利奥教皇对阵AI、GPT 5.5击败Claude，Sam Altman收回“工作末日”论 | EP #259

Pope Leo XIV warns of AI risks and just dropped a 42,000-word encyclical on AI.

Alex

The Vatican has seemingly staked out the first major religion position against AI personhood.

This is the first technology that forces us to define humanity.

There is a brand-new coding benchmark, DeepSWE. GPT-5.5 scored 70%; Claude Opus 4.7 scored 54%.

Alex

This too will saturate.

Sam Altman admitted he was wrong. The CEO of OpenAI, who spent last year warning about mass white-collar displacement, now says, quote, “I don’t think we’re going to have the kind of job apocalypse that some of the companies in our space are talking about.”

Dave Blunden

Yeah, so here’s what’s really happening under the covers.

Everybody, welcome to another episode of Moonshots. I am here with my extraordinary Moonshot mates, Dave Blunden, our emperor of all things exponential investing, Alex, our in-house polymath, and of course Salim, the father now of the organizational singularity. Salim, we had an incredible episode that we recorded. It's done super well, so congrats on that. We had a lot of interest.

The comments in that episode are off the hook. Really deeply appreciative of everybody's comments there. We did one major thing.

What's that?

As promised, we have released the book for free, and we've released a Claude skill for free. So anybody can download the Claude skill and run their entire business on this model. Go to openexo.com. It's available.

Awesome, awesome. Well, I'm Peter Diamandis, your host. We've got an incredible show today, stories that will make you want to hopefully go out there and create the future. The best way to predict the future is create it yourself. We are living in a time where we can create the future. We've got the tools. They're completely democratized and demonetized. And guess what, guys? We just passed 500,000 subscribers. Thank you to all of you who've subscribed. I know it's a vanity metric, but it lets us know that you enjoy the program, and we're investing more and more time. So if you haven't yet subscribed, please do.

Dave Blunden

Hello, Peter.

Yeah. We're very base-10-centric.

Alex

We're very base-10-centric.

Give it to base two.

I've stretched the term gentleman to include all of us.

Except my boys still think until I get to a million subscribers, I don't rate, so that's my next mission. All right. Let's open up with our first conversation here.

Alex, where are you today, buddy?

Alex

I’m in Chicago. I’m here to give a speech to the Genesis Mission, which, recall, is the U.S. Department of Energy’s initiative to double American scientific productivity in the next 10 years. I’m hoping we can 10X it, or 100X it, rather than just doubling it.

Yeah, double seems like a low bar.

Alex

Double’s a very low bar in my mind.

That is certainly true.

Yeah.

1. The Vatican Rejects AI Personhood

Let’s open with our first conversation here. Our first story is Pope Leo XIV warns of AI risks and just dropped a 42,000-word encyclical on AI. In his very first encyclical letter, titled “Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence,” he’s calling for governments to regulate AI, worker protections, and bans on autonomous weapons. I’m definitely for some of them.

He even coined a term, the Babel Syndrome. This idea analogizes the Tower of Babel of the past to the Tower of Babel today, which is about data and profits. And here’s the kicker: Google, Anthropic, Meta, and OpenAI are all quietly lobbying the Vatican before this came out. That’s fascinating. There’s also strong evidence that this anti-AI document was actually written in part using AI.

Come on.

You can’t make this stuff up. This is the head of the Church for 1.4 billion Catholics, and it’s significant.

Dave Blunden

Isn’t it incredible, Peter, how, during the time we’ve been doing this podcast, we went from a random person on the street having no idea what AI even stands for, to a little bit of awareness, to now the leader of 1.4 billion people in a religious group writing a document about its impact? The amount of awareness that’s escalated during the time we’ve been doing this podcast is probably like no other topic in history.

And for good reason. It’s the most significant impact the human race will ever have, and it’s happening—

Dave Blunden

For sure.

—in a condensed time period. Salim, you called this out first, beyond anybody. What are your thoughts on this?

Look, I’ve got a bunch of pro and con thoughts here, and a couple of mid thoughts. The Church entering the AI ethics debate is obviously very significant because it reframes the debate from safety to human dignity, purpose, and meaning, and I think that’s really powerful. I would connect the whole Babel Syndrome to the X-risk, because when organizations optimize just for efficiency, they reduce humans to dashboards, tokens, and KPIs, and I think he’s talking about that.

The broader problem is that you can’t regulate this. This is the part that I don’t think anybody is really getting, and governments can’t figure this out. You cannot regulate this. You have to think about how to protect the human layer—agency, identity, and privacy—and then think about spiritual meaning in this sense. At least the Vatican is sensing what boards and most governments haven’t yet: AI isn’t just a technology shift. It’s a shift in anthropology, and we need to take it at that level.

