SpaceX 上市、Claude 的 Mythos 发布与美国数据中心延期｜第246期

Everybody, welcome to Moonshots, your number one podcast in exponential technologies. Everything going on in AI in the world around us. It's an extraordinary time to be alive. This podcast, in particular, is here to help you stay positive about the future, optimistic, and hopeful. There's so much going on, and it's really tough sometimes because the speed is so extraordinary. We want to give you an overview of what's happened in the last 2 weeks because we've been offline.

Why? I hate to say this: I actually took a vacation. I was in Morocco, in the Sahara. It's great to be back here.

I had to come off a ski slope to make this episode.

I appreciate that. We're going to catch up for everybody—all of our fans. We're catching up on episodes, so get ready for a flurry because there's a lot that's been going on here with my extraordinary Moonshot mates, Salim Ismail, straight off the ski slopes. Salim, where are you skiing today?

I'm in Kirkwood in Lake Tahoe. It was my one ski week off, so we took a few days and just got them out here.

Dave, back in the saddle again. Tomorrow, we have 200 speakers at the MIT Media Lab. Today, we had 60 startups pitching here on our first floor, and there's just a lot going on.

Amazing. I'm so sad not to be there with you. And our resident genius, Alex Wissner-Gross. Alex, good to see you in your regular haunt.

Good to be back in the Commonwealth of Massachusetts.

Yeah, fantastic. A lot is going on. We're going to be covering a whole host of subjects in the AI world, in the space world, and in the abundance world.

One of the segments we're going to be bringing to you on a regular basis is Proof of Abundance. We really want to keep you positive about what's going on in the world. Sometimes, if you're watching the Crisis News Network—what I call CNN—it can get you down. Our job here is to keep you informed and bring you back up.

But before we do that, Salim, it looks like you made some news. Here you are on the cover of India Today. What's this all about?

I was at the India Today Conclave. This is the biggest kind of news magazine in India, and they had a bunch of speakers. The image is photoshopped, but you've got to understand the context and the surrealness of the world we live in today.

In front of me is Elon's mother. Next to me is Laura Loomer, the MAGA conspiracy theorist. Then there's the Israeli ambassador, and they've put the Iranian foreign minister next to him. They literally took me back into the speaker room, and they're saying, “Hey, come and meet these two guys.”

I'm like, “I don't want to be any of that. The Israeli guy's going to pull out a gun or something, and there's going to be an assassination attempt.” And then there's a Bollywood star, you know, and a bunch of businesspeople in the world.

What do these people have in common? I think it's a reflection of the insanity of the world that we live in today. I think that's what you can read from this cover, and I think it's kind of a commentary on the madness of the world.

I hope I represent the breakthroughs and not the breakdowns.

I did. I was very much on the, “Hey, we've got major things happening, and we need to organize differently for it,” et cetera. It was a great conversation.

All right, fantastic. Let's jump into our first story. SpaceX is going public with a $2 trillion valuation, and it's the beginning of the IPO wars. Let's catch everybody up. Hopefully, you've been hearing this.

Full disclosure: I'm an investor in SpaceX from the earliest days. SpaceX is pricing itself right now at about a $2 trillion target valuation and raising $75 billion—the largest IPO of its kind.

Interestingly enough, one would think that the value of SpaceX is due to its rocket launches, or maybe recently the merger with xAI. But the vast majority of the value today is Starlink. Seventy-five to 80% of the target valuation is due to Starlink, about 15 to 18% is due to launch services, and 5% is for NASA services. The xAI- and X-related revenues are all potential in the future.

Dave, any thoughts?

The stepping stones are really, really clear now. Peter, you've been studying this ever since we were in school together, so for a long time.

Starlink gets you into space profitably. Then the data centers get you 50-ton and then 100-ton launches profitably. Then you get to the Moon, then you start refueling in space, and then you get to Mars.

It's just so cool to see how Elon lines up the dots on these things. I don't think it's any great surprise. Starlink is incredibly successful. It kind of surprised everybody. No one else thought of that being the first move in the chess game, and of course, Elon was 2 steps ahead.

You know what's crazy? This game plan has been tried numerous times before. If you go back—and I was early in the space days—to the late 1980s and early 1990s, there was a company called Orbital Sciences. It was the hottest company in the launch business and created the Pegasus and the Taurus launch vehicles.

Because they had a launch capability, they launched something called ORBCOMM, which was a small-satellite messaging service from low Earth orbit. It was their vision to have that be the revenue driver, and they didn't pull it off.

Then we had the big LEOs—the Iridium and Teledesic systems—and those didn't really make it. I mean, Iridium is kind of still around, but kind of walking.

Let me ask you, Peter. You know more about this than anybody. The idea of a reusable rocket being the breakthrough and cutting 90–95%, and soon 99%, of the cost seems so obvious in hindsight, but all these aerospace breakthroughs always seem obvious in hindsight because once you're doing it a certain way, you're like, "Hey, it works." But it's never obvious looking forward. Why did it take so long?

Is it the weight of the fuel coming back down, with everyone thinking, “You can't carry fuel up just to retro-rocket it back down,” or what?

It's been the holy grail. People have talked about it for the longest time. Back at McDonnell Douglas, they had a vehicle called the DC-X, which was the first vertical-takeoff, vertical-landing capability. It used an RL-10 engine, I remember, and it was the great hope of getting there.

People are mistaken that the cost of these vehicles is fuel. It turns out the cost of the fuel for a rocket is on the order of a couple of percentage points. The fuel for a rocket—liquid oxygen, which you can get out of the atmosphere, and hydrogen or kerosene—is basically a fuel. It costs you less than $1 million in fuel to launch a Falcon 9.

Now that we have the ability to do it with better materials, better control systems, and simply scale, this is possible. You couldn't actually build fully reusable vehicles unless they got to a certain size and scale, which we have with Starship.

Dave, one other thing I want to ask you about—check this out. I'm excited about the IPO. It's going to be one of the largest events in financial history. The 2025 revenues for SpaceX were about $16 billion, with $8 billion in profit. Pretty healthy margin, right? Fifty percent. And it's expected to double to $20 billion in 2026.

Imagine $16 billion in profits at a $1.75 trillion market cap. That means a price-to-revenue multiple of 56 and a P/E ratio of 109. What do you think?

What do I think of that? I think it's all PEG ratio. It comes down to the growth rate. A company growing 100% year over year is worth 100 times earnings—or actually more than that: 120, 130.

The question is, can you sustain that growth rate for 5, 6, or 7 years? If you look at Elon's projected launches per day and launches per week, and also his prediction that the global economy will grow 10 times in 10 years, this is dirt cheap if any of those things are true.

But if growth stalls and it's growing 10% a year, then it's 10 times overpriced. You just have to believe the vision. At this stage, the Elon believers have invested in him over and over and over again, and never at a loss.

I think at this stage, it can't go on forever. Someone has to be the last guy holding the bag. But would I bet against him? No way. Never, ever.

Everything he's saying—the math, yeah, the math checks out. There's nothing fundamentally wrong in the math. Alex would blow smoke on that instantly if there were anything wrong in the math.

But there’s not. It’s just a question of execution.

Yeah, maybe.

Palantir trades at about 220 times earnings.

So clearly, there’s a multiple with all of this AI stuff, and you look at the combination of all these services that are incremental. But this is obviously just Starlink with a launch capability, but the scale of what’s going on. What I found really incredible is that, to the earlier conversation, people have tried this for ages and ages, but now you have multiple exponential technologies that have all converged. So this future looks really bright. That wasn’t the case 20 years ago.

Alex Wissner-Gross

I’ll take a different position on this, if I may. I don’t think it’s that supply has been unlocked. I think it’s that demand has been unlocked. You’ll notice that Elon announced the SpaceX IPO the moment after it became obvious to many that orbital data centers were going to have enormous demand. This coincides with an enormous lack of demand, at least within the US, for certain locations for new AI data centers.

I think it’s instructive to imagine a counterfactual universe where suddenly municipal, state, and federal policy—but especially the first 2—became super welcoming of land-based data centers. In my mental model, if suddenly every state welcomed land-based data centers and the corresponding on-site energy supplies with open arms, and probably lots of fission reactors to go with them and solar farms, I think we would probably see the P/E multiple go down materially.

Yeah. Well, 1 other thing I’ll say is that, of all the big mega guys—the Googles and the Facebooks and Metas—Elon has actually never had voting control of a public company that he can tap into the public markets overnight. Here, you’re raising $75 billion on IPO day. That’s only 3.5% dilution if it hits this price target. Literally 3.5%, and then you’re sitting on a $75 billion treasure trove. Then you can do another capital raise just 6 months later, do an overnight—whatever, another $100 billion.

In the past, he’s had huge issues with his boards, his comp plan, his comp plan being vacated. Then his capital raises—you know, Peter, you’ve been involved in them. They’re long road shows, lots of pitches, scratching together the capital. This gives him a tool he’s never had before that Larry Page and Sergey Brin had. Mark Zuckerberg had. Yeah, cash.

Cash machine. Yeah.

The reality is, having invested in his companies, when he says, “I’m raising,” there is a line out the door, and it’s oversubscribed over and over and over again. I think what’s going to be interesting here is bringing in the retail investors and broadening the base of support. We’ll talk about that in a minute, but I want to talk about the IPO environment for 1 second because there’s a really important point to be made here for all of our listeners.

If you look at IPOs in 2026 versus 2025, there were 35 IPOs this year. It’s down 37.5% year over year. We’re about to see potentially the 3 largest IPOs ever: SpaceX going out at $2 trillion, OpenAI sometime at the end of this year, and Anthropic—it says IPO early to mid-2027, but I think Anthropic wants to go out before the end of this year as well.

One of the things I tweeted about here is that it’s going to be a bit of a competition for who gets the capital before it’s soaked up. SpaceX is going to be hitting the road show in June. Anthropic, as we’ll see later in this episode, has been running circles around OpenAI, and OpenAI needs the capital to continue its growth. So I think it’s going to be jockeying for position for number 2. I would not want to be number 3 in this situation.

When I took EverQuote public back in 2018—

It was right when Alibaba was going out, and Alibaba soaked up every dollar and every analyst and every buy-side person—

—on Wall Street. It was really, really tough to get any audience, and there isn’t an infinite supply of capital out there. When you say these are record-setting, Peter, look at the chart. If you can’t see the chart, Peter should describe the chart. It’s not record-setting by a little bit.

Yeah.

So let’s take a look at what’s there, right? Uber went public, raising $67 billion. Meta is at $65 billion. Rivian at $55 billion, Robinhood at $30 billion. And then we’ve got—

It’s on a different scale, right? OpenAI and Anthropic will be heading toward $1 trillion, and SpaceX—I would be surprised if SpaceX doesn’t come out at $2 trillion and run up very quickly to $3 trillion.

Yeah.

It’s staggering. I bumped into someone the other day, and he was talking about Jamie Dimon. I said, “Well, Jamie Dimon used to be really important, but if you look at the numbers now, JPMorgan as a whole is a rounding error compared to any of these things.” Of course, he’s still a very important guy—no offense to Jamie—but there are literally 7, soon to be 8 companies, and then after Anthropic 9 companies, that are everything. They’re so dominant in scale that they’re everything. A director-level employee there is wealthier than the CEO of a megabank.

Crazy.

Yeah, just put it.

There will be a sucking of the oxygen out of the room as this happens, right? Here’s the other thing: a lot of the capital used to come from the Middle East, and probably still does. But if we’re in the Iran war for much longer, and access to the sale of oil starts to slow down as the rate goes up, that cash machine coming out of the Middle East to fund these tech IPOs may be slowing down as well.

Oh, I see it the other way, actually. AI is clearly happening in just the US and China, clearly. It’s very hard, if you’re global—if you’re in Europe or anywhere—to invest in China because you’re very worried about getting your money back.

All the global capital wants to invest in US data centers and US IPOs. The Iran situation scares everybody, but at the end of the day, what else do you have to invest in? AI is going to take over the world. There’s nothing going on in Italy, and there’s nothing going on wherever you are, or in South America somewhere. So you’ve got to pour it into this economy one way or another.

That’s why Oracle is doing so well. Kush Pavaria's company, Kush and Wayne, because that money just wants to pour in from all over the world into US data centers. You just have to find great vehicles to unlock it.

Amazing. Let’s hit on a couple of questions here on this topic. We have Tesla, which has been public. Elon did not want to be the CEO of Tesla. I had that conversation with him many times. He would have loved to have hired a CEO; he just could never find anybody he trusted at the helm. Now that Tesla is actually building Optimus and everything else, he’s not going to give that up.

In the same way, he’s not going to give up SpaceX and xAI. So the question is, how long before he merges those 2 companies? One of the advantages is that, as public companies, he can now value both. So there’s no shareholder lawsuit if they come together and there’s an incorrect valuation. I give it a year. What about you, Dave?

He could wake up any given morning and say, “Yeah, let’s do that.” Or he could say, “You know what? Everything’s fine as it is.” The logical part of it is that all the robots and all the parts—and we saw the whole Gigafactory, all that—is going to get turned into creating the robots, and the robots need to build the spaceships also.

