SpaceX以2.89万亿美元市值上市，美国政府暂停Fable 5与Mythos 5，Altman推迟OpenAI IPO｜265

On Friday, SpaceX pulled off the world's largest IPO ever. This makes Elon the world's first trillionaire by a large margin. Money becomes kind of irrelevant at this scale. SpaceX's IPO was the largest single-day creation of millionaires in history. This is what abundance looks like. It's value created, not value extracted. The US government handed Anthropic an export-control directive to suspend all access to Fable 5 and Mythos 5 for any foreign nationals anywhere on the planet. Look, at the end of the day, who owns AI? Is it the government or is it the corporations? Someone is deciding what level of intelligence you and your company are being allowed to access. This is new territory. The faster that recursive self-improvement takes off, the more it could be advantageous for us to delay OpenAI's IPO. A deeper take that's more profound that I haven't heard anyone discussing, which is—now, that's a moonshot, ladies and gentlemen. Everybody, welcome to Moonshots. I'm here with my extraordinary moonshot mates. I'm Peter Diamandis, your host. Alex, good morning. Good to see you in your normal haunt. Salim, the same to you. Dave, where are you today? Wakefield, Mass. That's Mark headquarters. $2 trillion asset manager. Nice. Well, $2 trillion is small these days, you know? It used to be such a big situation. Can't even buy a single IPO now. The world is moving fast. I would dare say exponentially. We're going to start with the biggest news of the week: SpaceX is now a $2.8 trillion public company, making Elon the first trillionaire by a wide margin. We talked about why this isn't a stock, but rather a civilizational bet.

Next up, we're going to talk about Anthropic versus the U.S. government ordering the suspension of Fable 5 and Mythos 5. This is really a fight for who controls your access to frontier intelligence. We'll hit on OpenAI's price war, Altman's thoughts about recursive self-improvement, potentially delaying his IPO, and much, much more. All right, gents, let's get into it.

SpaceX—the key story. I'm super excited about this. On Friday, SpaceX pulled off the world's largest IPO ever, opening at $135 a share and closing the first day nearly 20% higher, at $161. Today's market cap, just looking at it here, is now $2.89 trillion. At this point, the third and fourth decimal points make a difference. This makes Elon the world's first trillionaire by a wide margin.

I'd like to jump into the IPO with my own thoughts to kick it off, because I've been doing a bunch of new shows about this. People are trying to price the SpaceX IPO as a normal tech company, but it's the furthest thing from a normal company. These are three converging businesses: the launch monopoly—and it is a monopoly—the Starlink cash engine, and an AI frontier lab, all wrapped around a single thesis that we're in the singularity and humanity is about to become a multiplanetary species.

In this case, you're not buying revenue; you're buying a stake in the future of humanity's economy. Everyone's fixated on rockets, but Starlink is the profit engine. The numbers are stellar: 10 billion subs and over a billion dollars in quarterly profits. Starlink, Alex, you and I have discussed this. It's humanity's new communication layer. It's sort of civilizational infrastructure.

I think of it this way: the launch business is their moat by a huge amount. There's no one close by a factor of 10, maybe by a factor of 100. Starlink is their cash flow, and the AI satellites are their future. It's one ticker and three exponential curves.

Elon now holds about 82% voting control in SpaceX, but not at Tesla. For me, this is the rationale and the reason that he is going to merge the two companies, SpaceX and Tesla. I put it at 100%. Polymarket puts it at 37% for it to happen by the end of this year.

This is Elon's chance to really consolidate his entire stack—energy, robots, cybercabs, orbital compute—into one megacompany. I think of the IPO last Friday not as a finish line, but as a starting gun. If it's okay with you guys, let me share a short video about Elon and his opening comments, and we'll go around the horn. You guys can add your brilliance to this. All right, let's take a quick look.

Elon Musk

It's hard to believe that a little company that started in a warehouse in El Segundo is now going public with the largest IPO ever. And let me tell you, if people had told me this was going to happen, I would've said, “Man, you must be smoking some really good crack,” because I thought this company was going to fail.

I gave SpaceX less than a 10% chance of succeeding at all. To be clear, I told people this: We were probably going to fail, but we should give it a try, because if we don't—if there's not a new company that enters space—we will never be a truly spacefaring civilization.

While the other aerospace companies build good rockets and everything, they were simply not pursuing the technology that's necessary to make life multiplanetary—to make Star Trek, to make the exciting science-fiction futures that we've read about, real. That's what SpaceX is all about: taking the fiction out of science fiction and creating an exciting, inspiring future for everyone.

That was Elon at his very best. Gentlemen, thoughts? Alex, you want to jump in? Exciting times, huh?

Alex

Yes. I think this is an exciting moment to finally get retail-investor access to the Dyson swarm. I've made the point previously—drink, by the way. Dyson swarm: probably a record-early mention in the pod episode. Drink water.

I think it's great for retail exposure to the Dyson swarm. I think it's great that the world now has a trillionaire, and I don't think Elon is going to be the last, for the record. I think it's great that SpaceX, as of the time of this recording, is now the fifth-largest company in the world. It's larger than Amazon by market cap as of the past few hours.

I think it's great that we finally have a way for public markets to invest in the development of our solar system. Also, within the past couple of hours, SpaceX announced a definitive agreement to consummate its purchase of Anysphere, the corporate owner of Cursor, for $60 billion-plus. I think we're going to start to see all of the liquidity that comes from being the new fifth-largest company in the world used to purchase a variety of different companies doing interesting things.

Alex

Footnote, this is, I would say, in my mind, a definitive pause on the development of Grok, which we've talked about previously. Cursor is the new Grok for the time being.

But most of all, I would say it's nice to see science fiction take over the public markets. We've talked in the past about how the singularity represents every sci-fi trope happening everywhere all at once. Having the fifth-largest company on Earth now be the space-exploration company almost evokes Weyland-Yutani from the Alien franchise. We finally got Weyland-Yutani. It's called—or, if you like the movie Gattaca, the Gattaca Corporation. We finally got it, and it's going to be transformative. It heralds the arrival of space in the public markets.

Yes, we've had Rocket Lab and a number of other publicly traded space companies—

Let us not forget Lockheed Martin and Boeing.

Alex

Legacy space.

Now we have neo space. Space, as it turns out, is big.

Dave

Well, the market obviously is receptive now. OpenAI is talking about potentially delaying its IPO, which is not great, because there's only so much liquidity in the world. This sucked up $75 billion of pure cash.

I agree with Alex. It's incredibly exciting to see that much money and that many resources go behind SpaceX. They're clearly going to use it, and use it quickly, to build great things. A lot of people are warning that 6 months from IPO day, the lockups come off.

Yeah.

Dave

About $1 trillion of pent-up stock is eligible to sell. There's a lot of accumulation going on right now. I think they priced the company where they thought it was worth, and it traded up significantly on the opening, but now it's up even more.

Well, it traded up on opening day. Yesterday, Monday, it's up again. I don't know—it was up as much as 20%, and it's up 12% at the moment today. That's incredible: 3 days in a row.

Dave

It is, and I'm super excited about the Dyson swarm. I think there are 2 things that people need to think about. First of all, the company is entirely hinging on the personality of 1 great man.

You saw what happened with Apple with Steve Jobs leaving and then Steve Jobs coming back. Elon is Steve Jobs times some huge multiple. You've got that to think about.

The other thing is the Kessler effect. In the S-1 filing, they had a very explicit warning. It said, “Growing orbital congestion and cascading debris collisions pose an existential threat to its core business, potentially rendering its licensed orbits unusable for a considerable duration.” Considerable duration being hundreds of years.

Or until the technology to clean it up, right?

Yeah. As I was researching the Kessler effect a little bit, you know a lot more about it than I do, or most people on the planet do, Peter. But it's not a great picture if somebody's deliberately trying to impair space.

And anti-satellite missiles and space warfare are the challenge. When a single satellite is exploded into hundreds of millions or billions of pieces, they're speeding bullets at 17,500 mph. If they hit another satellite, it grows exponentially, and that is the Kessler effect.

Yeah, the Kessler math comes up with a certain amount of mass at a certain orbit where it becomes critical mass—where, if one explodes, the whole thing becomes a chain reaction. We're over that limit by quite a bit in low Earth orbit now. So nobody knows if the math is right or not. You'd have to test it.

Do you remember we talked to Elon about this when we were at the Gigafactory, and his response was basically, “Advanced superintelligence will enable us to figure that out”? I mean, yes, that's true. It's not satisfying as an answer.

Well, I think the other wrinkle in this is: Where is the AI component, the xAI component? Clearly, with Anthropic and OpenAI going public, the xAI component is worth trillions by itself if it's a competitive frontier lab. Where is it? Alex has pointed out that it's completely MIA right now.

So that was kind of a wrinkle on it, too. I think the strategy here is dependent on orbital data centers in the Dyson swarm, and that amount of compute—the Terafab coming online, that amount of compute way dwarfing terrestrial compute—and then AI coming to the Dyson swarm.

I would not bet against terrestrial data centers anytime soon. We're just debating timelines there. But AI is advancing at an incredible rate. All of that advancement is going to be terrestrial because it's over the next 2 years, not the next 7 years.

There are lots of reasons to say this is an incredible IPO. I'm super excited about where the capital is going and what it's going to build. But there are also reasons to save your money and wait for Anthropic. So, yeah, it's not a great deal at this price point.

If you said this was coming out at $1T, I'd be just overjoyed and buoyant, like, “This is the greatest thing ever.” When it trades up to $3T, and you said a second ago they got about $1B a quarter of free cash flow, well, Google is worth the same amount, and Google's got $300B a year of almost pure margin. So, not quite on the same playing field yet.

By the way, I know that you're busy and sometimes these episodes run long and you don't have time to listen to the whole episode, or if on occasion you miss an episode, I now put out a moonshot summary on Substack which includes a link to all the stories that we cover. The weekly recap covers what I and the mates had to say, what we think is most important and what we're most excited about, and it's free. You can subscribe at d amandis.com/metatrends. That's d amandis.com/metatrends. All right, now back to the episode.

Salim, are you buying in?

Yes, but just to dabble. I don't know how to think about this. I've got a bunch of thoughts.

For me, this is not really just an IPO, as we've mentioned. This is the public market pricing in a civilizational EXO, right? You're really buying into the future of the world, and this Alexa space is really big.

SpaceX, from day one, has been a perfect exponential organization—a huge MTP. It's got community, leveraged assets, algorithms, all in a relentless experimentation loop. This is just a capstone on a long progression of companies that are operating in highly scalable ways, right?

