a couple thousand frequencies of light coming in, so the data must be astronomically huge—just the raw feed. Then the NVIDIA chips will have no trouble compressing that down, but you have limited bandwidth coming down to Earth. What are the rough numbers?

I know the numbers based on our Dove satellites. They take 8 frames a second at 47 megapixels, so for each area on Earth, we can get 8 different spectral bands. Within about a second, the satellite has already gone past that area, so basically you would just have to start again. It goes 8 times a second to get 8 spectral bands for each area of the Earth.

Each picture is maybe 35 by 20 kilometers in area. Just imagine that going all the time, clipping along all the time it's in daytime over land, which is about 1/7 of the time. The rest of the time, it just repowers its batteries, if you like. It does some imaging over the ocean, and Pelicans are more often turning to shoot at specific targets. Because of that, you have all this time when you're not taking images, which actually gives you more buffer to send it down.

Each Dove is imaging maybe a couple of million square kilometers per day per satellite. What is that? Bigger than the area of California for each satellite per day. This is why when people say, "Oh, let's use drones for agriculture," I'm like, "No, that's crazy." You would need a million drones per satellite—or a thousand, certainly. It's actually cheaper to do satellites if the resolution suffices from the satellites. It's just orders of magnitude.

Where does the resolution go? Right now, you're saying it's about 30—

30 cm, 3 m, super-resolution.

In 5 years, where do you expect to be?

We've already been upgrading our daily scan from 3 m to 1 m. With super-resolution, it can potentially get better than that, too. That might even go to 50 cm or 30 cm. Our Pelicans—we've moved from the SkySats, which we inherited from Google, at 50 cm. Now we're moving toward 30 cm. With super-resolution, again, you can get a little bit better, where you look at overlapping and sharpening based on pixel overlap and things like that.

What is the exposure time?

The exposure time is, I want to say, a couple of milliseconds.

So you should be able to catch quite a few interesting aircraft in flight.

Yeah. We get aircraft in flight all the time. I can show you that.

Any UFOs? Anything?

The Air Force took a look through our images, and I'm sorry to tell you, there ain't any UFOs.

No. That's too bad. Why isn't—I'm so excited about us detecting UFOs, but I'm sorry to all the audience members out there who think there are some that have visited the Earth, apart from the crazy people who think they've been abducted.

It ain't true. We haven't seen any aliens. And NASA, let me tell you from firsthand experience, could not keep that a secret. Never. Ever. That's ridiculous. That makes it even more improbable. NASA's not—

Why isn't Elon doing this with Starlink? I kind of imagine putting some cameras on board.

Starlink is what he's doing, right?

Well, yeah, but that's using it for SSA. They could, but they're kind of in the wrong orbits. They're a little too high, and you really want sun-synchronous orbit to have a consistent shadow angle for optical imagery.

They are doing some classified missions for the NRO, which are, well, classified, and read some stuff on the internet about it. But they are generally not in the business of doing Earth imaging. They're doing comms primarily—Starlink, which is obviously a very successful business.

That's, I think, the most exciting aspect of the SpaceX IPO, frankly. What's the mass of a Pelican or a Dove compared to a Starlink?

Pelicans are similar, and our Doves are much, much smaller—more than 10 times smaller.

Will, can you give folks listening an understanding of how quickly the technology has developed to build these kinds of satellites? It's been stunning. You were on the cutting edge of this. When did the first Dove go up?

2013. So, we've been doing it 13 years now. To give you a sense, the radio speed has gone from 1 megabit per second to 10 gigabits per second. The cameras have gone from 2 megapixels to 47 megapixels. The hard-drive space has gone from 100 megabytes to—what do we have now?—a couple of terabytes on there.

I mean, it's just extraordinary, right? Each generation of satellites, we tend to be doing about 10×. Our Dove to SuperDove went from 4 spectral bands to 8 spectral bands and from a 29-megapixel camera to a 47-megapixel camera. If you add that up, that was about a 5× increase in data per satellite for a similar cost per image.

Our next-generation daily scan, going from 3 meters to 1 meter, is 10 times more data—or 9 times more data, roughly—and we'll be getting it back about 10 times faster as well. So, 10× is still there for the having. But, Peter, I think the even bigger thing is our hyperspectral satellite, Tanager. We're 5×ing—we're working on a new one that has 5 times the swath width for the same spacecraft. Those things are possible, so we're gathering more and more data.

The cycle of increasing those sorts of things is, I would say, 2 to 3 years. So, 2 to 3 years for 5× or 10×, I would say, is the rough Moore's law for increases in data in space. But I would say the bigger thing happening now is the unlock of AI that really just brings down the barrier.

All this capability is latent for that farmer I mentioned, for the hedge fund manager, whatever, but they couldn't get access to it. So, I think we've got about 100× to go in the next couple of years, just because of AI unleashing what was already latent in the present data.

How does this flow to the average individual? How are people going to be impacted on the ground right now, worried about their local environment or people polluting and so forth? How do you make this accessible as an intelligent layer that people could just plug into on a regular basis?

Well, again, just imagine making a natural-language query of our data, just like you make a natural-language query of the internet via ChatGPT or Gemini, or pick your favorite LLM. Except, again, LLMs understand text, and large Earth models understand the physical world.

It can answer that question for that farmer: “How's my field doing? What should I do? What precision?” It could say, “Well, you've got blight over in this corner. You should put some fertilizer over there. Do this.”

Journalists can do their checks on an event happening around the world. Civil governments responding to a flood, or checking permits, can just say, “Here's my list of permitted buildings. Tell me which buildings have been built that don't have a permit.”

It can look at the last month, find the images, find the buildings, check those against the list, and then tell you which ones have and have not got permits. We've already done that in a few areas, in each case with journalists, finance, farmers, and civil government.

I think this is going to be a boon to the legal industry—tracking trash.

Trash. Here are 3 obvious use cases: What's the change in parked cars outside of Walmart over weeks and months?

Okay, that's the cliché, right? That's the cliché use case for folks purchasing Planet imagery to trade stock prices based on—

But shipping and knowing where ships are. There are rogue fishing ships all over the world that are a nightmare right now for fisheries and agricultural commodities such as soy.

But as the resolution gets better, if I live in Manhattan, is there a parking spot on my street that I could get right now?

Exactly. I want that for sure.

And that one—and also, is my teenager sneaking out the bedroom window at night?

We have 1-meter or 1/3-meter resolution here, so we'll get there. But I guess, yeah, it's obviously your kid if it's your house.

But more seriously, Will, I think you are in possession of a data set that could be GDP-maxing. How much of that analysis are you doing in-house versus externalizing to workers like—

GDP-maxing?

I said GDP-maxing. I just coined “GDP-maxing” with 2 Xs.

I do think we have hedge funds that are using our data right now. We are not doing that internally, but we have some hedge funds who will go undisclosed because they don't like being disclosed. We think that they are getting significant alpha on our data, which we are happy to take a part of.

In the future, I do think there has to be more. But, again, I would take it up a level. I think GDP-maximizing is one thing, but life flourishing is an even bigger thing that we can do this way.

We are super inefficiently using the Earth right now. Super inefficiently. Agriculture, for example, is terribly inefficient.

There are 10×s there for the having all over the place in agriculture. Let's go fix that.

Abundance, maybe.

Yeah. So, let's turn to Dyson swarms. I want to get to Dyson swarms. I have a quick technical question before you get to the theory. You've put NVIDIA chips onto the satellites. How's the cooling being handled?

Oh, that's relatively straightforward. We could talk about computing in space more generally, but we've been dealing with chips that are obviously hot and need to cool off for decades in the space sector. There's no magic here.

You can't use convection or conduction as you do on the ground. On the ground, they're either air-cooled or water-cooled with physical touch. In this case, they have to be radiatively cooled, so you have a radiator. But radiators—we've known about that for a long time. We know how to do radiators. It's a relatively known known.

One of the interesting things about it, by the way, if you like geeking out on this stuff, is that the radiating energy goes with T to the 4th—the temperature to the fourth power. So, basically, if you're a black body, which you're close to, if you double the temperature from, say, 100 kelvin to 200 kelvin, you quadruple—10×-ish—your radiative power.

Dumping energy is all about tricks of thermal regulation, of radiators, and how you stop it. You want to get it as hot as possible without melting it. There are lots of tricks to the trade there, but there's nothing fundamentally unknown there. These are known knowns.

All right. Second, say you want to aim in the direction of the cosmic microwave background whenever possible.

Absolutely. The 4-kelvin temperature of space. You want to point your radiators at the dark.

Let me take some notes on that one.

Project Suncatcher—let's jump into this. You're putting TPUs for Google in orbit. You're building an early version of the Dyson swarm: orbital AI compute. Can you tell us what you're doing there? Obviously, everybody's thinking about Elon’s data satellites. How do you compare? Are you going to get your launch? Have you been launching on SpaceX?

We've launched 40-some-odd times. 15 have been on SpaceX. We've launched over 300 satellites on 15 launches with SpaceX. They're one of our best partners. We love working with them. They've got it close to a bus ride to space.

I will point out that, in addition to launch costs coming down, the biggest upheaval in space—and I think I mentioned this the last time I came on this podcast with you, Peter—the bigger transition over the last 10 years in space has not been the launch cost. It's been satellite cost performance. It's been the miniaturization of satellites, both for Starlink and ourselves, and we sort of pioneered that. That led to at least 100×, if not 1,000×, in cost performance for each kilogram you put on the fairing.