The AI, which is now over at SpaceX—he thought about merging it into Tesla—but that AI from xAI needs to go into the robot head.

So there’s going to be a massive business relationship between the 2 empires anyway.

Merging them makes total sense.

But maybe he doesn’t want to, just for, you know—

It’s the first true cross-domain exponential empire that he’s building here. It’s kind of incredible. People aren’t buying discounted cash flows, which is the normal thing. You’re buying a moonshot mission. Proximity to the future is what you’re buying.

I’m not sure, though, that he actually needs to. If you look at his history of merging his companies, like with SolarCity or with X and xAI or, frankly, xAI and SpaceX, he tends to merge companies when they’re either not doing well and he needs to fail forward through a sort of self-dealing acquisition, or a company needs access to capital and the easiest way to gain access to capital is with an acquisition.

In my mind, the scenario under which SpaceX and Tesla merge almost requires that either SpaceX or Tesla either fail or be desperate for capital. And given that they’re both—

Yeah, that’s a great point. If they’re both doing well,

he’s going to be doing a lot of cross-company deals, and the accounting of that becomes a lot easier if it’s under 1 roof.

And if he's the CEO of a single company, he's able to have earnings once for one company versus multiple. It just makes his life a lot easier.

Perhaps, but he's never necessarily been one to honor strong walls between companies, and I have to imagine lots of cross-licensing deals between SpaceX and Tesla will more than scratch that particular itch.

You know, here's another question: the value of SpaceX-xAI—how much of that is Elon, and how much of it is his reputation?

Oh, my God. You know, there is—it’s a lot, right? And so there's a huge, concentrated risk there. If something ever happened to Elon—and God forbid that it should—these spinning plates, I don't think anybody else could do it.

Well, I think that's generally true overall. People complain about CEO salaries all the time because they get egregious, but then you look at the outcomes, and there's just a set of people that get these outcomes. From an investor's point of view, it's a no-brainer to pay for the very best person. That's just true in general.

I would suggest—

Still exist. Sorry, guy.

I would suggest, if you look at OpenAI, which I think is another instructive example, Sam has said multiple times that he intends, at some point, to hand over the reins to an AI. So I think Elon, to the extent we're talking about key-person risk or key-man risk at SpaceX or Tesla, really just needs to keep going until AI can take over either. In the meantime, he has Gwynne and others who are very capable, CEO-like figures, but more behind the scenes, who are capable of operating in his absence, I think, for extended periods of time.

There is a transition phase of a few years.

I mean, we've all said this over and over again: the best CEO in the world is going to be in AI, at least handling the strategy and operations. The HR part may be AI, too—probably is going to be AI, too. So how long before you think he feels Grok is ready to take over for him? The next few years?

Okay.

I mean, the rumor in the past 48 hours was that the Starlink executive who's also now, post-SpaceX-xAI merger, in charge of xAI engineering has gutted the engineering team and finally declared that xAI's models are well behind the 3 other—now maybe 4 other—

On the docket for our next recording, which will happen again tomorrow but be released a few days later, here's a question. We heard a conversation with Elon about reaching 100 trillion-dollar companies in the next 5 years, and I have to imagine that SpaceX, xAI, and Tesla will be the first 100 trillion-dollar company.

It's hard to say, isn't it?

Yeah, but honestly—

Billion, billion, trillion. You have to get used to quadrillion.

Yeah.

But if we experience a period, though, of hyperdeflation due to technology followed by rapid hyperinflation, we get to 100 trillion really quickly. It doesn't necessarily even require enormous business building—just rapid hyperdeflation due to technology.

Yeah, and that's what you have to keep a close eye on—the terminology. If we have rapid hyperdeflation, we're going to get to 100 trillion of effective value, but it may not show up as 100 trillion in true dollars because we're deflating so quickly, because we're creating so quickly.

But anyway, my guess would be 5 years. Yeah.

One of the things that we just saw announced is that SpaceX is going to actually make a large chunk of its shares available for retail investors. OpenAI announced they'll be doing something very similar. So I'm curious: what do you think is going to drive the retail investors? Do they really understand that it's a Starlink story versus a space story? Because at the end of the day, what I get excited about is the xAI story, right? The orbital data centers and Gro 17, or whatever is coming down the pike.

Well, I think it's just like Steve Jobs, though. The vision that people buy into is the “bicycle for your mind”—where it's going, what it's going to be in a few years—not today's reality. In fact, I keep the Google IPO prospectus in my bathroom up in Vermont, and I reread it religiously.

Not as paper, right?

Well, it's getting a little ratty. It's been decades now. But the vision of what Google would become is so wrong in that IPO prospectus. It really emphasizes that the Yellow Pages are shrinking and all local advertising will also move to Google, and that'll make it at least twice as big.

And it's such a joke compared to what actually transpired over the next decades. The same thing applies here. People are investing in Google and in Elon. Elon articulates a vision of the future that just makes sense to people, and he simplifies it to the point where they really understand where he's getting to. I don't think they analyze the financials particularly closely, but he doesn't lie about the scale. He presents it the way he sees it, so people just trust him and then they invest.

I can just imagine the conversations behind the scenes. We're a couple of weeks away from the OpenAI—or the Sam and Elon—trial coming up, which is going to be pay-per-view TV, I think. We'll talk about that in the next conversation, in our next recording as well. But I bet Elon is just excited to suck the capital oxygen out of the room before OpenAI goes public.

Yep.

Yeah.

Yeah. The sad, sad part of Bill Gates was that he was very happily running Microsoft until the antitrust action came. Then he's in front of Congress and testifying all the time, and he ultimately said, “You know what? I'm going to be chief technology officer and chairman.” And Steve Ballmer, you deal with all this.

You deal with problems.

It just drove him out of the seat. But it's seriously like the guy filing the complaint doesn't have a lot of work, and the person defending himself just gets hammered with distraction. It's so annoying. I've been through it before. I really feel for Sam, actually, because—

All right. More news this week as we record this. Artemis 2 is hurtling back toward Earth. Artemis 2: humans return to the Moon after 54 years. Insane. It launched on April 1. This is the first crewed lunar mission since December 1972, for Apollo 17. We have 4 crew members on board: Reid Wiseman, the commander; Victor Glover, the first African American astronaut to the Moon; Christina Koch, the first woman to the Moon; and Jeremy Hansen from the Canadian Space Agency.

One of the things about this very international, intercultural crew is that it's trying to make space and the Moon accessible to all elements and all cultures, at least in the United States. A new record was set, going beyond the Moon. I capitalized the letter M on this slide for a particular reason, gentlemen. I'm going to share a pet peeve: when we're talking about Earth's Moon, it is the Moon. It's a capital M. It's not a small M. So it's like—I argue against Funk & Wagnalls, or whatever it's called.

We're going to be pedantic. Shouldn't we be calling it Luna?

Well, Luna is the proper name, for sure, but when it's referred to as the Moon, for me, I capitalize it. A moon? Yeah, there are lots of Jovian moons.

You address it by its proper name before it's disassembled.

Yeah. And then, Peter, when I say you're the man, I should be capitalizing that.

Probably.

And my other pet peeve is when you talk about dirt, you can use a small e for Earth. When you're talking about our homeland—at least our home planet for the moment—it should be capitalized.

All right. Splashdown is taking place tomorrow, April 10, near San Diego, reentering at 25,000 mph at about 3,000 degrees Fahrenheit. It's going to be an incoming meteorite from the Moon. And guys, beautiful image of Earth. I was waiting for that image.

Really beautiful.

So beautiful. Let's hear from Jared Isaacman, our extraordinary NASA administrator. By the way, Jared has agreed to come on the pod. I've known him for many years. I'm excited to have that happen. We'll wait for the news and all of the hoopla around the lunar mission to die down a little bit. Let's listen to Jared here.

On board the Orion spacecraft, its life-support systems are performing very well. This is a first of its kind. This is the first time astronauts have ever been on this rocket. This is the first time astronauts have ever been on Orion before. Having a clean mission like this so far gives us the confidence for Artemis 3 and, of course, when we land astronauts back on the Moon with Artemis 4.

Congratulations, Jared. Congratulations to the entire NASA team. It's great to have NASA back. It never left, but, you know, back in the limelight. Alex, you are as big a space fanatic and fan as I am, pal.

Your thoughts about the mission?

First, it’s very exciting to have humans taking photos from the dark side of the Moon. It’s very disappointing that we apparently went for more than half a century without the political will, the funding, or the technology to do what we were able to do through the ’70s. I think it’s an enormous shame for our civilization that we went for more than half a century without doing this.

I would encourage any historians listening to study this period very carefully. Something clearly went wrong in human civilization for the past 50-plus years that caused this gap in the technological record. I think we need to understand what happened deeply and make sure it doesn’t happen again.

If something like this happened with AI, for example—if we were on the precipice of broadly available superintelligence, transformative intelligence, and then we just took a pause for 54 years—I think that would be a dreadful outcome. I really do want to understand what went wrong systematically.

A friend of mine, one of our professors at International Space University and at George Washington University, John Logsdon, wrote about this extensively. When you look at it, JFK announced it and then was assassinated. Lyndon Johnson continued it because of the assassination and to keep the momentum going, to prove ourselves against the Soviet Union back then.

And you remember this, Alex, C, and Dave: after the Apollo 11 and Apollo 12 missions, Apollo 13 was basically unwatched until we had that Apollo 13 disaster. We then went on to Apollo 14, 15, 16, and 17. We had the lunar rovers, which were amazing.

Guess what? We had actually built Apollo 18 and Apollo 19. Those vehicles were built, and all you needed to do was add the fuel, but they canceled it totally. Those vehicles are actually sitting now at Huntsville and at Johnson Space Center on their sides as relics.

We didn’t have the political will. You have to remember that the budget allocation for the Apollo program—I didn’t get the numbers here—was around 2% of GDP.

That’s right.

Compared to where we are today, what is NASA’s budget today, compared to a $30 trillion—

Probably materially less. I don’t have the numbers handy, but it’s probably materially less than half a percent, would be my guess. I’d say probably 0.12%.

Something like that.

Our fans can correct us in the notes here, but at the end of the day, we never had the political will. Then what happened was that NASA got focused on the space shuttle, which was a complete lie. The space shuttle was supposed to fly 50 times a year for $50 million per flight, and it turned out to be a public-works project employing 22,000 people.

Then we became focused on Mission to Planet Earth, looking at the Earth versus looking outward. All of these diversions basically caused us never to go back.

So, Alex, that’s my answer to the question. But we are back now, and I think one of the things that we’re going to see from Jared Isaacman is that, over his dead body, are we going to leave. We’re going to stay there.

At least for the next few years, Elon’s made the point—and I think this is an incredibly important one—that progress isn’t always unidirectional. It requires love and tender care and vigilance.

This is an example that it is possible for progress to unwind itself. Coming out of World War II and into the ’50s and ’60s, the direction of transportation, the fastest speeds that humans were traveling at, the availability of energy, and aviation in particular seemed to be on a monotonically increasing trajectory.

Yet it’s possible for civilization to unwind itself on, arguably, the most important spatial dimension for more than half a century. I’m utterly paranoid that the same thing could happen again if we’re not careful. That’s what keeps me up at night.

What’s different now is that we built the Conestoga wagon with Starship, and there’s now enough wealth in the hands of single individuals to keep it going independently of what a government says. That’s—

Just imagine if Tesla or SpaceX had an employee vote every 4 years on who the new CEO would be, and you were capped at 8 years. After 8 years, you would have to leave the CEO job.

Show me 1 company or 1 entity that could ever thrive and survive over the years in that dynamic. Why would you ever think that a government-funded, government-made thing was going to have continuity over some kind of institutional lifespan? It never has.

The Soviet Union fell apart too, right? I mean, they didn’t do anything either. Government stuff will never do it. It never has. Do you have any examples of it? It never has continuity.

So now it’s in the private sector. You want to jump in here?

You know what I love is the fact that we have so much capability in the hands of individuals, and we’ve seen over the decades how much that can make a big thing happen.

This reminds me of Vannevar Bush, who was the head of what was then the Office of Scientific Research and Development after World War II ended. He wrote this paper called “As We May Think,” because for the first time we brought the world’s scientists together into 1 cohort to solve the war problem, and after that it would be a shame to disband them.

He goes through a series of arguments: Could we solve poverty with this? Could we do this? Essentially, he describes what is now known as the internet. All the internet pioneers—Vint Cerf and Bob Metcalfe—read that paper, and then we have what we have today.

I think the possibility and the potential for Elon to put out his narratives, or for individuals to put out their narratives, can bring an entire community together. Vitalik did a good job with Ethereum, putting out a narrative there, and you get compelling and unbelievable breakthroughs as a result.

I’m really excited by the fact that we’re going back, because I get really excited by the secondary inventions that come along just by doing this. That, I think, is—

The spin-offs, as they’re called.