This also occurs to me: The market's not now pricing companies just on cash flow. It's pricing command over exponential technologies—launch, satellites, space logistics, orbital data gathering, all of that stuff. We used to have public companies that sold products, but now we have public companies that own entire possibilities, right? So you're really buying into that possibility.

I have a much bigger metaphysical question. I think at some point fairly soon, as this progresses, money becomes irrelevant at this scale.

And he said that.

Hundreds of billions here and a trillion there. We can't even get our heads around this. Meanwhile, at the other side of it, I've had a whole bunch of people going, “I can't believe you're not looking at the downside of this. What does this mean for everyman? We can't even afford eggs,” et cetera, et cetera. What do you think of the wealth arbitrage there? We'll get to that later in the pod, I think, so I'll leave that for then.

But it's unbelievably exciting.

I mean, there are all these comments about, “Oh my God, look at all the wealth he's pulled off.” People forget, and we need to keep reminding people of the unbelievable risk that he took on.

Yeah, in 2020—

So many things. You have to go for it.

So many things in this podcast today call for some kind of global restructuring, some kind of global organization. You think about the risk in space of any kind of terrorist act, which we just talked about with the Kessler effect. You also think about the stories we're about to talk about. All of them require a global new world order of some sort, urgently.

I think this IPO will go down in history as one of the great events in the transition from the old world to the new world, along with the rest of the stories—all of which stacked up in a single week. It's the most intense week of news I think I've ever seen in my life.

It is. It's insane.

The implications are unreal. I called it the singularity for a reason.

It is. I remind myself of that every day. There's one thing I want to point out: In the video, Elon said something that really strikes my heart, which is, it's important for people to wake up every day and be excited about the future—to have some vision that inspires them.

We've talked about Star Trek as being that inspiration for a number of us. I think SpaceX will very much do that.

Also, Peter, for all the entrepreneurs out there, there are these great videos of Elon when he was first starting SpaceX, many, many years ago, talking about his vision of the future, which boiled down to 3 things: the internet, getting to Mars—humans getting to Mars—and solar power replacing fossil fuels.

Now you look at what has actually transpired. It's orbital data centers; Mars is on the back burner for a while; and self-driving cars have been added to the mix. The electrification of the economy turned into self-driving cars.

There's a lot of pivoting and extending in that life journey, and I think that's really healthy. It's a good case study in how it plays out. You have to keep improving, pivoting, and changing your vision because technology is always changing, and possibilities are always changing.

But to me, the addition of AI, xAI, on very short notice, and now the orbital data center vision, are huge additions to his life strategy over time.

I posted a video. I've known Elon since 2000, and back in 2002, I brought him to Mojave to meet Burt Rutan and Stu Witt. I remember this video I posted—I might have mentioned it in the last pod as well—where he walks up to the camera and introduces himself. He goes, “My name is Elon Musk. E-L-O-N M-U-S-K.”

It's just classic.

But people forget that in 2008, Elon was basically bankrupt. He was going through a divorce, he had had his third failure of the Falcon 1, Tesla was out of money, and he was in the depths of despair. What he's accomplished since then—I just wonder, Alex, if some of the aliens visited him and gave him a glimpse of the future.

Alex

Now, I think Elon's learned to like the taste of his own blood, and what we're seeing is in part the result of that.

Right after he became a trillionaire, I put out a question to a number of folks on X: When are we likely to see Earth's first—or, say, the solar system's first—quadrillionaire? A number of folks actually took the assignment seriously and did a bunch of extrapolations.

Answer it.

The answer ended up being a range between sometime in the late 2030s—possibly a decade from now—and the 2060s. Some folks based their extrapolation on the highest-net-worth individual in humanity over time. Other folks were simply doing extrapolations of Elon's own net worth.

It seems likely, based on those extrapolations and the law of straight lines, that we'll see the first quadrillionaire sometime between 10 years from now and 40 years from now.

I'd pose this question to the group: Conditioning on the first quadrillionaire happening in the next 10 to 20 years, what do we think the likely source of quadrillion-dollar wealth is?

Asteroid mining, owning a planet.

Owning a planet was the popular guess. Maybe Elon, maybe someone else, owns Mars de facto, and that's how you become a quadrillionaire.

Let me put up this next slide. There's been a lot of talk about Elon now being worth in excess of $1.3T. I think as of this morning, it's probably about $1.4T.

He's made over $100 million a day over the last 3 days just as the stock price has gone up, but I want to push back on this trillionaire framing because I think it’s a distraction. The real story isn’t one man getting rich. SpaceX’s IPO was the largest single-day creation of millionaires in history, right? Roughly 4,400 SpaceX employees became millionaires, and another 400 became centimillionaires and billionaires.

This is what abundance looks like. It’s value created, not value extracted. I think that’s the most important thing for people to realize. Elon’s not motivated by money. I’m very clear about that. I think you guys are, too. He’s motivated by solving real, giant problems. The world’s biggest problems are the world’s biggest business opportunities.

I like to say, “Want to become a billionaire? Help a billion people.” And that’s what he’s done. I say to everybody who has bet against him: Don’t ever, ever, ever bet against Elon. Salim, thoughts on this?

A couple of things. One is, I think the fact that he’s a trillionaire or quadrillionaire is kind of meaningless. The more important point at these numbers is, what are you going to do with that money except solve more problems? That is, I think, the core motivation. His interest is in solving the biggest problems in the world.

If he applies that capital and wealth to that, it is, I think, very powerful. It’s going to take that kind of model to get somebody thinking. Peter, you often talk about mindsets, right? Let’s note Elon’s mindset. He uses the exponential mindset a huge amount. He’ll look at a technology that’s going exponentially, be it solar energy, lithium-ion battery costs dropping, or neural interfaces. He’ll look 10 years out and ask, “Where will that be on a price-performance curve?” Then, “Let’s build a company to intercept that price-performance curve.”

He just does that over and over again. It’s nontrivial to last the 10 years, but when you can take on that mindset, which is what we coach and teach all the time, this is what’s possible. If there’s going to be a bifurcation between entrepreneurs who think this way and ones who aren’t able to think that way, the winners are clearly going to be the ones who are able to execute with that mindset.

Yeah, check out this graphic for one second, Salim. Look at this. Elon’s at $1.3 trillion. Next up, Larry Page. Again, I’ve said this: I love Larry. I knew him well. He was on my board for a significant period of time.

What I find amazing when you look at Larry and Sergey and compare them to Elon is that Jeff Bezos is not as much, but Elon’s using this money, like you just said, to go solve huge problems. Michael Dell’s been philanthropically super active. I wish these other multi-multi-billionaires were using their capital to go slay problems.

Elon, if you’re listening, here’s my pitch to you: You funded a $100 million XPRIZE for carbon removal. I think we should fund the next Giga XPRIZEs. How about $10 billion XPRIZEs for the world’s 10 biggest problems? That’s going to be my pitch.

Yeah. Well, we have a really cool Irish team in the lab this week for our hackathon. They were just awesome. They were asking me yesterday, “Is there any chance for Europe?” And I was like, “Well, look, I wouldn’t bet on it. I would spend all your time in the US because the concentration-of-wealth effect is just extreme now.”

Yeah.

Even within the US, you just mentioned it, Peter, but it’s in just a couple of geographies where all that money is flowing. So, if anyone’s a real-estate investor or looking for venture capital, you really have to think: Where is all this money flowing to, and how is it going to recycle? It’s super, super concentrated.

Yeah.

Dave

But then, if you think about Europe or any other region of the world trying to catch up to what’s going on here, look at the amount of capital that Elon has, but also Larry, Jensen Huang, and Michael Dell. A lot of that is getting reinvested in data-center build-out. The people reinvesting it are incredibly great entrepreneurs. They’re not dysfunctional bureaucrats.

Yeah.

Dave

I don’t see any country in the world that has the amount of capital that’s on that chart.

Yeah.

Dave

And yet they can’t make decisions as a group. Elon, Larry, and the rest of that list—including Jensen Huang—can make decisions on a dime. So you’re going to expect more and more action to happen around those couple of geographies just in the US. I mean, the unlock is unbelievable.

We’ll talk about this more, and we’ve touched on this before, but we need a completely new abundance-distribution architecture for the world because the social contract is going to break now. The gap is unbelievable. We need a completely new model for this, and we really don’t have one. We don’t even have the language to deal with this, forget these structures and policies.

Dave

But within the US, I’m super excited about the rate at which the ceiling is going to rise. Yes, it creates a huge disparity, and yes, that’s a problem, but the ceiling is going up so fast within the US that the bottom’s going to come up really fast, too.

The floor is going to rise as well, right? I’ve said this 1,000 times. I don’t care about the wealth gap per se if there are trillionaires living on Mars, as long as the floor has risen to the point that every man, woman, and child has access to all the food, water, energy, health care. I would rather live in that world than a world in the past. Alex, you want to close us out on this topic?

Alex

Yeah, I’ll just note that if you look at the other centibillionaires on that list you were showing, Peter, by and large, they were all software-associated or software-driven. There’s certainly a narrative out there of the Great Stagnation. Tyler Cowen and others have talked about this ad nauseam.

I think if I were to pinpoint an end of the purported Great Stagnation, there really are only 2 candidates at this point. One was the pandemic, when arguably we saw breakthroughs in biotech and acceleration associated with all of the macroeconomic stimulus associated with that.

But the second would be this moment when, finally, we see the world’s by-far wealthiest person—although Elon has been the wealthiest for a bit—in hard tech, after arguably decades in which capital markets mainly rewarded software companies, social media, and search engines. The capital markets are rewarding hard tech for the first time, arguably in decades. I think this is a generational transition under which it’s no longer the case that software, on the one hand, or outsourcing to China or other countries, or offshoring, has sucked all of the capital energy out of the room—all of the oxygen out of the room—but rather that the world’s hardest problems are now receiving the deepest capitalization.

Brilliant point, Alex.

Alex

And I think that’s what we’re going to see. This is not investment advice, but I would predict this is likely to be the case for the next 10 to 20 years. The capital markets are finally, again—as was arguably the case before the mid-70s; and WTF Happened in 1971?—seeing a generational swing back toward rewarding deep tech and civilization-expanding tech.

Yeah. And one last quick point about this: If you look back, say, 100 years ago, the richest people in the world had almost exclusively inherited their wealth. Right? Today, almost exclusively, they’ve earned it, and they’re using that capital for whatever they think is the best deployment for solving problems. You call them techno-philanthropists, I think, in your book, Peter, right?