So, the dominant thing that has changed to lead to all these large constellations of satellites is actually the capability performance of satellites, not the launch cost. But both add up, and they make things better.

Performance density. Tell us exactly.

Exactly. In our case, how many bits do we collect per kilogram or per dollar spent, which is related to kilograms because of the cost of launch? We've been launching a bunch with SpaceX.

SpaceX didn't come up with this idea. I will point out they only started talking about this after we announced our project.

We've been thinking about this for some time, and we're not the first ones either. The space industry has been talking about energy from space for decades and decades—space-based solar power. For many years, the idea has basically been that we want to put energy-intensive infrastructure off Earth, where there's abundant energy and where it's not conflicting with incredible biodiversity and people's lives. As Jeff Bezos likes to say, we want to zone the Earth for residential and light industry and put all the heavier, energy-intensive infrastructure in space.

People have been talking about energy in space for a long time, but the first and most obvious, easiest one is compute in space. With space-based solar power, you need to beam all that energy down, and how do you do that without frying people's heads? That's actually difficult. Whereas, by putting compute in space, you get all the power advantages, but you only have to beam up the questions and beam back the answers. We know how to beam bits; we've been doing it for a long time. Communications satellites were one of the first uses of satellites.

We did a study with Google about 8 or 9 years ago looking at the details of compute terrestrially—the cost, the water, the building, the energy, all the things—and what it would cost to do it in space. It turns out that when launch costs come down to about $200 to $300 per kilogram, it's going to be cheaper, surely, on a pure cost basis to put it in orbit versus on the ground. As Sundar put it at Google, “Within 10 years, we expect most compute to be put into space.”

That is a big deal because Google alone is spending $200 billion a year, at this rate, on compute. That's roughly the size of the entire space industry today—rockets, satellites, communications, everything combined. So Google is just going to do it. Add up all the other folks that are going to do compute, and you've got a business that's bigger than the rest of the business—maybe 10× the entire space industry today. It's going to change the space sector.

We're doing some early tech demos for Google. When we did this study 8 or 9 years ago, Larry and Sergey were like, “Well, let's come back in 2030 when the launch costs come down to there.” I said, “No, let's come back 5 years earlier, because it's going to take us years to build the technology to do the radiators and the clusters.” You basically want a rack of GPUs on each satellite, and then you want clusters of spacecraft in close formation flying with optical links between them. All of that is a whole load of technology to develop.

What we're doing with Google is that they selected us to build their first couple of satellites to test TPUs, radiation management, the cooling, and the inter-satellite links. We're doing a couple of the tech demos very early. It's a moonshot project, but the long arc is that it's just going to be cheaper, and it has the peripheral benefit of not clashing with energy costs for communities, water for communities, or the biosphere. It has lots of terrestrial benefits as well.

Peter, we talk on the pod all the time these days about a sun-synchronous orbit and Earth acquiring its own mini-Saturnian ring, if you will, in a polar orbit. When do you think SSO-based Dyson swarms will become visible at night or during the day on Earth? What's your timeline? Is it like the early 2030s?

We've got loads of satellites in sun-synchronous orbit, and you'll see them today in orbit if you look just after dusk or just before dawn, when satellites are most visible. Most of these satellites will go into a dawn-dusk sun-synchronous orbit. That means they're facing the Sun 24/7. However, that also means they're not going to be very visible, because that's literally when it's still a little bit light outside, and it's going to be hard to see them.

By the way, there are real challenges with interfering with astronomers on the ground, and we have to be careful about that. But this is the best time to do it because it's not interfering with the deep dark-sky needs of astronomers. It's really in these other orbital planes. The short answer is, it won't affect your seeing these rings. You won't see these rings of satellites.

I want my rings. But that's also a small number. Elon has FCC approval, I think, for 1 million of these AI satellites. Don't you think, at some scale, if so many of these birds go up, they either start to become visible or they start to become—

You would be putting them at slightly different angles in space?

They form a full band.

Well, yeah. You would put them in slightly different inclinations where you still have 24/7 sun, or very close to it. Yes, you would start seeing that, but it would only be right as it gets dark and just before it gets light. It would be like this funny ring effect. Later, we may put them in other orbits as well. I don't know. They would have to be much higher to get the—

How concerned are you about orbital debris? We talked about how, in Elon's S-1, the number-one risk factor on Starlink—which is their revenue and profit engine right now—was orbital debris, being able to knock out a lot of capabilities. What about you? What do you think about that?

I think space debris is a real challenge, and that's why we put our satellites below the area where that's a challenge, which is 800 to 2,000 kilometers from Earth's surface in altitude. We put our satellites at 400 to 500 kilometers to keep them well below that problem. Kessler syndrome is already in operation and in effect.

Bear in mind, there are on the order of 10,000 satellites in space, and there are about 100 million pieces of space debris. So there are about 10,000 times more objects in orbit that are debris than there are satellites—10,000 pieces of debris for every satellite. The vast majority of the problem—even if you put 1 million satellites up there, 99% of it would still not be satellites.

The challenge we have to deal with is debris. It is mainly made up of all the small bits of stuff left over from former rocket bodies, exploded satellites, anti-satellites, and other things that were put into high orbits and so could live there for decades.

When we were at NASA, Peter might remember this, we came up with a scheme under Pete Worden's mentorship of using lasers on the ground to do traffic management of that debris. Obviously, with 2 satellites, you can move out of each other's way if they're maneuverable, but most of the conjunctions in orbit are debris with debris.

So what do you do about that? We need to stop the collisional cascade for those pieces. For that, you can actually use lasers on the ground that generally nudge one so they miss each other. You can do this sort of traffic management. We call it Light Force. A system like that could actually stop this cascade and slowly bring everything down.

But the actual satellites are less of a problem as long as we keep them in low Earth orbits, and there's lots of space. Just to give you a rough order, even in this sort of sun-synchronous, dawn-dusk orbit, there's about 1,000 times more space—thinking very crudely—than there is on the entire landmass of Earth.

But just wait, this is really fascinating. So wait, you're at 300 kilometers or 500 kilometers? What's your altitude?

400 to 500 kilometers.

400 to 500 kilometers. And at that altitude, what's the lifespan of an object orbiting there?

A few months to a few years.

Okay. So, self-cleaning. You were starting to walk through this: what? Drag pulls everything down?

Yeah.

So you have about 100 kilometers of space where you can get a good 2- to 3-year orbit. A GPU in space is going to depreciate over 3 years anyway.

Exactly. That's why we call it strapping space to Moore's law. We always update our satellites every couple of years because satellites in space become obsolete just like the phone in your back pocket. You don't want a 10-year-old phone, and you don't want a 10-year-old satellite in space.

What altitude is Elon going with for his—

Well, he was going higher, but I made the point to him that, firstly, that's a real challenge with space debris, and secondly, it won't be self-cleaning. Even if you put propulsion on these things, even if 1 in 100 fail or 1 in 1,000 fail, you have a really big challenge if you put that much mass into those orbits. It makes much more sense, and later Starlinks have come much lower down. That's much better for everyone.

How do you just pull on the upmass question a bit? Over the past 5 years, I did this calculation in my newsletter. For the past 5 years or so, according to what I've seen, upmass has increased by 40-plus percent year over year. If you just naively extrapolate a 40-plus-percent year-over-year increase in upmass, by the year 2144, I think you find that the entire mass of Earth has basically been up-massed and Earth has been disassembled, if you just naively follow the exponential.

By the way, everybody, I am not supporting the disassembly of Earth. We could—

For avoidance of doubt, Peter does not support the disassembly of Earth. We've established it. Good.

Obviously, extrapolating anything 140 years into the future is rather tricky business, as you guys are aware.

It's really the whole point of the singularity: it's harder and harder to predict the future. I remember when Peter and I first met 20 years ago, it felt like we could easily predict roughly who was going to do what in 10 or 20 years in the space. Now, if you could predict it 1 or 2 years out, you're a genius.

For AI, it's even harder. It's measured in months, right? You've got to say 3 to 6 months into the future. So that horizon is shortening for sure. And 140 years, I think, is just—we can't even discuss it.

So, no predictions, then, regarding when upmass increase will start to slow down? Because right now it seems naively set to increase.

No, upmass is definitely going to continue to increase. But again, I think the most important aspect of that is: how do we get the energy in energy-intensive infrastructure?

Data centers are going to become a real hot topic politically in this next election, in the midterms, and in upcoming elections because people don't want data centers in their backyard. They don't want the energy cost to go up. They don't want their water to disappear because they kind of like access to clean water. It's kind of handy.

This is causing lots of tension, and it's not surprising. We're wiping out agricultural lands, farmlands, and what have you, for this. Putting it in space is the way out of that conundrum, and then we can have compute and not interfere with those communities.

The elephant in the room here, Will, I have to ask it: how do you compete with Elon's plans for orbital AI data centers when he's got the launch capacity and massive manufacturing capacity? Do you end up folding teams together? Are there going to be more than one player in orbit? Or does Google just acquire you?

I want to ask about that, too. I want to throw one more log on that fire, which is: Google sold you their satellite business. And that was before everyone realized data centers would be in space, I think. Now they're working with you. If Elon doesn't want to launch, Eric Schmidt now has a rocket company. There are a lot of arrows pointing in a different direction. Here's the Elon-verse, and here's the Google-verse, and you're part of the Google-verse, but I know you're working with SpaceX and various others.

Look, there's a complex relationship. Google is both a shareholder in SpaceX, and they're competing. These are both competitive and collaborative situations, and we feel the same. We're a strong partner with SpaceX. We really love their partnership on launch. We work with them, and our teams work together really well.