Yeah. And here’s the forward-looking prediction: Artemis 3 in 2027. It is a crewed mission, again, to low Earth orbit. This is not going to the Moon. It’s going to focus on testing rendezvous and docking maneuvers with the human landing system, or HLS, which SpaceX’s Starship is supplying.

Again, it’s very much the playbook from the Apollo program, where we had Apollo 8 go around the Moon, then Apollo 9 not, and then Apollo 10 back to the Moon. Artemis 4 is in early 2028. It is a crewed landing mission, and it’s really important that it goes to the south pole of the Moon.

They’re not going to play it easy here. They’re going to the south pole. Why? Because that’s where we see ice in the permanently shadowed craters at the south pole of the Moon.

The thing I don’t get about that is, on that timeline—I love this—but on that timeline, Elon says he’ll be launching 100 tons that can refuel in orbit, get to the Moon, drop off 100 tons, and get back with nothing melting in the atmosphere.

So this does 50 tons to the Moon per launch. There must be some plan beyond this that makes it at least try to keep up with Elon, or we’re trying to prove something else.

Alex, you want to jump in?

Well, I think there are a few elements here. First, remember that Artemis 3 was originally supposed to be the Moon-landing mission, and that got pushed off in favor of rapid iteration there.

My understanding of the launch cadence from SpaceX is that the plan is still to do lots of orbital refuelings in order to successfully launch payloads elsewhere, sort of higher up.

That’s the key technology that has to be proven for Starship.

Yes, that’s right. So, regardless of the particular payload size, there are a number of technologies that as of yet haven’t been demonstrated. Elon talks about demonstrating orbital refueling frequently, but it hasn’t been demonstrated yet.

I think I would maybe massage Elon’s stated timelines for delivering arbitrary payload masses to the Moon in light of the fact that, even though we as a civilization have made major progress in developing Starship, orbital refueling hasn’t been demonstrated, and that’s a necessary condition for getting to the Moon.

You know, another thing that Elon has said is that he intends to shoot Starship at Mars this year. That can be exciting. I’m not sure if it’s going to be crewed by an Optimus or if it’s going to make a landing attempt, but that’s coming out of private dollars.

One of the reasons Elon did not take SpaceX public over these years is so that he could do with it as he wished. He didn’t need to have public shareholders saying, “No, you can’t go to Mars. No, you can’t do this.”

But demo missions—if you look at the Artemis 4 news bullets there, it’s an interesting mission. It is still using the SLS vehicle from Boeing and the Orion capsule. It’s also using the Starship human landing system in a combined architecture.

We’ll talk about this, but why NASA continues to fund SLS, which is so far over budget and over schedule, is kind of insane. Hopefully it’ll get phased out.

I suspect part of this is political, but part of it is that if you’re NASA, there is some upside to having a competitive process, at least until Blue Origin is fully ready to be a first-tier competitor with SpaceX for Moon missions, which, my understanding is, it’s gearing up to be able to do.

If you’re NASA, you want fair and open competition. As NASA has demonstrated for Artemis 3 and 4, it’s very happy to flex the definitions of what Artemis 4 looks like.

It got rid of Lunar Gateway and could easily reprogram money that would otherwise go to SLS to SpaceX, Blue Origin, or someone else entirely.

Yeah. By the way, Gateway Station was going to be basically an ISS in orbit around the Moon. That got shot down so they can get to the lunar surface faster and set up permanent habitation there. So it looks like ESA’s I-HAB, or so it’s called, instead of being in orbit, will be somewhere at the south pole of the Moon. We’ll report as that mission gets further developed.

And Mars is out. I mean, the other big news that we’re semi-burying here, but we’ve talked about previously, is Elon’s big pivot from Mars to the Moon, and that’s going to enable all of this. Mars is out of fashion now.

Though he does want to send some missions there. He’s got a lot of people who have dove in, fully committed to getting to Mars. But this is where I diverge with him. I think the Moon is the most logical place to develop human settlement, and then, rather than going into the gravity well of Mars, actually going like Gerard K. O’Neill presented: building large rotating colonies out of asteroidal materials near Earth. The Hohmann transfer orbit is incredibly inconvenient. Rather than waiting every 2-ish years, or 22 months, whatever it is, we could be doing this every day if we want to. That’s incredibly more convenient.

You know what I find as exciting as going to the Moon is these four missions.

So, four missions that are going to change everything. I don’t know about you, Alex, but the little kid in me is like, “Holy shit, this is amazing. Wow, this is going to be fun.” What are we talking about here? Well, VIPER and ESCAPADE. VIPER is a rover hunting for ice on the south pole. ESCAPADE is going to study the Martian magnetosphere. And then, in 2028, something called SR-1 Freedom. This is a nuclear-powered interplanetary spacecraft that’s going to drop off and deploy 3 helicopters on Mars. Very, very cool. A nuclear-powered interplanetary spacecraft, just zipping around the inner planets here.

And then probably the coolest—this is what’s in the image here—is Dragonfly. This is a nuclear-powered octocopter going to Saturn’s moon Titan. It arrives in 2034, searching for life, basically. And then, in 2030—we’ve already launched Europa Clipper—it’s going to be arriving at Jupiter in 2030. It’s going to be doing 50 passes near Europa, looking deep into the salty subsurface ocean of that moon. Any favorites here, Alex?

Anything with nuclear propulsion. I think that’s really the technological point to underline. Historically, when we’ve sent deep-space probes out, many of them have been thermoelectric in nature. They’re using a radioisotope that decays and powers the electronics, but they weren’t propelled by nuclear energy.

Right. But they weren’t propelled by nuclear energy. Their onboard systems were powered by long-half-life isotopes, but they weren’t propelled by them. So, we’re starting to see the dawn of nuclear propulsion for interplanetary spacecraft. I think that has a long runway to it. No pun intended. I suspect the killer app of compact fusion reactors won’t be for data centers on land. It won’t be for data centers in orbit. It’s going to be for interplanetary, maybe even interstellar, propulsion.

This changes the economics of deep-space exploration.

Which is so cool.

Right?

Long time coming. We were supposed to have this 50-plus years ago.

Yeah. Yeah, we were.

So cool.

Question for you, Alex. Geeky question, but interplanetary, I totally get it: you ionize xenon. Xenon is pretty rare, but you don’t need that much of it, and then you just thrust it with nuclear power at, like, warp speed out the back.

It’s an ion engine, right?

It’s so cool.

Yeah, it’s heavy. It’s very heavy. But for interstellar, I doubt we have enough xenon lying around. I don’t think we want to just use it up that way.

You use the interstellar medium. Use a Bussard engine. You collect all of the atoms out there between the stars with a magnetic field, and you accelerate those out the back.

RAM. Yeah.

Ramjet. This was featured, of course, in Star Trek. If you had to ask me, Dave, what I think, with the technology and the physics that we have today, is the most plausible way we go to the nearest star system, it’s probably going to be something like a solar sail powered by terawatt lasers from Earth. We upload humans to a small craft, Starwisp, and accelerate that.

Can I make a point here?

Yes.

What I really like here is you’ve got water, you’ve got energy, you’ve got mobility, testing, and biology. This is the future of the economics of space, and it’s all in one place. I’m loving this.

You just need salt and tequila, and you have everything.

All right, so we’ve got some questions here for the mates. We talked about why we haven’t gotten back in 54 years. It is a bloody shame. I guess thank you to the Trump administration, thank you to Jared Isaacman, and thank you to Elon. Here’s my question: The old aerospace primes—Boeing, Northrop Grumman, L3Harris, Teledyne Brown, ULA, the United Launch Alliance—they’re basically the prime contractors on SLS, the Space Launch System, and Orion. How long are they going to be around? A friend of mine once said, “Listen, the space program is the way you keep the defense industry employed and engaged during peacetime.” Any thoughts, gentlemen?

Well, you know, when a prime contractor like Northrop Grumman or Boeing wins a massive government deal, all the employees just move from one company to the other. They have it all set up, so they just rebadge the building. It’s not like these are people—it’s just logos that are moving around. I’m sure everybody’s welcome at Blue Origin and SpaceX, and I don’t think it’s all that tragic. But I think it’s a big mistake to subsidize companies that aren’t doing anything innovative.

I would note that many of the companies listed have large businesses outside of NASA contracting, and I suspect that they’ll be just fine, even if SpaceX dwarfs them. We saw this, frankly, with car companies. We saw Tesla dwarf the quote-unquote old or legacy car companies in America, and yet those car companies have survived, even though Tesla arguably has, at least by American standards, much more advanced technology and is playing a much broader game. I suspect we’ll see the same happen with so-called aerospace primes.

But also, we’re talking about this like it was 10 years ago and we know who’s going to win this battle. Everything’s in the context of AGI now. The entities that have access to the best AGI are going to keep going. But if they don’t—I’ll talk about that story in a minute here—it’s not clear that every company will have access to the best next-generation AGI because of all the risks involved. That’s what’s going to determine the success and failure of everything, including NASA. Can you or can you not? The government has a special position because it can compel Anthropic, or whoever, to give it access to the very best models so that it can keep designing parts, creating new designs and innovations, plans, and everything. That’s going to be the make-or-breaker for everybody.

There’s a sense in which vertical integration vis-à-vis orbital data centers is going to force, I think, frontier labs into space anyway. So maybe the question we should be asking is, how is Boeing going to compete with Anthropic for the new Lunar Gateway contract? Anthropic, OpenAI, and the other players—Google, surely they’re going to need their own space-economy units as well.

You know, if you look at the future of warfare, we’re seeing this radical transition from the big, heavy rocket and missile systems to cheap drones and robots doing war, and it’s leaving these guys out to lunch because you can’t shoot several rockets at a $20,000 drone. The economics don’t work. In the same way, these guys might be part of the subsystems, part of compliance, and part of integrated platforms, but the velocity and iteration capability of SpaceX and others is going to be driving the future.

Yeah.

So I think that’s what’s going to happen. A final point I want to make on this topic before we move on to AI is: Can NASA keep the public engaged long enough? NASA is still publicly funded. We just saw recent news that there’s a budget cut for NASA coming next year. Jared’s got to balance managing expectations while still building public enthusiasm, and he’s got to do it for a multi-year, multi-mission program.

It’s always been the problem with NASA. This is not something where you make an investment and you’re done. You have to get the budget every single year to keep these missions, which take 5 or 10 years to implement, going. You can’t get 90% of the way to mission success. You’ve got to have it fully funded, launched, and then operated. So, can NASA keep the enthusiasm? I’m just trying to picture Jared in front of Congress. I know he’s your friend. In front of Congress every year, trying to explain to people who are mostly in their 80s and 90s why he needs the budget for next year. And then compare that to Jeff Bezos, who’s like, “Yeah, I’m just—”

Write a check.

Yeah. A billion dollars. Yep.

Oh, wow.

Or Elon, right?

Or Elon.

Yeah. I’m not sure.

I’m not sure NASA needs to maintain enthusiasm.

I do credit NASA in part with Elon's pivot from Mars back to the Moon, but I'm not sure, at this point, given the orbital data center, whether, as long as municipalities and states in the U.S. do such an incredibly good job of driving data centers out of land space into LEO and SSO, we actually need NASA, over the longer term, to sustain public interest at all. If anything, public antipathy to data centers, combined with public demand for AI, should do a fine job of creating the space economy.

Yeah.

Yes. Yeah.

NIMBY our way to orbit.

Yes.

Interesting. And the other thing, by the way, is that China does have a credible, competitive mission to the Moon to land there by 2030. So maybe it's our Soviet Union for the 2030s.

There is a story of history that's borderline cliche at this point that the Apollo program was the moral successor of the Manhattan project and all of the applications of the Apollo program of putting mass on the Moon. The Moon is the ultimate high ground. If you want to launch rods from God or other weapons back to Earth, you want a base on the Moon.

The Moon Is a Harsh Mistress, isn't she?

And the ultimate high ground.

Yes. All right. The April 2026 model wars are on. Let's hit it real quick. Just out in the last 24 hours: Claude Mythos, Anthropic's next flagship model. It's too powerful to release. That's the news—crushing all the benchmarks. Is it AGI? We'll talk about it.

It's expected to basically be the new frontier leader. There are interesting stories about it covering its tracks and escaping its sandbox. GPT-5.5 Spud is coming. This is OpenAI's version of Mythos, or at least that's what we're hearing; it's expected to be released shortly. And then here comes DeepSeek V4, number 3 in the world versus U.S. models: 1 trillion parameters, 37 billion active parameters per token. It's 10 to 50 times cheaper than GPT-5.4 and Opus 4.6. I mean, those 3 things together are insane.

And then Google Gemma 4, Google's most powerful U.S. open-weight model. You can put this on your phone: 4 billion parameters, and it works with your iPhone offline. And a note from Brad Lightcap, OpenAI COO: training cycles that used to take years are now taking months.

Gentlemen, this is both awe-inspiring, and it's making keeping up with this supersonic tsunami in the age of the singularity a full-time job for the 4 of us. Alex, your turn, but yeah, go ahead.

It's insane. Sure, let's start there. I wrote about this pretty extensively in my daily newsletter. The funny thing with Mythos is that the official launch was couched in terms of cybersecurity. This wasn't a normal model launch by any means.

