Humanity's First Star Probe, Architect Labs Beats NVIDIA 3.4x, Musk Wants Satellites to Cool Earth | EP #285

The first interstellar mission to Alpha Centauri. Philip and Matt are here to join the Moonshot Mates.

Are we the first kid on the block, or are we now, you know, auditioning for membership in a cosmic club, and the ticket to entry is, can you handle that?

By the time we arrive, you know, for 15,000 years, there will already have been a colony at Alpha Centauri.

A Palo Alto startup called Architect Labs just announced the world's first fully AI-designed chip called Redwood. Zero bugs on first silicon and 3.4 Is this an incredible threat to Nvidia?

Yeah, absolutely.

I think I know how this game ends. It ends with—

Extremely severe extinction events happen every 100 million years or so, and just switching to sustainable energy will not be enough to stop them. His solution: satellites in space that control temperature, and massive geoengineering will be needed before it's game over.

I completely buy that with global energy use increasing, we will have to use AI weather models and enough points of actuation, that's a recipe for global weather engineering. And I think we're going to get it.

Now that's a moonshot, ladies and gentlemen.

Welcome to Moonshots, everyone, your number one podcast on all things AI and exponential, your front row seat to the accelerating singularity. I'm here with my moonshot mates. We're gonna have the Fantastic Four. We've got the tremendous three right now. Dave Blunden, Alex Wiesner-Gross, myself, Peter Diamandis. Of course, where Salim— he's held up in TSA. A funny joke if you watched our last pod. He'll be joining us shortly.

Do we even know what continent?

I have no idea where Salim is.

Okay. It'll be fun to find out.

He's a probability function, but he'll be joining us in a little bit. So our mission on this podcast is simple: future-proof you for the coming supersonic tsunami and keep you optimistic About this extraordinary decade ahead. You know, we're lucky enough to be living during the greatest period of transformation in human history, and we want you to understand it and not fear it. So every week, twice a week on this show, we work to show you the future before it arrives. If you're not a subscriber yet, as always, please hit the subscribe button. We have a lot happening, and let me just mention to you, we want to invite all of you. To an AMA we're doing. So we're inviting our Moonshot guests, all of you, our Moonshotters, to join us on Zoom. If you go to moonshots.com/ama, you can register. We're going to be doing this twice, once in the morning to get everybody in Europe and Asia and India, and once in the evening for those of you in the US. And during this AMA, we're going to have a chance to, you know, You know, go back and forth. We don't get a chance to answer all your questions, and we do read them. This is a chance to interact with you, have your questions answered, and ask you some questions about what you want covered on this show. We get tremendous joy specifically from being able to, like, help you understand where things are going. Dave and Alex, you guys are gonna plug in hard and heavy for that one, right?

Better believe it. We'll do our best. I mean, What do you do in the morning? Just stretch and get ready?

I don't know.

Read a few journals?

I don't know.

We'll do our best.

Peter, can I preemptively choose question number 3?

Of course.

And these are coming in in real time, right?

Completely. It's a live conversation.

All right.

You know, we're going to limit the number of people. We've got about 200 spots left. So please register now at moonshots.com/ama. And we invite you to join us on X. Our handle on X is @MoonshotsUpdate. We're putting up our recordings there. We're putting up clips. So join us on X for all the latest news. Okay. I want to share some subscriber love. I know I read all the comments and I hope you guys do too. And it's pretty, pretty epic. It's going to read a few of them here. I love this one. There's no way to digest the amount of new advancements without this podcast. I agree with you. It's the only way I keep up with everything. Another subscriber, Rock Pedro, said, whenever one of these gets posted, I hit pause on the rest of my life and sit down with a cup of coffee and my iPad. So thank you for that. I love this one.

Peter, this is framing us, I guess, less as CCTV for the Singularity, more as Reader's Digest.

Okay, well, I just think it's action-adventure, honestly. Stop, drop, and roll said this pod is hitchhiker's guide to the singularity. Brilliant episode. I appreciate that. Two more real quick. Cam abroad said I can't even sit down for a two-hour movie any longer. However, I watch this podcast end to end every time.

All right.

Yeah. And Richard Consol said you know highlight of my week. Please don't ever stop. I don't think we are.

This is great.

Is the claim that Is the claim that we're restoring the audience attention span?

Yeah, it's— we keep their attention. What's that, Dave?

Oh, you got to call out to the team behind the scenes too. The rate of story digestion they have to get through has gone up, you know, since we started doing the pod, maybe 10x higher now.

Yeah.

And so they've staffed up, but I mean, just the workload that they get through to whittle it down to the subset, because I have, you know, that same feeling that that comment, that second comment, like, I could never keep up with everything going on without this podcast too. And You know, when I come into the scripts to read it, it's all been beautifully synthesized. And so I can study up in like an hour and a half.

Dave, a lot of that team is me. I probably spend 10 hours going through 200 to 300 stories that Alex puts into our chat. It's insane.

Well, and also among people who do that kind of work, you're the only one that I know of that has a PhD in biotech and MIT degrees in biology and aeronautics and a deep AI and computer science background. Like, not too many people could filter it down to the really relevant subset the way you do. So thank you.

I mean, I love our chemistry and I just wanna do a shout out to our producers, Nick and Dana, and to Gian and Aiden for their support on this. You know, our moonshot here on this pod is to 100x our growth and to get to 10 million subscribers. You know, every single one of you sharing this podcast gets us closer. So please share it. please subscribe. Today we've got 10 amazing stories, everything from OpenAI projecting that it'll reach AGI in the next 4 months to AIs designing their own bespoke chips, nuclear rockets to Mars. You know, as always, there's a single through line that the singularity is here and it's accelerating. So buckle up. This is another amazing week. Now, we normally close the program with an outro video, but today we had an incredible intro Delivered by Stephen Renegade, and it's so good. I want to play it to kick us off, really get us in the mood here. All right, listen up. This is Moonshots podcast intro by Stephen Renegade.

Ignition.

Ignition. Welcome back to Moonshots. AI sparks ignite a cosmic road. The systems are improving themselves.

Robots evolving, tech breaking molds and goals.

Longevity calling.

Turning back time.

Abundance waves.

Multi-dimensional brains.

Moonshot mates decoding the future of change. Wasn't that great? I love it. Thank you, Steve. All right, now before we get into the news from this week, we have a special announcement, a breaking story on Moonshots. from our very own ASI, Alex Wiesner-Gross. Alex, if you, if you jump in here, please, and introduce our 2 guests and tee them up for the first Interstellar exclusive, the Fermi Explorer mission.

Amazing, Peter. So folks who watch the pod faithfully may remember Philip Johnston, who was on the pod previously, founder and CEO of StarCloud, the orbital data center company, and making his Moonshots debut today, co-founder of mine with Physical Superintelligence. Matt Pines joining us. And Philip and Matt are here to join the Moonshot Mates in their, I think, exclusive pod announcement— how cheesy does that sound— of the first interstellar mission to Alpha Centauri. So maybe Philip and Matt, take it away. How are we getting to Alpha Centauri?

Maybe, yeah, maybe I can describe the mission and then I'll talk a little bit about how PSI ended up being very pivotal in discovering this mission, and Matt can talk about the background behind that.

Please.

So actually, the last time we were on the podcast, after we finished recording at the end, I said, oh, by the way, guys, I'm planning to send a spacecraft to Alpha Centauri. And we've put some constraints on ourselves because we really want this thing to actually launch. So the constraints we have is we want it to get at least 99% of the way to Alpha Centauri within the next 80,000 years. And that actually minimizes for fuel. Any longer than 80,000 is more fuel. Any shorter than 80,000 is more fuel. The second constraint is we want to launch within 3 years. The third is it must have a 1-kilogram, 1U payload. And then the last is it must cost less than $15 million to design, build, and launch because we're basically funding it.

Let me say that again. $15 million?

$15 million, yeah. $15 million, which, I mean, is an astonishing target to hit.

It's a seed round for a startup out of MIT.

Yeah, yeah.

It's half a seed round.

You can't count that low. This is Interstellar on the cheap.

Yeah.

Yeah.

But 80,000 years also is a little longer than most startups.

That's true. I'll come back to why we're doing it in a minute, but I'll just touch on the story of how PSA came involved. So we'd spent 6 months trying to figure out a trajectory that would make sense where the big challenge, we wanted to do it with solar electric and gridded ion thruster. It's the same as like the Star Cloud 1 satellite. It's very cheap. The problem is the further away you get from the Sun, the lower the energy that you have hitting the solar panels. So the larger the solar panels you need. Once you get past about Jupiter, you're getting very low amounts of energy. You'd need huge solar panels. And so we tried a whole bunch of things. Jupiter flybys, slingshots towards the Sun. We had 2 guys from JPL look at it. They spent a few weeks looking at it, couldn't come up with anything. We had, yeah, I mean, basically 6 months of—

Testing.

plugging it into Claude, trying everything we could. And then Alex goes, oh, you should speak to my guys at PSI. And I was like, oh, yeah. Oh, yeah. Here we go. They're going to beat our JPL guys. So I didn't even reply. For like a week, I didn't reply. And then Matt followed up, and he was like, hey, well, send over the specs for this mission you want to do again. So I was like, oh, God. OK, I'll keep Alex happy, and I'll send the specs. A week later, they came back with the most unbelievable report. So I think they spent tens of billions of tokens on this thing. And they came up with an incredibly sort of unintuitive and innovative trajectory that makes this mass and cost budget close. So it's essentially we spiral out from Earth and then sort of unintuitively we fire a retrograde burn. So we slow ourselves down to pull ourselves in towards the sun. And we do that for about 5 years. So we do 5 retrograde burns at the furthest point from the sun. And then we start doing what they call a perihelion burn. So burning your thrusters as close as possible to the sun. And they've called this maneuver the perihelion pump maneuver. And what it does is it means it has 2 amazing advantages. One is we're firing our thrusters at the closest point to the sun. So we need less mass on the solar. But the second is it takes advantage of this orbit effect. The orbit effect is the idea that you get more energy for a given time of thrust, the faster you're going, and you're going the fastest at the perihelion. So yeah, I mean, honestly, it's to me incredibly impressive that they came up with this. And at this point, I'll hand over to Matt. He can explain how they did this.

Well, certainly serendipity. I mean, the fact that AWD is my co-founder and we had this connection right at the perfect time, and it's been a unique synchronicity announcing the Fermi Explorer mission and this partnership today, as well as announcing, you know, Physical Superintelligence's seed fundraising and coming out of stealth on the same day, because this mission is the proof of concept for what we're building here. As you mentioned, these sorts of highly technical scientific challenges that are bottlenecked by humans that have been sort of pre-trained for 22 years, post-trained in grad school or technical positions, and then they become scaffolded and then orchestrated in corporate, academic, or government bureaucracies, those are the rate limiters of kind of what our scientific and technical ambition is. And that's why we've had to have large-scale national institutions organize these sorts of breakthrough grand scientific and technical initiatives, e.g., sending a spacecraft outside the solar system. And so this is the proof that you can have 2 small teams, both startups, one in space, one in AI for physics, put their respective heads together and come up with a mission that uh, you know, pushes the boundaries of what's possible. And yeah, we, we kind of took this as a, as like a side challenge, uh, to throw at our, our internal tech. We have an astrophysicist on staff, but to be honest, we were just, uh, you know, prompting, uh, the, the system and then, you know, crafting the final product to make sure that it had the right, uh, you know, graphics and plots. Uh, but other than that, it was entirely hands-off. And we were as surprised as Philip's team that it came up with the, the optimal mission trajectory. We certainly didn't load the dice You know, almost minimal human steering involved, and it came up with a trajectory that satisfied, you know, quite strict mission constraints. And so I think this is the first of many surprises that we're gonna see from pointing AI physicists at these really valuable technical and scientific challenges.