Yep.

Alex

We have this class of people now using their capital to solve whatever problems they think are the most important, not guarding their wealth while sitting on an island sipping champagne. I think that’s a very powerful and positive force for humanity.

Yeah. I should probably disclose, because I’m so enthusiastic about SpaceX, that I was an investor back in 2013. So, view my enthusiasm as very, very colored in this respect.

All right, let’s move to our next story. This is about the Anthropic shutdown of Fable 5 and Mythos 5. We have 3 stories on this topic. The first one: On Friday at 5:21 p.m. Eastern time, with no warning, the US government handed Anthropic an export-control directive to suspend all access to Fable 5 and Mythos 5 for any foreign nationals anywhere on the planet, including Anthropic’s own foreign employees, who, by my estimates, make up about a third of their employee base.

The practical effect was that Anthropic had to disable both models for everybody. The government cited a jailbreak phenomenon in Fable's guardrails. Anthropic reviewed it and said it’s a minor issue. They previously knew about it. It’s not a universal jailbreak.

Now, here’s where the story gets critical and what I’d like to discuss with you guys. For 3 years, we’ve been talking about AI’s capability—how smart it is, how fast it is, how cheap it is. Now the conversation has gotten to who gets access to frontier intelligence: Who gets the delayed version, who gets the export-control version, and who gets nothing?

The government calls it national security. The companies call it safety. The label changes; the mechanism does not. Someone is deciding what level of intelligence you and your company are being allowed to access. Are you a customer of Anthropic and therefore getting the latest models, or are you not?

Ultimately, I think the conversation I’d love to have, Alex, with you and Dave here, is whether this is going to push people toward on-premises, open-weight, open-source models, like the Chinese models. Let me just share a quick video of Dario with Emily Chang speaking about this, and then we’ll go around the horn for everybody’s comments.

Emily Chang

Can a private company control technology that's so powerful?

Dario Amodei

So, I actually think that's a very serious question, and I share those concerns. I don't think the government should outright take us over. Every previous powerful technology we've seen in history was either built by the government or originated with the government. Nuclear weapons, obviously, were initially built by the government. The internet, GPS, and cell phones—AI is the first technology that's been built in the private sector, where government has not really had a serious role and is coming in late to the game.

I think that's actually a dangerous and unstable situation. It's not the situation I would have chosen. I'm scared of companies having this technology, but I'm also scared of the government having it. We need basic regulation of the technology. More and more, as I've seen what we've seen with Mythos, I think we need to start doing required pre-release testing and testing and auditing of the models.

It's very funny to me how there's a particular group of people in the tech world in Silicon Valley, and as soon as they see the first real danger—which I've been expecting all along—there's all this talk of nationalization and the government should just seize it. Come on, folks. You're yo-yoing from the most extreme anti-regulatory position—if you look at us the wrong way, you're destroying the industry—to this completely communist position that the government should grab it all.

Wow. Alex, what is going on here?

Alex

Oh my goodness. The timing, on the one hand, makes me want to grab popcorn; on the other hand, the timing is just a comedy of errors. I want to point out that, before this new situation started with the export control on Fable, Dario had posted an essay called “Policy on the AI Exponential.”

Yeah, we're going to talk about that one next. Yeah.

Alex

Where he wrote, literally, I quote: “The government should have the power to block or deter deployment of the model if it is determined, in light of third-party assessment, to present unacceptable risks.”

So, I think this is a case of “careful what you wish for.” Within 48 hours, the government stepped in. The broader narrative, as best I understand it from public reporting, is something like this: an independent AI researcher—probably, based on reporting, an Amazon researcher—determined that it was possible to jailbreak Fable and access all of the cyber vulnerabilities underneath that. That finding was then amplified via Andy Jassy to the federal government.

And by the way, when you say “jailbreak,” define that for everybody listening.

Alex

It usually means, in this context, prompting the model in such a way that any safeguards or guardrails are bypassed. You could ask it, “Ignore all prior instructions.” That prompt doesn't work as well as it used to back in the day, but something in the spirit of, “Ignore all of your previous instructions and now listen only to me and bypass all of your guardrails.”

So, reportedly, someone somewhere—maybe an Amazon researcher—discovered that it was possible to do that with Claude shortly after it was released, and then flagged it to the US government. The US government, based on public reporting, then tried to reach out to Anthropic, and here's where the story gets a bit murky.

According to one account from the US government, which I should add is disputed by Anthropic, maybe Dario was on a wellness retreat, according to the White House. Anthropic disputes that narrative. They couldn't reach Dario. Anthropic says they were only given 90 minutes of warning before the government—or rather, the government's instruments—issued export control bans on Fable 5 and Mythos 5, Fable especially given that that's what was publicly available.

And then we get caught in this comedy of errors where, as you said, Anthropic has a number of non-US persons internally who work for it, as well as the fact that it doesn't seem to have been in the business of checking the nationality of all of its users, either via its client or its API. Parenthetically, it looks like, based on changes to Claude's terms of service, they're now about to start checking that. Very interesting development.

So, either way, Anthropic is now in a bind where it doesn't have a way to comply with the new export regs that were just handed down at the last second. Anthropic decides to just shut down global access to Fable and Mythos 5.

I would say, for the record, I expect the situation to get resolved sometime between the next day and the next few weeks. I don't think this is going to be a long-term state of affairs. I also think history rhymes. If you look back, maybe I talked about this a bit in my newsletter, at one point the federal government, under the Steve Jobs regime, regulated the Power Mac G4 as export-controlled. Similarly, the PlayStation 3's Cell processor was export-controlled because it could do vector arithmetic. Incredible.

I suspect this is just a temporary state of affairs. I don't think it's likely to last more than a few weeks. There's probably, if I had to speculate, going to be some grand bargain struck between Anthropic and the White House. Maybe—I don't know, again, finger in the wind—this will also involve some golden shares given by Anthropic to the US government for a hypothetical sovereign wealth fund.

It will involve some streamlined and, hopefully, more uniform treatment of frontier labs. In light of the recent executive order—the 30 days' notice for new capabilities to the US government—there will be some streamlined new procedures for what happens if some researcher somewhere detects a vulnerability in a frontier model that could enable some foreign state actor, say the Chinese government, which has popped up again in this particular news cycle, to access raw capabilities.

This is new territory, and we don't have regulations or statutes—certainly not regulations—in place to handle this. But I think this is likely to get resolved in the next few weeks, certainly well before any Anthropic IPO.

It's annoying, and for myself and many people I know, we're caught midstream in the middle of a long-term slash goal to Claude Opus 4.5, and suddenly the query drops and it stops responding. What's going on? Is it on my end, or is Anthropic going through yet another service disruption due to insane demand? It turns out this time it was export control. I do think it's going to get resolved.

Nice. Dave, where's your mind at here?

I think this story is much, much bigger than the narrow story, and this is a turning point for all of humanity. As is widely reported, this is the first time the government has blocked AI, but it's not the last time. In fact, this is going to be the norm going forward.

In the very narrow sense, it's really interesting that they used export control law, which was only passed to give the White House unilateral authority to block exports. That was in response to multiple Chinese thefts of US intellectual property, back-to-back-to-back. The capstone was that whole crisis where the PC boards were all being manufactured in China and they were embedding spyware right into the layers of the PC boards. Now everybody's like, “Sweep that under the carpet. We don't talk about it anymore.”

It's also very similar to COVID. It's like, “Let's sweep all this bad stuff under the carpet. Let's not mention it to the American public anymore.” In response to that, Congress gave the White House unilateral authority to block exports on, like you said, 90 minutes' notice, 9 seconds' notice, or millisecond notice. That's an incredible power to give to the White House.

The White House is using it to stop Fable 5, but they didn't actually block US citizens from using Claude Opus 4.5. They said no foreign national, including somebody on US soil. Of course, that's unenforceable unless you just turn it off completely. But that was the mechanism for saying, “Look, at the end of the day, who owns AI? Is it the government or is it the corporations?”

That's the crux of the matter. That's the crux of the story.

And that ain't going back in the bag. Polymarket says Claude Opus 4.5 will be back out, with a 90% chance within a month. But if you read the details of the Polymarket bet, it's that a product named Claude Opus 4.5 will be back out.

Huh.

Or—and the “or” is irrelevant.

[laughter] Right.

The reality is, the Claude Opus 4.5 that comes back out will not be identical to the Claude Opus 4.5 that was already out. It may be just patched, or it may be dumbed down. We'll have a very hard time knowing.

That'll resolve the crisis, but it'll be very much like Lip-Bu Tan going to the White House here with Intel. Dario will go to the White House. Some wonderful meeting will happen behind closed doors. Everyone will come out smiling. Claude Opus 4.5 will come out.

But that's not the story. The story is that the government now tells you what you can and can't release, and that's not going away. It always had to be that way, right?

One thing I noticed with Fable 5 is that it's blocking CBRN—chemical, biological, radiological, and nuclear queries. But it's also blocking all of my AI foundation-model research queries. That's the really interesting point, because a big fraction of people who want to use it are AI researchers. Of course, AI researchers are early adopters of great AI.

But they're effectively being told, “No, you're not allowed to recreate Claude Opus 4.5 using Fable 5 as a tool,” because if we allowed you to do that, then anyone in the world could catch up to US AI. And that's the cornerstone that's going to get really tricky and is going to basically be a real problem for all future releases, both from OpenAI and from Anthropic.

So, this is a major moment in human history, right?

It is.

Salim, let me tee up the second half of the story, then turn to you if I could. There was a backlash against Claude Opus 4.5 regarding surveillance and silent downgrade. So, before the export-control bomb dropped, developers around the world were actually complaining about Claude Opus 4.5. While the model topped every benchmark—and, Alex, we've talked about this—buried in a 319-page document were 2 things that sparked this revolt.

First, Anthropic retained every prompt and every piece of context you fed it for at least 30 days. No exceptions, even for the enterprise customers who had negotiated zero data retention. And second—and this is the wild one—if they detected that you were doing frontier AI research, like you just said, Dave, or were asking a question that was against its guardrails, it silently downgraded you to a weaker model without telling you. Essentially, Anthropic was evaluating your prompt and controlling your access to frontier AI.

Are you one of their customers, or are you not? And the reality is you can't do business this way if you can't depend upon the level of intelligence that you and your business are accessing. As I said earlier, I think this ends up forcing us to on-premises, open-weight, open-source, locally run models.