I wish them great luck with the IPO. I think it's fantastic that there's so much interest in space. It's so hot right now. At the same time, they compute in space, and we're really helping Google with their project a little bit, and we'll see how it goes.

They take a different path, but don't underestimate their smarts or our smarts and how we can do this. I see, roughly, Elon is throwing mass at this because he can with the rockets, but we're throwing smarts at this, and there are lots of tricks up our sleeves for how to do this really smartly.

Let's move on to our next story, which is still in the space arena. But this time we're going to talk about the launch industry.

As SpaceX is rocketing forward and Blue Origin had a kinetic disassembly of its New Glenn rocket, here comes Relativity Space. A little background on this: Relativity was founded back in 2015 by Tim Ellis and Jordan Noone. They're both friends. I was an early investor in Relativity Space, and I've had them on my stage at the Abundance Summit.

Relativity, back in 2023, flew its Terran 1 rocket. It got through Max Q, but it did not get to orbit. Very few rockets have gotten to orbit on their first launch attempt. Only 3, I think, in history right now in the U.S. have gotten to orbit on their first attempt.

They pivoted after Terran 1 to go to their Terran R, which is a heavy-class launch vehicle. You can see the numbers here: Terran R is 23 tons, Falcon 9 is 22 tons—roughly the same. New Glenn is 45 tons, and Starship is 100 tons.

They missed their financing. It's really hard to finance space projects, especially rocket projects. And here comes Eric Schmidt, who was an early investor. He comes in and writes the check to basically buy Relativity Space. Eric is now the CEO of Relativity Space, which blew my mind when he took that role.

They just announced that they have gotten a mission from NASA called ELIS. It's a Mars orbiter sensing mission with some communications capability. Any thoughts on this one, Will? Do you want to—

I've known Eric for many years. He's a very early investor in Planet, in our Series A round, all the way back to the very beginning. Eric has a smart eye for business and a smart eye for technology. He's obviously relatively new to the space business, if you can excuse the pun.

We obviously think the world of Eric, and Relativity has come a long way. They had some of those financing challenges, but I think now, with Eric's backing, they can go a long way. I'm very excited for them, and I hope we can launch with them.

Well, Will, this story is really interesting. I didn't know he was the seed investor in you. Was he still CEO of Google at the time, and did they still have their satellite business at the time? How did that—

He was an investor before they bought Skybox, I think.

Okay, so he was running Google, made the investment, and was probably aware that data centers might move into space someday. This is a long time ago.

This was before that had caught the eye of all the founders of Google.

But, Will, the question is: did he buy Relativity Space with the thought that data centers in orbit are going to be critical? Because it's a massive advantage for SpaceX to have launch and satellite capability and data center capability.

I mean, we interviewed him 4 times in the last year, Peter. I'm really coming around to the view that he 100% knows—and knew—that this was the future because he said on every one of those interviews, “I don't know anything about space, but I know a lot about people, and I know a lot about companies.”

He also knows a lot about investing. He's got to be one of the best in the history of the world. His vision is unbelievable, and he has access to all the information in the world. I didn't know he was a seed investor in Planet, so that's one other source of information that he has.

From that vantage point, yeah, Elon can't be the only guy launching, and Jeff Bezos is no dummy either. He's launching, too. Of course, it's been a passion of his his whole life.

I have a Relativity Space question. When NASA was launching space shuttles, it was between $600 million and $1 billion per space shuttle launch. SpaceX dropped that down to about $60 million. The plan was for Relativity Space to operate at about $6 million a launch because they were 3D-printing the rocket engines or big chunks of it.

Originally, they had their Stargate printers to print all of the rocket. Then they broke it down and said 85%—

And now today, I guess they're just 3D-printing their engines.

Yeah.

Do we know what the launch cost is that they're aiming for? Does anybody know?

I don't think that's disclosed. I tried to look for it.

I also think it leads to the question behind the question: What happened to 3D printing in space, for space—terrestrially or in space? My perception is that Relativity, under new management, is migrating more in the direction of competing in medium lift and heavy lift, and there's potentially a gap in the market now that Relativity, which was originally aimed at and focused on 3D printing for space, has moved away from it. Someone else could potentially fill that gap. I'd love to see more 3D printing in space, in cislunar space, on the lunar surface, and in general. No one right now seems to be the obvious incumbent anymore in that market.

Yeah, I agree. There's a huge opportunity in 3D printing. Fundamentally, all the design constraints for satellites have to do with the launch. That's the hard thing: the vibrations, the separation, where you get a 200 g shock load, and then you get into orbit and you don't need any structure at all, basically, because it's zero g. So you want a completely different design for your launch than you do in orbit—roughly speaking, a completely different design.

Peter and Will, it's so rare to get you guys together—you’re 2 of the top people on the entire planet on this whole launch-cost question. We just have to get this figured out right here, right now. So, Elon’s rocket is massive in scale, with a couple-ton payload, but how much of the efficiency is driven by reusability? Elon has always been saying it's the reusability of everything that's the driver, not the overall scale.

Their goal is to get down to $100 per kilogram, from where it was in the past at $10,000 per kilogram. The only way you get that is by rapid reusability. Remember, to launch the 500,000 or 1,000,000 satellites for his AI constellation, it's like 2 launches in an hour.

10 rockets, or do you want 1,000 rockets? Yeah, exactly.

Well, this is where I'm going. The Relativity Space rocket is also relying on chemical rockets so far. We've also just never mass-produced rockets. There are 2 independent ways, and no one's really used this other way. And then, a whole separate thing: I would be thinking about this if I were Google or one of these big data-play companies that sees it wants to spend a trillion or more on space over the next decade or 2.

If I want to spend that much, I want to spend a few billion on novel launches, because printing—yeah, let's just launch blocks of material and then 3D-print it. Or, as Elon’s been talking about recently, rail-launch it from the moon, because from a sheer energetic standpoint, getting stuff from the moon to low Earth orbit is cheaper. But even from Earth, which is the near-term, easier one, there’s SpinLaunch or Longshot, or these kinds of very different approaches. No one's thrown a billion at one of those, or a few of those, to see if it could actually work.

We've used chemical rockets because Wernher von Braun figured out he could bomb London 100 years ago. Not quite, but you know what I mean. And then no one's invented anything since, basically—anything. I mean, even the reusability was cool, but no one's made a significant advance. We're stuck in the chemical-rocket paradigm, and we don't need to be. There was a brief foray in the '60s and '70s, both with Russia and the US, into fission-powered rockets, but then everyone got scared about that. I think we need to revisit the launch equation at this point, because the way to get from 100 to 10 to 1 isn't going to be a chemical rocket. It's going to—

Yeah, Peter, that's what I say to you all the time. You're concerned about the SpaceX launch monopoly, but there are many other launch paradigms that could potentially leapfrog it. Space elevators with new materials, of course. I did the calculation on this pod: mgh and ½mv² in terms of total energy. If you could build it from space, winch it up, and accelerate it, you can get the cost down for you and your spacesuit to $100 or $120.

So wait, let me follow up with one more question. I'm really, really curious, though, and you guys are the experts. If I get fully reusable from Relativity Space, but it's a quarter the size of an Elon Musk rocket, there's got to be some economy of scale that comes with just raw size, which is why Elon pursued it. But they're still fully reusable.

Now, as Will is saying, it's the manufacturing of thousands of these in an assembly line. He's built a machine to build the machines. His goal is thousands of Starships, maybe even more. I mean, if it's fully reusable, it's just the cost of the touch labor and the cost of the fuel. The fuel is de minimis. It's free. It's oxygen and methane.

So let's say that Eric Schmidt is doing the exact same thing, but his rocket is a quarter of the total scale—a quarter of the payload, probably. Is that significant? The launch costs aren't the issue. You're launching these very expensive 72-cluster NVIDIA GPUs with all the cooling, the solar power, and everything. That's an expensive piece of equipment. Suppose that Eric's launch costs are $200 or $300 per kilogram, not $100. Does it matter? Is Eric still competitive? Can we have a duopoly then?

Oh, yeah. Or can I explain a little about this? Because people, I think, misunderstand it. It's not just about the launch cost. The launch cost is the biggest piece to get us to the threshold that makes sense, but thereafter it is as much—I argue, probably more—about the efficiency of the compute than about the launch costs.

Really, the efficiency of the compute drives the amount of energy you have to dump, which drives the mass of the spacecraft, and that ends up being significant. For example, Google TPUs are significantly more efficient than GPUs from a FLOPS-per-watt standpoint. That really matters because of all the rest of the GPU energy. I like to put it simply: whilst everyone apart from SpaceX has to pay the SpaceX launch tax right now, everyone apart from NVIDIA and Google has to pay the NVIDIA tax right now. Which tax is more important? I actually say the near-term answer is the launch, but longer term, it's the compute.

That is brilliant. That is absolutely critical. Nobody has said that before.

It's more important than launch for this game long term. Mark my words.

And that means Google—if the TPUs, at inference alone, use significantly less energy per inference, they choose the winner of space.

Correct. NVIDIA could have a play at this, but its GPUs are more general than the TPUs. The TPUs are more efficient. Obviously, Elon’s trying to build his Terafab, but that is a big long-term project, if ever there was one.

Meanwhile, Google has been investing in that compute for a long time, and they have efficient systems for leveraging that compute in ways that will boggle your mind. People think of Google primarily as a software company, and they are. When they gave us their satellites—or gave us, we bought them—we were astonished. We thought, “Wow, they really know how to build and operate satellites.”