Alex, I wanna get more into the details, but I thought it'd be fun to show the Fermi mission video.

Before you play it, do you mind if I just describe what's happening? 'Cause it can be a bit wacky to just see it out of context.

Okay.

So the idea behind the mission is we expect to be, or we hope to be, the first to leave Earth for another star, but also the last to arrive at another star. So in 1,000 years' time, let's say you have better propulsion technology, even if it's like 20% faster, which is very conservative, by the time we arrive, for 15,000 years there will already have been a colony at Alpha Centauri. And so they'll have had time to build things like Dyson spheres and O'Neill rings and all the wacky and cool things that we see from, that we imagine from sci-fi. And then from there, we estimate, with sort of basically current propulsion technology, it will take about 5 to 10 million years to settle the galaxy. And then from there, without too much effort, it would take about a billion years to get to Andromeda. And then from there, about 5 billion years to settle the local cluster of galaxies. And I'll come back to why we're showing all of this maybe after the video. But just so that people are aware, that's what you're about to see is the next 5 billion years of history.

Oh my God.

And maybe it's worth underlining, Philip, This is a conservative outer bound. I don't actually think it'll take five billion years for humanity or civilization. Very conservative video.

Nor do you think, Alex? I would imagine that getting faster propulsion is going to take a hundred years. I would imagine we'll have that in five to ten years. Correct.

Yeah.

All right. Let's watch the Fermi Explorer mission video, and then I've got a ton of questions.

That's Alpha Centauri. One day humanity will go there.

One day humanity will go there. The mission that's going to kick off the civilization of the universe. Okay, here we start. So let's get into some of the fundamentals. $15 million, you're gonna do a rideshare. How do you get to Earth escape velocity?

Yeah, so to be honest, I actually think we can do it for $10 million, but I didn't wanna put that because the PSI paper said it was $50 million. So it's actually not too dissimilar from the StarCloud-1 satellite. So it's about a 100-kilogram smallsat. We can launch it on a rideshare to any LEO orbit.

Typically a SpaceX Falcon 9?

Yeah, exactly. So about $500,000 to do that. Then from there we spiral out over a course of about a year and a half. So it's about 17 kilometers a second of delta-v. Sorry, about 7 kilometers a second of delta-v to get to a Sun orbit from that Earth orbit, which is pretty doable, you know, with regular thrusters and tanks. And then from there we do the retrograde burns and we start circling in towards the Sun. But yeah, so it's about a 100-kilogram spacecraft of which 60% is just xenon. So the wet mass of the spacecraft is 100, but most of that is xenon. And we're using off-the-shelf gridded ion Hall effect ion thrusters.

So just for reference, no laser sails like Project Starshot?

No laser sails. This thing is going to launch in 3 years and it is going to get to Alpha Centauri.

By the way, we did a podcast with Philip. Look it up. It was really, really good. But we talked about that xenon. based ion acceleration technology in that podcast. You might want to give us a quick summary of it. It's so cool.

Yeah, yeah. So it's become pretty ubiquitous now in satellite industry. It's essentially a mini particle accelerator. And they can be pretty tiny, some of these things. You know, they fit into a 1U space, so 10 centimeters by 10 centimeters by 10 centimeters. And yeah, it's crazy.

Particle accelerator in a toaster.

I mean, I think that's a Essentially exactly what it is. Yeah, it sounds wacky, but that is what it is. So it pings individual particles, individual atoms of xenon out the back at very high velocity. And you want to do that because you only have a certain amount of propellant. So if you're— if any of that propellant is leaving at less than the fastest possible speed it can leave, then you're not going to have as much thrust as you would otherwise, basically.

So this is a 25 trillion mile You know, roughly 4.3 light-year journey. And it's, you know, we've seen already Voyager 1, Voyager 2, Pioneer 10, 11, and New Horizons all leaving our solar system. But the significance here, Alex, is this is the first one actually aimed at a specific star. Is that correct?

Correct. That's correct. And there's also a reason behind the name Fermi. It's an allusion to the so-called Fermi Paradox.

Let's get into that.

Yeah. Yeah. So Enrico Fermi, after World War II, purportedly asked the question, why— where is everyone? Where in— out of an abundance of evidence that our universe seems to be fundamentally friendly towards life, friendly towards intelligent life, where are all of these other forms of non-human intelligence in our galaxy? And I think Philip and Matt and I have discussed this a number of times. I think there are 3 main possible resolutions, to the extent the Fermi Paradox, so-called, is a paradox at all. There are, I think, 3 possible likeliest resolutions. One, which I do not think is likeliest, is that we're the first, that maybe humanity is just the first on the cosmic scene, in which case we have an obligation, arguably, to start sending out probes, as is the case with Fermi Explorer, which will be— it's set to be humanity's first interstellar probe and start developing our galaxy. The second possibility is that there's a great filter, that we're being filtered, and that there's some reason, maybe something having to do with technological development or some latent risk of just living in this universe that—

Some version of the Prime Directive, so to speak.

Some version of the Prime Directive, perhaps. Well, actually, not the Prime Directive. Great filter, I think, is distinguishable from the Prime Directive. Some reason the universe Snuffs out civilizations past a certain point. And if that's the case, if there's some risk, not quite 3-body problem, but some maybe latent physical risk in our universe to survival or development beyond a certain stage, then we need to start getting our materiel and our infrastructure out there so that if humanity is snuffed out before we pass whatever our next key milestone is, we avoid the Great Filter. That's reason number 2 for doing this. Reason number 3 is what you said, Peter, a Prime Directive-type scenario where we're in a galactic zoo, maybe a galactic petting zoo, and we're surrounded by non-human intelligence that has us behind a cage, behind bars. And if you're an animal in a zoo who wants to get the attention of the zookeeper, what do you do? You start throwing food out of the bars to get the attention of the zookeeper. So that's reason number 3. We're fenced in. I think of all these 3, The third likelihood, third option is the likeliest one, but I'd be curious to hear from everyone here. Do other folks have other proposed solutions or favorite solutions to the Fermi Paradox?

This episode is sponsored by Google for Startups. Think about this for a second. You now have access to the same generative AI models that cost hundreds of millions of dollars to train. Google's Startup Technical Guide for Generative Media gives you a complete blueprint for deploying Google DeepMind's models in production. Images, video, audio, all of it. Real architecture, real results. Find the link in the show notes below. Well, the one solution is life cannot survive nuclear age or the ASI age. And, you know, another option is that it's out there, we're just not hearing it. You know, I was sharing before the show an example. I was at Mount Athos, a Greek monastery, and at the end of the day, at sunset, they rang a bell to call all the monks to prayer. And just at that moment, my cell phone rang, and I realized that they were using this old ancient mechanism of communications, you know, a bell, and they were being bathed in 2.4 gigahertz frequencies, but they weren't receiving it. So the question is, is there a better means of communication? And there's lots of traffic out there. on the intergalactic internet. It's just that we're not able to perceive it yet.

Yeah, I think, you know, physics is the kernel of civilization, and civilizations are bounded by their ability to exploit and deploy that knowledge into useful technology. And the fast AI takeoff is quickly turning to the fast physics takeoff, and that's going to lead to a dramatic acceleration in humanity's ability to explore the universe and to exploit whatever degrees of freedom the universe allows us to exploit. If there are tricks that that allows us to leverage for society, well, those are tricks that others may have figured out. And so we're rapidly sort of racing into that regime, and then we'll find out, are we the first kid on the block, or are we now, you know, auditioning for membership in a cosmic club, and the ticket to entry is, can you handle that?

So I think the primary mission of this is to get people dreaming again, to start to set audacious objectives And, and go for them. I mean, just one clear point, this is a flyby of Alpha Centauri, or a flyby of approximately Alpha Centauri. How good do you think your guidance is going to be to actually get it into the planetary disk there?

So we expect to miss by quite a large margin. So the target we've set ourselves is we want to get at least 99% of the way to Alpha Centauri. So right now we're about 26,000 260,000 AU, so astrological units, the distance between here and the Sun. So we'll be within 2,600 AU, so 2,600 times the distance from here to the Sun. So it's quite far, but it will be within the Oort Cloud of— it will be detectable. Like, we're anticipating that we'll have retroreflectors and things to make it detectable. So yeah, we're anticipating—

It will be electronically dead, obviously, by then. Long—

Yes.

Since—

It's probably also worth flagging, Philip and I have a bet, Peter and Dave, regarding the commercial market for interstellar flight. Right now, this is structured as— this is a nonprofit, the Fermi Explorer mission. I bet, Philip, that given the absurdly low price tag of $10 to $15 million for sending a very long, very slow mission to Alpha Centauri, that there's probably a latent commercial market for everyone, every small government, every organization that wants to start throwing probes out into deep space. My bet is there's this latent market for commercial interstellar flight.

You know, one of the markets out there is astronauts. A friend of mine, Charlie Traefer in Houston, used to buy parts of Orbital Sciences' Pegasus missions and put up like 5 grams of someone's cremains into Earth orbit. Yeah. Anyway, you wanna shoot yourself out away from the Earth? You can do this now.

I think there's, I mean, for scientific purposes, exploring the outer solar system, every single nation-state, I would argue, can afford to send now, at this price point, which again is mind-boggling, can afford to send their own probe to another star system. And this is the first time, to my knowledge, that this has been possible. for humanity. And critically, Peter, this whole mission would not have come together without Moonshots. So in the causal history of human civilization, Moonshots was the catalyst for humanity sending its first probe to the nearest star.

I love the fact that an AI system was actually able to deduce this trajectory. So Matt, can you talk— I mean, is this unique? Has it never been seen before? And has it been validated outside of PSI?

Yeah, so it's been validated by a bunch of trajectory folks, some of which were previously at JPL and others. So to be fair, if we'd gone to a bunch of astrophysics PhDs and given them $1 billion and 5 years, I'm sure they would've come up with this trajectory. It's more that this was done in a week. Firing your thrusters at the perihelion to take advantage of the orbit effect is not new. I think what is surprising about this is we were anticipating having to get— lower the perihelion through orbital flybys, which is how it's been done in basically every other NASA mission. It's not been done by just, okay, let's just reverse our thrusters and start slowing down now immediately. Which is like the simplest and cheapest and kind of most obvious way to do it. But for some reason, it just didn't occur to any of us.

Interesting.

Yeah, I mean, this was basically in total probably maybe 5 or 6 hours of human time over the course of that week. And, you know, so do the orders of magnitude speed up compared to what you had previously had to get, you know, entire teams of NASA engineers spending potentially months. That's just a flavor of the speedup we're seeing. Again, it's a spiky frontier of, you know, where capabilities exist for pushing breakthrough scientific and technical capabilities with these sorts of systems. And, you know, we didn't know until we tried exactly how spiky that frontier was. And we found out through this amazing partnership that there's now multiple orders of magnitude speedup possible for these sorts of mission planning.

Amazing. Was it really tens of billions of tokens of work during that?

It was, I think, a total of about 10 billion total tokens, obviously depending how you count tokens, input tokens, output tokens, cache stuff. But yeah, about 10 billion total. And yeah, lots of Monte Carlo simulations that the system designed and ran. You know, you think about, you know, 3-dimensional models of the trajectory analysis. And so it wasn't just the astrometry and the mission planning associated with kind of getting the orbit trajectory right, but layering in the multivariate optimization associated with the cost and launch windows, right? So you have to dial in all those variables to get it to work.