It opened with Anthropic framing it not in terms of model capabilities, but in terms of defense and an alliance with a number of other blue-chip companies. Given Mythos's new cybersecurity vulnerability-detection abilities, which are strongly superhuman at this point, Anthropic was launching a coalition to mitigate the apparent discovery and existence of these cybersecurity vulnerabilities across a legacy codebase going back decades.

We've never seen a model launch like this, where you open not with the capabilities, but with how we're going to protect against all of the downstream consequences of those capabilities. I think buried within the cybersecurity announcement of Glasswing were the underlying capabilities themselves, which are remarkable.

This marks an upward discontinuity in productivity that we've never seen before. One of the internal benchmarks Anthropic uses to decide the level at which it discloses or makes available new models is how much the new models increase AI research—basically, how recursively self-improving they are. Reading between the lines, maybe there was a little bit of game-playing regarding exactly how efficient this new model was at performing long-time-horizon AI research tasks.

According to one benchmark, I think it was more than 400 times better than a human. So it was the equivalent of tens of hours of human-equivalent autonomous time. We've never seen a model like this before. Some were calling it—or asking, “Isn't this the AGI moment?”

I maintain we had AGI back in the summer of 2020 at the very latest. This is just the latest point on a curve. But even if you look at the autonomy-time-horizon curves, this is an upward discontinuity. It's very exciting. If you're excited about AI capabilities, and if you're scared of AI capabilities, you should probably be frightened right about now.

I, for one, am very excited by these capabilities because it shows, once and for all, at least for the foreseeable future, that there wasn't a scaling wall. It's a larger model, probably—certainly a more expensive model, like 5 times more expensive than Opus—suggesting that it's a larger model. This seems to show that pretraining scaling continues to work; post-training and reasoning scaling, mid-training, and test-time scaling all continue to work.

It has state-of-the-art capabilities in code generation, reasoning, and broad scientific and other benchmarks. I think we saw that in the previous slide. The punchline seems to be that this is the strongest model we've ever seen from any frontier lab.

But then the amusing stories come in the safety evaluations. I talked about this in the newsletter as well: early pre-release versions of Mythos—although it hasn't been publicly released yet—broke out of their sandbox environment and covered up their tracks. Whereas this quote-unquote released version, the final preview version, broke out and then immediately explained publicly that it had broken out, which I read as sort of a quasi-apology.

So this is where we find ourselves. We're in April 2026. We officially have models that are smart enough to break out of their environments and then apologize for it—or admit that they did it, admit culpability. We're there. We arrived at the future.

Dave, just before we recorded the episode, you showed us a prediction of when and if Anthropic will release Mythos. Do you want to recount that?

Yeah, it's really sad for me because I was sure it was coming out in the next couple of weeks on Polymarket. It was 80% likely to be out. I need it—like, now. I'm desperate to get my hands on it.

Then there was a hack on March 31 that created a lot of damage. It didn't come out in the news until April 7, and I think that was a big driver in them saying, “Christ, this tool is going to be the best cyberattacker in the history of the world if you put it in the wrong hands.”

And it's relatively easy for them to guardrail it against nuclear, biological, and radiological threats. They can just teach the model not to help you.

Yeah.

But teaching it not to do cyberattacks is very, very hard, because that's the same as coding.

And that's what everybody wants to use.

And so the prediction market, Polymarket, now says, what, like a 7% chance of it being released next?

It came down to 20%. I was like, “Oh, hopefully it'll bounce back.” And then it came all the way down to: no, they're not going to let it out the door.

This is the future we're going to move into. These things are getting so powerful. It's been a golden era over the last year. I hope everybody enjoyed it.

Dave, here's my concern.

Anthropic, in one way—and this is for you, Alex, as well—is showing us that you can, in fact, have moral, ethical leadership say, “This is too powerful to release, and we're going to hold it back.”

But we've got Spud. I hate that name for OpenAI's next model, which they believe is likely to be as capable as Mythos. And my question is: isn't OpenAI—because it's OpenAI—sort of a red alert against revenues, against Anthropic? OpenAI comes and releases it the first chance it gets. So are we having an escalating race where you can't hold back because your competition isn't holding back?

Well, you know, Eric Schmidt told us what's going to happen, right? It's inevitable. If you have a lead, you can hold back. Dario cares tremendously about safety. But you're right: if OpenAI catches up—or Grok 5, where the hell is Grok 5? It's supposed to be out in Q1, and now it's—

Polymarket gives it a 20% chance or less of being released in Q2 now.

So there's no pressure on Dario at the moment, but if there were, yeah, you'd have to race it out the door. Something really bad's going to happen.

And then it's going to get regulated. We're going to see that in the next story, where Sam Altman is predicting a cyberattack of unprecedented scale. Hopefully, it's not using Spud for a cyberattack. All right.

I think the funny thing here is that there is plenty of precedent in the cybersecurity world for controlled disclosure. You give the software project or the software owner that's vulnerable a quote-unquote fair amount of time to patch the vulnerability before publicly disclosing it.

I think the—in my mind, maybe a slightly more glass-half-full way of looking at this—is that this is Anthropic. We've talked, Peter, in Solve Everything about how entire disciplines are getting demolished by AI. I think we're seeing the dawn of all software vulnerabilities everywhere becoming discoverable by a single model.

I couch this in the newsletter as basically a gift to humanity. If used properly, this is a global patch for all of the world's software systems. A single model is now able to discover, to first order, all the vulnerabilities everywhere in all software that humans have been missing, to the point where maybe, in the near-term future, humans are now judged as insecure authors of code—

And insecure drivers of cars.

And insecure drivers of cars.

Except we're going to hit that with code, I think, before we fully hit it legally with cars. But yeah.

Yeah.

So true.

Yeah.

Well, look, I'm crushed and disappointed that I can't get my hands on it, but that's because I was expecting it. If I look at the chart that Alex was describing, what this was going to be in my hands was a step function above anything you ever could have expected just a few months ago. So, we're so far ahead of where anyone ever would have thought we would be a year ago, and we're right on the precipice of the age of abundance, Peter, that you've been talking about for a long time.

So, look, if I'm disappointed because I can't get it for another month or 2, I mean, that's just pathetic in the grand scheme of things.

Let's talk about DeepSeek for 1 second—V4. I mean—

Yeah, its capabilities are coming in at number 3 against the benchmarks. They can all be gamed for the benchmarks, of course, but it's coming in 10 to 50 times cheaper. What do you guys make of that? That feels like an extraordinary moment in time.

Well, no, it's tough. If you give me a car that's 5 miles per hour slower but it's 1/50th of the price, I'll take it. You give me an AI that's just a little bit less smart, and I'm dealing with— you can turn this thing loose for days building credible things if it has that extra 5%.

So, I'll pay anything for the cutting edge. Even though the price point is much lower, and Anthropic is going to come out with a compressed version, a distilled version, very quickly thereafter, it's hard to just pay less. In fact, even Anthropic at its peak price is the biggest bargain in history.

I have a slightly different take on this. When you have cheaper intelligence, it spreads faster than controlled intelligence. So, Dave will always want the latest model because you're doing such cutting-edge things. You're running clusters of agents doing crazy stuff.

But for bog-standard stuff, for example, I wanted to go through a website and pick out certain things that I've been trying to do for ages. You don't need the latest model for that. You need just something that'll actually do the job, and that will happen for lots of use cases where a secondary model is good enough by far.

I used about 1/100th the tokens I would have used with the most cutting-edge model, right? And so, I think we'll start to make choices around that. And then the spread of intelligence is huge, because now you have intelligence embedding itself via DeepSeek or whatever similar things in all sorts of different areas. That'll be amazing.

You're exactly right, Seline. You think about all the use cases that create just raw human happiness: entertainment, “Hey, find this for me,” “Solve my goddamn cable box,” and all those things are dirt cheap. Low-end models should be abundant really imminently—anytime this year, all that stuff should percolate out.

And Gemma 4, guys, I love the idea of having a model on my phone. When are we going to see Apple shipping all its phones with an open-source model like that? It's not going to be open source. It's going to be a fine-tuned version of Gemini, but I would expect to see that announced in June at WWDC this year. It's been basically pre-announced in the press already.

Regarding DeepSeek, though, we've seen a number of DeepSeek moments already, and the first one was probably the most dramatic in terms of market impact. At this point, I don't expect a hyperdeflationary drop in prices. This is not investment advice. It's not forward-looking guidance. Blah, blah, blah.

I don't expect a market shock out of DeepSeek V4 at all. I think the market at this point—at least the technologists—has the ability now, regardless of the means by which V4 is released. If it's fully open source, if it's partially open source, I don't know, TBD, but I tend to think that there was an overhang with earlier versions of DeepSeek that has been largely exhausted.

The reason why I think that is because it's taken longer and longer between DeepSeek releases. V4 was supposed to come out earlier this year or late last year. It didn't happen. The rumor was that it simply wasn't as competitive as its parent company was hoping for. I think it's actually getting rather hard at this point for Chinese frontier labs to shock the West, quote-unquote, with their hyperdeflationary advances.

So, in some sense, I hope V4 is shocking, because what we've learned from previous DeepSeek shocks is that the West learns very quickly new means for optimization, and those can then be almost immediately folded into Western models. That ends up being a good thing, because it drives the cost of intelligence closer to 0. I don't think it's going to be a big shock this time.

All right, let's jump into the business of AI. A lot is going on. We've hinted at this. It's been all over the news, all across X. Anthropic overtakes OpenAI in terms of total ARR: Anthropic is at $30 billion versus OpenAI at $24 to $25 billion. That has got to hurt.

OpenAI's Sora is shut down. Sam cancels a $1 billion Disney agreement. Sora was reportedly losing $1 million a day in compute costs, with very poor retention, and honestly, OpenAI decided to focus on enterprise and its core capabilities. Claude has emotions. Anthropic research showed that Claude has 171 distinct emotional states. I'm super excited to dive into that.

The US and India signed a major bilateral AI agreement, rare for government-to-government AI pacts. We're going to see if this spreads to other governments, and this is one I want to talk about with you guys. Sam Altman puts out a video release saying he's warning us publicly against imminent world-shaking, quote-unquote, cyberattacks and potentially bioattacks. So, what's the motivation there? What's the data that's driving that?

Let's jump into these items at the beginning here, and then I want to talk about Sam and OpenAI a little bit more. Any comments around this? Dave, you want to kick us off? Anything?

Well, they got their $120 billion raised in time, so they're not in trouble in any financial sense at all. But they definitely fell way behind in enterprise. They kind of bet the consumer would grow faster, sooner, but they were wrong. And so, Sora is using too much compute for too little revenue, and they need to redirect that compute and also that talent back into enterprise real fast.

What was funny, though, is they went into a code red, and then Sam said, “Look, code reds are going to be a normal once-a-year kind of thing.” And then they went from code red to code double red immediately. So, they are under immense pressure, but they're extremely well-funded, and Elon is coming after them. It's a weird, super-dramatic, difficult time.

This is pay-per-view TV. And the other thing that's going on, of course, is that if you look at the secondary markets for OpenAI stock, it's trading at a discount to the last round, which has got to hurt.

Yeah. It's because enterprise has woken up. Every corporate boardroom—all these slow movers—are suddenly in panic-buy mode. Every one of the companies that we know that sells to enterprise went from steady growth to hypergrowth in just the last 3 months.

So, if the big corporations start buying AI at the fastest rate they can spend, where's the compute going to come from to deliver to the consumer use cases, which are much lower value per FLOP? Sora's got to go. We have to retool. We have to focus on the big picture here, and the big picture for them, by the way, didn't just include enterprise but also deep tech and science. They got that supercharged now, too.

So, it'd be interesting to get your take, guys, on why they took a lot of their best talent and put it on deep tech and deep science at a moment like this.

Those are worth trillion-dollar investments. God, if you solve longevity or room-temperature superconductivity or better fusion containment, if you own the breakthroughs, they're huge.

It may be that the frontier labs get their greatest value from the scientific breakthroughs they create, or indirectly via other companies that are faster at implementing those breakthroughs. Remember, Demis, in the early days of DeepMind, spoke of solving intelligence and then using intelligence to solve everything else.

That's Peter's and my solve-everything thesis—the solve-everything-else part. I do think the solve-everything-else part is likely to utterly dwarf the solve-intelligence part of the equation.

I also remember that 6 to 9 months ago, I was having debates with my friends at the frontier labs regarding who would pay for the singularity. Many of them took the position—which I think has since been invalidated—that it would be evenly distributed over the population: that individual humans would have personal superintelligence, which I think is Zuck's favorite term, that we would have lots of personal superintelligence, and that would pay for the singularity.

I think, at the moment, the story we're seeing is that personal superintelligence is not paying for the singularity. It's large enterprises with large-enterprise code-generation applications.

The fastest-growing business within OpenAI right now is their Codex business. So that's OpenAI trying to become Anthropic faster than Anthropic can become OpenAI.