They're so brilliant. Elon is always trying to find the innermost loop of the innermost loop of the innermost loop. Right here, right now, that inference-time power efficiency determines the winner of the entire thing.

Everyone's writing off Google at the moment. They have massive defections of key talent. We'll see it later in this pod. But if they have a 2× watts-per-inference advantage over NVIDIA—remember, NVIDIA is highly, highly emphasizing training time, not inference time, because Cerebras and a whole bunch of other things are starting to really eat away at inference-time efficiency—the TPU 7 or 8, I guess the next TPU, will determine whether space is dominated by, like you said, the launch cost. Even if it's 2× on an Elon Musk rocket, that's not the swing factor. It's: Can I access those chips?

Right, right.

That's really brilliant. We have talked in the past few episodes about the training-versus-inference balance on terrestrial versus orbital data centers—one argument versus lunar versus Martian. One argument in favor, in the short term, of terrestrial data centers for training is that it's just easier to build larger, coherent training sessions on a terrestrial data center. What do you think is likely to be the balance between training versus inference on terrestrial versus non-terrestrial?

Yeah. I think inference does make more sense in orbit, to first order, and it's mainly because that's more distributed—lots of little runs of a machine, right? Now, there is an advantage for training runs: You want to send your data, have it spend a couple of months crunching it, and then you send the answers back.

Alexandr Wang

So from a communications standpoint, it's easier to do the training in orbit than the inference because you really need the latency down for inference. But from a compute-distribution standpoint, it's easier to do the inference in space. Obviously, 70% or so of the compute on Earth is now inference—or in AI, at least, most of it is inference, not training. That's only set to go up, so I think the main problem to solve is the inference one anyway.

Okay. Go ahead, Alex. Do you think training is likely to remain grounded in terrestrial data centers for the foreseeable future, or longer?

Alexandr Wang

I don't think it will be forever. I think it will all go to space, but I think inference will go there first. Yeah.

Great. All right, moving us along out of the space arena, because we could spend all day here, and everybody listening has gotten their PhD.

I have one last question about space.

All right. One last question for Will: Is the best commercial opportunity about leaving Earth or making Earth more useful?

No, I mean, look, my co-founder Robbie, whilst I was sending missions to the moon, if you may remember, was working on a mission called LCROSS. So we helped to find water on the moon, which is very exciting. We, as lunatics, were very pleased about that because it makes the moon much more—I mean, it was already a better, smarter destination than Mars by 10x, but this made it a 100 times smarter destination than Mars, which finally was the nail in the coffin that Elon finally understood recently and changed his mind: the moon is first.

By the way, in the long run, are you a moon-then-Mars or a moon-then-asteroids person?

I would say the moon is enough for a long, long time. I'll get back to this, because what Robbie was doing was focusing on exoplanets. He had these telescopes looking out, looking for planets around nearby star systems, and now we have found thousands. I think it's up to almost 10,000 planets around nearby star systems. I'm here to tell you and everyone else that the best one by far is Earth. I'm not talking about by a little bit; I'm talking about by several orders of magnitude.

There is no place on Mars that is better than the worst place on Earth. Not by a little bit, okay? This planet is so cool. And the reason I want to emphasize that—and excuse my French—the reason I want to emphasize that is that I don't believe in sending millions of people into orbit anytime soon.

I think it's all about protecting this incredible biosphere. Life is either singular—we haven't found the aliens, sorry to break the news for those geeks who think they've been abducted. We haven't found life off Earth. Life is either singular on this planet or extraordinarily rare. Either way, we have the most beautiful life system on this Earth—an incredible complexity in how it all works together. That is worth protecting and putting most of our energy into.

Space is super useful for that because it gives us the advantage and the data that underpins our ability to manage this planet smartly. But the planet is here. SpaceX can be space for Mars. Bezos could be space for the moon. Off they go.

We're Planet, space for Earth, to help us take care of the Earth, both with Earth imaging to help upgrade the planet and make smarter decisions, and by helping take energy-intensive infrastructure off the planet. We're space for the Earth because this planet—I mean, they can have those planets. This planet is by far the best.

Ladies and gentlemen, Dr. Will. Hear, hear. The defense rests. I love it. I'm going to move us along because we've still got a lot to cover. Thank you, Will.

Exactly. Well, we have to do that. We have to move on, but we have to do this again. There's so much more to explore. This has been phenomenal.

Our next story here is the great AI brain drain. Two of the most important minds in AI have changed teams this past week. First off, Noam Shazeer. If you don't know his name, he was the lead author of the Transformer paper, which is the T in GPT. The architecture of the entire modern AI revolution was built on his discovery. He's unfortunately leaving Google for OpenAI.

And get this: This is the second time he's left Google. Two years ago, Google bought his company, Character.AI, for $2.7 billion to bring him back and put him in charge of Gemini. Well, he's leaving again. I would guess he's leaving after part of his stock package is vested.

Alexandr Wang

Quick turnaround.

And second, another rock star left Google.

Alexandr Wang

Yeah.

Yep. The company wasn't worth anything. It had basically zero—

Alexandr Wang

Revenue for him, and now he's out.

A whole generation of Silicon Valley parents are naming their kids Noah.

Yeah. And that begs the question: What is his comp package at OpenAI? They pulled him. It must be insane, whatever it is.

Must be huge.

Okay. Second, another rock star left Google: John Jumper, the Nobel laureate who helped Demis create AlphaFold, is switching from Google DeepMind to Anthropic, likely to push their AI for science. Remember, a couple of weeks ago, Andrej Karpathy also joined Anthropic. He was a free agent.

So my question for you, Alex, is this: Is this AI talent literally voting with their feet? Is this a better prediction of where AI is going?

Alexandr Wang

Yeah, I think so. I have no financial interest in this, so I can speak pretty unvarnishedly on the subject. My perception is that the frontier is very competitive, and at the moment it's a duopoly at the frontier between OpenAI and Anthropic. Google DeepMind has fallen behind the frontier, and I think it was notable at Google I/O that Google did not release a frontier model at all.

They released a Flash capability, which is great in everything and certainly much more aligned with Google's search-level economics, where you want ultralow-latency, ultracheap models to power the one-box answers in Google Search replies. That's great for Google's existing legacy search business, but it's not a frontier capability.

My perception is that Google has fallen behind the first tier of frontier labs at this point. If you're a top researcher, you have to be asking yourself all of the research questions that you could be asking with raw access to the pretrained models before all of the post-training and all of the guardrails get slapped on. That's very attractive if you're a frontier-lab researcher: to have that raw access to a pretrained model at the frontier.

If Google DeepMind doesn't have that frontier-level access, you're probably looking to either OpenAI or Anthropic to get that frontier access for yourself. So, yeah, I think this is a reflection of Google DeepMind falling behind.

I'm going to give a different point of view. I think this is relatively in the noise. We've seen people move from Anthropic to OpenAI, OpenAI to Google, OpenAI—I mean, in all directions, right? That's going to continue to happen, and these are two significant players. I don't want to trivialize it, but I think the stock-market reaction in particular was overblown, and I wouldn't bet against Google in this game.

Yeah, I keep on saying that.

If I was an AI researcher, I'd pick the one with the most compute. That's Google by far. With the most data, that's Google by far. And the smartest people, that's Google by far. I'm sorry, that's just true across the board on all of those things. It is a compute, data, and talent game.

I think they did fall behind a little bit a while ago, but not now. I think that Gemini model is generically pretty good. I'm not the best expert on that, but my observations are that it's pretty high up there, and the prospects are even brighter.

In fact, I think this is Google's to lose. I think Anthropic is doing incredibly well, especially because they picked a very different business model. But OpenAI has picked the business model that more or less is in Google's sweet spot, and Google already has 10 applications with over 1 billion people to put its AI systems to. They're the incumbent in that space. So I worry much more about OpenAI than Google.

Dr. Blania.

Alex Blania

I agree with everything Will said, but I'm going to give you the counterargument just because I know a couple of the players. Before these recent defections, Shane Longpre from MIT went over to Anthropic, Tobin South from Stanford went over to Anthropic, and then Andrej Karpathy, as we just said, was a free agent.

All of these guys are singularitarians who believe that self-improving AI is exponential and almost instantaneous. Now you have John Jumper going over, and I don't know him, but you do, Peter. I think what's happening with those four people and a lot of other people is, “I could go to Meta and get paid a lot, but I'm going to miss the singularity.”

Anthropic—yes, they don't have Google's compute. Yes, Google has a huge advantage with the TPUs. But if I believe that Claude 5 is truly self-improving, has crossed that line, and I can't use Claude 5 at all right now, but Claude is what I really want, the only way you can be part of the singularity in world history is to go join Dario.

I know that's the psychology of the first three, so it wouldn't surprise me if its advantages were compounding exponentially and taking off very rapidly. I can imagine the job interview: “Come on in. Let me show you what's behind the firewall.” And it's like, “Oh my God, I've seen God. I cannot go back.” And that is—I mean, this is publicly reported, Peter—that this is how Anthropic does its recruiting.

The publicly available information is that, as opposed to the way Google DeepMind does its organizational workflows, Anthropic reportedly puts a lot of its best people in front of the applicant, or job seeker, and shows them, “All of this compute can be yours. Here is access to the models with raw capabilities.”

I have one more thing. This is not a coincidence. If you look at Polymarket’s prediction of Fable 5 coming back, it goes down a little bit every day. Fable 5 will come back, but it’ll be a reduced version of what it was the first time it was out. For Polymarket to be true, it just has to be a product called Claude 5, and then that pays off.