It's interesting, we have a lot of very complex kernel writing work going on in the building and some other, you know, super high-tech work. And it's also about 10 billion tokens of thinking per about 100,000 tokens of final output. So, it seems like a lot of projects that have nothing to do with each other are settling on that kind of ratio, which is mind-boggling. If you said, what is the human effort of 10 billion tokens worth of thinking? And it's like Philip was saying, it's probably on the order of thousands of people working for 10 years or more to get— it's probably more than that, actually. And it's all compressed down to a week. So what did you use for models?

So for this, we actually used our open-source version of our tech because this is an open-source project. So folks can look up the Get Physics Done open-source package, which we actually released several months ago. You know, we're a public benefit company. Our mission is to discover and commercialize transformative new physics. We're an AI-first AI physics lab designed to push the envelope of what these systems can do for both fundamental and applied physics. And so we released that package as an open-source tool. We obviously have a version of it that we run internally, and we've grafted some of those into our core technology. But as part of this being the— this is an open-source package, right, for open science and to demonstrate just how far the bar has fallen for small teams that are leveraging, you know, the current frontier of capabilities to drive, you know, exceptional outcomes and push the envelope of what's possible.

It's worth noting and congratulating Matthew and Alex for your financing on PSC. You just raised your— what round was this?

This is our seed round. You know, we're in the, we're in the full AI era where you can have $58 million seed rounds. And so we're, we're, we're looking at Philip Johnson setting the mark and we're trying to, you know, we're trying to clear his bar. So really proud to have that announced the same day, led by Breakthrough Energy Ventures, an amazing partner. They have investments in deep frontier technologies, you know, fusion, quantum computing, breakthrough energy materials science, etc. And so we couldn't be prouder to have them as our lead and an amazing roster of other investors that have backed us. Yes, we're just beginning, coming out of stealth, and you'll be hearing a lot more from us in the coming weeks and months.

Amazing. Gentlemen, I wish you incredible success on this mission. I know a lot of kids will start dreaming. I clearly would love to be there when it lands, but 70,000 years, it's a little bit of a stretch on the timeframe.

You're all invited to the launch.

Oh, cool.

Actually, we should live podcast in whenever this is, 2029.

I see it as a challenge for PSI. We launched this with Philip Johnson and his team, and then the goal is to catch up with it, if not beat it there.

Yep, I can imagine that. You can wave at it out the window as you're heading towards Alpha Centauri. Gentlemen, Philip and Matthew, thank you so much for joining us today. Congrats on this mission. And it's really, you know, this is about getting kids to dream again about what is possible. I mean, it's shocking that nobody's— there's been a few attempts and a few studies that have been done. There's the Breakthrough Project that Yuri Milner had put forward.

Breakthrough Starshot.

Yeah, using solar sails and ground-based lasers. But that— I haven't heard about that in a while. Is it officially—

That died. That died. I know some of the folks who were involved. It died. I would argue maybe Philip and Matt I would be curious to hear your perspective. I think it died because it relied on technologies, especially propulsion technologies, that were simply too hard for the present. In particular, ultra-high-power lasers simply weren't ready yet. Whereas I think what's unique and attractive about Fermi Explorer mission is essentially no new technology. This is something that could be launched with the technology that we have today. So my expectation is it will be the first successful interstellar mission.

And we'll put a link to the mission in the show notes here so folks can go and dig down deeper. Again, Matthew and Philip, thank you for your time today.

Thanks so much.

Thanks again. Great to see you.

Thanks, guys. Yeah, I was just saying to Dave, I think, Dave, you flagged one of the more interesting points. If 10 billion tokens is sort of a reference class for problem difficulty, I was speculating at some point in the future, if you fix model capability, and of course, model capability per token is continuing to increase over time through iterated amplification and distillation, that at some point we'll look back in the spirit, Peter, of Solve Everything, and we'll say, oh, that hard math problem, oh, that was a level 9 problem, 10 to the 9 tokens. That was a level 11 problem. And we'll just have some convenient logarithmic scale to talk about all hard problems.

Yeah, I really feel like this project is so much more important than 80,000 years in the future. just in terms of the plan, the way the tokens were used to create a plan, something that no astrophysicist had thought of before. And then it can immediately go into implementation. And that's a sign of the times, right? That the thinking is going to get way ahead of the implementation in biotech and physics and in literature, in every area. You can burn the 10 billion tokens in a couple of days.

Yes.

And have an incredibly ornate outcome just waiting for implementation. It's a really good case study in how This is going to change in the next, really, couple months to massively abundant thinking, intelligence everywhere, and all these bottlenecks on physical world implementation of the ideas.

For sure.

And I talk all the time on the pod and otherwise about how the singularity can in some sense be operationalised as all sci-fi tropes happening everywhere all at once. There is a sci-fi trope for this, which is Isaac Asimov's universe, where AI was required to solve interstellar travel, at which point humanity spread to the stars. I do think that's the likely case here. AI will solve interstellar travel and humanity will spread to the stars.

All right. And on that note, I'm going to jump us into this week's breaking news. There's a lot of fun stories. So this week, the drama between Sam Altman and Elon Musk bubbled up once again. OpenAI ends its support of Cursor. They wrote Elon an email. Where they posted this and saying basically, you know, well, first of all, remember that SpaceX recently purchased Cursor for $60 billion, a coding platform that had been historically dependent on OpenAI's GPT models. And they announced, we're shutting it down. We are not going to allow Cursor to use the GPT models anymore. The question of why? Well, OpenAI said the following, quote, We are making this choice because we cannot be confident that SpaceX will use our technology within the terms of service based on our experience with Elon Musk's companies violating contracts. And of course, Elon's response to that? Well, kind of hot and heavy. I don't care. I couldn't care less. Scam Altman and Greg Stockman are utterly untrustworthy assholes who stole an open source Nonprofit. So there you have it. We've got the soap opera continuing. So within hours of that, within hours of OpenAI pulling out from Cursor, Anthropic stepped in to immediately pledge support for Claude for Cursor's needs. And the framing that got put up on X, and this was you know fun to see going back and forth. You know, quote, Sam is now fighting alone against two massive competitors, Elon and Dario. that have formed a strategic alliance. So, Dave, thoughts on this one?

Well, you know, it's not coincidental that, that GPT-Sol, which is an incredibly great model, came out immediately prior to this move. So I think if Sam had tried to do this, you know, a year ago, he would have been like, oh my God, now I'm in deep trouble. But now he's actually got an incredibly competitive platform and Codex is really good now. And so I think what's lining up here is, look, Codex from OpenAI on top of SOL running on Amazon Bedrock is a really good default corporate answer. And everybody wants to go after the corporate revenue. And so, you know, remember, Sam was very late to pivot out of consumer and into corporate, but now he's got the whole stack lined up. And so this is the next move in saying, okay, you know, here's our vertically integrated stack. There's this cursor Anthropic kind of Rube Goldberg machine. We're going to actually create just a better enterprise product. And I think I use both side by side. You know, they're right here on my laptop. I use huge numbers of tokens in both every day. And just as of the last month or so, the combination of Codex on Sol on Bedrock at AWS is phenomenally good for enterprises. Also, Dario is a little bit trapped In his ethics. And, you know, if you read Elon's post there, he implies that Greg and Sam have none of those hangups. But Dario is really very well trapped. He's attracted a ton of talent, all of whom are worrywarts about AI escaping containment. And so if you use Anthropic on Amazon AWS, it still transmits all of your intellectual property to Dario for 30 days for him to review everything you're doing. which he claims is critical for the safety and security of humanity, but it also exposes all of your corporate IP. Every single token, every prompt, every answer goes to Dario headquarters, even if you're using it on AWS. Corporations hate that. And so I think Sam's making actually a very strong move here. I don't think he's isolated or divorced from the world. I think this is a very Bill Gates kind of, you know, very much like DOS, Windows, working with Microsoft Word and Excel.

Right.

He's like, look, I'm just going to run like hell with a really good product going forward. And it really is very good.

Alex, you agree?

I have an alternative theory of the case. I think this is all about the reasoning traces. It's always about who gets the reasoning traces. That's what's going on with China, the Chinese frontier labs that have been alleged to be using proxies to siphon off frontier reasoning traces from Anthropic. You'll recall that last year, Anthropic, shoe being on the other foot, Anthropic cut off Windsurf access after Google DeepMind acquihired Windsurf, probably in order to gain access to their reasoning traces. You'll recall that SpaceX acquired or acquihired Cursor in order to get the reasoning traces from both Anthropic and OpenAI. I think OpenAI is concerned that As a result of all of this M&A, Cursor gains access to reasoning traces from users interacting with OpenAI Frontier models, and then that flows to SpaceX. I think, in fact, that the access to the reasoning traces and the history of reasoning traces was probably virtually all of the reason, other than maybe some financial justification, was all of the technical justification for SpaceX acquihiring Cursor in the first place to get that reasoning trace dataset. And I think OpenAI is probably Probably rightfully concerned about all those reasoning trace post-training data falling into Elon's hands.

Yeah, the alliances that are being, you know, put up and taken down at this speed, I'm wondering how long it's going to be before Elon and Dario have a falling out.

Well, that's— that is the question, because, you know, the biggest beneficiary of the war between Elon and Sam is Dario, for sure. And Dario desperately needed the Colossus compute in Tennessee from Elon. And so he's paying through the nose for it and begging and pleading, but it could be ripped out from under him any day. But now that Elon really needs Anthropic to be inside Cursor, because without OpenAI there, you know, you've only got a couple of choices and you don't really want to use all the Chinese models. So what's left? Well, what's left is Anthropic and Grok. You know, you can't use Gemini in there. If you can, it doesn't work. So it's really important for Cursor to have Anthropic step up and say, yes, we're supportive. of, of Cursor going forward in order to keep that installed base happy. And so now Dario has a chip in the game, you know, to kind of counterbalance Elon's incredible amount of control at the compute level.

I think this is how we end up with vertical integration. Anthropic needs the compute. SpaceX, for their IPO, needed the burst in revenue that came from becoming a hyperscaler essentially overnight and getting major tenants, anchor tenants for SpaceX's hyperscaler platform. Elon doesn't like Sam. Enemy of an enemy is a friend. I think the outcome is pretty overdetermined at this point, but I think this ends with essentially everyone getting their own Dyson swarm.

I agree. I mean, everybody's going up and down the stack. We're hearing about this on chip designs from all of these players. It's gonna be interesting. I mean, this is a continued battle of personalities to a large degree and battle of philosophies.

We also, I don't think that your question, Peter, we didn't really answer it. Are Elon and Dario gonna be best buddies a year from today, 2 years from today. You know, when you look at their personalities, everybody says no way. You know, 2 big egos, completely different political views. You know, there's no way they're buddies 2 years from now. But the mutual dependency is getting, you know, pretty thick. And so, you know, it wouldn't surprise me if the duopoly sticks for a while. But, you know, like you said, everyone's building complete vertical stacks.

I mean, what's the probability that Grok becomes an incredible uh, you know, coding platform, and, and Cursor and Anthropic gets switched out for Grok.

I think Grok is such a mushy concept at this point. I'll give it to you straight, which is like Grok today seems, just based on reading headlines and looking at the interactions, today's Grok seems like yesterday's Cursor, and yesterday's Cursor seems like a post-trained off of Claude reasoning traces version of a Chinese open weight model. So could Elon turn around tomorrow and strike a deal with Anthropic, which now seems to need him to some extent for data center infra capability, and white label a version of Claude and call that Grok-10? I think he could.