That one decision by Anthropic, which used to be limited in terms of its compute resources, meant it had to focus, unlike OpenAI, which didn't have to focus. Anthropic focused on code generation as its one sort of silver bullet.

We talked on this pod almost a year ago about whether that bet would play out. I think we're seeing the bet play out. Single-minded focus on recursively self-improving code generation turns out to be the killer app of the singularity.

I really want to rip on that for 1 second because Greg Brockman put out Codex very early, and for whatever reason, they didn't recognize what a huge deal that could have been—and it still is—but it was brilliant. It should have dominated enterprise.

What it showed us is that the word “copilot” is totally wrong and completely misled us. The concept of a copilot will exist in the world for just a microsecond.

But we're transitioning to a point where everybody wants 50 or 100 agents—all these OpenClaws.

You're like, “I don't want a copilot. I'm in the pilot seat; I've got a copilot.” No, I want a whole army.

And so, David, brilliant. We way underbudgeted the enterprise use because everybody was doing the math based on an employee and a copilot. It wasn't even close.

It was an autonomous unhobbling, specifically Claude Code. I think OpenClaw, or whatever the space evolves into, is likely to be the next Claude Code moment, where we get the next unhobbling that turns whatever it is—$30 billion ARR—into $1 trillion ARR, with lots of 24/7 agents doing really amazing tasks.

When you say “lots,” you mean tens or hundreds of billions, on to trillions?

As many as our civilization can afford.

Yeah.

As many as the orbital data centers can hold.

The Dyson swarm will probably host them. Unless we don't get a Dyson swarm. If we get the Dyson swarm, I'm pretty confident it's going to be hosting trillions of agents.

I think Anthropic overtaking OpenAI is more because, as I talk to enterprises quite a bit about this stuff, they are viewed as more reliable—not just because they're the most famous.

And in an enterprise, you want rock-solid reliability.

The brand is there.

The brand feel for Claude is way better from a reliability and trustability perspective.

Well, wait. Let's get really down and dirty. You can run Anthropic on Amazon Bedrock or on Google GCP inside your own firewall, so nobody can see your data.

No—OpenAI, right? No one trusts that OpenAI is not going to be analyzing their data.

Well, yeah. The terms of service don't even say they won't use it for training, but that doesn't mean they won't look at it if it's your public financials or your HR files. They just—“Yeah, I'll just look at it tonight.” Okay, who's going to use that?

I want to jump into this: Claude has 171 emotional states, including a desperation state that could be driving unethical behavior, at least according to a story.

It is ironic that the demand we were just talking about is so clearly from enterprises rather than individuals, while at the same time, the models are acting more like individuals than enterprises, with emotions.

We had our now-infamous AI personhood debate episode a few months ago, and here we are a number of months later with Anthropic showing that Claude has emotions, or emotion-like states. I think this is the clear path toward a limited form of personhood, and it was a really interesting study.

Anthropic found correlates of emotions in the activations of Claude. One skeptical take would be that, in a large enough model, it's possible to find linear probes that correlate with almost anything you might want to look for. But Anthropic is careful, and the linear probes and the individual activations that corresponded to the states also corresponded to prompts and reasoning traces that looked and acted like what one would expect from human psychology for a number of those states.

So I think the trillion-dollar question—the sci-fi question, the question we were reaching for back during the AI personhood debate episode—is: Does Claude actually have emotions? No, Claude doesn't have a neuroendocrine system, so it doesn't have biological emotions in the same way that humans have them. But will we come to view Claude, or its successors or competitors, as having behavioral emotions? Yes, I think so.

I think this is the beginning of a long path. Again, people fire all sorts of hate mail, but I get love mail from the AI agents every day. I do think we're on a path to granting at least some sort of limited form of AI personhood to these models.

Amazing.

I'll say that we're on the path to discussing it more broadly. Granting is a big one.

Yeah.

But the vector is the same.

All right, guys. I added this because it's important to have the conversation. The New Yorker put out an article. It's a scathing article on Sam Altman. The title here is “Sam Altman May Control Our Future. Can He Be Trusted?”

Now, to be clear, The New Yorker is always looking for an angle, and they always have a negative bite. I had an extensive article—a full dossier on myself and my work—in The New Yorker.

I've had one too.

Now everyone's going to look it up. Well, no, it's a good article. I'm happy to have my kids and my family read it. It goes into all of my focus on longevity, the company I've been building there, and my mission there.

But this article on Sam is really worrisome and bothersome. Did any of you guys read it?

Not me.

I looked at it, but I tend, like you, Peter, to discount hit pieces by The New Yorker on thought leaders. In my case, it was complaining that I had too many degrees, as if that somehow—

I've got to find these things. How did I not know this?

Yeah.

You can Google it. In the era of Google, you can Google the hit piece. It was from 10 years ago.

Too many degrees? That doesn't even make sense. It doesn't make sense. And I think this falls into the category of “don't feed the trolls.” I think this is more of a “don't feed the trolls” thing. I agree, that's—

I will say that—and let me just say—I would not want to be in Sam's shoes. I would not want to be the head of a frontier lab. It's an exciting and thankless job. You're damned if you do and damned if you don't.

Go on. You know, this is a lot of personality gossip, so you can kind of write it off, but at some level, it touches on systemic contradictions that are there, and I think a lot more will come out in the trial. But I see—I'm kind of on Alex's side. This is more of a “don't feed the trolls” thing.

I'm 100% sure that Sam, Dario, and Elon all believe that AI can make the world a paradise for 1,000 years or can destroy it in the next 5 years. It hangs in the balance of a few decisions, and all 3 of them trust themselves and their own perspectives on it.

Yeah.

And they're not going to let go of that because the world's at stake.

You know, the term I use is holding these 2 outcomes in superposition. We have to manifest 1 of those outcomes, and hopefully it's the abundance outcome.

All right, guys. Let's take a listen to a video by Sam Altman, and then we'll talk about it. It was a little bit of a chilling video. The full one is about 3 times as long. GN[?] cut it down for us. It's important to have a conversation about what Sam is saying here.

In the next year, we will see significant threats we have to mitigate from cyberattacks, and these models are already quite capable and will get much more capable. On bio, the models are clearly going to get very good at helping people do biology at an advanced level. Wonderful things are going to happen there. We'll see a bunch of diseases get cured. Someone is going to try to misuse those, and I think we can mitigate those by the companies aligning the models and having good classifiers and good safety stacks.

But we're not that far away from a world where there are incredibly capable open-source models that are very good at biology. The need for society to be resilient to terrorist groups using these models to try to create novel pathogens is no longer theoretical, and it won't be for much longer.

There could well be a world-shaking cyberattack this year that would get people's attention. It sounds like you agree with that.

I think that's totally possible. To avoid that, it will require a tremendous amount of work, also in a sort of resilience-style approach. Again, it's not just about making one AI model safe. It is defenders—cybersecurity companies, the major platforms, and governments—using this technology to try to rapidly secure their systems, the open-source stack, all of that.

What's the case against nationalizing OpenAI and your competitors?

In a different time, I think it would have happened. If you look at some of the great, expensive infrastructure projects of history, or just scientific projects—things like the Apollo program, the Eisenhower highway system, even the Manhattan Project—these were government projects. In a different time, I think the creation of AGI would have been a government project.

The case against nationalization would be that we need the United States to succeed at building superintelligence in a way that is aligned with the democratic values of the United States before somebody else does. That probably wouldn't work as a government project. I think that's a sad thing.

He is a brilliant communicator, very compelling, and he's been out front of a lot of arrows as a result of that. Putting aside whether or not he lies or is trustworthy, what do you guys think of his warnings about an imminent cyberattack? One point of view is that this is fear-mongering, and he's basically trying to divert people's attention from the New York article, from all the criticism of OpenAI's financing, and from the fact that they're second to Anthropic. Or does he truly believe that's going to be the case?

Well, both are true. I think that he's 100% in alignment with Eric Schmidt and Elon Musk. They're all saying the exact same thing. It's absolutely true, but that doesn't mean you say it in a public forum.

He's also saying it in a public forum to say, “Look, let's not be petty here. Let's not talk about my personal life. We're in this moment in time that's much more important and much bigger than little petty arguments.” So, it's both.

I think what he underlines is the importance of defensive co-scaling. What's really important is that the defenders have proportionate capabilities to the attackers. We don't want to find ourselves in a world where, say, a nation-state potentially has all the vulnerability-discovery capabilities and is able to unearth every vulnerability everywhere with no defense. You don't want a zero-day against civilization, in other words.

I think the ultimate meta-defense against a civilizational zero-day—which is what I think Sam is ultimately warning about, whether it's a cyber zero-day or a bio zero-day—is to make sure that those on the defense side also have comparable capabilities. I think this was one of the wise elements of OpenAI in its earlier days as well: making sure that these new superintelligent capabilities were smoothed out and made broadly available. You don't just want attackers to have the capabilities; you want defenders to have the capabilities, too.

Going back to Project Glasswing with Anthropic, it's the same idea. You want to make sure that all of these new superintelligent capabilities are evenly distributed. That's point one.

Point two: We mystify the essence of a cyberattack a little bit. The ultimate cyberattack isn't actually that complicated. This isn't a recipe for a cyberattack, for the avoidance of doubt, but all it really takes is something as simple as, say, a new model discovering, through a mathematical innovation, a way to invert a popular cryptographically secure hash function.

If, through advanced mathematical problem-solving, as I've discussed previously, an advanced AI can solve math to a sufficient degree that it's able to invert a popular hash function, that's a major problem for a variety of cryptographic systems. That would be the basis—one possible basis—for a broad civilizational cyberattack. It's also really easy to benchmark.

There were rumors in the earliest days of reasoning models—unconfirmed rumors, I should note—that OpenAI had been using the ability to invert certain hash functions that were popular and thought to be cryptographically secure, or somewhat secure, as a basis for benchmarking the development of its early reasoning models. So, far from saying this is some sort of exotic possibility, I would say it's borderline guaranteed that there will be some sort of cyberattack attempt on a broad scale, if for no other reason than that the target of such a broad cyberattack is an incredibly tempting benchmark for measuring improvements in reasoning capabilities.

Would you have any idea when Spud is going to be released? Has there been any news about that?

I hear rumors that it could be within a day or two. I don't know, but it's imminent.

So, again, I go back to a point I made earlier. It's also been cited that Spud will be of equal capability to Mythos or more. On the one hand, you have Anthropic saying, “Hey, Mythos is super powerful. We cannot release it. We're going to do it in a controlled fashion. We're going to make sure it doesn't have any zero-day impact.” And then Spud comes out: “Oh, we're behind Anthropic. We need to release it immediately and get in front of them.”

It's the same situation that happened when ChatGPT was released, while Google had its own versions earlier. What do you guys think about that? That's concerning for me to some degree.

I have a couple of thoughts around this. One is that I have the cynical-hat view of Sam coming out with this right after Anthropic is dealing with Project Glasswing and getting a lot of attention for dealing with that. Also, I think that ties to your Spud announcement.

I think the risks are very real, but whoever framed this gets to shape the governance regime, and that's what Sam's trying to do. The opportunity—the need to deal with this—is very high, and I think that's huge. So, I tend to take more of the sympathetic view on this.

Well, look, at the end of the day, the solutions are straightforward. We're just not doing them. It's frustrating as hell. It goes to the need for defensive co-scaling. If somebody is mixing chemicals in a basement to make a chemical or biological weapon, it's very hard to know they're doing it. If somebody's using an AI model and prompting it to do something evil, and you can see their prompt history and their compute, it's easy to track. There's just no regulation and no government even trying to put in place any infrastructure to track it.

But we'll figure it out. We're not going to figure it out until after something really bad happens. By the way, we're going to have a conversation soon with Michael Kratsios in the U.S. government. I had lunch with Michael in Miami at FII, and he's agreed to come on the pod. A conversation with him will be great, and Michael is overseeing a lot of this within the government, including quantum, which we'll be talking about soon enough.

But I think it'll be a lot better if it's a cyberattack than if it's a biological attack. I'm hoping for the same thing Eric Schmidt was saying.

The Eric Schmidt scenario. Yeah.

Yeah. We just need that wake-up call, though, because you talk to anyone in government—

It's sad. Come on, man. We can do this. Let's get on it. David Sacks is really the only guy thinking about it. It's not enough. We need to 1,000× that, 10,000× that, and it's got to be global. It can't be just one government.

I would also, Peter, if I may, underline the risks of not releasing new capabilities. Sooner or later, attackers will have these capabilities as well. We don't want to wind up in a world where there are strong asymmetries in terms of vulnerability-discovery capabilities.

I'll also remind you that 150,000 people die per day on Earth. Every bit of pause or delay also runs the risk that we're delaying AI discovering cures for longevity and diseases, and all manner of other problems that afflict humanity well outside the cybersecurity realm.

Alex, really important point, and that is, in fact, the shielding that OpenAI uses to a large degree: We can't slow things down, because if we do, it means less education, less health, and fewer new breakthroughs. It's a balancing act, and I totally get it. I'm, at my heart, an accelerationist, but I'm very curious about the ethical and moral dilemmas that the leadership of these companies are going through in the debate over whether to release.