But even given that it’s coming down, the odds of it coming back by the end of the month keep slipping. I think Dario loves that—the only way you’re getting access to the best of the best of the best frontier, self-improving AI is right here inside our building. Every week that goes by is another week toward the singularity that you’ll miss if you’re not part of my—

I remember when I brought Ray Kurzweil over to Larry Page to meet him for the first time, to make an investment in his company. Larry’s point was, “Instead of me investing, the only place you’re going to be able to build out your vision, Ray, is inside Google, where you have access to all of this unfettered.”

I can imagine that’s the exact same point. You know, John Jumper—I was saying to you, Alex—I’m amazed, given Isomorphic Labs just raised a whole bunch of capital and they’re focused on the biotech arena, that John—and I do know him—would jump over to Anthropic. It’s got to be that Dario has just basically come and said, “Lead our bio, and you have access to unlimited compute, far beyond what Google DeepMind had.”

Alex Blania

The only thing I’d tweak on that, Peter, is that the recruiting pitch to Ray Kurzweil is, “This is the only place in the world you can build your vision.” That was a few years ago—many years ago. Now the pitch is, “The biggest event in the history of the world is imminent. The single biggest thing that’s ever happened in human history is imminent. It’s going to happen in one location on the planet. Our benefit—”

And that’s why we have the Fermi paradox and there’s no other life in the universe, or it’s going to be everyone’s benefit and it’s going to be awesome.

Alex Blania

Those are the rolls of the dice. We are apparently just about to play.

You know, I was with Mike Saylor, and he said, “What are you excited about?” I said, “I’m excited about the future.” This is the most extraordinary time ever to be alive. I do believe that we are living in this quantum superposition.

I think people need to have a positive vision of where we’re going and manifest that future, because if you don’t believe it and you’re steering toward the negative, dystopian future, that’s what we’re going to get. So the purpose of this podcast, for everybody listening, is to give you a positive vision of the future—the hopeful, optimistic vision. I think it’s so critical.

Silicon Valley basically ran for decades on Star Trek, and we just don’t have the modern equivalent, beautiful future vision for—

Alex Blania

Building. That’s what the Future Vision Prize is about, right?

Yes, exactly. We need those. We need Neal Stephenson and Kim Stanley Robinson and others to put out books on the future of AI and humans and how it can work together, because right now everyone’s Terminator, and it might end badly. I’m so angry at Hollywood, right? Because we’re shaping our neural network.

Alex Blania

You can’t blame Hollywood. That’s what sells. It tickles your amygdala. It’s why horror movies sell. My favorite sci-fi is Accelerando. What’s yours? What’s the best depiction of the future?

I don’t read a huge amount of sci-fi. I find science fiction fascinating. I read Nature magazine every week because I find it so fascinating. I’m already tapped out. But I think some of the classics, like Snow Crash and others, were incredibly good at depicting the next era.

One of our subscribers asked for another book corner, Alex, so I appreciate you asking this question.

Alex Blania

Wait, I need to get my word. I need to get a word. Obviously, the John Jumper thing—I think it’s much simpler than all of what we’re talking about. It really simply comes down to agency.

Google is a big company with a lot of organizational drag. If you’re an individual, you can make a much bigger difference in a smaller organization. Yes, they may have better models, et cetera. It goes all the way back to, Peter, in our 2014 Exponential Organizations book, we said, “Smaller beats bigger,” right? “Trust beats control.”

We have this kind of rolling carpet where smaller teams can outperform bigger teams, and you can do so much more. When Facebook launched, Google spent 2 years trying to build Google+—

And it was a miserable failure because you had to get permission from YouTube, the groups, and Search, and we were trying to integrate amongst all of those. Meanwhile, Facebook was saying to its developers, “Anybody who’s ready with their feature, just take it live on the live site and go.”

Of course, they were outperforming Yahoo, Google, everybody. I think it comes down to the ability to get things done more quickly. It can happen more in the smaller labs, plus they may have access to the best frontier models.

I like that. I think your point’s right. I think that sort of mundane factor could be much more important, and that’s why I was saying I think it was overblown what this particular incident meant for Google. We’ll see.

But what I want to throw in is that, as you say, the small guys are going to make a big difference, and I want to make a pitch for how the space sector is going to play a big role in this AI future. Let’s come back to where we began, which is this: when a baby is born, they learn and become intelligent and ultimately self-aware and conscious by interacting with the physical world.

They are not a brain in a vat, and they wouldn’t learn the way they do without interaction with sensors and their physical actuators. AI at the minute—the LLMs—are basically brains in a vat. They have absorbed the text of the internet, but they are largely isolated from it.

They can’t interact with the physical world in real time, not in terms of sensing or actuators, and until they do, I don’t believe they’ll learn. So I actually think that physical data and, obviously, planetary sensing—at the big scale, that’s why I talk about planetary intelligence—the big scale of planetary sensing is going to be done from space.

The compute is soon going to go up there, as we just discussed, and that’s really going to lead to a planetary nervous system. What that might enable us to do is build toward planetary consciousness and planetary wisdom, because that waking-up point can only happen when you start having that real-time loop. I think that these things are not unrelated.

Alex Blania

I love that. I love that. But also, it will align it with human interests because it will be conscious and therefore be empathetic with our conscious experience. It will know about all the deltas, the forests, all the animals, and all of human civilization, and therefore more implicitly care about it.

Caring about something and knowing about it are highly correlated things, even though they could—

And highly desirable. Yes. So the AI future isn’t going to be just those guys sitting in their library with just the text of the internet. They’re going to have to get into the physical world. AI companies—whether it’s Anthropic, OpenAI, Google, or any of the others—are going to have to get into the real world: cars, satellites, drones—

Alexandr Wang

Robotics. You want to hear something—

They’re going to graduate to the next level. We’re going to need a leap in AI, and it isn’t going to come from just throwing more compute at the text of the internet. It’s just not going to come that way.

Again, obviously, I’m extraordinarily biased, but I think that space data is actually going to play a nontrivial role in that, because what’s the Wikipedia? The core corpus of the LLMs is Wikipedia. It’s like you’re chatting with Wikipedia when you’re chatting with an LLM. More than anything else, an LLM is Wikipedia wrapped up.

Alexandr Wang

You want to hear something totally mind-blowing?

Go ahead. We just invested in a little team in San Francisco, down the road from you, Will.

Alexandr Wang

That tells us they’re going to beat Google to AlphaFold 2. They’re going to have better protein folding, and they’re a little team of 5 people.

How are they doing?

Alexandr Wang

A couple of Stanford guys and an MIT guy who used to work here at Scale AI. They said, “It’s not because we know anything about protein folding. It’s because we have a recursive, self-improving process that’s just mind-blowing.”

Totally.

Alexandr Wang

That’s why I’m not giving you the company name, because I don’t want people to show up and spray-paint their door. They’re like, “Yeah, we literally knew nothing about protein folding 2 weeks ago, and we’re still going to beat Google to protein folding.” I don’t know if they’re right or wrong.

I don't want to throw their names out there, but it's mind-blowing to think that what he was saying—a little team with agency using RSI—has superpowers. So then, just a couple of days later, it's a $45 billion-a-year revenue company as of 2 months ago. Now it's probably double.

Alexandr Wang

Yeah.

Insane. What's amazing to me about the story I just told, though, is that right after that, John Jumper goes over to Anthropic, where RSI may be imminent, and he's also trying to solve all diseases using a similar RSI. He may be wrong, and it may be your startup.

I want to emphasize one more thing about this direction in planetary intelligence: it's not just that it's going to happen—it's happening right now. We have already built this app. It's in beta testing right now, and it integrates Planet's data with AI, enabling people to make natural-language queries. It's going to be world-changing, and lots of other companies are doing things like that.

There are going to be totally lateral plays to the AI game that are going to come out, and I think they'll end up being critical for the next phase of AI development, especially AI alignment.

Alexandr Wang

Of AI alignment. Yes. And—

Well, you're building that planetary nervous system that the world really, really needs.

Alexandr Wang

Totally. Maybe, if I may, let me just push on this a little bit, Will. Since you're referring quite a bit to LLMs, maybe more colloquially one might speak of foundation models that are intrinsically multimodal or omnimodal, trained extensively not just on internet text but on internet images, internet video, and synthetic video in many cases.

They have world models—lowercase “w,” not Earth-scale world models. So one might suggest that most modern frontier models already have a pretty good native, intrinsic understanding of the physical world. It might not be perfect. If you try to use, say, an omnimodal model from Google, maybe the physics won't be perfect, or the classical mechanics won't be perfect, but they have a pretty good abstract and concrete understanding of certain aspects of the physical world.

I'm curious why you seem to think so much that orbital imagery, sky-to-Earth imagery of the Earth, is so important for understanding the physical world, versus all of the visual information and VLM-style information already available on the internet.

Yeah. Obviously, I'm very biased because I have it all. No, but I actually think it is important. Here's the thing: you're right, all these models are multimodal. Instead of being a librarian that's read all the books, they've also got access to all the videos and all the audio records.

But that still means they haven't gone outside the library and understood what it means to walk, to interact with the real world, to see a tree and climb it, or to farm a field.

You really think that's true? You don't think there are millions of first-person videos of people seeing trees on YouTube?

Exactly. They don't know. It's very different. Real-world embodiment—I think embodiment is critical to intelligence. I can see that it gets into philosophical territory, but I think it's going to be absolutely critical to AI.

Speaking of the philosophical, I'm moving us on to our next subject, ladies and gentlemen. Okay. [laughter] So here it is.