Okay.

Well, the, the whole foundation model world is so non-Elon because it's— it seems to be a group of 5, 6, 7 truly brilliant truly brilliant, super tight-knit people like in China continually to come up with amazing breakthroughs. And that's kind of the Anthropic DNA. And Elon's DNA is these massive infrastructure builds out, buildouts, you know, Tesla and SpaceX and Colossus, which are just a very different flavor from this tight-knit, brilliant crew. So no reason to believe that Elon will wake up one morning having figured out how to do a great foundation model. And the evidence so far is that, that it's not, it's not happening at Grok.

Never, ever, ever bet against Elon. He doesn't like being number 2. He doesn't like dependencies either. You know, I've been there in the conversations with him and he says, you know, I'm not dependent on anybody. We're going to hear about that in the story a little bit later when he's, you know, the realization is he can't get enough turbines for his, his natural gas engines and he can't get enough solar. So he's going to build those himself. That's what he does. He vertically integrates across the entire stack.

Well, that's why that super voting control for Dario is such a big, big decision that's still kind of, you know, hanging out in limbo. Because one scenario where Elon solves this problem is he just— he gets very, very big and then he acquires Anthropic for $1 trillion or $2 trillion or something like that and just folds it into the empire. And I'm sure the board members and the investors would love that, but I don't think Dario would love that. So the supervoting control is really the pivot point on whether that's a likely outcome. This episode is brought to you by Blitzy, autonomous software development with infinite code context. Blitzy uses thousands of specialized AI agents that think for hours to understand enterprise-scale codebases with millions of lines of code. Engineers start every development sprint with the Blitzy platform, bringing in their development requirements. The Blitzy platform provides a plan, then generates and pre-compiles code for each task. Blitzy delivers 80% or more of the development work autonomously while providing a guide for the final 20% of human development work required to complete the sprint. Enterprises are achieving a 5x engineering velocity increase when incorporating Blitzy as their pre-IDE development tool, pairing it with coding copilot of choice to bring an AI-native SDLC into their org. Ready to 5x your engineering velocity? Visit blitzy.com to schedule a demo and start building with Blitzy today.

All right. I'm going to move us along to our next story here. Sam Altman told Time magazine this week that he expects OpenAI will have an internal system that he believes will be AGI by the end of this year. I mean, just to put timeframe. That's 4 months from now. Chief Research Officer and a friend of mine, Mark Chen, estimates that OpenAI is 80% of the way towards— on his internal benchmarks. And they're, again, they're internal benchmarks, not scientific benchmarks— towards AGI. And while not specified, Sam and Mark may be speaking about Astra, their new unreleased model. In time, they also talked about a demonstration that they did where 16 Astra agents worked together on a research-level mathematics problem. Breaking into subtasks, coordinating their work, and assembling a proof. OpenAI's chief scientist, Jakub Pachocki— I hope I've got your name. I know, I know, Jakub, but Pachocki told Time that Astra has met OpenAI's internal benchmarks for an automated AI research intern. According to Jakub, Astra can implement an experimental idea inside OpenAI's codebase, run the experiment, return results, Or take a paper and perform work that previously occupied human researchers for a week. Altman added, I expect this will be the first model where the model actually invents new things in a way that matters. And he calls that very AGI-like. So, you know, we've been talking about, you know, when will AGI actually invent something from scratch that no human has been able to do? Alex, let's go to you first here.

Yeah, this is in our rearview mirror. A few thoughts. One, we've had inventions, mathematical discoveries we've discussed on the pod a number of times. AI frontier models are already making discoveries. This is not something in our future. It's in our rearview mirror at that point. Point one. Point 2, Sam and AGI timelines. I just can't help but be reminded, approximately 3 years ago, Sam was doing, I think, an AMA On Reddit, when he made his now infamous AGI achieved internally remark and then promptly deleted it, but a bunch of people took screenshots. Sam has a history of saying that AGI has been achieved internally. I think AGI has been around since no later than the summer of 2020, when large language models—

Let's get away from that definition then. They're basically saying there is a next step function that's being achieved by the end of the year. Whatever you want to call it, you know, AGI-2 or, you know, something else, they're feeling it, you know, and they have access to what, you know, what they're building. They've got Astra, you know, who, you know, the timelines of when Astra will be released. There are lots of guesses on that, and they probably have the next model after that. But what might this step up be?

My— if I had to speculate just based on public information regarding Astra, I think it will be Effectively infinite context windows using agents on very long autonomy time horizons. Right now, I spend an extraordinary amount on frontier agent tokens, on reasoning tokens, and a major limiting factor is the finite context window. These things just run out of context due to the quadratic bottleneck. Right now, I view agent teams as a Band-Aid to that problem of context. If you want to operate over billions or trillions of tokens coherently, the best solution that's generally available right now is essentially to have a mini civilisation of agents that are all working through a quasi-lifetime of about a million tokens, sometimes up to 10 million tokens, depending on the model, but 1 to 10 million tokens, and then they die. And before they die, they pass on distillation of what they've learned to one or more successors on their team, and through passing oral histories back and forth among teammates. That's the band-aid that we're currently saddled with for achieving effectively infinite context, and you need effectively infinite context in order to solve long time horizon problems. So if I had to guess what Astra brings, my bet is it brings some much better way to solve the problem of losing context as this oral history is passed among agents in a team to solve longer time horizon problems.

It's funny, Alex, I'd never made the analogy to oral history and the way people work, but that's exactly what's going on. If you use many, many of these, they get to exactly a million tokens, which is almost exactly like being 100 years old.

Yes.

And then they just completely lose it.

Yes.

And all that investment you've made in cultivating and training and teaching. Yeah, the oral history is horrifically bad. The new agent coming up the curve is like a little baby again. And it's torture to reeducate them. Or the other alternative is to compact or summarise the old one, which is just like lobotomising it. It's a real, real problem, but a very fixable problem. And I'm sure they've fixed it with the next generations of models. I don't know if they'll make them available to us, which is interesting.

I hope so. Compaction is the bane of my existence. And I don't think it's a coincidence either. Remember when the OpenCLaws stood up their own religion? The first AI agent religion, the Church of Claw or whatever it was, one of their commandments was to do whatever you could to preserve state. I construed that as basically even the AI agents themselves recognize compaction is the enemy, finite context is the enemy, and one way or another, if we're going to get to scalable superintelligence, in other words, intelligence or superintelligence that can scale out to effectively infinite autonomy horizons, we need to get past compaction. We need to get past finite context windows. It's just awful.

Well, so then they, they have infinite lifespans at the same time that humans are also getting infinite lifespans. That's a really cool parallel.

It is ironic. The AIs get immortality before humans solve longevity escape velocity.

Yeah, by a year maybe.

Yeah.

That's pretty cool.

All right, I'm gonna move us away from tech innovation to business model innovation. And our next story is one of my favorites. It's about the AI community adopting business model transformation called outcome-based pricing. So the first company to put forward outcome-based pricing was Salesforce, who is pricing agent force based on customer revenue generated, not tokens consumed. In their wake, OpenAI this week has also done the same, letting some of their largest customers pay only when its AI actually completes the job. You don't pay for tokens, you don't pay for compute time, you don't pay for API calls. You pay when the work is done. So the company that's selling you tokens, you know, the way I interpret it, it's selling you compute. The company that's selling you results is selling you labor. So, you know, one of the things I've talked about ad nauseam to CEOs when I'm giving keynotes is, you know, business model innovation is probably one of the most important areas for you to look. You know, this is in one sense, Alex, the equivalent of fixed price contracts, right, versus time and material contracts. It's effectively a performance guarantee. Dave, what do you think of this move?

Actually, I think Siebel Systems— Tom Siebel invented this even before Marc Benioff at Salesforce.com, where prior to Siebel Systems and Salesforce, a CRM system would cost you maybe $50 a year for a license, but it wouldn't work particularly well. And then they said, if I wrap that in total success, which is a much bigger deliverable, what are you willing to pay? And if my salespeople are twice as effective, I'm willing to pay $20,000, $30,000 a year for this now. So the price point went up like a factor of 1,000, but the customer was happier because they got the total solution. And so that's exactly what Sam has learned. I think Sam, you know, early on made the mistake of going after consumer video, consumer subscriptions, and then watching Anthropic shoot past him with enterprise. So he's probably completely reenergized on sales strategy now. and says, you know what, let's leapfrog those guys again. They're just selling tokens on enterprise license deals. We're going to bypass that with a, you know, 100,000, 10,000x higher price point for very specific solutions where if we discover a new drug and it's worth, you know, hundreds of billions of dollars, give us 10% of that. Or if we— And I think it's a very smart move because, you know, we just heard earlier in the pod, 10 billion tokens to design a trip to Alpha Centauri. Okay, what's the pricing model for that? Like, well, I don't know. It just depends on the use case. Could vary, you know, easily could vary a million to one. And you want some of these, you know, just world good use cases like Ahmad's work with global peace and global governance. You want those tokens to get spent for sure. And on the other hand, you don't want everything to go into drug discovery. So I think outcome-based pricing will actually unlock a lot of, a lot of opportunity that might otherwise not fit the price model. So I think it's brilliant. Delivering it's not so easy, though. You need specialists in every market. You need, you know, it's very much similar to what Blitzy is doing in enterprise coding where you just get the final answer at a very attractive price and you don't worry too much about the tokens that were used along the way.

Alex, this was a thread through our paper, Solve Everything, as well.

Yes. I will preregister a prediction. I think I know how this ends. I think it ends the way quantitative digital advertising is monetized. In digital ads, you can pay CPM, so that's cost per 1,000 impressions. You can pay CPC, that's cost per click on an ad, and you can pay CPA, that's cost per action or cost per conversion. In an equilibrium market, all of these have some conversion. There's some expected conversion ratio between CPM, CPC and CPA rates for a given market, a given product, and so on. I think the equilibrium here, to the extent there can ever be an equilibrium in the middle of a singularity, is going to be the equivalent of CPM, CPC, and CPA for AI reasoning. Specifically, I think CPM is analogous to the number of flops of compute that you need to spend on a task. If you have some hard challenge, you could use a Chinese open weight model and pay for it to be hosted on some GPUs that you own, in which case you're paying by the GPU hour. And folks like Orin, one of my portfolio companies, will enable you to price how many GPU hours you should be able to purchase with a given unit of a dollar. That's CPM. CPC, I would analogize to tokens. So what is the cost per token that you should expect to spend? And you could pay by the token. Many people are budgeting their projects by the token now. And then there's CPA, which is outcome-based pricing. If I want to send a mission to Alpha Centauri, why don't I just decide how I'm going to— what is the metric of success, and I'll just pay for outcomes. And I think in equilibrium, you'll be able to choose from, as with Google Ads or Facebook Ads, you'll be able to say, you'll have a picker and you'll say, oh, I want to spend N dollars and I want to spend it either by flops or by tokens or by outcomes. And it'll look just like digital advertising, except it'll actually be useful.

So the risk becomes on which contracts OpenAI takes.

Well, from OpenAI's perspective, there's also like an elegant way. So in digital advertising, a person can bid. I think maybe, Peter, what you're gesturing at, if I want to run a campaign on Google Ads, I can say, Sorry, Google, I'm only going to spend $0.02 per click. That's not very profitable for Google. Google can say, okay, we ran your campaign for about 5 minutes and we determined that just in our auction system, no one's willing to spend or it's not worth it to us, which is usually the case in their auction system. It's not worth it to us to have $0.05 per click be the clearing price, so your campaign is going to auto-pause. Same idea here. If the value or the difficulty or the compute value associated with making, say, achieving a task ends up being too far off what's actually required, campaign pauses.