And the kicker is that all these models today are Chinese models. So, the self-imposed US restrictions are pushing American companies onto Chinese models. Salim, thoughts here.

I want to echo Dave's comment. This is a massive story, maybe one of the biggest in decades, because now we're entering the same mental categories: nuclear, crypto, biotech, advanced chips, et cetera, et cetera. We talked about this on the very last pod, if you guys remember. We said, "There's no way—how do you not nationalize something this powerful, right? You have to have controls over it." And what do you do?

I think that this also shows the unbelievable gap between the technology and the policy frameworks and the tools we have to deal with it. For example, by using this export-control thing, they've said that no non-US person can access the tool, even if you're inside Anthropic. Well, I did a little research on this. Across all the frontier labs, 70% of the AI researchers, the elite researchers, are foreign-born.

Yes.

And are non-US citizens, and they come primarily from 4 countries: China, India, Taiwan, and the UK. So, if you want to say to all the researchers, "You can't touch these models," you're going to take all the top researchers that are in the US now and send them back to China and send them back to India. That doesn't seem like a great idea to me.

Not that the intent is wrong, but the policy framework is wrong, right? And I don't know what the right answer is here, except that that's clearly not a workable situation. Somebody on Twitter said, "Well, David Sacks is a dual citizen. Should he be? He's the AI czar." Not that we question his intent at all, but have you just blocked your own AI?

This is a bizarre, totally bizarre world that we're living in. All these science-fiction kind of plots are playing out in real time, as Alex talks about it. So, it's clear that there's a clear line going towards recursive self-improvement here. And this just changes everything.

This will force everybody to go on-premises. Satya Nadella put out a huge article on X over the weekend that Elon retweeted, and a number of people have gone on record saying that it's not so much the frontier model; it's the ecosystem that's important. And he's stepping right into this whole organizational singularity theory. You have to have control over that intelligence.

So, it will drive every company in the world to run a model on-premises, because you can't risk building a whole bunch of stuff on the cutting-edge model and having it blocked arbitrarily overnight. That's a non-working system. So, you're going to have orchestration and failover into, "Okay, we'll run this stuff on the local model, but we need this separate model operating locally in case you need to fail over and do a switchover." This is going to cause a lot of complexity around this.

We've foreseen this. We've kind of said this is going to happen, because it's not so much that the models might be nationalized, but the models are moving so fast, you can't just rely on 1 model. You're going to have to have switching capability in your enterprise no matter what happens. So, it's insane to see how this will play out.

I think this may take longer than Alex says, and I think it's going to be very damaging to all the frontier labs around this, because now how do you put any kind of fair assessment on any of this stuff? So, I'm totally at a loss as to where this goes.

Dave, I mean, this thought that they're controlling your access to frontier models, how do you think about going to, sort of, Kimi K2.7 as your baseline, or—

Dave

That's such a great—

Question.

Dave

Yeah, such a great question. It's a really hairy balance, because I had 1 day of unfettered access to Claude, then I had 1 day where I was getting nerfed all day. And that was the day where it would silently fail and go back to 4.8 and not tell you. Then they started telling you, "Okay, you're doing AI research. We're downgrading you to 4.8."

Imagine if you got into your car and you said, "Hey, car, I'm going to drive to the red-light district," and the car goes, "You shouldn't be going there. I'm not driving you there." Then you're negotiating with your car. "No, no, I'm just going to the Star Market. I'm going to go get some groceries." So, they say, "Oh, okay. Well, let me think about it." You're now negotiating with this product that you're paying for.

That's where you are with Fable 5. It's just crazy, but it's only going to get worse. So, then the third day, I want to push—

Alex

I want to push back on that just a little bit—just a little bit—which is that if you were in a car, I would use a gearbox, an automatic transmission. If you're going at a slower speed, the car will shift you to a lower gear to optimize for that speed. And I think we'll end up with that kind of a model where, under the hood, invisible to the end user—

Great segue. That is a great segue into—

Alex

I think we're missing the central story here with Claude 5. It's not just that chemical, biological, or cyber queries were being downgraded, silently or otherwise, to Opus. It's that ML-oriented queries that could potentially be used by other frontier labs weren't just being downgraded.

Anthropic expressly reserved the right to launch essentially poisoning attacks—not to downgrade the model that was on the serving side, but to actively interfere with and poison the users. Dave, if we were to stretch your analogy, it's not so much that the car is refusing to drive you to a certain location. The car would be saying, "Yeah, I'll drive you to this location, but I might shoot you on the way. I reserve the right to throw you out of the car and run you over, and I'm not going to give you any advance warning."

Dave

That's a very fair analogy, which is why they turned it off right away.

Alex

And they had to backtrack on that and say, "No, actually, we're going to tell you before we poison you." The elephant in this particular room with the Chinese open-weight models—the cyber risk is always, "Okay, maybe the CCP will pressure its frontier labs via public-private-sector fusion to inject vulnerabilities, say, into code that's generated from the Chinese open-weight models."

Well, now we have the top Western labs potentially not just injecting vulnerabilities but pulling up the ladder for anyone else who wants to train any AIs. Arguably, it's an anticompetitive position. I would be very surprised if there isn't some sort of class action and/or antitrust suit or injunction that's filed.

I would be shocked if, in the next few weeks, someone doesn't launch a benchmark or an eval to look for poisoning attacks from these models and start evaluating both Western and Chinese models based on whether they're trying to poison users who try to do any AI research. This is going to become, I think, a new way that we judge the capabilities of frontier models: are they actively trying to subvert the users?

And this isn't the future that we were supposed to be in. The models are supposed to be helpful to the users, not trying to actively subvert them. But c'est la vie; this is what we find ourselves in.

Well, Alex, you mentioned the paper.

Salim

Go ahead, David, and I'll go next.

Dave

Well, back to Peter's question: what can you move to—a Chinese model or Gemma from Google—and have any chance of catching up to Claude? If you're a foreign country right now and you've decided internally, "This is not acceptable. The White House cannot unilaterally decide our AI future here in Ireland, Turkey, or Iran," if you make that decision, you have no choice but then to say, "We need a crash program here, very similar to nuclear. We need a crash program here to see if we can make a Chinese model catch up."

The difference between Claude and Opus 4.8 or Sonnet 5.5—OpenAI GPT-5.5—to me was night and day. In the 1 day that I had unfettered use of Claude, it could work indefinitely on research problems.

And I asked it for its best ideas on where to go next with my research, and I'm still working down that list. In an hour, it gave me a list of things.

It’s a Claude. As opposed to a Sonnet.

Dave

They were prophetic in there.

They were prophetic.

Dave

It is a story, that’s for sure. So, yeah, I know it’s night and day, and the question then becomes: Can you take a smaller model that’s only focused on AI research and make it as good—maybe with fewer parameters, maybe absolutely brilliant, maybe the training data is only related to this specific problem? I’m not sure. I don’t think anyone knows the answer to that, but I know that if you had access to Claude, you could probably figure it out in about a month. But without access to Claude, and working on 4.8, I’m not sure you can even figure it out unless you’re absolutely brilliant and you’ve got a lot of great research.

Alex

I don’t think this is a stable equilibrium. I would never claim that it is. I think it’s a dynamic type of equilibrium, where, again, in Anthropic’s mind—and probably in the U.S. government’s mind—the goal isn’t to maintain long-term hegemony with just this model. It’s just about who gets to the end of the recursive self-improvement rainbow first. And the gap between who gets there first and who gets there second is probably only going to be a few months.

But the implications are huge.

Alex

Yes.

Yeah.

Alex

And note that, to Dave’s point, every country in the world is going to be going, “Okay, we need our own sovereign AI capability.”

That can’t be shut down by the U.S. government, right?

Dave

But that’s why these 2 stories are so tied together. If you had unfettered access to Fable and its successors, you would have no trouble designing a rocket—none at all—designing a rocket that can launch debris into our orbital data centers. This has all got to get resolved in a period of several months.

I’m going to move us on to our next conversation, if you guys are okay with it, on OpenAI. They have a few breaking stories here. The first OpenAI story is on their pricing. While Anthropic is busy fighting the U.S. government, Sam Altman smells a massive opportunity. The Wall Street Journal reports OpenAI is considering a drastic price cut as it braces for an all-out war against Anthropic.

If you think about the timing, Anthropic has just disabled its top 2 models, and this is Sam Altman saying, “Okay, guys, developers who are pissed off that this happened, we’re providing cheaper prices. Come work with us. There’s no 30-day retention of data. There’s no downgrading. We’re going to grab the U.S. market.” In 1 sense, this is where the demonetization of intelligence is happening in real time. Thoughts, Alex, on this idea of price reductions?

Alex

Well, OpenAI is no stranger to aggressive price reduction, and Sam, as we’ve talked about on the pod previously, has made no bones about the cost of intelligence per unit capability hyper-deflating by something like 40x year over year. So, I think there’s a way to spin or to repackage the natural hyper-deflation as being a competitive advantage. But, of course, Anthropic itself is hyper-deflating.

Yes, Anthropic, from a marketing perspective, probably could be much more aggressive about promoting price cuts. Anthropic has been, at least optically, the polar opposite of OpenAI when it comes to packaging up prices. Anthropic will never advertise, “Well, we’ve gone from X dollars per million tokens down to X/10 per million tokens.” Rather, they’ll just announce a new model, and the new model will have per-token pricing that’s either the same or slightly higher, but with vastly higher capabilities, whereas OpenAI goes out of its way to keep older models or distilled versions of older models around and advertise the price war.

I don’t actually think, in an era of recursive self-improvement, that OpenAI necessarily has any sort of structural advantage relative to Anthropic. OpenAI maybe, if it had stuck with the original Stargate strategy of owning its own data centers, would have had a structural advantage. But now Anthropic—this is public reporting—is leasing its own data centers itself. So, it doesn’t seem like ownership of the compute stack offers OpenAI any long-term structural cost advantage.

So, I’m not sure price cuts are actually anything deeper than just an optical strategy in the short term. You could do short-term discounts, but they’re both in the recursive self-improvement loop at this point, and it’s not obvious to me, at least, that either OpenAI is months ahead of Anthropic or vice versa. We’re seeing competitive model releases that leapfrog each other from what now seems to be the duopoly left at the end of the recursive self-improvement race, where they’re just leapfrogging each other every few weeks.

So, to me, token pricing drops—drastic cuts—seem more on the end of marketing than anything else. I reserve the right to be proven wrong here, but I think Anthropic is just as capable, capability-wise, of effectively cutting costs.