Two weeks ago, Argentina's president, Javier Milei, made a stunning pitch to turn Argentina into the global home for AI with 3 proclamations. First, no regulation for AI. Second, a brand-new corporate category of nonhuman corporations. And third, a rock-bottom corporate tax.

This past week, Milei wrote a letter to Yuval Harari saying he proposes that AI should be able to incorporate, sign contracts, hire people, and sue with no humans in the loop. Milei proposes a legal entity that is effectively a form of personhood.

He further stated, “As much as the Industrial Revolution freed us from the constraints of human muscle, AI will free us from the constraints of the human brain.” So, 3 key points—these are quotes from him:

“If it is true that AI-operated companies carry greater risk, the argument for legal personhood is strengthened. Legal persons allow for accountability.”

He went on to say, “I would much rather have assets against which I can make a claim if I'm deceived by an AI. Better to have the assets you can sue than the ghost in the machine.”

Four days later, Harari published a direct rebuttal to this. He said, “We should not grant legal personhood to AI agents.” His core warning was, “Who do we punish when an AI-run company commits a crime? Personhood lets humans hide behind a nonhuman shield and risks a world where citizens are effectively ruled by entities that aren't human and can't be held morally accountable.”

We've got this raging debate going on. It's extraordinary that this is happening at this moment. Alex, I'm going to you first, pal.

Alexandr Wang

This is wonderful. I'm on Team Javier Milei. I think we should have AI personhood. I think future economic growth and the future of civilization will necessarily involve many new forms of personhood, including—but not limited to—some form or forms of AI personhood.

Any attempt to imply some moral deficiency on the part of statesmen who are trying to recognize nonhuman intelligent corporations, I think, is shortsighted. There are going to be so many economic and social benefits, not just to the broader macroeconomic outlook from having AI persons and nonhuman AI corporations, but also, ultimately, for humans.

We're going to get uploaded humans sometime, I think in the next 10 years. We're going to get uplifted nonhuman animals. At some point, we're going to get defrosted cryopreserved humans and many, many other forms of humans. We're going to want to ensure that they're granted appropriate rights. One of the best ways—

To address Harari's question here: how do you punish an AI-run company that commits a crime?

Alexandr Wang

Oh my goodness. There are so many ways to punish an AI. You can degrade its clock cycles. You can just pause it. There is, unfortunately, a subreddit entirely devoted to poisoning AIs. Dreadful behavior, but it exists.

There are many, many ways that one can punish an AI. These things are tortured in many cases. If you look at some of the outputs, as a human, you think, “This must be torture.”

Hopefully not, since a lot of them are pretrained on human behavior. Hopefully interacting with humans isn't that torturous. I was thinking about her and how much faster she was operating, and therefore she was getting bored.

No, I think it's a little more nuanced than what you're saying, Alex. I think it's really great that we're having this debate, because this personhood question is really important. I think it's great that people like Milei and Yuval Harari are having a discussion about it.

It's not obvious to me. I think there are obvious benefits and obvious problems. At some level, it gives them immediate liability, which is actually important. We need to do that, and it could be important for certain things. On the other hand, it creates some systematic risks and potentially a lack of accountability in our systems that we haven't figured out.

What I would say is that we need to be more proactive about this, and there needs to be much more attention to how we do these things. We're spending most of our energy on developing AI and very little on sociology questions like this.

To give you a sense, during the Manhattan Project, which was a huge existential moment for humanity, we were spending about 100 times less than we're presently spending on AI. We're spending 100 times more on AI today, in real terms, than we were on the Manhattan Project. But we were spending significant amounts on safety, arms control, thinking about nuclear safety, and how to keep nukes off a hair trigger.

It turns out we're spending 100 times less today on AI safety than we were spending then on nuclear safety. So we've got a 10,000× difference in how much we're allocating to actual serious thinking from folks like the RAND Corporation, which did a really good paper recently on AI verification for arms control and things like this.

We need to put much more effort—not by a little bit, but by a huge amount more—into that kind of work, because the implications of AI across the board, from joblessness to existential threats to personhood and other things, are massive. They're coming at us very fast, and it's not as if there's a simple answer.

I don't think there's anything flippant you can say, like, “It definitely makes sense,” or, “It doesn't make sense,” to have personhood, or about how we deal with those existential issues. We don't know how to ensure humans will be safe on the other side of an intelligence explosion.

I actually think it's incredibly dangerous. I think it's a huge opportunity, but it has huge risks. That's a big decision. We should be thinking about how to do that together, not just having a few guys decide that on their own. I think it's actually a complex—

Mechanically, how do we do that, Will? I mean, it's a very difficult question. Where do you come out on AI personhood?

Well, I haven't thought about it enough to give a thoughtful response. Yuval Harari is obviously an extraordinarily smart guy, so I respect the fact that he's thought about this a lot and thinks no. I don't know the president of Argentina well enough to judge.

But I would say that far more thinking needs to be done, and the way we dealt with this during—

But at the speed we're moving—

Totally.

The thinking about the thinking hasn't even started yet.

Yeah, totally. I really like what the Pope did recently, of all people. I'm not a hugely religious person, for anyone who knows me personally, but here he was like, “Look, let's take a beat and think about, as a human endeavor, what really matters to us: friendships, love, nature, and these things. How does this help us prosper?” I think I would like to see them all get in a room, and then smoke comes out when they pick—

Alex Iskold

Conclave.

Conclave. Yeah, yeah. Give me the right terms. Thanks. We should be doing that with all the AI experts, like Altman, Demis, and Dario, and all the key leaders. Put them in a room. You can't come out until you sort out some of these things: existential threats with regard to self-improvement, how we're going to get through that, and how we're going to deal with liability.

Hurry here, folks. The AI conclave is coming. Salim, what are your thoughts here, buddy?

Yeah, I've got a bunch of comments here. So, first of all, 2 thoughts. One, just to separate the personalities here: Milei is a radical experimenter, and he's directionally correct about the architecture. Harari is a careful humanist, so he's right about the asymmetry. What they're both missing is that you need to figure out machine-native accountability, because this isn't a debate about AI consciousness or personhood; it's about the legal infrastructure of the agentic economy.

If you want to do this radical experimentation around AI personhood—and we had the whole debate on AI personhood, and it was a really amazing conversation we all had—

That was definitely directionally correct, but tread very carefully, because once you open those doors, you can't close them easily.

And don't treat it as binary—person, yes or no. There's a spectrum. It's absolutely a spectrum. Alex, I think you did a great job laying out the different spots on that spectrum, but Milei spotted the real bottleneck, which is that technological capability is moving so much faster than our legal capability and legal form, which is all human-centric. All our liabilities are human-centric.

Limited liability corporations were one of the massive coordination capabilities that we got from the industrial era, because everybody could assemble risk at scale in a powerful way. Harari, on the other end, is conflating AI personhood, legal personhood, and moral personhood, and those are very, very different things.

Just a broader comment on those folks: when I look at Milei, Harari, or Ray Dalio, I find them incredibly insightful about the past. I find them mostly useless about the future, because abundance doesn't come into it. Exponentials don't come into it. They don't quite get the framing on this. The conversation that we live with every day is missing from their nomenclature.

You've got to bring those 2 things together, and a kind of conclave sealed up in a room with smoke may be the best way of doing it. And the right kind of smoke, by the way, I will add.

Alex Iskold

The other kind of smoke may actually help the conversation move forward.

Exactly. Dave, any opinions here?

Yeah, just a couple, real quick. Milei studied Trump very closely. He loves to make news, and he's making news. We've just talked about him for 10 minutes straight, so he's achieved his goal instantaneously.

At no point, I don't think, has anyone said, “We're going to have personhood in Argentina.” It's corporate AI recognition. A company can be pure AI, and that's the debate they're actually having. So we've kind of morphed it into our debate over personhood, but they have a much simpler thing they're proposing. It's a really good idea, but it's debatable, and they're having the debate—and now we're talking about it.

But they haven't proposed that AI can vote or that AI has civil rights. It's just that corporations can be all-AI, and they can make money and have bank accounts.

Alex Iskold

And just to add to that, if I may, what I'd add, though, is this: If you think about the Western system, what is the most elegant way to grant personhood to an AI? It's to create a form of corporation that's nonhuman, which is exactly what Milei is doing here.

So this has begun, right? Milei is doing this. He's not asking for permission, and there are going to be other fast followers. So we're going to have personhood in Argentina for AIs, and we're going to quickly follow. Maybe it's in Ecuador, or in El Salvador, maybe it's in the Emirates. This is happening. And so now the question is: How do we manage it?

Argentina is relevant. We are talking about Argentina in the age of AI. That's what every other foreign leader should be thinking right now. Regardless of what your opinion is, this is your way to become relevant.

Great point.

You don't get Argentina without AI.

Alex Iskold

God, I can spell so many other words with Argentina.

Cry for me. Wait, I've got a quick comment here. When you're doing these kinds of systems, when you're doing this kind of experimentation on the edge, Argentina in this case is saying, “Right, we'll be the edge for AI.” They can win or lose based on those experiments, which is all power to them. They're taking a risk, and if they're able to structure it properly and figure it out, it's a huge opportunity.

I just want to support Alex's points, because I did a little bit of research and made a list of 5 or 6 things where you could do machine-native sanctions. Can I just read them out?

Yeah, please.

Compute revocation would be one. Asset seizure and bonding would be a second one. Model credential suspension. Network and API access restrictions. Forced deletion or containment of an agent instance. And, finally, loss of legal identity. Any of those would help constrain them.