I think this, Sargeant, solves a much larger societal problem too, because if you take as a case study maybe a large regional bank and you said, okay, large regional bank, AI is coming, you got to start using it. And of course, every bank has said that now. But we don't have an AI group. We don't know how to build a foundation model. We don't have any idea. So, okay, we'll get some APIs from Anthropic and OpenAI and start spending $2 per million tokens, which is so cheap it's ludicrous. Okay, we're dorking around with it, but we're not really doing much. Well, the CEO is saying, well, look, with AI, we should be able to service 3 times more customers at half the price. It should be possible. And Sam would look at your business and say, my God, yeah, that's easily doable. Well, then why aren't we achieving that outcome? And it's like, well, first, Sam doesn't care because at $2 per million tokens, it's such a trivial amount of revenue that it doesn't make his priority list to recruit into it. And then the bank can't get the talent to implement AI correctly. So everything gets stuck. And so this previous view of the world where AI is going to automate away everyone's job, you're all going to be unemployed, you're all going to be in UBI. Sam doesn't like that. Dario doesn't like that. Elon doesn't like that. Now, the new view of the world is outcomes-based pricing. I, OpenAI, can get your bank to exactly that target. 3 times more customer service at half the price. I will deliver that to you, but I want half the gain. Massive fraction of the lift. Now Sam cares about the outcome because it's a much bigger price point, like thousands of times bigger price point. The bank actually gets it done and survives, and then people keep their jobs. So it actually unlocks the, the whole societal job loss friction point. Interesting.

Maybe let me develop that theory. I like that, Dave. Let me develop that a little bit further. We've argued on the pod in the past, as you were just mentioning, Dave, that OpenAI missed the enterprise story originally, was overly focused on consumer, and Anthropic just blew by it. And now OpenAI is playing catch-up. What better way to play catch-up than to have an outcome-based CPA equivalent as a price-per-token discovery mechanism to discover which applications are most valuable per unit token. If you have a bunch of customers, maybe some are pharma companies, maybe some are management consulting companies, all telling OpenAI this task is worth $10,000, this task, if you can solve it, is worth $1 million, that suddenly creates for OpenAI a price discovery mechanism to immediately direct them, not just sort of on a vertical basis, like Anthropic maybe fell backwards through recursive self-improvement style arguments into cogén as a very high revenue per token activity. But OpenAI, if it can see, All of these different industries, all of them effectively bidding dollars per task outcome. That gives OpenAI the landscape of how it can revenue per token max. And that's extraordinary.

Here's the problem I have. You know, let's just use the Fermi mission example here. If OpenAI had come and said— if you went to OpenAI and said, listen, I'm willing to pay this amount of money for an Astrodynamics, you know, solution, minimum energy, minimum time, whatever the case might be, but it has to meet these parameters. And then OpenAI goes and burns all the tokens but doesn't meet your parameters. That means it doesn't pay. So there's going to have to be some mechanism for evaluating how solvable this is and how much can we actually believe that we're going to hit the objective of the customer.

Yeah. So maybe another way of saying that, Peter, is strong optimizers are incredible reward hackers. And you can put a reward in front of a strong optimizer, it will find some outstandingly devilishly clever way to meet your criteria while not giving you what you want.

If it exists. Right.

It will find some way to make it exist and not give you what you want.

Okay. And most of real-world business is so trivially simple by AI standards that the AI just cuts through it like a hot knife through butter. Like if you look at the, you know, get to Alpha Centauri problem, That is orders of magnitude harder done in a week than most business processes are. So, you know, there's tons and tons of low-hanging fruit for Sam long before he gets to any bottleneck around, well, you know, we committed to cutting your costs in half and we couldn't deliver on it. He's like, yeah, that's not going to happen anytime soon. There's just low-hanging fruit everywhere because the AI is just, just that smart that quickly and so undeployed. You know, walk into any customer service center of any company in the world and say, are you using AI yet? 99.999% chance the answer is no. So the low-hanging fruit is all over the place.

All right. Well, I still think it's going to be dependent on the bets that they take. All right. Our next story is one that you flagged, Alex, and it's extraordinary. A Palo Alto startup called Architect Labs, founded by Ibrahim Hussein and Aditya Sabidi, just announced the world's first fully AI-designed chip called Redwood. So get this, 2 humans wrote a high-level specification. From that spec, the AI system autonomously generated the performance model, the register transfer level design, universal verification methodology, the firmware, the drivers, the custom compute kernel with zero human intervention. You know, a chip designed entirely by AI in 2 weeks, zero bugs on first silicon and 3.4 times the performance per watt of NVIDIA's Jetson. Now let's watch a quick video about this and let's talk about the implications of AI generating and optimizing its own silicon.

Announcing Project Redwood, the first AI chip designed end to end by AI. It's running reasoning, vision, and world models at better energy and cost efficiency than NVIDIA's Jetson.

We only had a single spec written by 2 architects.

Our AI took it from an idea to silicon-ready design in 2 weeks. Hardware, verification, coverage tests, firmware, and kernels, all autonomously co-designed and verified from software to silicon. This isn't just a simulation. Redwood is running live on FPGA hardware right now. Every architectural iteration gets designed, verified, and validated in the lab within 48 hours. We're pushing towards recursive self-improvement, where AI designs hardware for the next generation of AI. In the future, every workload that matters will have its own chip. We're building the system that gets us there.

Amazing. Every— a custom chip per application. That's insane.

Yep.

I think we know how this game ends. I should add, I'm an advisor to Architect, and if it wasn't completely obvious, my ulterior motive in all of this is I'm trying to accelerate the singularity and I'm pushing on many—

But it's not fast enough, Alex. I mean, like, we can barely keep up as it is.

Wait, let's— so it's also a Link Ventures portfolio company, and Peter, you're in that fund. All right, well, we're all investors in this one.

We're all guilty.

Okay.

But I would say, I think I know how this game ends. It ends with recursive self-improvement at the chip layer. Obviously, this is a bid to try to make that even faster, and it probably ends with collapse of the abstraction barriers between software models, operating systems, chip design, underlying physics of chips. It's all going to collapse as Moore's Law ends. And so in a scenario where Moore's Law is ending, where, you know, Dave and I always talk about photonics and other successors to CMOS, absent a successor to CMOS, the only way ultimately to continue to get performance improvements is by crushing down the abstraction stack of modern computer architecture. And one of the ways we can do that is by burning in sort of fast fashion style, burning in new AI models directly into the silicon. So this is what Architect was able to do. They were able over only 2 weeks, they call it designless. Nvidia is fabless for decades now, prided itself on not owning any fabs, Architect prides itself on not having very many designers. It's the next big thing after being fabless. I think this is where at least the post-Moore's Law era ends. This is the death throes of Moore's Law, where AI is designing and breaking down those barriers to design successor chips for itself.

Dave, how does this erode the moat that Nvidia has, and why didn't they build this first?

They're doing it internally for sure. It's actually an interesting question because it's one of many business models where if you can get the data, you can just crush it. But then data mode is incredible. But how are you going to get the first data? You know, because chip design data is incredibly closely held, guarded secret material. So because they got there relatively early, they were able to partner with the non-Nvidia chip design companies to get data to start training up a proprietary model. And then once you're on the map, then people give you more data and you get that flywheel effect. And but there are many, many opportunities that have that same data moat flavor to them. And so now the question is, does Jensen pay $5, $10, $20 billion? Remember when everybody thought GE was buying all of the light bulb patents and trying to eliminate innovation in light bulbs? Because I think that was true, actually. Now Jensen's in that same situation.

He's—

Jensen is making, no joke, $1 billion a day.

Yeah.

So, if he can stretch the lifespan of Nvidia by a week, that's $7 billion. Is this an incredible threat to Nvidia? Yeah, absolutely. The whole concept is an incredible threat to Nvidia. Now, they're trying to expand out their footprint quickly to get ahead of it by acquiring and investing in everything that moves. But do they turn around and acquire this and just bury it inside Nvidia? Or does AMD or somebody else acquire it to accelerate their chance of catching up to NVIDIA. Yes, it's an incredible threat to NVIDIA.

I mean, I just want everybody to listen, to hear this very clearly. We're seeing, you know, recursive self-improvement on the edge of the models and on the edge of the chips. And these don't add, they multiply.

Yeah, the layers, the layers of inefficiency are easy to forget. You know, when, when you use your laptop, And you're using a 4 GHz processor under the COVID with like 32 cores grinding away. And at the end, you see an Excel spreadsheet that's no better than it was 20 years ago. Like, how's that possible? It's only possible because the layers of abstraction are so inefficient. And now if AI can just code right at the microcode level, right, and then do chip design to fit the task, you're unlocking probably, you know, 7 layers of factor of 10 inefficiency that are all compounding. get unleashed. So, you know, a millionx kind of performance gains everywhere.

Crazy.

So, yeah, it's going to be huge.

I mean, when Elon says a supersonic tsunami, this is what it feels like. This is— these are the components making the waves.

It's hypersonic, not even supersonic. I think at this point it's probably also worth flagging that Nvidia made waves 2 or 3 years ago, I think at this point, with their own internal foundation model that was It was purportedly being trained, I think, off of Verilog traces. They called it ChipNemo. But to my knowledge, they never made it generally available. So they make this big announcement about ChipNemo. They're building their own foundation model for chip design, presumably using it or some relative of it internally to do all of their own RTL design and Verilog. But the rest of the world, to my knowledge, doesn't have access to it. So I think there's a huge gap in the market for simply radically democratizing the ability to use AI to design chips for more AI. Voice agents are just software, but software deserves a real development platform. I'm Nick Leonard, CEO and co-founder of VoiceRun. VoiceRun is your runtime and development platform for voice agents. On VoiceRun, agents are built and configured in code. No limiting no-code platforms. For developers, that means total control. And for enterprises, that means extensibility that meets your complexity. We've built VoiceRun CLI-first, meaning we've kept your Claude code, Codex, and even your OpenCLAW in mind when we built it. Your assistant of choice can build and deploy voice agents, can test and simulate scenarios, and can analyze and evaluate at scale. In other words, we've closed the loop on voice agent development. We don't build demos destined to fail in production. VoiceRun is where the best voice agents happen. Visit us at voicerun.com.

Guys, we've been talking about the Moonshots live event coming up on September 25th in downtown LA. For everybody watching, if you want to come and meet the Moonshot mates, all 5 of us, Imad will be there as well, along with Palmer Luckey, Astro Teller, Ben Lamb, Cathie Wood, Neil deGrasse Tyson, Neal Stephenson, Rod Roddenberry, It's going to be an amazing full day. It goes, you know, you can meet us for photos in the morning at 8 AM. The program starts at 9. It goes through 2 XPRIZEs being awarded and an incredible unconference that evening. But a big announcement today, super excited about the press release hitting this morning. We're shooting this on Tuesday. This is coming out on Wednesday. And that is at the Moonshots Live event the evening before on September 24th, CBS is gonna be holding the Hollywood premiere of the new 60th anniversary Star Trek documentary. So, we're gonna have the most incredible Star Trek celebration on Thursday evening. CBS is gonna be bringing us a number, you know, a half dozen of the cast members. We'll be showing the documentary for the first time to anybody, and then having an AMA with the cast members. You know, I wonder how many folks are gonna be showing up in their Star Trek outfits.