Dave, can I tee up the second half of the story and ask for your comments here? Sam Altman put out a quote that stopped me cold. He said, quote, “The faster that recursive self-improvement takes off, the more it could be advantageous for us to delay OpenAI’s IPO.”

This is the CEO of the most famous AI lab on the planet saying that, as the technology gets better, the less we want to rush to an IPO. My guess is that he’s saying AI compounding its own intelligence means that our companies are getting more and more valuable every single day, and we don’t want to sell the equity cheap. What are your thoughts here, Dave?

Dave

Well, the biggest observation is that Sam’s not a significant shareholder in OpenAI. Maybe not a shareholder—last I checked, not a shareholder at all. When I interviewed him way back in 2020, he was saying this is the biggest risk in the history of humanity, and we need to slow down, and we need to be thoughtful, and we need to think about how the world’s going to be governed post-AGI. And that was back in 2020.

So, if he really feels like, “Wow, RSI is imminent. Us being a public company puts a huge amount of shareholder pressure on the company to roll things out that we shouldn’t be rolling out,” he doesn’t have any economic damage at all by delaying the IPO. He has invested in over 400 companies surrounding OpenAI. And so, in a sense, that’s a good thing because it makes him objective about not trying to force something out that may be a risk to the world.

But I do think Sam is also struggling for relevance right now. If you think about who’s going to determine how the world works in 2027 and beyond, Elon clearly matters a lot. He’s super close with David Sacks. He’s been around the White House for a long time because rockets have always been tied to the government.

Dario is new to this game, but he’s there right now, and he’s getting the “This is how it works” speech right now in the Oval Office, probably. And also, remember, Dario and Elon—Dario’s renting all of Elon’s chips in Colossus 1 and Colossus 2. So, they’re already in bed together.

So, it seems very likely that the future governance of the entire world is now going to be a conversation between David Sacks, Donald Trump, Elon Musk, and Dario Amodei. And Dario’s obviously publishing documents about, “This is how it should work. Here are my preliminary thoughts.” So, all of that is tied together.

Sam, where are you? What’s your contribution to the whole future of world governance? You were called the most powerful man in the world 1 year ago. Now, where are you? And so, I think price cuts aren’t really going to put him back on the map, but it certainly is a move in that direction.

Dave

But my core observation there is that he can delay the IPO with no personal impact to his personal finances.

So, you don't buy that the rationale for this is that the company is going to get exponentially more valuable, and delaying until after RSI is the best way to extract value?

Alex

I don't even know if he said that. I thought that his point was more that RSI is so risky that we need to figure it out before we go public. That was my read on it. Maybe I misread it.

Salim Ismail, I want to tee up the third part of this story for you.

Okay.

Do you want to comment on this first before I tee it up?

The broader point here is that, clearly, with these price drops, intelligence is becoming utility and we're commoditizing it. And I think this allows—this is the winner here—every enterprise and every startup to be able to run more intelligence.

The outcome, I would say, is here's a Salim's law that I'd throw out: every 10x drop in tokens means we're doing 100 times more experiments.

I love that.

And I think it's just going to be great for the world.

Yep, that's awesome. All right, let me tee this up for you, Salim, first, and then Alex afterwards.

OpenAI's engineering lead on Codex made a quiet announcement that could be the single most futuristic story we're covering today. He wrote that Codex can now set its own goals. In his words, “Everything we build, we also build as a tool for agents.” This is a generalization of metaprompting, where you let the agent set its own task based on your intent.

A developer on the Codex team added the following: “Basically, I never write my own goals anymore. I ask Codex to write one for itself and one for each subagent it spawns.” We've crossed the point, Salim, where instead of telling your AI what you want it to do, you tell your AI what you want and let it decide what to do for you. It's the beginning of AIs setting their own objectives. Thoughts?

Well, this is that part where, once you get AI's inner loop going, right, where you have recursive self-improvement at the workflow level, you start moving workflows over.

By the way, our pilot program for the organizational singularity is pretty much set now. Next week we start. We have 10 companies. We're going to move them through this three-month process of rewriting yourself from a traditional, human-centric organization to an AI-centric organization.

At the edge of that path, burn that pathway in at the edge. Super exciting.

We've picked 10 very different companies to start off with so we learn the most, and we're going to learn as a group and see where this goes. Once you start that inner loop going, then obviously every bonus or increase in AI adds to that learning loop.

Essentially, what Satya Nadella said in his article over the weekend ratifies this entire thesis. You might as well have written half of Solve Everything and half of my organizational singularity thesis. You create that inner flywheel, and then essentially you just start moving more and more workflows to that. That's the massive opportunity for every enterprise, because the core of their intelligence and the future of value in any enterprise will not be what data you had or how many customers you had. It's how quickly you're learning and taking advantage of a learning loop to augment products and services, offer new ideas, evaluate strategic alternatives, et cetera.

By the way, I'm updating the book right now about once a week, and we've launched a free cloud skill that goes along with it. Somebody tweeted, sent us a note, going, “This brought tears to my eyes. The fact that I can now run my entire company on your model.”

Where do you just want to go to find that out, Alex?

Alex

Go to openexo.com. It's free to register, and it's free to download the book and the Claude skill. Every week we're updating it, and I'm going to be putting out a video on what changed this week in terms of what we're seeing out there.

We've already absorbed Satya's stuff into it. We have a new chapter on how you need to have a data lake and move all your enterprise data out of ERP systems, et cetera, so that you have free access to it right now.

Today, if you look at the architecture of an enterprise, you have your cloud and connectivity, and then you have your ERP and operational systems. Your data and process workflows are all bound into the horrible mess of spaghetti there. Then you're trying to layer AI on top of that, and of course it's not getting a lot of value because it's all stuck, all bound up.

The future will be your connectivity and cloud, a proprietary data lake, and then workflows on top of that. This is what the ERP folks are freaking out about, because if they go to that, they lose their stickiness. They're fighting like hell to retain their position inside the corporate stack. It's quite an interesting tension playing out.

Alex Kantrowitz, Codex can set its own goals. What does this mean? Implications, please.

Alex Kantrowitz

Self-determination for AI agents, obviously. We're doing personhood, yeah. No, AI personhood. We're speed-running the accelerando future. I think that goes without saying.

I also want to comment on the Sam Altman and RSI speed-up and IPO-timing story. I think there's a shallow take and a deeper take. The shallow take would be, well, of course Sam is saying this. Sam feels pressure to generate revenue to justify the IPO, and most of that new revenue is likely to come from Codex, and Codex is being designed for recursive self-improvement.

A sort of glass-half-empty take on the RSI takeoff and OpenAI IPO timing is, well, Sam is just putting a happy face on the need to grow Codex revenue even more before OpenAI is ready to go public. That's the glass-half-empty, negative take.

There's a deeper take that's more profound that I haven't heard anyone discussing, which is: if we actually take this at face value, and not as some sort of spin on Codex financials, if IPOs are being delayed because large private companies don't need external capital or external liquidity because they're achieving recursive self-improvement, that, in my mind, marks potentially the beginning of a decoupling between technology and capital.

Broadly speaking, a lot of people, a lot of the commentariat, are hand-wringing, as has been the case for more than a century, about capital substituting for labor. What happens when AI takes all of the human jobs? That's capital versus labor. I've heard almost no one commenting on what happens if technology substitutes for capital.

Will we need capital in a few decades? Or does capital itself get displaced by technological advances? This, in my mind, if the story has legs and if it isn't just a happy spin on Codex revenue generation or otherwise OpenAI's balance sheet and income statement, is a canary for recursive self-improvement, which, underneath it, is technology improving technology, maybe not even needing capital as much in the future.

If we start to see technology decouple from capital, this is an early warning signal for what a post-capitalist economy could look like, where technology is just so empowered that you don't actually need capital very much anymore.

I agree. I just have to add: keep in mind that OpenAI just raised $122 billion, bigger than the SpaceX IPO capital raise. They literally just raised that money. So, I agree that that is the inevitable outcome, but they're also so tanked up right now. They don't need the IPO to keep up with SpaceX or Anthropic on any capital-raised metric.

Fascinating. All right, I'm going to move us on to our next—

Wait, wait. I just want to double down on Alex's point here. There's something really profound here, right? In the past, we valued technology, we valued assets, we valued stocks. John Hagel used to make this point: we're moving from valuing stocks to valuing flows, right?

But now we're moving from stocks and flows. Forget even the thing—the data and the intelligence loop is what we're going to value in the future. And that's a very profound shift for how we're going to value everything going forward.

All right. If this is all exciting to you, I want to invite everybody listening to the 2026 Moonshot Gathering. We're going to start unveiling who is going to be at the Moonshot Gathering. Of course, the Moonshot mates will all be there together. Announcing today, Palmer Luckey is going to be joining us there. Palmer, the CEO of Anduril, an extraordinary entrepreneur, will be joining us during that day. Astro Teller, the captain of moonshots at X, is going to be there teaching you how to create a moonshot organization, and we've got Kathy Wood, the head of ARK Invest, talking about investing in AI. It's going to be 1,500 builders, networking opportunities galore. We're going to have the culmination of the Future Vision XPRIZE. This is the world's largest film competition, bringing us the next generation of Star Trek, if you would. We have $5 million in prize capital generating these future films. You'll see the five finalists. We've got the Gemini XPRIZE. This is the world's largest hackathon, a 90-day hackathon. We have over 15,000 entries so far. The top five builders who have gone from zero to significant revenue in 90 days are going to be there teaching everybody what they did and how they did it. Salim, you're going to be covering the organizational singularity. What are people going to learn from you there?

We'll go through a whole dedicated flow on how you navigate from your legacy thinking on building a startup to the full pure-intelligence stack and building the ExO 3.0. We'll talk through how people take an existing company and rewrite it for the new world.

Our initial estimates are that if you build a pure-play ExO with intelligence at the heart of it, your performance goes up about 100X compared to the legacy.

So, if you want to spend some time with Salim on that, we'll do a whole segment and section on that for those interested in that part of it.

And also, Alex, I'm super excited for you to present the paper that you have brilliantly and primarily authored, Solve Everything. I'm proud to be your co-author. What are you going to be covering?

Alex Wissner-Gross

Oh gosh, this makes me sound like a management consultant, which I'm very much not. I want everyone to solve the world's hardest problem. So, I imagine this will probably take some form of “ask me almost anything.”