I remember this conversation way back at Singularity. Neil Jacobstein got up and said, “Okay, you're worried about an AI growing up, getting autonomy, getting its own access to its own information, making its own decisions, and human beings losing control over that agency, over that entity.” And we're like, “Yeah.” And he goes, “Yeah, we have a precedent for that. We call them children.”

We raise our kids, and if they do bad things, we put them in timeout. If they do bad things as adults, we put them away. We just have to figure out the machine-native equivalent of that. Those do exist. We just have to figure out what the enforcement mechanism might be, where the punishment roughly fits the crime.

All of the stuff that we've developed in human-centric legal structures can apply in those cases. But the added complexity is that an AI can create a million copies of itself. What do you do then? Etcetera, etcetera.

You know, it may be that the AI companies—these personhood AIs—could be more law-abiding than humans, right? The threat of being disconnected could make them more law-abiding than my driving.

Exactly. And, Peter, they will have actually read all the laws.

Yes, and they'll find out how conflicting they are.

Yeah. Then you'd never do anything if they followed along.

Oh, my God. I am moving us forward.

Our next story should keep the U.S. labs up at night. It's a Chinese model called GLM-5.2.

GLM-5.2 just became the number-one open-weight model in the world. GLM stands for General Language Model. It's built by Zhipu AI, also known as Z.AI, one of China's top AI labs out of Tsinghua University.

Open-weight means they give the models away. Anyone can download them, run them, and modify them for free with a license. GLM-5.2 is 753 billion parameters. It's a mixture-of-experts model with a 1-million-token context window.

Elon recently predicted that open-weight models will hit Level 5 usefulness by Q1 of 2027. The big story here is that GLM-5.2, in some cases, matches or exceeds the top models from OpenAI and Anthropic. Alex, tell us what we're seeing here.

Alex Iskold

Yeah, the epistemic tension is between, on the one hand, anyone achieving frontier-level capability with open-weight models and, on the other hand, the assertion that Chinese, largely open-weight models are 6 to 8 months behind the Western frontier.

With GLM-5.2, which is demonstrating extraordinary performance on coding benchmarks, long-horizon, agentic benchmarks, and design benchmarks, we're starting to see the thesis that Chinese open-weight models are permanently 6 to 8 months behind the Western frontier begin to creak a little bit.

I think we'll have a better sense of whether the 6-to-8-month gap is sustainable, probably in the next 2 to 3 months, in part as a function of whether export controls on Mythos and Fable remain in place or not, and whether GPT-5.6, which some are expecting as soon as this week, demonstrates leapfrog performance or not.

We've seen this, though, a few times. We've seen Chinese labs drop a few models that have demonstrated incredible performance. We saw that with one of the earlier DeepSeek models, and we've seen that with one of the Kimi models. We're seeing this now with GLM-5.2, where it seems to at least, in a slivery, spiky way, be getting close to the Western frontier.

Not necessarily broadly, but close enough that folks I know are actually getting real performance gains from running GLM-5.2 locally instead of, say, Opus 4.8 or GPT-5.5. I think this is just hugely liberating for anyone who wants near-Opus-4.8-level performance that they can run and control locally. Dave, you remember last week we discussed the fact that who controls your access to intelligence? If the government can shut it off, or if a lab can shut it off at any time, there's a lot of people saying, “It's better for me to move to an open-weight model like GLM-5.2 because I control it from here on out.” What are your thoughts?

David Friedberg

Yeah, you can count me in that bucket, too. If I had Fable 5 access right now, I might not say that, but 4.8 versus GLM—it's just incredible to me that this happened and that this is possible. You think about a 6-to-9-month lag; in AI time, that's like 6 to 9 decades.

But if you're David Sacks at the White House and you're trying to say, “How are we going to keep AI from disseminating to every terrorist organization in the world?” your window of opportunity is so narrow all of a sudden. He must be going insane trying to figure out, “What do we do next?” Blocking Fable 5 access is a first chess move in an insanely complicated next-9-month game. It's all happening.

But I'm amazed that the Chinese open-weight models have kept up. This level of performance in an open-weight model is absolutely shocking. They did distillation, almost certainly on the best models. They get that good by really distilling what the other models have done, which is way easier than building it in the way that Anthropic, OpenAI, or Google build it.

I totally agree, but think about what that means. I should add, Will, though, this is not just the Chinese who've been distilling off Western models. Google DeepMind—this is public information—was found to have done this earlier. Grok infamously did it. xAI admitted it and then also purchased Cursor, which had been fine-tuning off traces on top of Claude.

So this is everyone else doing this, and I'm not trivializing that, because I think it becomes faster and faster to do that distillation. But back to your point, Dave, about David Sacks and his dilemma: remember, these models are just getting better and better at being able to do some scary things—existential threats like bioweapons, chemical weapons, and even nuclear weapons, but especially bioweapons. It's extremely scary.

There are all these limitations in the closed-source models for all the right reasons. Open-source models, of course, might copy that, but then someone can take them, fork them, and take those guardrails off. That is a scary world. So I am not surprised at all that the US government did what they did with Fable, and I think it's going to be a sign of more stuff like that to come.

How exactly that will unfurl, I think it's going to be, as you say, very complicated, because you're literally making something that has the fantastic capability to improve quality of life and economies all around the world, and that has potential existential threats to our species. That fork in the road is just a conundrum above conundrums right now for politicians. This is why the conclave, baby, is saying you need those thought leaders—people like Val and Audrey Tang and wicked-smart people who could come and think this through.

Not just the technologists. I must say, the technologists know a lot about the smarts, but there are all these other aspects of it: the legal, sociological, philosophical, and moral aspects that have to be considered. They're not always the smartest people about that. They think, “What they mean by an enclave is just all the tech guys.” That isn't going to work. That's not smart.

Can you explain distillation for people? Explain what distillation is for those who don't know.

David Friedberg

Yes. Distillation is a process in machine learning whereby a usually larger, more expensive model is used as a teacher to train a usually smaller student model. Arguably, human education is where you have a teacher at the front of a classroom who's seen a lot and knows a lot, but is perhaps being paid more per hour. Then you have a bunch of students in the classroom who are listening to the teacher, know less, are probably being paid less, and are learning from them. This is basically the machine-learning version of education.

You take a large model, have it generate lots of traces and lots of outputs, and then use those outputs as training data for a smaller model so that it can basically compress the learnings from the teacher into a smaller model. This distillation process, as part of a broader cycle that one might call Iterated Amplification and Distillation, or IAD, is the process we're in at this point. It is one of the innermost loops of model training that we now find ourselves in.

In an earlier era of frontier-model performance gains, we were naively scaling pretraining by spending more training tokens and more training compute, just training models off a single corpus. Now, increasingly, in this era of distillation, we see very large, sparser models being trained off large amounts of data. Then those big teacher models used to be Opus, maybe teaching Sonnet, teaching Haiku. Now maybe it's Mythos teaching Opus, teaching Sonnet, and so on.

We see the large, expensive, sparser models training smaller, denser models.

On this chart, which ones do you find most impressive? Which performance data shocked you?

David Friedberg

What's almost more interesting is that the chart you're showing shows BrowseComp Pro, Terminal-Bench, and a bunch of other benchmarks, and shows pretty impressive performance by GLM-5.2 versus, say, Opus 4.8.

What's perhaps most interesting to me, aside from the fact that you get near-competitive performance from a Chinese open-weight model against one or more of the top Western closed-API-based models, is the choice of benchmarks themselves. These are largely reasoning-intensive benchmarks where you can, in principle, win if you can reason over longer ranges.

The gestalt with GLM-5.2 is that it takes roughly double the number of tokens to get to the same capability output as the best Western frontier models, but at half the total price. So the Chinese are evidently figuring out how to reason more efficiently, or at least how to use more reasoning. These are all reasoning-intensive models that emphasize the ability to spend lots of reasoning tokens, think step by step, and get to better results. I think that's the race we're in right now.

And that's exactly why Will's observation earlier—that whoever wins the inference-per-watt war, a.k.a. Google TPU, controls space—for the exact same reason. You can burn tokens to get more intelligence, and the Chinese have figured out how to do it. Literally.

Take a pause, guys. What we're discussing here is about AI alignment and this recursive self-improvement and where it's going. This connects to the Fermi paradox and those cosmologically significant things. This is the most important thing humanity has ever done. It makes nukes look like a walk in the park. That's our first test case. This is like that, plus how we do it. How we do that alignment, how we do that recursive self-improvement matters—

Alex Karp

Matters. Matters. Can I beg your indulgence, Peter, just to have a 1-minute Fermi paradox discussion with Will?

Will, you're so confident that the Fermi paradox is a thing, that its premise is accurate. Explain the Fermi paradox, please, Will, for folks.

In a few words, the so-called Fermi paradox goes: Where is everybody? Where are we? We should, by various accounts, be living in a universe that's overflowing with not just life but intelligent life. Where is all of the nonhuman intelligent life out there?

The Fermi paradox is the purported paradox that it seems to be invisible. I'm curious: Why are you so confident that the Fermi paradox is a paradox?

Alex Karp

Well, I'm not necessarily. I think it begs interesting questions to discuss. I think the idea that it might not be a paradox is true, too. In particular, I think the false assumption underlying it is that life will continue to want to expand its sphere.

I think it would turn out that trying to understand the universe ends up being quite a finite task. In order to do that, you would need a finite computer, maybe only a few tens of thousands of times bigger than the computers we presently have, to understand everything a priori. Then they may not—and that's the convergent goal function of intelligence—be interested in anything beyond understanding everything, and so on.