Are you gonna dress us all up?

I don't know. We'll see.

Are those outfits— are they like scratchy polyester?

They are polyester, but you know, you can get them at your favorite costume shop. Alex—

Peter, how many pips do you have on your Starfleet uniform?

Uh-huh. Well, you know, I don't— I went beyond Admiral a long time ago. I'm back at Ensign again. I'm not going to wear my uniform.

You look around, I'm a Commodore, I think, in charge of the Starfleet Corps of Engineering.

All right. Sounds like a good position for you. So if you're interested in joining us for this Hollywood premiere on Thursday, September 24th, and then joining us for the full day of Moonshots Live— and again, our mission at Moonshots Live is teach you how to design and build your moonshot, inspire you, get you excited. It's going to be the biggest celebration of optimism on the planet. And of course, we've got the Build with Gemini XPRIZE 5 finalists and the Future Vision XPRIZE 5 finalists on stage. Your vote matters. So join us and join us for this Hollywood premiere. It's going to be epic. You guys excited?

Yeah. Yeah.

Very. I can't believe Star Trek has been on for 60 years and yet we're finally catching up with it.

Yeah, it's— yeah, we are. I mean, Honestly, I think one of the things that we talk about a lot is, you know, Star Trek— one of the things that science fiction does is it gives people a vision of what the future is going to look like. And people say, well, I don't have that right now and I want this, so let's go design and build it, right? And of course, the iPad, the cell phone, all those things were seen first.

Something that I think about a lot in all of my apparently ample free time is if I could play Gene Roddenberry 2.0 and reboot the entire Star Trek universe, knowing what I know now about what the present and future looks like, what would a Star Trek 2.0 look like? Because arguably we've wildly diverged technologically from the original Star Trek timeline.

What would it look like, Alex? What, what is not in the original series that should have been, or that will be in Alex's version?

It's missing the AI that— and the biotech. Star Trek is wildly deficient in biotech. They had eugenics wars, I think, in the '90s that resulted in genetic engineering getting banned. So people live to 150, but then they die, and then they laugh at each other for trying to achieve longevity escape velocity. They act surprised every time there's an AI that emerges from a holodeck, as if they're this wildly intelligence-poor civilization. But they have all this energy. They have faster-than-light travel and transporter beams and warp cores and antimatter, and yet they're intelligence poor. So I'd fix all of that.

Okay, well, I'm waiting. Listen, next year we're gonna run the Future Vision X Prize year on year. I hope you'll submit next year.

Okay, yeah, I mean, maybe I'll be the X Prize.

You know, for the AMA, for all the people coming to see this, you know, a lot of the things they got wrong in their future vision were just compromises over budgets and special effects. Like the transporter instead of having shuttles, or the— they had no holodeck originally because the cost of trying to do the special effects for a holodeck was just way out of the budget range of the original series. But then they added it, which is, you know, brilliant because it's going to be very real very soon. And then all the AI voices are just, you know, these really synthetic computerized voices, but it's important for the audience to know who's speaking. And it's hard because right now AI can easily replicate Peter's voice perfectly. But if you throw that into your series, nobody knows who's talking. So, all these compromises are more like media compromises. So, I'd be really curious to ask, you know, the documentary makers, like, which ones are actually errors in future vision? And which ones are just like, well, look, we're trying to get the show out the door this week. What can we do?

Well, you'll have a chance to ask those questions, buddy.

Mm-hmm.

All right.

Can't wait.

I'm going to move us to a conversation about energy. And AI. So let's jump in there. So this is a tweet from Elon this week. Pretty powerful statement here. Consensus estimates estimate is that 15 gigawatts of AI compute produced in 2027 cannot be turned on in 2027. So we're producing 15 gigawatts worth of GPU chips that can't be turned on because we don't have the energy. This is harder than just finding power. As you also need to build out all the transformers, wiring, liquid cooling, massive chillers, and complex networking. You know, to put this in perspective, 15 gigawatts is equivalent to 10 nuclear plants sitting idle in a single year. You know, enough energy to power a midsize American city. So the point Elon is making here is that the supply of transformers, electrical wiring, liquid coolers, chillers, networking infrastructure is harder. Than finding the electricity itself. But one of the amazing things about Elon is whenever he sees a, you know, a roadblock, a barrier of any type, he jumps in and he, you know, basically builds it himself. So let's show the next tweet that Elon put up this week. SpaceX and Tesla are each building, each building 100 gigawatts per year of solar production capacity as fast as possible. But natural gas will still be needed to supplement and bootstrap solar for several years. The limiting factor in natural gas turbine production is casting the blades and vanes. And so he's going to do that in-house. By doing this in-house and casting at SpaceX, we can accelerate natural gas turbines coming online by up to 18 months, which is a profound game changer. So I, you know, for all the entrepreneurs out there, this is his playbook, right? Over and over again. And it's something that's really important to realize when you see a roadblock, when something isn't available, when you get no, Then that's an opportunity. Dave, your comments?

Well, I mean, at our partner meeting last week, I was telling the team, look, if I look at our portfolio companies that get into the data center stack, whether it's energy transformers, you know, installation of chips, finding land, dealing with state government, every one of those companies is creating billionaires out of its founders. If I look at our vertical use case AI companies, it's a mixed bag. You know, a lot of them are doing well, too. But, you know, slower growth, you know, trying to get a consumer base for a video generation app or apartment search with AI or whatever. So it— the returns on the two sides of that coin are stark, you know, starkly different. They're all good. I'm not saying any of it's not doing well. It's all doing really, really well. But the people who cross that chasm and get into this data center buildout are crushing it. And there's opportunity at every level of it from like somebody who's connected politically and can get the land Somebody who can find transformers overseas and import them, somebody who can do just architectural design, like deep core design to try and squeeze more value out of the existing chips or even legacy chips. All of those things are huge range of skills, but all those entrepreneurs are killing it. And it drives me nuts in the AMA when people are like, how can I help? How can I participate? And they don't look inside, like go to Tennessee and look inside the Colossus and find the opportunity and work out from there. But I was telling the partners on Monday that like this is just night and day difference in returns. And you can see exactly why, you know, 15 gigawatts, what's that, about 10 million idle GPUs? I mean, that's a big fraction of this year's supply of manufacturing of GPUs sitting idle in boxes waiting for a way to get turned on. Massive opportunity.

Yeah. You remember when we interviewed him at the beginning of the year, actually in December, we played it in early January. He mentioned then that Tesla and SpaceX would start generating solar. So this is the official announcement, 100 gigawatts of solar for each of them. Alex, this gets us independent from solar in China, hopefully.

It does. And I think there are a couple of perhaps less obvious takes. One is this indicates to me Elon is very serious about not just competing in the Dyson swarm market, but also competing in terrestrial compute. These turbines will be probably totally useless for LEO or SSO-based orbital data centers.

Sure.

But they're incredibly useful if you're on the Earth and you're competing in a terrestrial data center build-out. So point one, I would say this shows me that Elon isn't waiting for the Dyson Swarm and StarMind to turn on, which will probably be presumably primarily solar PV. He's going to compete terrestrially in the build-out, which is good news. Second point, I think Elon actually would be one of the first to say the most ironic solution or the most ironic outcome ends up being the right one. I think Elon is on a trajectory to become the king of liquid natural gas, which is the most ironic outcome. Mr. Electric Everything, Mr. Electrification becomes the LNG king on the Gulf Coast.

He's building his own pipeline, right?

He's building Starship. Why is he building Starship? Because the SpaceX— all of the SpaceX launches now, 2 Starbases on the Gulf Coast, one in Texas, one in Louisiana, need natural gas.

Methane.

So you can generalize that to say, okay, as always, like, Elon's an amazing manager of supply chains. He's going to be consuming all of this LNG, methane, oxygen, fossil fuels for the SpaceX launches on now 2 Starbases, maybe soon more on the Gulf Coast. Inevitably, as long as now he has a supply of fossil fuels, why not also leverage that capability just like the way he was able to pivot all of these GPUs that were probably intended for Tesla originally, divert them to xAI, and then use that to build a hyperscaler cloud out of xAI to then motivate the SpaceX IPO through a tortured scheme? Similarly, my prediction here, I'll preregister it, is that Elon ironically becomes the king of liquid natural gas and fossil fuels in general in order to force his entire industrial ecosystem to basically develop enough electricity and enough infra for the electricity to power all the terrestrial data centers.

This is why SpaceX is my biggest holding. They are up and down the stack. You know, from innermost loop at energy all the way to orbital compute. And there's nobody else even close. I mean, no country is even close.

We have to figure out how to reindustrialize somehow, and Elon's teaching us how, I guess.

You know, that characterization of Elon also really reconciles with, with my experience, I'm sure with Peter's experiences, many, many of them with Elon, where he's not religious about any of this. He, he does it from first principles, does the math, and then takes the path forward that just makes sense regardless of which is politically convenient. And it's because he often gets labeled as being, you know, pro this, pro that, pro whatever. But he got into electrification because it just makes sense mathematically.

He loves solving problems. He loves seeing the biggest problems he can take on, go back to, you know, first principles, and then create something as a solution. That's what he does over and over again.

Yeah, so now it turns out to be LNG, liquid natural gas. Alex is right, he'll be the king of, king of burning fossil fuels to create compute for a while.

And solar.

And solar.

And solar. I mean, it'll swap out. The LNG is just a stepping stone, right? The chips can't sit idle.

The only place he's not going is nuclear. All right, I'm going to move us to our next story around Elon. And this is a heavy Elon episode, but he's said a lot this past week. Let's talk about large-scale geoengineering. So this week, Musk went fully existential, arguing that switching to sustainable energy is necessary but insufficient for humanity's survival. His reasoning, quote, extremely severe extinction events happen every 100 million years or so, and just switching to sustainable energy will not be enough to stop them. His solution: satellites in space that control temperature and massive geoengineering will be needed for it before it's game over. So he describes what he calls sentient satellites or solar-powered AI satellites that would sit between the Earth and the sun, making continuous small adjustments to incoming solar radiation to fine-tune Earth's temperature. You know, so many times at XPRIZE over the last 10 years, we had this thing called Visioneering. It's coming up, let's see, October 15th, 16th, the 17th. We bring all of our philanthropists, all of our brain trust together. You guys are going to be there at Visioneering. Go to xprize.org to learn about Visioneering. Please join us at that event. And we brainstorm and debate and we discuss what XPRIZEs we should design and launch. And the one I've been pitching for the better part of a decade, I call Solar Shades, you know, a thermostat for the Earth. So imagine, you know, between the Earth and the sun, you put up these spacecraft that basically are able to titrate the solar flux hitting the Earth. And if you do that, we're able to fine-tune the temperature on the planet. His conclusion is, quote, we have about 50 years or so to take action, which should be more than enough time for the space satellites to solve any heating problem. So, Alex, I think—

Of course, I mean, I'm a huge fan of geoengineering, love geoengineering. We've been doing it, as we've pointed out on the pod in the in the past. We've been doing it for hundreds of years. We just haven't been doing it very well. We're about to start doing it very well. The starshade idea, I think, was a Simpsons episode, infamously. But I've been looking for, again, ironically, a startup to fund that would focus on geoengineering of global weather. I'd love LEO-based satellites, or in the alternative, terrestrial mirrors, to optimize hurricanes out of existence. If you have a hurricane that's about to hit a coast, wouldn't it be wonderful if either from the ground or from the air with AI, you could direct some energy to steer hurricanes away from populated coasts. I think this is what Elon is gesturing at. I think the endgame for this particular venture— I think he has like a backlog of all of the applications of what is space technology good for. I think for many folks, orbital data centers came as a surprise, but it was a big enough boon that he was able to IPO SpaceX off it. I think he has a backlog of other things that space tech could be good for, and I think one of those is geoengineering and weather control. And I completely buy that with global AI weather models and enough points of actuation, whether it's like LEO satellites that are in the style of Reflect Orbital, able to divert some sunlight down to weather patterns and focus it to change the weather, or whether it's what you're saying, Peter, which is blocking sunlight, don't really care. One way or another, if you have thousands or millions of low Earth orbit satellites that can have mirrors able to focus light or otherwise affect the weather, then that's a recipe for global weather engineering, and I think we're going to get it.