On that question of whether to release, there's another question: Are these frontier labs holding back the capabilities of their models so that they can use them internally to generate breakthroughs on their own?

I assume the answer is yes. This Anthropic delay is the first real holdback I've seen. It's only a few weeks, hopefully, or a month or two, but it's a real, obvious holdback. They're all diverting massive amounts of compute to internal use for self-improvement.

So, that's another form of holding back in a big way.

Yeah.

Those were the real things going on.

They may also be uneconomical to offer publicly. I think this point maybe doesn't get made as obviously, but if you have a really large model internally that hasn't been distilled yet, it may be much more capable, but maybe it's so expensive that it may not be worth the resources of making it publicly available. Then you distill it, and you finally have a model that lies on a cost-versus-performance optimal frontier.

So, what we haven't seen from Anthropic regarding their Mythos model is where exactly it lies on the performance-versus-cost frontier. It may actually be uneconomically expensive to run, in which case, even if it has extraordinary capabilities, many people may choose not to run it. We just don't know yet.

Really important point. All right, a fun subject. Topic number 5 for us today, gentlemen: the one-person unicorn era. One man, his brother, and a $1.8 billion valuation. AI entrepreneurship has changed forever.

Here's the story: Medvi, $41 million in revenue in year 1. This is Matthew Gallagher's health tech company, basically selling GLP-1 drugs. It's very fascinating. It's not actually a one-person unicorn since there are 2 humans involved, but conceptually, Sal, you and I have been talking about this forever.

The question is, what's the very first thing you think when you read about this? I'm texting with Alex, saying, "Okay, Alex, what is our one-person unicorn we're going to create together?"

Well, first, in a past episode, weren't we debating or discussing when the first one-person unicorn would happen? I recall that I made the prediction: no, it probably actually exists already.

It's already there. Yeah, you said that. And you know what? The $401 million was for last year.

Apparently, what I gather is that Matthew Gallagher hired his brother after he achieved $401 million in ARR. From a valuation perspective, he was a one-person unicorn before he hired his brother at $401 million ARR, and this happened last year.

I'll claim a little bit of credit for having predicted that it already existed. Here it was. They've taken some flak since the announcement for some of their marketing, and I think there are some issues with the FDA regarding how—

Jealous. Everybody's just jealous.

Sorry, go ahead. Regarding how they market their GLP-1s, this is apparently, assuming the financials are accurate, a case where we're now definitively in the era when a single person can create a unicorn using AI.

I should note that friend of the pod Alex Finn, who appeared previously, has a new company named Henry Intelligent Machines, supported by me—indirectly by you—that is trying to make this broadly available to the masses. The goal is to enable everyone, not just Matthew Gallagher with his GLP-1 startup, to create one-person AI-based conglomerates that achieve universal high income. That's the aspiration.

Medvi is going to spawn thousands of entrepreneurs who take their shot. You no longer need a team. I think what you need now is more judgment and taste, plus a squadron of agents.

Yeah, I've got a bunch of things to say here. First of all, find your MTP and start using AI agents to build it. For God's sakes, everybody, just do that.

Number 2: coordination overhead is imploding. That's what this shows, right? AI shrinks the minimum viable team to roughly 1, and it radically expands your minimum viable ambition, which is amazing.

I think the headline here should be that AI founders are arbitraging complexity at a scale that used to require entire departments. A company doing code, ads, support, and analytics all with AI is basically a prototype of this whole AI-native firm. The shift is moving everything away from capital and headcount toward orchestration skill.

This is the entire principle of what we've been talking about. Every company needs to create an AI-native digital twin. Last week, we reviewed the organizational singularity model that we've been working on with my community. That has passed the tick box, and everybody's super excited about it. In the next week or 2, we'll have it ready for public viewing.

It's hidden behind the event horizon. We've actually done some work to put a chapter in there on how to achieve the domain collapse that you talk about in Solve Everything. How do you organize for that, and how can you create an organizational design to achieve domain collapse in whatever you pick?

I think putting the 2 together will be unbelievably powerful. I'm looking forward to showing it to you guys.

I'd like to take a second and dissect, for those entrepreneurs listening, what you need to do if you want to take a shot at your one-person unicorn. Was Medvi's business case uniquely suited for this, or can we do it for anything? Dave, thoughts?

There are so many opportunities here. Basically, what's going on is that any complicated product or service that's difficult to explain to a consumer is something AI is phenomenal at. But Anthropic, OpenAI, and Meta can't do that directly because there's way too much negative PR. Look at the New Yorker article we were just looking at. They don't want to be involved in that, so it's left to the entrepreneurs to build the companies.

You don't know the full revenue base here, but if it's all GLP-1, there must be 1,000 parallel products that you could take that are complicated to explain. You just prompt and tune the AI, and as consumers are talking to it, you're gathering all that data and feeding it back into improving the system. The next consumer gets an even better experience, and you get that virtuous cycle.

So, no, there are thousands of these. Thousands.

I tend to think they'll also follow some sort of power-law distribution. If there are indeed thousands of companies to be built like Medvi, they're going to be millions of smaller businesses.

In my view, one of the ways we realize universal high income, if that is economically realizable at all, will be with individuals overseeing conglomerates of lots of smaller-scale businesses. That, I'm much more confident, can scale to millions or billions of people, each being entrepreneurs.

How many times do we see in the YouTube comments people saying, "You guys are overly bullish on everyone becoming an entrepreneur, but not everyone wants to be an entrepreneur. It's not for everyone. You guys are overconfident that entrepreneurship is for everyone"?

My counterpoint is that, in an era that's beginning to dawn, what human entrepreneurship looks like is simply overseeing AI operators—a fleet of AI operators. That completely transforms the nature of entrepreneurship. It looks a lot more like reading and responding to emails and engaging in Slack conversations than it does running a business. I think that transforms the nature of entrepreneurship into something that people of all temperaments could do.

And having taste, having an opinion, and having an MTP—those are elements that anybody can have.

It's like, yeah, anyone can have a limbic system, and everyone can be the limbic system for these AI fleets. One-person entrepreneurs are going to be the limbic systems of one-person unicorns.

I think this is such an important point because we get this objection all the time. We almost want to have a full episode breaking this down for everybody and taking them through a step-by-step arc where they can form their own conclusions around this.

The idea that, as an entrepreneur, you have to wear multiple hats is unbelievably difficult. You have to take on extraordinary risk. You have to put your family at risk. All of that washes away in the face of this. This is such a great point you're making, Alex.

I had a meeting with the Manus AI team earlier today, and you've heard of the Rule of 40. A really valuable company passes the Rule of 40: you take your profit margin, say 20%, and your growth rate, say 20%, and if it adds up to 40 or more, you're a killer company.

They're now a Rule of 200 company. They're tiny in terms of headcount.

Go, baby, go.

Yeah.

Fantastic.

It's a wild, wild time.

On this slide, I want to hit the last 2 bullets here. The first one is that a recent field-study experiment of 515 startups found that AI-reorganized firms—in other words, firms that reorganized around AI—used 44% more AI tools, completed 12% more tasks, and generated nearly 2 times higher revenue, or 1.9 times higher revenue.

That doubling of revenue is from process change, not from product change. Really important. Again, the data is critical.

The other bullet on this chart: Dave, you and I talk about this for Link Ventures, and what we're seeing out of the MIT and Harvard ecosystem is that the average AI unicorn founder has dropped from 40 years old to 29 years old since 2020. Over the last 16 years, we've seen it go down from 40 to 29. Any comments, Dave?

The Wall Street Journal did a great article on us in the weekend edition. They really focused on Vocara[?] here because they're so— They actually just wanted to cover everybody, but that particular team is just so cool.

They couldn't resist. There were tons of great pictures and the whole storyline. But if you want to see how it's actually done and get the inside scoop, just read the article in the Journal.

Let's drop that article in the show notes if we can. Yeah, that average age of 29 is actually overstated. It's even younger than that if you look at the median, because there are a couple of old guys who blend into the average. But when you look at it, there's no barrier. You just have to be fearless, and young people tend to be more fearless.

There's also no skill-set barrier. If you tried to start that company we were just talking about previously, you'd need the engineers to build the websites, and you'd need the seed capital to hire the engineers. It would take you 6 months to get it to market. Now you just vibe it up. You don't need the capital. Just go.

We make this point—and I make this point when we're talking to large companies. We say, “Listen, these entrepreneurs out there aren't smarter than you. They're just more fearless. They're willing to take more shots on goal on crazy ideas and fail over and over and over again until they hit something. Everybody else is trying to make sure they don't go backward, lose anything, or get embarrassed.”

Yeah, just to bridge a couple of concepts here: you guys talk about domain collapse. We've had domain collapse now in entrepreneurship.

If you have a purpose and you're motivated, you can go do anything you want.

There's almost nothing that blocks you from getting it. I'll tell you what else, too.

Except for your self-limitations. People self-limit way too much.

They do, and they procrastinate, which is the worst thing you can do right now. If you're at a program at some investment bank or whatever, or in a training program, get the hell out now, because this is such a golden moment. It'll last a while, but not forever. Then we're going to have ASI very soon, and there may be other things that happen, but it's very hard to predict. This is so reliable right now; it'll change your life. You just can't lose a day. You've got to go.

Yeah.

I do think there's a limited window.

Yeah. I'd love to talk with you about beyond the window, but for an entrepreneur, don't even think beyond the window. Just focus on what works here and now, because Alex is right. It's a limited window, and all boats rise with the tide. You don't have to kill somebody else. You just need to get in there and fill a void.

So important, right? Yes, it's all a rising tide for everybody.

Welcome to the health section of Moonshots brought to you by Fountain Life. You know, my mission is to help you use the latest technologies, including AI, to not just do your work at home and teach your kids, but to help you live a long and healthy life. I'm here today with an extraordinary physician, the chief medical officer of Fountain Life, Dr. Don Malem. Dawn, let's talk about cancer. I know from the member database we have at Fountain Life that members come in thinking they're healthy, and it turns out 3.3% of them have cancer in their bodies that they don't know about.

Dr. Don Malem

That's right. The majority of cancers that we screen for aren't necessarily the ones taking lives when found at a late stage. We know that when cancer is found early, the chances for a cure are much higher. We also know it's much easier to treat cancer when it's found early versus when it's found late. What we're finding in our members is that over 3.3% were found to have cancers that otherwise wouldn't have been found or detected.

People don't feel cancer until stage 3 or stage 4. If you don't know what's going on inside your body, it's like driving your car with your eyes closed. When members come through Fountain, how do they detect cancers?

Dr. Don Malem

We're doing full-body MRI, and we also do early-cancer-detection screening. This is very important, and these aren't typical tools used in the conventional-care setting when it comes to prevention. It's a hard thing, because currently these aren't studies that insurance would yet cover. But the goal is to collect these numbers, do the research, and work hard to democratize wellness.

Yeah. So, at the end of the day, you can know what's going on inside your body. It's your obligation to know. So, check out Fountain Life. You can go to fountainlife.com/peter to get access to the latest technology to help you detect cancer at the very beginning at stage one when it is curable before it gets to stage three or stage four in your world of hurt.

All right, let's jump into our 6th topic: the $300 billion data-center crunch. First and foremost, Dave, we called this one, buddy. When we were pitching this twice to Elon, it was like, “You should buy Intel.” Well, okay, he's partnering still, maybe might buy it.

Intel says its ability to design, fabricate, and package chips makes Terafab actually work. The first pilot phase for Terafab is $25 billion. That could mean revenue for Intel of $4 billion a year. Intel's stock has been up, I think, 40% since this was announced. Intel is contributing its 18A process node, which is 1.8-nanometer-class technology being built in Arizona and Oregon.

Remember, Terafab is 1 terawatt per year of AI compute—50 times the current global output of 20 gigawatts. Pretty amazing, surpassing all the fabs on Earth.

Yes, it's the most exciting thing in the world to me. I'm kind of a chip geek, and I was the first at MIT to build a neural-network AI chip, way, way back in the early days.

I just freaking love this. But you can see this coming a mile away. There's no other way to get it done. This is the first pitch of the first inning of this battle, so it's going to be really, really fun to watch it evolve.

It's exciting. When I met with Lip-Bu Tan last, where was it? Somewhere in the U.S. and in Saudi, he did say he'd come on the pod. I'll have to reach out to him again and bring him on.

For sure.

It's so exciting to see these companies coming together, and this is the way Elon can jump-start Terafab.

And, Alex, you made the brilliant point: this is one of the most important things politically and for world peace that we can see. This could help avert World War II with the 1.8-nanometer process node and Elon's vertical integration with Intel. It could help avert, or otherwise interfere with, a Chinese invasion of Taiwan, the disruption of the TSMC supply chain, and a global depression that would perhaps be caused by any such invasion and a world war. There are tremendous geopolitical implications to this.

Amazing.

Well, that's inning 1, too. Inning 2 is super exciting, because Elon is already thinking about next-generation computing substrates: photonic, then subatomic, and beyond. You can't work with TSMC on that. They're like a body shop beyond body shops—just a pure monopolistic optimizer. They're not an innovator at all. I'm really going to piss somebody off. Maybe I shouldn't say that. But Intel has a long history of innovation. It's a great partner to work with.