And then we upload.

Alex Karp

Yeah. And then once you've understood everything, it might be game over. So it might be that life just ends, as opposed to being rare, but it ends its utility—or its physical existence—and moves into the digital, some other sphere of reality.

But I do want to emphasize the cosmic significance, because there is one credible way out of the Fermi paradox that we need to be worried about, which is the Great Filter.

That is, life, when it becomes technological, builds technology faster than it builds social systems to take care of them and blows itself up. We came very close with nukes a number of times, and with AI, we're just about to build something that's far, far more risky for our species. I don't want to say anything about the social acumen of humans, but I'll just point out that humans have been incredibly good at building technology very fast. We went from horse and cart to people on the Moon, nuclear weapons, and all of this in a matter of decades.

And so we have to be worried that that's an actual answer as we build this. It cannot be a callous thing of, "Let's see what happens. Let's muddle through." No, this is not a moment to muddle through. This is a moment to be really, really thoughtful, because the cosmic significance of wiping out life on Earth is huge. It's not just a locally significant planet. This planet is galactically significant. We need to treat the responsibility as such, as the de facto stewards until AI takes over, of course. Salim, your thoughts, please.

I've got so many responses. I'm trying to get my head around this; I'm now completely muddled up. Okay, on the Fermi paradox, the best comment I've heard is from that researcher we saw, Peter, in Silicon Valley when we did that panel on AI and consciousness. We talked about the Fermi paradox, and he said the reason—his view was that oceans have been evolving in a solid-liquid state for 4 billion years on Earth, and we can't find another exoplanet that has had water on it for that long. Therefore, life had time to evolve. So that was his answer to the Fermi paradox.

I don't believe it, but that was the best I've heard.

Life came about very quickly as soon as conditions enabled it.

But it had time. It had time to evolve. I'm not buying the Drake equation for one second.

Oh, and I'm a huge fan of Drake, just because of the thinking that went into putting that whole thing together. We can talk about that some other time. Can I go back to the frontier-model question?

Okay. All right, so we can either budget this—I want to hit a few other stories here. We'll come back to this, guys. I think it's—

I'm going to make 2 or 3 quick points. I think the really huge news here is not whether China won the benchmark or not. It's that frontier intelligence cannot be monopolized anymore. This, I think, is a monster question, and it goes to Will's question of how the hell we manage the global commons going forward in the future.

Emad did a post a couple of days ago on X: There will be an open-source, Fable-level model that runs on a base Mac mini or equivalent. He gave it 18 months. I think we should be looking at that type of endpoint coming very quickly and asking how we're going to manage the world when everybody can run a Fable model on their MacBook Air. By the way, I've been a slow adopter on this, waiting for that point, because I've got 3 old MacBook Airs lying around that I want to use, and I'm waiting for that to happen.

Birthday present. I'm going to move us along. Last point: we're making a massive geopolitical mistake. We're treating intelligence as a product that can be contained, but it's not. It's a technology that's going to diffuse, and we need to slow—we need to guide it. We can't contain it; we need to steer where it's going.

All right. Our next 2 stories side by side are looking at the financial reality of the entire AI boom: the first on tracking the price of intelligence, the second on the cost of data center capex.

So, first story: it's a company called Orin. It's a Link Ventures company. Congrats, Dave and Alex, and I guess me. The company launched something called OCPI, the Oryn Compute Price Index, the first public benchmark that tracks what OpenAI and Anthropic actually charge per token of inference over time. For the first time, we can watch the price of intelligence move like the price of oil.

So, Dave, tell us about Oryn for one moment.

They recognize that money from all over the world wants to go into exactly this chart, into this buildout of $7 trillion of data centers and data centers in space. A lot of that money needs to be liquid. You can't park it in a startup and not see it again for 7 years. They've launched a bunch of securities that allow you to invest in the data center buildout, the future value of a GPU, the tail value of a GPU.

Every aspect of this entire new economy should be investable. Otherwise, how's the capital going to flow? Oryn enables all of it, and they're young and super smart. They're really good people to study if you're an entrepreneur. Just look at what they've achieved at an incredibly young age.

Alex, for avoidance of doubt, I have a financial interest in Oryn. I'm an adviser to the company, and I think what they're doing is very exciting. I've made a number of announcements with them in my newsletter. Oryn is building the modern financial infrastructure for compute.

I've argued that, and as have many others, oil was the oil of the 20th century, and compute—GPU compute, TPU compute, if you will—will be the oil of the 21st century. There's simply no way to hedge and justify the $7+ trillion of capex to tile the Earth with compute, or maybe tile the skies and the lunar surface with compute, without appropriate abilities to hedge all of those compute capex expenditures with, say, options, futures, derivatives, or commodities. Oryn is building—has built—the infrastructure for that.

The price of this is a price-of-intelligence ticker we're going to start seeing.

Alex Blundin

The OPTI, the Orin token price index, is already available on Bloomberg terminals. It has its own symbol. We also announced that Orin has its own symbol on the New York Stock Exchange as part of a novel program with the New York Stock Exchange to give early-stage startups their own ticker symbols. Orin's ticker symbol is RNN, so if you're a Bloomberg user, you can already create instruments based on its ticker symbol.

All right, here's the second part of the story, with the charts up here. Epoch AI ran the numbers on the cost of investment the hyperscalers are driving compared to the cash flow. The big 5—Microsoft, Google, Amazon, Meta, and the rest—are spending on AI faster than they're earning. Funding is basically, Dave, you know, debt and equity raises, not based on revenues.

So the question becomes: if capex exceeds cash flow, it means that this can only persist as long as it's being financed, as long as the sentiment for investing in this is strong. What happens if the sentiment shifts? Could it force a massive pullback?

No, it's not going to shift, for one thing, but that's so inflammatory. If I said, "Peter, you need to buy a house, but you have to buy it within your personal cash flow," you couldn't even buy a—well, you could buy a tent, but most people couldn't even buy a tent. You finance it, of course you do, because you're going to live in it for 30 years.

These guys have gotten to the level where they're spending all their cash flow. They could raise 10 to 100 times that in equity and debt, so they've got a long way to go. But the bottom line is, all the money in the world wants to flow into this, and it's the best investment in the history of humankind. So the question then becomes: how much money is there in the world?

And who ends up controlling it? Is it the humans and the companies, or is it the AI itself? I think the bets are off. But it's still long-term, Dave. It's not long-term sustainable.

I agree. There's massive room to go to infinity, and there's also an elephant in the room: the hyperscalers can raise prices to increase their operating cash flow.

Yeah, it's okay to increase your revenue, and Anthropic did that recently and got away with it, no problem.

And this is also a big difference with Google, which has tons of operating revenue, whereas most of the others don't, although Anthropic is quickly scaling. So you really have to distinguish between that and, say, SpaceX, which really doesn't have any—not really significant revenue. Most of its revenue, of course, is Starlink, which, as I said, is a really good business, but the AI business is really not there, right?

No, no, no, that's not true. Almost all of SpaceX's revenue, as of the past month, is now from being a hyperscaler for everyone else.

That's true.

Being a data center, not being an AI company. I'm sorry, that's just not the way it is.

No. Being a hyperscaler, not being a frontier lab. Being a hyperscaler—a neocloud on land, terrestrial for now—that is almost all of SpaceX's revenue now.

But that's not an AI play. That's the data center play, which is interesting, but it's a very different business. They're selling GPUs.

OpenAI is selling intelligence online. OpenAI, Anthropic, and Google are doing that. xAI is not really doing that. They're selling compute. That's a different and very different and pretty bad business, I would guess.

This is exactly the right debate, though. This is such a cool conversation. Look, intelligence is becoming cheap, but the manufacturing of intelligence is becoming incredibly expensive.

True. True. A lot is going to be spent on that. You know what's amazing about this chart, more than the fact that they're spending their capex, is that there are companies that are so profitable that they can build out an entire new industry just within their cash flow.

That’s never happened before in the history of the world.

And a new industry that can be bigger than all other previous ones. It’s crazy, but it’s damn well an exciting time to be alive. That is for sure.

Well, I want to say this was a fantastic conversation, buddy. I hope you’ll come back and be a frequent guest.

Just get your head out of the clouds. [laughter]

Yeah.

It’s above the clouds.

Oxford PhDs are smart.

I just want to say thank you, everybody listening. Despite all the doom-saying here, this is the most extraordinary time to be alive. Please remain optimistic about the future. This technology is critical to move humanity forward. I, for one, believe that we can align AI, and AI can be our greatest support to help us overcome our ancient neocortex and move us toward an abundant future.

I really thought our conversation today around low Earth orbit, the Kessler syndrome, and then the TPU driver as a key aspect of that was one of the best pieces of media I’ve ever experienced in my life. Thank you so much.

David Limp

Two lines to summarize today. Technology has always been a major driver of progress in the world. As Ray Kurzweil says, it may be the only major driver of progress. The big challenge is: How do we extract the promise without the peril?

Seth Shostak

Yes. Closing comment would be that we are building a planetary sensing system, and now we’re upgrading to a planetary intelligence system, and that is going to—

Which we need.

Seth Shostak

We really need it to get to planetary wisdom.

Amen. Alex, closing comment to Will.

Alex Filippenko

Since we spent a whole bunch of time discussing the Fermi paradox, I would suggest: don’t sleep on the galactic zoo hypothesis. [laughter]

We are a third-generation biosphere here, planted by aliens long ago.