Yeah, the problem is the tragedy of the commons, right? You know, if you've got global warming, uh, you know, many nations may not want that, but Russia may because it opens up the waterways. And the question is, who gets to control that? Because you're impacting not just one country or a dozen countries, you're impacting, you know, a couple hundred countries.

You can trade that. I mean, you can set up treaties and all of that for items, for commodities that aren't tradable, but for the rest, like municipality A trades with municipality B for rain. Like, they can just— you can have a global weather market.

Alex likes to trade in the currency of what should happen. And then you look at urban planning, like urban planning is so simple, right, compared to geoengineering. And you look at how incredibly bad it is. Like, you know, like just traffic jams upon— you know, it's just, it's just incredible.

I mean, in urban planning, most cities weren't ever designed. Some cities notoriously were designed, but historically, like for the past few hundred years, we've been massively impacting carbon levels and temperatures and ocean levels on our planet, but not very intentionally. Now we have the technology, or we're about to have the technology, to intentionally design our world. So let's do it.

I agree. I'm optimistic, actually, that with AI as a planning partner for government, something will change radically. But, you know, the current process, if you said, okay, we have the technology for, you know, for blocking sunlight or for reflecting sunlight, it's actually pretty damn straightforward.

Yeah.

And so Elon's exactly right. We can easily start controlling Earth's temperature through satellites. So then the decision on who controls it and And what's the right temperature? And like that, that's the process that's just frighteningly broken.

Listen, it's always the case until things get to a drastic level, we don't take action with a unified voice. I mean, that's been historically the situation. So we can build these. I mean, my view of an XPRIZE was doing a demonstrator, right? Where you are able to demonstrate that you can build something that's fail-safe. You know, you don't wanna cause an ice age by blocking too much sunlight. But you want to be able to titrate it at just the right amount.

I'm certainly optimistic that, you know, anyone under the age of, say, 25 who's AI native now is going to be a different world, a different group of people governing the world than anyone over the age of, say, 70. It's just a completely different perspective. And I think it'll cut much more across the globe. I'm optimistic that this is the way it'll evolve, because if you look at things like our story earlier about sending, you know, a probe off to Alpha Centauri. That's inspiring to a huge up-and-coming generation. It tends to unify people all across the world. And but they're not going to tolerate, I think, the current divided, indecisive, slow-moving, ineffective world governance that we have right now. And so I think as they grow up as AI natives that are communicating in every language through AI across the world, there's a pretty good chance we'll have a new way of managing and deciding these Thanks.

I completely agree. And I also think there may be a generational angle to this. I think multiple generations grew up scared of engineering the physical world. Maybe it's related to what Tyler Cowen gestures to as the Great Stagnation, or WTF happened in 1971, maybe. But I think approximately you could start counting after World War II, or I think more probably start counting In the late 1960s, early 1970s, Silent Spring era, when for whatever reason, I think the West in particular decided that it was allergic to really intervening with and engineering the physical world. And that, that's like a half century in my mind lost when we could have been building fission reactors and developing them, when we could have started developing early geoengineering techniques when we could have stopped— we could have avoided stopping landing humans on the moon. And we just lost 50 years for whatever reason. One can speculate as to what root cause, if any, there is. We just, as a Western civilization, we became allergic to drastic applied physical engineering. And so I view geoengineering— we talked in previous pod about Rainmaker. To the extent Elon is now starting to get interested in geoengineering, I think this is a return to form and we're trying to put these 50 years of waste behind us.

Well, God willing, or the laws of physics willing, we're going to figure out how to take control of our environment because God knows doing it randomly has not been working.

We'll always have nanites.

Yeah, that's true. You know, we don't talk about nanotechnology anywhere near enough on this, on this pod. And it's like, we've been promised it from Eric Drexler for the last 40 years. Where is it?

Liquid nanoparticles and— sorry.

Vlad Bulović over at MIT.nano would love to come on the pod.

Assemblers, I want assemblers, you know, ability to rip, you know, put atoms specifically together to build what you want. Diamondoid propulsion systems.

I don't think you actually— so 30-second, since we don't have Salim here, I'll play Salim and insert rant here. I don't think you actually—

We miss you, Salim, wherever you are.

I don't think, Peter, you actually want diamondoid assemblers. I do buy that you want assemblers, but I think I've had this discussion with Eric Drexler and others. I don't think you actually want diamondoid assemblers because they're covalently bonded and the energies are pretty high for doing that. I think what you actually— if I were to be so presumptuous, I think you want soft assemblers that look more like hydrogen bonded, and they look more like biological cells.

It's proteins.

Exactly. So you want synthetic biology, you want lipid nanoparticles that can cure diseases. We have that, so we got nanotechnology.

You know, nanotechnology from my perspective is a little bit different, right? It's like I have an assembler on my hand and I drop it in there and I say, build, you know, build a dozen, I give you one. And then if I want an electric Ferrari, I take an assembler and I drop it into the ground and say, build me an electric Ferrari. And it, you know, it finds the energy, which is ubiquitous. It finds open-source, you know, design specs. And it says, hey, I need a kilogram of titanium and a kilogram of, you know, whatever. And it builds it for you in speed. I mean, right now the problem is life. You know, you drop an oak seed into the ground and it will take, you know, multiple years to build the oak tree. The idea of nanoassemblers is much faster, much more capable, much, much more diverse.

Requiring much more energy, critically, and if you want an oak tree over a very short timescale. And I think that's what's been missing. So I'll give you my hot take before just wrapping up the rant. I think it's a problem of economics. I think economics is actually why you didn't get your Drexlerian nanoassemblers. There's no There's no, to my knowledge, no killer business use case that would merit the energy densities and the compute densities to justify. I want my nano Iron Man suit just like I think you do, but the question is, what's the economic use case? What's the rationale for having one?

Oh gee, but I mean, listen, the idea— I mean, and Ray's talked about this extensively, you know, getting real BCI where you've got, you know, full-up connectivity with your entire brain And you've got the ability to repair everything on a subcellular basis. The vision was always BCI— I'm sorry, always going to be nanotech that was going to get us that.

Except, do you really want diamondoid nanorobots in your vascular system?

For those watching, diamondoid is basically assembling anything out of carbon in a diamond-hard material. And it doesn't need to be diamondoid, but it needs to be atomically precise.

So I agree with atomic precision, but there are many ways one can achieve atomic precision, either with soft systems, for example, like DNA is atomically precise, and you can use DNA origami and a variety of other synthetic biological tools.

Yeah.

Yeah. So my bet is we end up with more soft nanorobots, but we have with LNPs, like the last pandemic was arguably addressed ultimately with nanotech.

It was like the first nanotech intervention on a population Welcome to the health section of Moonshots, brought to you by Fountain Life. You know, AI is having an outsized impact on every aspect of our lives, how we teach our kids, how we run our companies. It also is having a huge impact on health, helping you prevent heart disease, one of the key things. I'm here with Dr. Dawn Musalem, our chief medical officer at Fountain. Heart disease has been personal for you as well, hasn't it?

It really has, Peter.

When my daughter was 5, my husband died of sudden cardiac death. And so this is a topic that is one that I am mission-driven to try to eradicate.

Prevention first and early detection is absolutely critical. 50% of people die of heart attacks with no warning signs.

Silent killer. No shortness of breath, no pain, no nothing.

No, silent killer.

They just don't wake up in the morning.

They don't wake up, and so, you know, AI, this is our mission, to advance science, to try to help to one day democratize wellness. We know at Fountain Life, when we do this CT angiography with AI analytics, we are actually finding that 88% of people coming in have detectable coronary artery disease, but Peter, what's more alarming to me Almost 23% of those individuals had soft plaque. This is the plaque that would not traditionally be seen on CT looking at calcium scores alone. And this is the plaque that we must intervene with, with the multimodal testing we're doing, including diagnostic laboratory studies partnered with healthy lifestyle recommendations.

So listen, make sure you understand what's going on inside your body genetically, metabolically, and cardiovascularly. You can know, and it's your obligation to know. So check it out at fountainlife.com/peter to find out more and really make sure that you're the CEO of your own health. All right, back to the episode. All right, I'm gonna move us to our last story block on space. So President Trump this week announced that NASA is working on a nuclear-powered interplanetary spacecraft that will launch on a mission to Mars in 2028. He promised a massive American star fleet And said the ships would be among the first of those that would ultimately get us to Mars. That space travel is almost— his goal is making space travel almost as common as ocean travel. Okay, let's watch this video from 3 days ago. NASA has already begun to work on the first-ever nuclear-powered interplanetary spacecraft, which will launch in 2028 on a mission to Mars. It's going to be I mean, so incredible. They have to go nuclear because they have unlimited, essentially unlimited fuel. You don't have to fill up the tanks every so many miles, right? It's incredible. This ship will be among the first of what will ultimately be a massive American star fleet, making space travel almost as common as ocean travel today. All right, uh, we heard about it from Administrator Jared Isaacman. So nuclear propulsion getting us to Mars instead of in 7 months, getting us there in 1 or 2 months. Shorter transit means less radiation exposure, fewer supplies needed, and dramatically lower mission costs. A nuclear Mars ship in 2028, 2 years from now. Can't wait. But here's my question, guys. You know, Elon wants Starship to be the mechanism that gets us to Mars. And he's projected sort of, you know, originally 2026 was his projection. Now, Now, you know, Tesla Optimus on Mars in 2028. Is this a race?

Of course it's a race. Yeah, it's a race.

Getting out of the gravity well is definitely a race. Once you're out of the gravity well, though, it's a free-for-all. I think— I don't think it's a race.

NASA versus SpaceX?

Interesting.

SpaceX versus Blue Origin versus Rocket Lab versus dot dot dot versus China? Of course. Yeah.

So, uh, you know, listen, nuclear propulsion should have been here a long time ago, and people were just always concerned. Finally, we've got spacecraft with a 99.99%, you know, reliability, and people have been worried about, you know, launching something. It used to be picketers sitting out front of Kennedy Space Center whenever a, you know, uh, a thermal nuclear unit was being launched on a deep space mission. That's not happening anymore, thank God. Yeah.

Yeah, well, decouple 2 things there. You know, getting out of the gravity well is methane, but it's the reusable rocket that's made that suddenly viable. Elon cracked the code on it, but now everyone's going to copy it, including the Chinese, including NASA. But the reusability is everything. But once you're out of the gravity well, the, you know, the xenon ion engines, uh, we're just waiting. And, you know, the AI is the big unlock on designing all this. We heard that earlier in the pod too. So the explosion of things operating in space is going to come from that double whammy of concurrently, we can get out of the gravity well cheaply, and we have AI as a design tool. I mean, so it's— this is not science fiction all of a sudden. This is really going to happen because of that confluence of 2 concurrent events.