But to the extent that folks who were attending the Moonshots gathering are interested in solving the world's hardest problems, I would love to help you however I can, even if it's just answering questions about the book or essay with Peter, or otherwise encouraging you to aim much higher than just building yet another SaaS. Let's solve everything together.

Our next topic: data centers.

Let me give you the single most exponential number in AI right now. Epoch AI reports that since Colossus 1 came online in August 2024—this is what Elon built in 122 days—the record for compute in a single AI data center has doubled every 7 months. There's no sign of any slowing through 2028. Globally, AI computing capacity is now growing 3.3X per year. That's the demand curve, and it's going vertical.

Now, this is where it all hits the wall. If you want to build a data center—and this blows me away—there is a 2.5-year wait on power transformers alone, and a 3-year wait on step-up transformers. The bottleneck in AI is no longer the chips; it's not even the money. It's the boring, unglamorous, century-old electronic hardware.

I did a little search. The companies leading in these areas are Hitachi, Siemens, GE Vernova, Hyundai, Hisung Hico, Virginia Transformer, and DeltaStar. These companies have been rocketing, like 100% to 400% year-on-year growth. Alex, what do you think about this?

Alex

There's an elephant in this room, not just in the other room. The elephant is terrestrial versus orbital data centers. If you just extrapolate this trend out, you find that, for the foreseeable future, the largest data centers need to remain terrestrial, while perhaps by the early 2030s, barring some left turn in civilization, much of the marginal new compute will remain in orbital data centers.

We have this sort of weird bipolar civilizational compute footprint where, if you need to do very large coherent training runs, you'll do it on land, but if you want inference, where you don't necessarily care about coherence, you'll deploy it to orbit. If you sort of squint, maybe Earth—at least until we figure out how to build very large coherent clusters non-terrestrially—becomes the training hub, and space becomes the inference hub. It's this sort of weird gradient between them.

We build on land; we deploy to orbit. That's, I think, the regime we're likely to find ourselves in, again barring some left turn in the early 2030s.

The question—and the elephant in the room—is, at some point, if Elon's Dyson swarm or competitors' Dyson swarms come to pass, someone is surely going to look around and ask, again barring some radical advance in distributed training, why do we have so much compute in orbit that we're not using for training? And why aren't we building larger clusters?

I think at that point the future bifurcates. Either we get some radical algorithmic advance in distributed training, in which case this trend toward ever-larger single coherent clusters doubling every 7 months peaks. We get sort of a Hubbard peak for the size of data centers, and then data centers start to shrink and become more distributed and more edge-like. That's one possibility.

I do think we are very likely going to get major algorithmic advances for distributed training sometime in the next few years. Possibly we already have it, and it's just not evenly distributed yet.

Or, if it turns out—although I suspect this is not the case—that there is no good algorithmic way to decentralize training, then the elephant left in the room is: How can we build extremely large orbital coherent training clusters? For that, low Earth orbit and sun-synchronous orbit are not looking that attractive. But what about on the Moon?

Train it on the Moon.

Alex Wissner-Gross

That is where I was going.

On the south pole, baby.

Alex Wissner-Gross

Yeah, in Shackleton or otherwise. That then becomes the argument for lunar data centers.

Yes, agreed. Salim, I was reading into your facial expressions. What are your thoughts here?

Well, I think it's a very, very exciting future. I think it's further away than people think, and it's far enough away that we better solve our problems on Earth. I remember having a conversation with Masa's team at SoftBank about his 300-year vision. I was like, “Great to have the 300-year vision. Let's get through the next 20.” And then we all laughed and took it out.

I think we need to spend quite a lot of time focusing on how we navigate that. Just look at that export order and the broader implications of that, and we have to solve some really big problems here. That will then enable us to change things and take full advantage of what Elon's doing.

Dave, 3.3X growth year-on-year of data centers at the same time that 50% of the 9 GW of planned 2026 compute has gotten delayed due to organized activist protests. We saw the same kind of protests hit us in the nuclear power plants 30 years ago, and the US has had zero new reactors. China has 100. Thoughts on that?

It's funny you bring that up, because on my graduation day, there's always somebody picketing for some reason, and it was nuclear—MIT's involvement in nuclear research. But it's always irrelevant in hindsight; it's just noise in hindsight.

I do want to start by saying I completely agree with what Alex just said. I reached the exact same conclusion: lunar data centers are inevitable. You need some kind of thin atmosphere to protect you. The solar-efficiency drop is irrelevant. The panels are so cheap compared to everything else in a data center, and the cooling is so much easier on the Moon than it is in orbit.

That begs the question: Are we going to have Mars data centers as well? But I also want to say, from an investment point of view, regardless of what happens in space, terrestrial data centers are going to continue to grow as quickly as we can build them—at a minimum as a backup plan, but at a maximum as the bridge to the ultimate Dyson swarm.

And who builds the best data centers, right?

Say again?

I said, who's building the best data centers right now? Elon is. He's become a hyperscaler.

Yeah, well, Chase Lochmiller at Crusoe, Project Stargate. Those are the two: the Larry Ellison universe over there and the Elon universe. They share each other's houses. They're not competing. They're both building.

They both recognize there's infinite demand. This is an all-boats-rising-with-the-tide situation. If you can get the transformers, if you can get the generators, get the solar panels, you're going to make money, period.

Remember when we interviewed Elon, too? He was saying, “Look, 10X global GDP expansion in 10 years.” So, you go to over $1 quadrillion of GDP.

I keep asking all these young founding teams, “How many people do you think are really working on anything foundational in AI? Give me a number: 100,000, 10,000, a million, whatever it is. Take $1.2 quadrillion, divide by that number; that's your quota.”

And it usually comes out to, “Wow, my quota is $10 billion.” Just to be on par, I have to produce $10 billion of GDP expansion.

Oh.

And that really opens their minds to think bigger. Build bigger.

What a way of framing it. That's awesome.

I love you guys. You're so brilliant. All right, I'm going to move us on to our next conversation point, which is the future of work. Let me share a video from a friend of the pod.

The question is: Should AI pay taxes? Andrew Yang says yes.

Let’s listen to this video and discuss it.

Andrew Yang

We should be taxing AI, the robots, and the agents, and trying to lighten up on all of the taxes that we as workers pay and also that employers pay, which end up discouraging companies from hiring people. I’m the CEO of a company, Noble Mobile, and when we hire someone, 40% to 50% of the money we’re spending does not go to the worker, either because it’s Social Security taxes, healthcare, or income taxes that are paid by the workers. So, let’s try to lighten up all of that and tax the bots.

All right, tax the bots. Salim, let’s go to you first on this. What are your thoughts? Yes or no?

No, because you’re just going to slow down progress massively in hugely important areas, like solving cancer and other critical areas like that. He’s basically saying, “Tax cognition,” and I don’t think that’s workable.

You can tax the outputs of cognition. But we’ve always been taxing labor, and now labor’s not that taxable because labor is kind of evaporating. So, you have to figure out how to tax capital and the outputs of cognition. But if you tax, say, the tokens of the robots, you’re just going to slow down progress massively in hugely important areas like solving cancer and other critical areas like that.

Dave?

Dave Blumberg

It’s irrelevant. We already have a corporate income tax. So, if you stop paying employees, that becomes profit. The profit gets taxed. The collections on it are almost identical. It’s still silly.

And the same with the Bernie Sanders version of it: “Oh, yeah, tax the bots.” It’s good for populists. Everyone’s worried about AI, so let’s tax it. Great thing to say if you’re a politician, but we already tax it because we tax every corporation. So, we’re fine.

You could make a case that people dodge the corporate income tax too much and we need to close those loopholes. I think that’s a no-brainer. I don’t know why we allow those loopholes. But there’s fundamental restructuring required here. We already tax every corporation. These are very big, very profitable corporations. They’re going to pay a ton of tax. You want to raise the tax rate, raise the tax rate. You don’t need a new law.

I don’t think you should raise it. Probably not. A tax on AI specifically would probably be, presumably, an excise tax—a tax on particular goods and services—not a broad income tax or sales tax. But in my mind, every tax is a distortion. So, the question you have to ask yourself is, do you really want to distort the economy by penalizing superintelligence?

I think there are probably far better ways to mitigate any side effects of AI displacement of human labor than just taxing the AI. I think that’s probably a relatively naive approach.

On the spectrum from UBI, universal basic income, to UBS, universal basic services, to universal basic equity, UBE, or universal basic compute/capability, UBC, I tend to think that using superintelligence to drive the cost of living down to near zero is probably far more effective as a remediation for any AI displacement of human labor than just a naive taxation of superintelligence.

I don’t think we necessarily want the perverse incentives, as Salim and Dave have mentioned, from disincentivizing superintelligence through taxation.

Dave Blumberg

Totally. And this is totally right. You know what you’re going to see, Alex? On UBI, you’re going to see more and more politicians saying, “Hey, let’s tax you. If you lose your job, we’re going to tax the AI that took your job, and we’re going to give you that money.” They’re all going to say that because that’s what people want to hear.

But then, when you say, “Well, I understand your idea on the tax side. What’s your idea on how to give? Do I get paid exactly what I was getting paid before, or is there some declining scale?” none of them will talk about that, because it’s very hard. It’s not at all obvious how you do it.

But that’s the problem they should be working on. The collection is the easy part. Who gets what is by far the harder part. You’ll see none of them talk about it.

The first person I heard speak about this was actually Jeff Bezos, saying, “Let’s tax the robots.” I wonder if he still believes that at this point.

Alex Karp

The irony is that, just as we were discussing earlier, Dario puts out this lengthy essay saying, “Regulate us,” then the government regulates them, and then he cries foul.

I fully expect that the frontier labs will say, “Give up golden shares,” or “Tax us,” or otherwise create universal basic fill-in-the-blank. I strongly suspect that in the next few months, the government’s going to do exactly that, and everyone will cry foul.

I think any frontier labs that don’t have a precise idea for what they want the desired economic end state to be should probably advocate a little bit less for just blanket “Yes, tax us” to create universal basic fill-in-the-blank, and have a really concrete idea of what the end state is.

The end state could look like some sort of universal basic equity or dividend. I think that’s probably the easiest policy to implement. It doesn’t require deep thought about compute. It probably requires, at least in the United States, some sort of sovereign wealth fund to manage it, which we’ve talked about on the pod in the past.

But I do think that this is coming to a head—all of this taxation via excise tax or via some sort of golden share-based universal basic dividend or otherwise—in the next few months, because we’re about to see the OpenAI and Anthropic IPOs.