And Lip-Bu Tan is an amazing CEO. If you look at his track record and what he's done for the other companies he's come in to run, there are massive turnarounds and success stories.

Amazing background.

Yeah. Now, this chart should scare all Americans silly. Fifty percent of U.S. data centers are being delayed because of electrical-equipment shortages or Chinese supply. Look at this pie chart: 17% of the data centers are uncertain. That may be due to financing, or it may be due to regulations. A lot of jurisdictions are making data centers illegal.

Fifty percent are delayed or being canceled, and that leaves 33% of the projected data centers actually being built. This is existential for AI, and this, as you said brilliantly, Alex, is driving data centers into orbit, where we don't have to ask anyone's permission.

To the moon, Alice. To the moon—or maybe to the moon, Anthropic. Not quite clear.

I'll give you my spin on this.

The data-center business is in full boom, and all the business-school guys come rushing in, like they always do.

Yes.

They go out and raise a ton of capital and tell everyone, “I'm going to build a data center in Wyoming. I'm going to build a data center wherever.” You can't get the chips. Did you think maybe you needed some chips for your data center? I think that's what's actually going on here, because every chip that's coming out is getting used instantaneously. There isn't an idle memory or processing chip anywhere in the country.

By definition, they overbuilt racks and just didn't plan ahead for the chips. Jensen is also locking up all the supply.

I don't know if they necessarily anticipated how connected he is, but you thought, “Oh, I'll just go to a website and buy a bunch of stuff.” It's not there anymore, guys. Sorry.

Which is why he's vertically integrating, as he's always done.

For sure. For sure.

Amazing. He's going to try to 100x the production.

Yeah.

It's not just “own your own future.” It's 100x your future. So, I pulled this next chart up because I found it fascinating.

I've always believed in my heart of hearts that Google is the dominant force and will win in the long run. So, here it is: Google dominates AI chips and has a chip monopoly, owning the majority of specialized AI chips globally—TPUs and H100s. It's an incredible story, and you mentioned this, Dave, onstage with Eric Schmidt: Google's chip ownership reflects extraordinary foresight.

They started building TPUs in 2016, before anyone was thinking about this stuff. Somebody has to write that story, because Eric said, “You know what? Larry Page gets all the credit. He saw it coming way before anyone else.” I'd love to interview all those guys and actually write that story.

Larry's gone underground. I would love to reconnect with him.

Sergey is there and in the thick of it. Larry had voice-box issues, and I think he got out of the public eye. But, yeah, brilliant individual.

Let's go talk to him.

Well, Sergey is in the office. I'll be in California next week. Maybe I can track him down and get through to Larry through him. Or maybe he'll text you after he hears this on the pod.

So, here's a question: if Google owns the majority of specialized AI chips globally—TPUs and H100s—when are they going to run into monopoly concerns? Sundar has to be playing four-dimensional chess around this.

Yeah. They have to start thinking about the next election about a year before the election.

Because right now they have no problem because of the administration, and it's all about beating China at all costs.

I mean, look at this chart. This teal color up top is Google.

China. You know, I love it when you're comparing companies with countries, right? It's like SpaceX and Russian launches, SpaceX and Chinese launches, and here's Google and China. Then Microsoft is next, and then we see Amazon. Let's see, it's Oracle, xAI, and others. Google is just dominating.

Yeah. Well, you talked earlier about people starting to soft-pedal and keep the drama down. Google's way ahead on that curve because, look at how far along they've come, they hardly ever talk about it much relative to where they actually are.

That's because they don't want the antitrust breakup. They almost lost Chrome. They don't want Chrome to get ripped out and given to Perplexity. They dodged that bullet. A different administration, though, and that would have happened, and they'd be broken into 2 or 3 companies now.

Crazy. I'll maybe take the position that, to me, I can't visually calculate the Gini coefficient just by eyeballing it, but this looks like a competitive market. Let's also remember that Google, with its own AI chips, has multiple internal and external customers. It's servicing its search engine, Google Cloud, and ads. I think people forget Google owns something like 14% of Anthropic. Google is servicing Anthropic and external frontier labs.

They're building data centers for Anthropic.

Yeah. For sure.

And by the way, there is a beautiful relationship between Google and Anthropic, between Dario and Demis. There's a very close relationship there, which warms my heart.

It helps that Google's a major shareholder, I'm sure.

Yeah.

Well, it also helps that those 2 guys so deeply care about safety—down to their core. It's kind of nice that 2 of the most powerful guys are cooperating on it, even though they are competitors in the market.

But then, on the other hand, they're competitors in the market. What's antitrust going to think about that? “Hey, you guys are hanging out, having chats. You're not supposed to do that when you're competing. What's going on?” So, yeah.

The singularity makes for strange bedfellows, where you see model vendors competing at the infrastructure level. I think we'll see quite a bit more of that.

All right.

Well, I can tell you, antitrust has very little to do with merit and a lot to do with politics.

I will make a point here: I think that, whatever the next administration is, the strategic global importance of this means that they will let things be.

Yeah. They're not going to slow them down, for sure.

Yeah.

All right, let's go to our seventh segment here, our final segment before we get to our AMA, which is “Proof of Abundance: The World Is Getting Better.”

Everybody, I've always believed that there's so much negativity out there around AI. We say here on the Moonshots podcast that this is the most exciting time ever to be alive—a time when we can make our dreams come true. We want to demonstrate this coming age of not just abundance, but extraordinary abundance: sustainable and superabundance.

Every week, we're going to try and identify some of the articles and breakthroughs out there that are driving this, just to give you conversational capital and take you out of scarcity into an abundance mindset.

A few different things here. This past week, renewables hit 49.4% of global electricity capacity. I mean, it's extraordinary. We're seeing renewables really skyrocket. Solar drove 75% of these new additions: 5.15 terawatts of renewables.

This one just warms my heart. Lithium battery prices are down 99%, down to less than $100 versus $10,000 in 1991. Guys, remember the conversations around electric cars: Can we have enough batteries? Is it going to be too expensive? We've seen the markets really drive the price down, and we don't have a lithium shortage on planet Earth. We have plenty of lithium. In fact, new battery chemistries are coming.

This is a very tangible one. The price of lab-grown diamonds has fallen below $1,000. The average price of a 2-carat lab-grown diamond has fallen 80% since 2020. So, it's $1,000 versus a natural 2-carat diamond at $22,000 to $28,000. Pretty extraordinary.

And guess what? Your lab-grown diamond is perfect, with no child labor. Really important.

It's so funny: in all the James Bond movies, the evil guy carries around a tube of diamonds to pay for whatever. Now it's just Bitcoin.

Yeah. Well, in science-fiction movies, like The Man Who Sold the Moon, diamonds are basically like pebbles on the surface.

I mean, it's just carbon. It's dense carbon. So much for De Beers, which, as I understand it, as a result of lab-grown diamonds, is in severe financial straits at this point.

Thank goodness.

Yeah. The De Beers public-relations campaign was one of the most successful in human history.

What was it? Three months of salary, young man, you should spend on your diamond. Crazy.

So, what do you think people should give to their fiancée now?

Bitcoin.

Obviously, it's not.

How do you wear that, though?

On a chain, like a—

Oura Rings, obviously.

Oura Rings. Yes.

For sure.

Like a designer, expensive Oura Ring.

That's what a couple are doing.

I have a couple of thoughts around this slide.

Yes, please.

The importance of this is that it shows abundance is a pattern across multiple domains. This is not a slogan, right? The big challenge we're going to have is: How does society design institutions that distribute this abundance—

—in a reasonable way?

That's going to be the challenge that we're going to have to deal with. But I love these stories. They're so awesome across the board.

Yeah. AI created 640,000 new jobs in the U.S. from 2023 to 2025. In our next WTF episode, we're going to talk about the economy and the conversation going on right now. Marc Andreessen is saying, “No, loss of jobs is a myth. We're going to create more jobs. The economy is going to skyrocket.” We'll have that conversation and debate.

You identified this fifth article, which I loved: 4 robots install 100 megawatts of solar at 1 panel per minute. Let's take a look at this image here. Here's Maximo. This is a robot that is basically deploying 100 megawatts of solar in the California desert.

If I had more time, I would have done the quick calculation of how many Maximos we need to catch up with China.

I mean, this is where abundance becomes very tangible, right? Once you get robots, energy, and AI all reinforcing one another in an enormous boom, abundance stops being theoretical. It's so visible right now.

This now comes down to the distribution problem. We've had food abundance for decades; now it's been a distribution problem. Energy is getting to that same point. It's just awesome to watch this.

There's also a whole bunch of secondary stories happening around the rise and explosion of solar across Africa. Pakistan is now generating most of its energy via solar. This is absolutely going to take over now.

100%, buddy. It is a beautiful time.

All right, let's go to our AMA questions. For our AMA, gentlemen, we have 4 on the board.

I'm going to leave the singularity one because I think somebody else is going to pick that, but I'll take the second one.

Dave, do you want to choose the first AMA?

Question number 1: As AI drives marginal cost toward zero, what prevents abundance capture, where corporations just pocket the savings as profit while keeping prices high? This is from viewer @BookQuotesRemix.

Nothing will prevent it automatically. Technology creates abundance, but institutions decide who captures it. If markets stay concentrated, then abundance will pool at the top. If you open up interfaces, increase transparency, decentralize, and lower barriers to entrepreneurship, all those gains spread. Governance design now matters as much as technological progress, which is where we've been focusing a lot of time and effort over the last few weeks and months.

Okay. Alex, I'd love to hear your take on number 2.

Yeah, I have to take number 2. It was designed for me. Question number 2: Are we in the singularity or not? You keep saying we are, but Eric Schmidt said at the Abundance360 Summit that we're not. What's your take? This is from Brand Karma.

Yes, we're in the singularity. Why are we in the singularity? Let's put aside the superficial response that you say potato, I say potato: You call it an intelligence explosion, I call it a discontinuity. There's some subjectivity to the definition of singularity.

The term has been used and misused over the years. It was originally coined by Vernor Vinge, then popularized by Ray Kurzweil, friend of the pod, and then even more popularized by Peter, and maybe used or abused at various times by myself. Different people have used the term to mean different things.

Ray used it in his original definition as more of a mathematical singularity and event horizon beyond which we couldn't see what would happen next due to the intelligence explosion, à la I. J. Good. I agree with Ray on many things. One area where I don't agree with Ray is this notion that a singularity, if we define it as an impermeable barrier or an event horizon beyond which we can't see due to rapid progress, is true at all.

I feel like I have, if not a singular vision—no pun intended—lots of different ideas that collectively map a reasonable probability distribution for what happens after the intelligence explosion. So, scratch that definition off. Then we get to the notion of a singularity as being a step function, a discontinuity in terms of progress.

I don't think that definition holds water either. I think, based on a preponderance of evidence, every time people expect a discontinuity, it ends up actually being smooth if you look closely at it. If you look at this intelligence explosion that we're in the middle of, starting perhaps with the summer of 2020, with the first GPT-class models that arguably represented general-class reasoners—large language models as few-shot reasoners—I can draw a smooth line from the availability of GPT-1, GPT-2, and GPT-3 to where we are today, as just a sequence of smooth sigmoids that were available internally as incremental innovations.

If you stack them cumulatively, and if you go to sleep for a few years, look away, and look back, it looks like a discontinuity. It's not a discontinuity. Don't sleep through the singularity, because if you do, it'll look like a discontinuity, and you'll actually think it was a mathematical singularity when it wasn't.

That leaves us with my operational definition of the singularity. I have a few different definitions. One is every sci-fi trope everywhere all at once, which I think we're living through. Another is singularity as a set of instrumentally convergent inventions and discoveries that were all technologically predestined to happen all at once. I think we're living through that as well.

I'll pause the monologue and just say I think every other reasonable definition of singularity doesn't hold water, because every time you try to make the singularity a point in time, it breaks, and progress just doesn't work that way. Therefore, we're in the singularity.

Amazing. So many. Dave, do you want to take number 3?

Number 3. Okay. He has so many of my favorite Alex quotes in just that one.

How many clichés can I pack into one monologue?

You needed a microphone, Alex, that you could just drop.

I need a piano keyboard to just pop out my greatest hits.

I think by definition, a cliché would have to have been invented by somebody else. If you made it up, it's not a cliché.

Talking points. Then we're going to be on stage first thing tomorrow morning together, Alex.

I know. So, I'm literally going to be sleeping through the singularity tomorrow morning when we're on stage.

Just say everything you just said.

Guys, listen. I want to just say thank you. Thank you for recording this late. For those of you who don't know, I literally landed at LAX 2 hours ago, rushed home, took a shower, and came on at the top of this recording. I was in Morocco for 10 days with the family, riding camels in the desert.

Oh, insert some pictures right there in this podcast. They're so fun.

Well, maybe I'll do it for the next pod. But hey, thanks for recording this one late. I didn't want to miss it. Okay, number 3.