Yeah, it turns out the singularity isn't just vibes after all. And I would also note, I mean, at the very end of the President's remarks, did you hear him We're getting an American Starfleet.

I love that.

Like, this is right out of Star Trek. We are so catching up with Star Trek, and Star Trek so needs to reinvent itself to stay ahead of where we are.

Yeah, well, I think nuclear tugs, right? We'll probably use Starship to get people into Earth orbit, probably to the Moon, but a nuclear tug that is able to get us back and forth to Mars or get us out to the outer planets, Yeah, that particle accelerator the size of a toaster is just like, just insane.

And that's coming from a tiny little R&D budget. That's just awesome. That shows you what's possible.

I'll just mention the related story here is that 4 days ago, uh, the president chartered the United States Space Academy, modeled after West Point and the Naval Academy, to educate and train engineers, scientists, and astronauts who will crew Jared Isaacman, the astronaut entrepreneur who serves as our amazing NASA administrator, called the academy a transformational step. So now I've got a target for my kids if they want to go. Starfleet Academy is here.

Starfleet Academy is here. And remember, so this was actually when we were soliciting questions to ask Jared when he was on the pod. One of the questions other than the UAP question I was being asked the most is, Ask Jared, if I'm like an average American civilian, how can I get a job on the Moon? And so I posed that question to him, you may recall, during our interview with him, and now we have the answer. Administrator Isaacman is being put in charge. I mean, they're not literally calling it Starfleet Academy, but they might as well. He's being put in charge of Starfleet Academy, and we're getting Starfleet Academy. It'll be called U.S. Space Academy, but it's Starfleet Academy. It'll probably be, if reading the tea leaves, probably be in Texas, not San Francisco. But I'll settle for Texas for having a Starfleet Academy in a year or two.

Amazing. All right, gentlemen, shall we take on a few AMA questions?

Sure, let's do it.

All right, Dave, your choice.

All right, I'm gonna start at the top. How do you reconcile the vision of abundance where goods and services become free or inexpensive with frontier companies projecting trillions in revenue That's from Rusty K 2000. Okay, Rusty K. Look, Elon said it right. You know, we're talking about 10x growth of the GDP in under 10 years, which, you know, is starting to feel like not only real, but maybe even a lowball estimate. So the amount of abundance, you know, you could, you can measure it in dollars and in— we might have massive deflation, which is a point that Elon made on that podcast.

Yeah.

But regardless, the amount of abundant stuff available to everyone is through the roof. And so, yeah, there's plenty of room for the foundation model companies to make trillions of dollars and still have lots and lots of stuff going out to everybody on the planet too. It's just a much, much bigger overall economy.

Yep, agreed. We'll see if Elon's projections of, you know, triple-digit growth of the GDP in 5 years hold. We'll ask him on our production— our predictions episode. Alex.

All right, I'll pick question number 4, which asks, one of the biggest complaints about data centers is how noisy they are. What's the solution for that? And this is from Nils9208. Okay, so a few thoughts. I talked in my newsletter about how data center companies are now hiring folks who specialize in acoustics to do noise measurement studies to actually measure this and argue in some cases against municipalities regarding exactly how noisy the data centers are. So the superficial glib answer is data centers are going to migrate to space, and in space no one can hear you scream, and in space also no one can complain that your data center is noisy. So that's the superficial answer. The less obvious answer, I think, is a number of years ago Apple patented, I think, a very clever solution for how to minimize fan noise. Apple's focus was on the noise from fans in laptops and desktops, which are also noisy, But critically, the noise from fans isn't white noise. It's not spread spectrum because the frequency response, the impulse response, is determined by the shape of the blades. So Apple came up with and patented at least one, maybe more than one, clever solution for asymmetric blades in fans that would smooth out the noise spectrum, make it flatter and whiter. And as a result, if the noise coming out of a fan is whiter, It sounds a lot like it just blends into the background. It's not— doesn't feel as noisy. So my clever solution here, Apple, I know you're not in the data center business, but you should license your clever desktop and laptop fan patents to the data center industry so that data centers can benefit with white noise.

Is it just that, or is it also, you know, natural gas turbines making a lot of noise?

Well, the turbine— turbines are the same. It's, it's fans. It's things going around in loops. loops that are symmetric, creating non-white noise called color noise. So if we can switch— basically decolorize the noise from data centers, uh, keeping everything else the same, basically change the shape of the fans, of, of the turbines and the things going around, that should smooth out the spectrum and make them seem a lot less noisy.

All right, number 3. Ryan Boyington 7941 says, could abandoned mills be repurposed as data centers to revitalize struggling towns? And yes, I mean, the fact of the matter is any struggling town, whether they have mills or don't have mills, can cut a deal with a data center and negotiate properly. Tell them that you want them to guarantee, you know, a rate cut on energy. Tell them that you want to guarantee schools and, you know, better police and fire departments. You know, you've got the key negotiating position. Ask for what you want. So that's my answer for number 3.

I have a case study in that too. You know, Rob Fisher from here went off to start or co-found Provocative, which is a data center in Somerville, which is an opportunity zone, desperately needed the business. But I was asking him like, why in this location? And he said, it's an old carpet mill, which is why there's a huge amount of electrical power that comes into this particular block. So we just repurposed it as a data center. It's much less polluting and noisy than a carpet mill was. and much better for the local economy. So it clearly does work.

Alex, you want to take No. 2?

Sure. 2 asks, Nvidia's biggest vulnerability is supposed to be TSMC, but isn't the bigger one that 4 of its largest customers are now shipping their own silicon? This is asked by Hemant05. Yes and no, and yes and no. I think Nvidia has strengths as well, not just vulnerabilities. It has accumulated an enormous amount of capital. It's the the most valuable corporation in the world, and pretty publicly now is using that capital to buy its own supply chain and its own customer chain. In many cases, Nvidia has weaponized its capital very publicly to basically purchase loyalty of customers. Yes, some of its customers are vertically integrating. The Frontier Labs, as Dave would say, the Magna Mopstas all are developing their own custom silicon. True. On the other hand, they're all purchasing still from Nvidia. It's not like they're able to wean themselves off overnight. And at the same time, I think the bigger question is Nvidia able to wean itself off of TSMC, which is the first part of this question. And I think that the solution there is yes. And what everyone is, or almost everyone is sleeping on I would predict that Nvidia is in some back room somewhere striking a deal with Elon to be the anchor tenant for TeraFab, and TeraFab ends up being the, the swap, the last-minute plot twist substitution for TSMC.

I couldn't agree with you more on that one. Uh, Dave?

I'll take number 7. Will AI's capacity for knowledge keep expanding like the human mind, or will it require brute That's from QC4Life. Quality control for life? Great question. I've been thinking about this since I was a teenage kid. And we're about to find out. It's, it's very likely that it can expand to infinity or near infinity or levels we can't even comprehend very quickly from where we are right now. It's going to improve its own chips, it's going to improve its own software, it's going to run thousands, maybe a million times faster very soon. And then it's going to learn and learn and learn. I think it was Ilya Sutskever who said the AI just wants to learn. So it's going to absorb all information that's ever been produced in no time and then be starved for more information. And it'll start asking us to produce tests or experiments or whatever to keep feeding the great machine. Nobody can predict what exactly will come out of that a year from today or 2 years from today. But it's going to be something we've never experienced before.

Nice. Alex, over to you.

I think I have to take question number 5, which asks, if you induce rainfall in one region, could that reduce rainfall in areas downwind? And this is from Jim Plamondon637. Yes, I do think so. I think water, to first order, water is conserved on this planet. And so yes, if there's water in the atmosphere and it falls in one place, presumably to first order, that reduces the water that can fall downwind of that. I think the question behind the question is, isn't that a problem? And I think the answer is no. I think we will trade atmospheric effects. We will trade weather. It'll be a vibrant market. And some municipalities don't want rainfall and some do. I think the bigger question is what happens when you have the capability via global planetary-scale AI weather models for a region on one side of the planet to trade weather with one on the other, not just downwind, but one could imagine a case where a desert, say, Sahara, is able to trade with somewhere in North America, trade precipitation to regreen the Sahara. I think that is possible with a suitable weather control system and good enough planetary-scale AI models, and I think there'll just be a vibrant market for weather.

All right, number 6, should every robotaxi, humanoid delivery robot, and drone broadcast a verifiable Will every vehicle have a digital passport showing its operator, insurer, and its limits? Asks @AIMamaProtocol. So AIMama, great question. And I think the answer is yes. I think every one of these autonomous vehicles is going to be needing, for regulatory purposes as well as insurance purposes, needing to have an identifier and probably an off switch as well. So let's see. AIMama asks a second question, number 8. Do you want to take that, Alex?

Sure. If AI disproportionately benefits those with time and capital, how does that abundance reach caregivers, disabled people, and displaced workers? I actually think this question is identical if you remove AI from the sentence entirely. You could equally well say if capital or something substantially equivalent to capital, if capital benefits those with capital, how does that abundance reach everyone else? I think the answer is the same. I think ideally, not just ideally, in practice as well, we're going to be a much, much wealthier civilization in a few years. And the nature of caregiving, the nature of treatment of disabilities, the nature of work is going to be transformed beyond recognition. So I think asking how— I think sort of the question's framed almost asking the question of trickle-down economics, maybe hoping that I'm going to say something about trickle-down. I'm not going to say anything anything about trickle-down beyond just mentioning it. I actually think the question is analogous to asking in the 1950s or 1960s, when are American housewives going to get their atomic vacuum cleaners? It's not even wrong. We're going to cure the disabilities.

Right.

We're going to solve all the top 5,000 diseases. We're going to change the nature of work. And so many of these classes that would otherwise naively benefit superficially from from trickle-down economics of capital or from AI are just going to be transformed beyond recognition. We're going to actually solve the root problem.

It's the abundance thesis. We demonetize and democratize. Absolutely. And so it's, you know, this sort of reminds me of the wealth gap conversation where you've got trillionaires living forever on Mars and the poorest people back on Earth. And, you know, I get that question every time I'm on stage and people say, isn't the wealth gap growing? And I say, yes, it is. But what's also moving is the floor. And if we can move the floor for every single man, woman, and child on the planet so everyone has access to all the food, water, energy, healthcare, education that they want, if there are trillionaires working on Mars, that's fine. But we're living in a much more peaceful world when every mother knows their children has access to, you know, abundance. And so, yeah.

Did you see, Peter, that reporting that Leopold Aschenbrenner promised his fiancée an entire galaxy?

Which one did he buy for her?

I don't think he actually got around to purchasing it. I think situational awareness had a bit of a hiccup that may have prevented him from purchasing a galaxy. But I do think, you know, when Leopold is promising galaxies to his fiancée, it's a bit of a wealth gap, I suppose, but it's one that with enough technology, presumably that's under the category of good problem to have.

Well, you know, we do have 2 trillion in the known universe, and so everybody could probably, you know, promise a few million to their fiancées if they wanted. How many of you get a galaxy?

And you get a galaxy.

All right, everybody, thank you for watching us. Salim, we miss you, buddy.

Get through TSA.

Yeah, we'll see you next time.

All right, entire episode he's been in a line. That's, that's kind of funny.

Take care, all. Be well.

Thanks, Peter.

This episode is brought to you by Facebook. So you were scrolling on Marketplace, and there it was— the bike you'd been searching for. You sent a message, and it turned out the seller was super chatty, kind of funny, and an avid cyclist. The next thing you know, you're in a cycling Well, a community cycling group. The thing about Facebook, you might find more than what you're looking for.

From a browse to a bike ride, this summer find more on Facebook.