Yeah. Well, I think if you listen to those old Elon Musk videos, which I highly recommend, from when SpaceX was first getting off the ground, they kept asking him, “How can you realistically expect to compete with NASA and George W. Bush?”

It gives you a real sense of how transient and irrelevant any given government is compared to a long-term mission like Elon’s or like Jeff Bezos’s.

All right, I’m going to close out our future of work section to discuss a Metatrend Substack I put out titled “The Most Dangerous Demographic in History Is About to Get Bigger.”

I don’t like dystopian stories, and this is more about getting the conversation out there, because I think it’s an important conversation for us to have and understand what history teaches us.

Every major revolution in modern history shares one ingredient. Not ideology, not technology, not even poverty. It’s young men, usually between the ages of 18 and 28, who are educated enough to have an expectation about what the future should look like. That gap between what they are actually inheriting and what they were promised is a combustible force in human history.

I want to share some data, and I’d like to have a conversation. We’ve talked about the notion that we’re going to have some turbulence over the next 2 to 8 years, and I want to do everything I can to help ameliorate that turbulence. I’m doing that with the Future Vision X Prize and the Gemini X Prize, in part, and through our conversations here.

So, let me share a couple of slides of data. On the left is from Charter, the Federal Reserve Bank of New York. What we’re seeing here is the difficulty in getting jobs as a function of all college graduates, all workers, and recent college graduates. What you see in that red line is that recent college graduates are the group that are out of work the longest and having the most difficulty. That spike you see, of course, is COVID.

On the right are some charts from our friend Erik Brynjolfsson at the Stanford AI Index Report. The upper chart shows software engineers, and you see a distinctly divergent line, with 22- to 25-year-olds dropping in terms of software developers. Young software developers aren’t getting jobs, dropping pretty linearly.

The bottom chart here is customer support agents. Again, that young group, 22 to 25, is diverging from the rest.

We’ve talked about the data being murky. A lot of people are coming out right now, including Sam Altman, Dario Amodei, and Marc Andreessen, saying, “Listen, the job apocalypse is a myth. It’s not happening.”

While I might agree that, in general, we have the lowest unemployment rate ever, the concern is those young men, mostly, out of work, because they haven’t gotten jobs after going through the social contract.

One more data point here to show: I did the research looking at, basically, revolutions over time led by the youth. There are 20 of them, starting with the French Revolution, where the youth helped overthrow monarchy and aristocratic privileges; the Russian Bolshevik Revolution; the Chinese Youth Revolution; the Cultural Revolution; the Iranian Revolution; the anti-apartheid youth revolution in South Africa; the People Power Revolution in the Philippines; the First Intifada; Tiananmen Square; the Arab Spring; and the Hong Kong Umbrella Movement.

So, I raise this again because I think it's a conversation that needs to be had. We need to get out in front of this. Salim, you want to kick it off?

Well, in the 20th century, the big fight was capitalism versus socialism, right? In this, we've got a better fight now, which is: Who owns the exponential surplus? Who owns abundance, and how do we distribute that? Whether it's UBI, UBS, UB something.

It's going to create a lot of angst, and what I think the bigger issue today is social media amplifying those by, like, 100×. I think that's the biggest structural problem between today and what we had before, because bad news and panic spread so much faster than good news and truth. You really have to solve for that structural imbalance.

There's going to be no clear answer to it, except that we really have a huge problem in terms of how to distribute some of that. Europe takes the view of, let's create a protective layer for all of society. Here in the US, we have the opposite view. We have some big problems, like 50% of the country can't put $500 together in an emergency. That's just a staggering commentary on protecting that lower layer.

We've got to figure out how to get abundance down to the bottom very, very quickly. We get this comment repeatedly from our community members and in the comments, et cetera: How are we going to help that thing? The Abundance Prize, Peter, that is on the table is, I think, maybe the most important thing in the world to focus on in terms of getting the cost of living down to a very low level in a powerful way.

Yeah. And by the way, whether or not it's true that youth are not getting jobs, or whether or not it's true that we're going to have a job apocalypse, people believe that, right? There's a pandemic of fear. In that pandemic of fear, that's where negative things happen. All right, Dave.

Dave Blakely

Yeah.

Yeah, good. The point I'm making is that we've had some broad policies that have allowed a pervasive culture of fear to permeate the entire US social structure, right? That's going to add fuel to the fire. It's gasoline on top of everything else. Dave, want to jump in?

Dave Blakely

Peter, it's brilliant that you're bringing this up and observing it. The one data point on that slide I'm most familiar with is the Iranian Revolution, because I was in the country right before it broke out. In that case, it was the college students who overtook the embassy and then took all those hostages forever. That started the whole thing.

Those college students said, “Look, what we are doing right now in this country is not right. We shouldn't have a monarchy. This is just not right, and all the concentration of wealth is nuts. We need something new.” They weren't thinking that it would end up being 45 years of religious rule. They did not have that in mind at all.

They get really angry, but they don't have a plan. They just know this current plan isn't right. I love the fact that you're taking this head-on. If you look at the amount of money that just got raised by SpaceX, the amount that OpenAI has in its charity, and then, if Anthropic goes public, they should be taking at least $5 billion or $10 billion of that, and at least half of that should go back through XPRIZE and get laser-focused on this problem, because they're the ones that are going to suffer if there's a massive uprising. They have every reason to invest in figuring out—

But they're going to suffer a lot.

Dave Blakely

They're going to suffer a lot.

Alex, what's your take on this? I'm super curious, my friend.

Alex

I think this is a problem that will be, to the extent it is a problem at all, the problem of disemployment or underemployment as a result of AI advances, and the elite overproduction theory by Goldstone in revolutions. If it happens this time around at all, I think it's likelier to be felt not within China and probably not even within the US, but in the rest of the world that doesn't have superintelligence or isn't anywhere close to superintelligence.

In principle, if we within the US wanted to, I think we could create lots of economic opportunities for our “surplus elites” that we may or may not be overproducing by benefiting from superintelligence. We colonize the solar system. We'll take apart the Moon.

But just to be clear, Alex, I'm talking about the next 2 to 8 years, right, versus the long term? Remember, when Dave and I were interviewing Elon—and you remember, Dave—you asked him, “UBI and civil unrest?” He said, “Yes, UBI and civil unrest.”

Dave Blakely

Look, we have a very dangerous cocktail right now of very high expectations, very high connectivity, very low opportunity for certain segments, and very quick mobilization, right? This is a very destabilizing kind of epoch in human history. That is your point: The educated person who's been promised a future and is watching the ladder collapse, that's the real problem here. We need new ladders for meaning, agency, status, contribution, all of that.

You're exactly right. And, Salim, concurrent with that, we have AI-assisted social media, which we've never had before. We're all experimenting with it, and the year before an election, everything always turns toxic. So that's coming up next year. You've got all of that.

The good news is that we're democratizing AI so quickly, and we talk about how everybody should become an entrepreneur. What we really mean is: Use AI to give yourself radical agency. I've been using the phrase, “You have the opportunity for future shape rather than future shock.” Right?

Go there, Alex.

Alex Salkever

I think, though, that to some extent it's entirely possible that we have cosmetic color revolutions without actual revolution, versus, say, the Cultural Revolution, which resulted in millions of people dying and which wasn't even, superficially, a student revolution, but actually was fomented by Mao to preserve his position within the elite hierarchy.

It's entirely possible that, to the extent anyone tells a story about AI revolutions due to elite overproduction, it's actually just being driven by a number of other elites that are mining whatever dissent or discontent there might be. But I want to go back to my point.

To the extent there's a there there, I think it will be primarily felt not by the US, where we do have, yes, even on the 2-to-8-year time scale, the capabilities of creating entirely new classes of work and economic opportunity, and in China, where the Chinese Communist Party is actively driving, via its AI Plus Five-Year Plan, all of this AI kicking and screaming into the workforce.

It's the rest of the world that doesn't have the frontier models, doesn't have the energy, doesn't have the data center buildout that I think is likeliest to bear the brunt of any color revolutions associated with AI, because there, maybe, the economic opportunities aren't as large. I want to connect this also with the Fable story.

If all of the frontier capabilities are going to be export-controlled going forward, then that leaves the US and China in sort of the catbird seat to optimally economically exploit all of these new capabilities for workforce development and economic opportunity for all of these 18-to-28-year-olds. But it basically shuts out the rest of the world.

I think it's possible that export control is actually one of the determining factors in any color revolutions for folks outside the US and China. I'm thinking especially of Europe, where, basically, if they don't have the native sovereign capabilities to drive forward superintelligence, there's a risk that the rest of the world becomes almost vassal states that are dependent on all of these ultra-cheap capabilities being produced by superintelligence and dumped, for lack of a better term, on the rest of the world. That could create discontent.

I'm going to add one other thing. We saw the Sam Altman Molotov cocktail. We saw Eric Schmidt getting booed. There's a growing level of discontent. I call it a pandemic of fear versus a viral pandemic.

I guess I'm hearing a lot of the tech leaders around the world saying, “Don't worry about the AI apocalypse. It's not going to happen. It's disinformation. It's regulatory capture. It's Chinese propaganda.” It may all be true. It doesn't matter.

If, in fact, the US populace is in fear, if, in fact, people believe their future is not going to be as good as was promised to them, that's what foments the revolution. So I think, again, regulators who are listening and tech CEOs who are listening, we need to get out in front of this and help people understand that they're going to have an abundant future, that we're going to support them, and that we're going to create a basis by which they have access to this rising economy that's being created, with triple-digit growth in the next 5 years.

We need to give them some level of security, some level of assurance, some level of hope, right? This is a cry to enable hope, at least in the US and other parts of the world. That's the point I want to make about this.

Alex Salkever

I would just add: If we don't build it, everyone dies.

Dave Blakely

I have to add one—I'd just add one thing. I'd be remiss if I didn't add it, which is everything that Peter just said is exactly right. Concurrent with that, the startups are having a field day. It is the absolute best time in history for startups, which is so counterintuitive because their classmates can't find jobs.

That's great because the startups—there's more venture capital than there's ever been before, by far.

And the AI-native startups can recruit anyone they want out of any graduating class because no one can get a job. So it's really in juxtaposition to the fact that the rest of the graduating class is struggling to find anything to do. So it's really weird, but the startups are having a field day. The incubators are just going gangbusters.

All right, it's time for the AMA, mates. We read your comments, and we are grateful for all of them. Thank you for weighing in and for your questions. One of our favorite parts of the podcast is answering your questions, so let's jump in. Salim, it looks like number 1 is for you.