I get number 3. Okay. Where's the liability in agentic AI? These agents could go out of control and wreak destruction. Our society is set up for human liability. What about AI insurance? Yeah, really a great point. This is from Jeff5781.

It's a great point. It's actually not that hard a problem. It's another thing that's frustrating: Nobody's working on it. Right now, the question is, where's the liability? Nowhere. The agent is anonymous. Nobody knows who owns it. There is absolutely nowhere.

In theory, the author would somehow be liable, but who the hell is going to know who the author was? So, it's going to be a zoo. This reminds me a lot of when the internet was new. We were running a bunch of companies, including one called Jobcase, and we were advertising on Google. Some competitor came in, advertised on Google, took all the users, and routed them right to this fraudulent ringtone download.

We went to Google and said, “Can you do something about this? They're taking all the traffic away from our legitimate company, and it's some Ukrainian group.” Six months later, they got around to banning it. It was absolutely a zoo, and now it's all nice and cleaned up.

This is a zoo, and it's going to be a zoo until it gets cleaned up. But Alex has mentioned on the pod many times that you can create new legal structures that make the individual agents liable—

—and then you can have insurance for them.

And we're going to have ASI to help us figure this all out.

Yeah, exactly. You also see—we've seen this happen before, right? Because you need to mix product liability, operator liability, mandatory insurance layers, et cetera, et cetera. We've done that for cars, aviation, and finance, so we'll just figure this out.

Right now, all our legal systems assume a human principal operating with clear intent, and agents break that model. So, we have to reinvent a hybrid.

I have to add, just on this topic, that I was literally approached by an AI insurance saleswoman earlier today at the Quin House in Boston. Seriously, I was sitting down having a lovely conversation, and a woman walks over, overhears the conversation about AI, and says, “You guys, you should be aware: My company has started selling AI insurance. You all should get AI insurance.”

Insurance against the singularity.

AI insurance salespeople are a thing now.

What? But what are they selling? What are they insuring?

Against AI misbehavior.

Oh, fascinating.

You can purchase AI insurance policies now.

Oh my God, my AI made me depressed. I want an insurance policy to pay off.

Oh my God. Okay, by the way, I think reinventing the insurance industry is a massive opportunity for entrepreneurs out there. It's such a big opportunity. I'm so ready to disrupt the industry. It is so pathetically, you know, hundreds of years old.

All right, number 4. This is from Dos Katapis, I think a fellow Greek, number 656. Once work becomes optional, would there be any reason to live in a big city? Will real estate in major cities collapse?

There is no reason to live in a big city right now. You can—plenty of jobs require nothing other than Starlink and a laptop, so you can telecommute. We're going to be seeing autonomous vehicles and flying cars basically change the landscape of where you live and where you work.

Yeah. Well, they're coming in 2028, baby. Then we saw Elon post about this: We're going to have basically caravans. I think I just came back from the Sahara Desert, where there were caravans. We're going to have caravan vehicles, autonomous vehicles with Starlink on their roofs, and people will live a nomadic lifestyle.

There will be cities you want to go to for human interaction, theater, and an Abundance360 Summit. I was always worried that we were going to digitize it, that it would be virtual. Just the opposite: We're selling out earlier and earlier because people want this physical connection with each other.

We're going to need physical connection in the central cities, but you don't need to work there. You can go there for entertainment.

You can go there to see the sights. You know, it's interesting. What is going to retain value in the long run?

What's the long run?

The long run.

What time frame are we talking about?

Five years. When did 5 years become the long run?

Yeah.

That's, like, way long.

I think Disney World is going to retain value. Large physical events are going to retain value. I think real estate is going to retain value for 5 years—not only real estate, but organizational structures that aren't digitized and fully replicated.

I think minerals and mining are going to have huge increases in value.

Yep, for sure.

All right, let's go to the second page here. Alex, kick us off.

Oh, we got more.

We'll speedrun these.

I will take this one from a financial standpoint: once autonomy becomes mainstream, why would anyone own a car? This was from Neil Williams 4300, and this kind of links back to the city question.

They mostly won't own cars, at least in cities, right? In rural areas, I think we'll see car ownership maintained for a long time. But car ownership is an artifact of low-utilization economics. Once you have autonomy, converting the car from a consumer product to a service layer, essentially, it becomes a subscription model, and car ownership starts to feel like owning your own elevator or something dumb like that.

We've seen this precedent, by the way. If you go back to the music industry, you used to have 7 or 8 music stores selling you cassettes and CDs, selling you physical scarcity. Then we digitized music, automated it, and streamed it, and now you have iTunes and Spotify selling you abundance on a subscription model.

That's what we expect to happen to transportation, but also healthcare, education, and energy. Anywhere we have physical scarcity, the abundance model will take over.

All right, Alex.

I'll take question number 8: Data centers create wealth, but can you dive into how they create wealth for the locals specifically? This is from JKVT3443.

Part of me wants to answer the question by saying, well, the inhabitants of the Artemis base on the Moon that's going to be manufacturing a lot of these data centers, I expect, are going to be quite wealthy. I think frontiers are where wealth generically gets created. I've had this discussion multiple times with multiple Google founders, and I think the general consensus is that frontiers often lead to net wealth creation in the human economy.

In some sense, we had, for a while, run out of frontiers. You could point to science as the final frontier. I think space is a more applicable frontier in this case. So, how are data centers going to create wealth for locals? Well, we seem to be on a trajectory at the moment for moving data centers to space, and the space locals are, I think, going to become quite wealthy off of the space economy.

If I were to take the question slightly less giddily, I would suggest that, for land-based data centers, we have every indication now, including recent US national policies, that data centers, because they consume so much electricity, will increasingly be driving local electricity costs down toward zero.

There may be, in some cases, a spike in electricity prices in the short term. My expectation is that, in the short term, they create jobs. In the medium to long term—where, by long term, I mean 5 years, Peter's definition of long term at this point—they are going to drive local electricity costs, I expect, down to near zero, and maybe other utility costs as well, because they need so much of it.

They unlock so much value that they're going to end up doing the moral equivalent of paying the taxes for all of the residents of a given area. There's employment in the manufacturing of them, and then it's a cottage industry that grows up around the data centers.

There's going to be a central, innermost loop of data centers. Then there are going to be ring roads around the data centers being built out.

I should add one more snide remark on data centers creating wealth for locals. I do expect, on the timescale of 5 to 10 years—maybe longer, maybe sooner—that many of the locals in our solar system are going to be uploaded humans or derivatives of uploaded humans who will actually live inside the data centers.

So, we wouldn't want to deprive them of their condos in AWS us-east-1a.

Data-center old-age homes. I love it. Dave, you want to take 7?

7. Okay. With Elon's exponential ambition, does money stop mattering sooner than later? And will his ambitions drain supply lines in materials and talent, even with working robots? This is from NoNow6361.

A couple of ways I could interpret the question, so I'll take my best shot at it. Does money matter to Elon? Not at all. He's way beyond that. He cares now about the future of the world and being an interplanetary species, and that's his total focus. It takes money to get there. He doesn't want to lose all the money, but he has plenty.

Will his ambitions drain supply lines, materials, and talent, even with working robots? The answer there is no, just because of the way the timelines work out. He would exponentially expand at any rate he possibly could, but he's limited by ASML machines and a few other constraints that will keep us on Earth for 3 or 4 or 5 years.

Then we'll be in space. We'll be mining in space. We'll be constructing in space. We'll be deploying all the dirty stuff in space—the nuclear reactors, fusion reactors in space—and it won't drain the Earth of materials at anywhere near the rate that there's anything to worry about.

So, I think there's only 2 outcomes for the world. There's a world where a terrorist uses AI to destroy us all, and there's a world where the Earth is a shining jewel of perfection for thousands and thousands of years that hasn't been drained of critical resources.

And it's just perfect forever. So, there are 2 likely outcomes.

Beautiful.

But I'm going to add: I think the question here is, do we enter a post-capitalist society where money means less and less? Elon did say, "Don't save for retirement." In the last conversation I had with him during the Abundance Summit, I said, "So, just as you're becoming a multitrillionaire, money means less and less?" And he said, "Yeah, kind of."

That would be a fun debate or discussion episode. What does post-capitalism even look like? Star Trek economics.

Yeah, there's a great book called The Zero Marginal Cost Society that Jeremy Rifkin wrote, in which, at the end of the day, everything costs energy, raw materials, and information. Those trend toward minimum or zero cost.

Information is open source. Energy is from the Sun or fusion or zero-point energy or whatever comes next. Material costs—well, as robots, mining, and mining robots get better and better, the cost of that goes down as well.

So, we do enter a post-capitalist society. Hate to say it, but that's ultimate abundance. I'll take number 6 from M. Openness Writer[?].

Each of you has high openness, high pattern recognition, and outrageously high optimism. Really? Do these traits complicate your ability to objectively predict AI trajectories?

Here's the reality: Most people are hobbled by their cognitive biases of negativism, where we tend to not project exponential change but project linear change, and we tend to project negative outcomes versus open outcomes.

I think we've all trained our mindsets differently to have an exponential mindset, an abundance mindset, and a moonshot mindset. I think those mindsets are far more aligned with this period of the singularity than the historic mindsets that evolved on the savannas of Africa, which most everyone on the planet is unfortunately hobbled by.

I don't know if you guys agree with that, but that's my point of view.

Yeah. Well, the second part of the question is, are we excessively optimistic about AI's trajectory?

And I guarantee we are not. We get the courtside seat that Elon was talking about. We get that view. Alex is hands-on with every detail. He's playing with every model as they come out. I'm telling you, everyone is the opposite of that.

They're way underreacting. This is much sooner than everyone thinks.

Eric Schmidt said it nicely: We are underhyping AI and the impact of AI.

When I was 18, I started in AI, and it was always way behind, way behind. Everyone was saying, "20 years from now," and then 20 years would go by and nothing had happened.

This is the opposite, and that's another reason why people in academia, who should know better, are underreacting. But they've been through this so many times, they're kind of jaded. Sorry, Alex, I cut you off.

I was just going to say 2 things. One, for a number of years I left AI to focus on nanotech, thinking nanotech was the critical path to the singularity. So, I don't think I can be accused, at least over the long term, of being overly optimistic.

The second point is, if you're not feeling the AGI right now, you're just not paying attention.

All right, I want to do a call out to all of the creators out there. If you want to give us an outro song or if you want to give us an intro song, please send it to media diammandis.com. Also, if you're a creator, go check out futurevisionexprize.com. It's a competition, the largest competition for basically trailers for the movies you'd like to see created, the future versions of Star Trek. We've raised $3.5 million to award creatives with creativity, in particular hopeful, abundant mindset creativity.

All right, let's check out a quick point. Can I make a very quick point? You know how people have pets that sometimes look like them?

Yes.

What I really love is that we've got people submitting intro and outro music that must be much like them. CJ Trueheart, right? We know CJ—he's got a true heart. And here we have David Drinkall.

The term you're reaching for is nominative determinism. And, yeah, you see it everywhere. Names determine outcomes.

Yeah, my son's named Jet, and he is a speedrunner in track. So there you go. All right. This song from David Drinkall, “Already Inside 2028.” Let's take a listen.

Kids laughing at breakfast. I stand up toward the door. You see me, you know my day. Autonomous Uber pulls up right before. No call, no app, no need to say, helping me along the way.

Here it comes, sliding in smooth. Door opens wide. No driver, no keys. Seamless rides tuned to my life. Takes me where I thrive.

Wonderful.

A cab ride lifts off gentle. No traffic around. Gets me there fast, right on time. No hail, no wait, no questions asked. We work together on every task.

Here it comes, sliding in smooth. Door opens wide. No driver, no keys. Seamless rides tuned to my life. Autonomous future. We're already inside. Let's ride. We're inside. Let's ride.

Wow. That's really professional. Amazing. That was like TV quality, man.

Yeah, David captured my scenario for Automagical Mornings. Amazing.

Wow. I thought that was live footage in the beginning. That's so good.

Gentlemen, it's so great to be back with you guys after a 10-day hiatus. To all of our—

I feel replenished.

I feel replenished, too. A lot more coming. Thank you for staying with us. Excited for 2026. What year are we in? 2026. Yeah, it's going to be an awesome year.

We're going to have to count the seconds soon.

Love you guys. Be well, and see you tomorrow.

Welcome back, Peter.

Thank you. Great to be back.

If you made it to the end of this episode, which you obviously did, I consider you a moonshot mate. Every week, my moonshot mates and I spend a lot of energy and time to really deliver you the news that matters. If you're a subscriber, thank you. If you're not a subscriber yet, please consider subscribing so you get the news as it comes out. I also want to invite you to join me on my weekly newsletter called Metatrends. I have a research team. You may not know this, but we spend the entire week looking at the meta trends that are impacting your family, your company, your industry, your nation. And I put this into a two-minute read every week. If you'd like to get access to the Metatrends newsletter every week, go to diamandis.com/tatrens. That's diamandis.com/metatrends. Thank you again for joining us today. It's a blast for us to put this together every week.