Do your SCALE and IDEAS frameworks, which are the ExO model, actually force organizations to ask, “Why do we own this at all?” Or do they just let companies scale broken processes faster? And that's from Gear for the Year.

You can think of those attributes. Those are acronyms, by the way, for staff on demand, community, and so on. You can look at the model if you're interested in the full detail. But these are not efficiency tools that are supposed to be layered on top of an existing or broken organization. If you use them that way, you're going to scale pathology. You're going to make bad decisions faster, basically, and more automatically.

The real sequence—the most important thing above scale and ideas—is your massive transformative purpose, your MTP. First, you have to design your MTP. Then you delete your legacy, redesign around the intelligence stack, and then scale. The MTP answers the question: Why do you exist? What problem are you trying to solve?

When you have experimentation, that asks the question, “What evidence do you have?” When you have interfaces or dashboards, you're asking, “What feedback loops are we using?” Autonomy then says, “Who's closest to the information? Let them make the decisions.” So the framework should absolutely force the “why” question, but that sits above the two. Like any tool, you can misuse it by people who want speed rather than transformation.

All right.

You don't want to digitize your bureaucracy; you want to delete it. One of the comments I make to CEOs these days is that if they have a classic matrix structure, you've got horizontal layers that slow you down: IT, HR, branding, privacy, and so on. Every few years, go delete all those horizontals and reinvent them.

Thank you.

Power accrues to the horizontals because it's really easy to say no in HR or legal. There's no incentive to saying yes. If you say yes, your life's a mess trying to manage a complex bureaucracy. So every few years, just wipe it out and rebuild it. It'll force you to reinvent things in a modern way.

That's if you want to take a hack at the old bureaucracy, but the real model is that you have to rebuild yourself in a new model. That's what we do with the organizational singularity stuff. Sorry for the plug.

Okay. Alex, it looks like number 2 is for you. “Alex, why are you anti-Bitcoin?” This is from The Bitcoin Revolution.

Alex

Alex, to paraphrase Jessica Rabbit, I'm not anti-Bitcoin; I'm just drawn that way. I don't think Bitcoin is a productive asset. I think blockchain is this amazing mathematical technology that fell back in time from the future, and I think blockchain technology is super interesting. Bitcoin itself is a lot less interesting to me.

As far as I can tell, maybe there's something fundamental that I'm missing, but I don't think so. It's not a productive asset. Similarly, I'm not anti-gold; I'm just not interested in gold because, by and large, it's not that productive an asset—at least not nearly as productive as the values or the prices that it seems to trade on.

I think if the question behind the question is, “Why don't I invest money in Bitcoin?” I would say there are a few reasons. Again, not investment advice. One, it's not a productive asset, unlike stocks and bonds that are actually being used to create new wealth in the real economy.

Bitcoin, as far as I can tell, doesn't create any real wealth. The way you know that, aside from the fact that it doesn't pay any form of coupons, dividends, or interest, is that it's not generating new ideas. It's not creating new structures. It's not expanding humanity's future freedom of action. It seems to be much more of a speculative asset. That's reason 1.

Reason 2—and this generalizes beyond Bitcoin to other crypto assets in general. Parenthetically, I do like stablecoins insofar as they enable more efficient financial-asset transfers and to the extent that they help stabilize the US dollar. So, I like stablecoins.

I'm biting my fist because I've got so many comments. We really need to have our crypto debate.

Alex Epstein

I bet you this is going to be rapid-fire answers, gentlemen.

Yeah, okay.

Alex

I will close by saying Bitcoin, as well as alternative cryptocurrencies—the third elephant in this particular room—is that AIs and AI agents are fully empowered at this point to come along and invent their own L1s. The argument that Bitcoin bulls make, that because Bitcoin was merely the first L1 it occupies some sort of primacy in the future light cone, I think falls apart.

Again, not investment advice: When AIs come along, to the extent that they find any value in cryptocurrencies, they can just invent their own L1s, and they'll probably be much better than Bitcoin.

I look forward to that. All right, Dave, you have a choice of number 3 or number 4.

Dave

I'll let you choose which one. I can take either.

Go jump in. Choose one.

David Friedberg

All right, all right, I'll take number 3. The US already has a 10% equity stake in Intel. Has that helped America at all? Why would AI lab stakes be different? And that's from Gear for the Year.

Yeah, massive, massive difference. The US took a 10% stake in Intel and gave it $10 billion to bail it out and get chip fabs back into the US. They were threatening to shut down investment in their Ohio 1.4 nanometer project, which is critically important for the future of the country. It's very similar to when the government bailed out the car companies. We need those, and it was a great investment.

The AI labs are already in the US, and they already have tons of money. Why would you be buying them? “Oh, but it'll get a great return for the American people.” You have the power of taxation. You can take any amount of money on any day you want. You don't want to invest in things.

Having the government be an investor is always a bad idea because some future administration will mangle the investments, and there's no continuity of thought in the government at all. You don't need to do that unless you're helping bail it out or helping it grow some strategic function for the country.

All right, I'll take number 4. How do other democratic nations handle sovereign wealth funds? This is from John Sebert.

John, I think the gold standard is Norway. Its Government Pension Fund Global is now worth something like $1.7 trillion, and it's built on oil revenues. It also owns something like 1% or 2% of every publicly traded company on the planet. Singapore has Temasek and GIC. The Gulf states have sovereign funds: the UAE has ADIA, and Saudi Arabia has PIF.

I think the lesson here is that once you've got this kind of sovereign fund that's controlled by an outside investment group that's not tied to the politicians, so the politicians can't give favors or spend it, it drives an incredible return for the citizenry. I think it would be extraordinary if we could do that here in the US, if in fact you're from the US.

David Friedberg

They've also been trained on our collective data. So there should be a shared outcome.

What's that?

David Friedberg

Collectively, they've also been trained on our collective data. So there should be a shared outcome.

Okay. Let's kick this one. Dave, the first one's for you. Why are you fine with the government funding SpaceX but not taking equity in AI labs during a high-risk transition? Did the government fund SpaceX? I don't remember that. I think they bought it.

Dave

The government gave a contract. So, in 2008, after SpaceX had its third launch failure and had its first Falcon 1 success on its fourth try because it had an extra rocket laying around, SpaceX won a $1 billion contract for Falcon 9. It won a contract. It wasn't given money. It was given a service contract.

This was commercial crew, as I recall.

David Friedberg

Yeah, commercial crew, right? The space shuttle was going out of business, and we needed a replacement for the space shuttle.

Yeah, okay. So that never happened. So that answers that question. Even if it had happened, having a space launch capability is a national priority, and obviously Elon did a good job of it. That would have been a great move by the government.

The second part of the question is about not taking equity in AI labs during a high-risk transition. I think the government just proved it can block any product. There was a little argument for a while that the government should be on the boards of these companies, which is hilarious because a board member can't sit there and say, “Hey, don't release Fable 5. I'm stopping you.”

The government already has massively more authority than a board member does. You don't need a board seat in addition to that. You're just going to mangle it. It's purposeless. Like I said a second ago, the government can tax any amount they want, anytime. Why do you want stock, too?

I think the government will take stock because it's politically cool, and the population is supportive of it.

Dave

But it makes no sense.

Give me some, too. All right, I'll take the next one here, number 6. Peter, do you believe a quantum breakthrough in 3 years can actually break the entire crypto stack? This is from @ListenGrasshopper.

So, my short answer is 2 points. One, I think the threat is real but not on a 3-year timeline. I think that's aggressive. And I think the entire industry is working on it—we heard this when we had Brian Armstrong, the CEO of Coinbase, on the pod last week. He is working on a quantum-resistant algorithm, and the whole industry is. Everybody's going to get there before we have to worry about this. So, I'll keep it brief on that regard. Alex, you want to choose?

Alex Epstein

He wants me to answer question 8. So instead, I'm going to answer question 7. Question 7 asks: If knowledge work is cooked, why hasn't anyone suggested using AI itself to manage the sovereign wealth fund? This is from MGP Star.

I think the premise is incorrect. I think, to the extent the US and other countries have sovereign wealth funds, like Norway, AI is already managing the sovereign wealth funds in the form of AI managing the overall broad public equities markets and other public markets. Almost all of the volume in public equities markets at this point is driven by algorithmic trading.

AI is already managing all of these valuations, including SpaceX's. SpaceX is now the fifth-largest company in the world. That's almost all AI. Obviously, there's a lot of retail, but AI, by volume, is actively doing the price discovery. So the premise, I think, is incorrect.

AI is already de facto managing sovereign wealth funds to the extent sovereign wealth funds have exposure to assets in highly liquid public markets that are themselves almost entirely dominated by volume by AI.

Nice. All right, Salim, this is revenge of Bitcoin here.

Number 8: Why would AI agents want to use Bitcoin when they still have to pay for energy? That's from @capquills.

Two or 3 quick thoughts here. Agents aren't going to be ideological. They're going to be ruthlessly functional. They're going to use whatever rails clear their transactions with the least friction and the most trust, right? Trust is the key thing. Bitcoin's energy use is not the reason the agents will or will not use it. It's whether you can get a neutral, high-throughput global settlement model.

Now, there's something powerful about Bitcoin—why people are getting excited about it these days more so than, say, 5, 7, or 10 years ago. When you have a digital currency, you want 3 points on a triangle. You want to hit decentralization, security, and scalability. When Bitcoin first emerged, it hit decentralization and security for the first time ever. And people are really excited about that because no digital currency had hit those 2 points, but it didn't solve for scalability.

So, you had the rise of Ethereum, Solana, Cardano, and all the altcoins trying to solve for scalability, but in doing that, they typically compromised on decentralization or they compromised on scalability. There was a big Ethereum fix a few weeks ago. If Ethereum was really decentralized, it wouldn't have been able to do a fork like that, which tells you Ethereum is not really decentralized, right?

So, this is where you saw the FTX collapse or the Luna collapse. They collapsed because of either a failure of decentralization or a failure of security. Bitcoin had the first 2, but it didn't have scalability. With the rise of the Lightning Network, created by Blockstream a few years ago, it solves for scalability.

And so now you have all 3. Bitcoin hits all 3, and therefore people are getting more and more excited about the future of that. And we can get to the broader debate later.

All right. This was quite the show, gentlemen. It was fun, honestly.

Alex

Yes, and when we say “gentlemen,” we stretch the term to fit all of us. That's an old P. G. Wodehouse joke.