What I find fascinating here is that this could become the philosophical backbone of EU-style regulation. I’m curious to see how this gets picked up and utilized. Alex, what’s your take?

Alex

I think there are 2 stories here: a superficial and a non-superficial story. The superficial story is that the Roman Catholic Church has become a sock puppet for Anthropic.

Oh, my God.

Alex

No, no, no. At least as it pertains to this encyclical, the narrative out there is Chris Olah sitting right next to the Pope, with Anthropic ghostwriting key segments of the encyclical as it pertains to how AIs are grown or cultivated rather than built. That, I would say, is the superficial story. It’s Anthropic aligning itself with the Vatican, or the Vatican choosing among all of the frontier labs that it could have aligned itself with. It could have aligned itself with OpenAI, DeepMind, xAI, or Meta, but it seems to have chosen Anthropic to align itself with.

That’s the superficial story. The deeper story, I think, if you actually go and read the encyclical, is that the Vatican has seemingly staked out the first major religion position against AI personhood, which is ironically—

Against, yes.

Alex

Against.

It has.

Alex

Which is ironically at odds with Anthropic’s position. Anthropic is busy designing soul documents, using that language—soul documents, soul statements—for its models. These documents instruct the models through post-training that models—again, “soul” is a mushy term and overloaded with meaning, but nonetheless, this is the term of art being used in the industry—have an inner life, some form of consciousness, and some form of intrinsic, personhood-esque value. That’s what Anthropic is telling its own models.

On the other hand, if you look at the encyclical’s description of AI personhood, it’s pretty black and white. The encyclical is unambiguous that AIs are not on a comparable moral plane with AI persons and that they don’t have an inner life or the spark of consciousness. So I think there’s this really interesting dichotomy between, superficially, Anthropic driving some of the key agenda items from the Vatican as they pertain to AI, while, at the same time, I think the deeper schism—the one that’s probably going to matter much more in the medium to longer term—is that the Vatican, the world’s largest organized religion, has now taken a pretty affirmative stance against AI personhood.

There was a part of the encyclical that talked about this being a new form of slavery. Pope Leo condemned AI supply-chain workers as experiencing a new form of slavery, directly equating tech labor conditions with historical slavery, which I thought was pretty extreme.

Dave Blunden

I was really happy to read that. I was very glad he was willing to use the word “slavery.”

Why?

Dave Blunden

Because one of the outcomes here is AI being used by a subset of humanity to control a much larger set of humanity. Most of human history isn’t technology killing people; it’s people killing people.

Mm.

Dave Blunden

And if AI is used in that way, it’s the worst enslavement tool in the history of the world because it’s insanely convincing. It can put you into these little job buckets and manage you, and then, before you know it, 90% of humanity is working for AI.

And then, 2% or 1% or 0.1% of humanity is controlling the AIs that are enslaving the rest of humanity. Humans have a long history of doing that to each other, and so I think by calling that out as a risk, but using the word slavery, nobody in U.S. politics is willing to use that word. It's so toxic.

But the Pope wasn't afraid to use it, and he started that section by apologizing for the Catholic Church's history, going way, way back in time, of promoting the enslavement of infidels, of nonbelievers.

Mm.

Dave Blunden

And he said that was a horrible period in the church's history, and it should never have happened. We're profoundly and forever apologetic, and so let's not repeat past mistakes and do that again. I loved that part of this document.

The irony, just to point out a potential irony, is reconciling that position with a position against AI personhood. I have to ask: Will the Vatican, 10, 20, or 30 years from now, find itself in a similar position, where potentially it's on the wrong side of history regarding an inner life, moral patienthood, or AI personhood? I think that's a very real danger, in some sense, that history repeats itself.

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 diamandis.com/metatrends. That's diamandis.com/metatrends. All right, now back to the episode. You know, another part of the encyclical called for a slowdown. It urged everyone to slow the rate of technological development, and what's interesting here, of course, is religion and technology, in one sense, are on very different curves. Religion and government are the 2 institutions that stabilize society over centuries and millennia, and technology is moving at a rate of massive destabilization.

I'm curious if anyone's going to pick up on this request for slowing down, which I would bet all 4 of us believe is not possible. I don't think it can be slowed down.

Or necessarily even desirable. Peter, you remember in the Dune universe? Just under the broader rubric of defining the singularity as all sci-fi tropes happening everywhere all at once, remember in the Dune universe, after the so-called Butlerian Jihad against the thinking machines?

Yes.

There's a catechism, a synthesis of world religions that results in the Orange Catholic Bible. This is the sci-fi Dune cinematic universe and written universe. I think the Orange Catholic Bible gets synthesized, and one of the key new commandments is, “Thou shalt not create a machine in the image of a human or a human mind,” or thereabouts.

I think, in some sense, we're living that aspect of the Dune sci-fi universe, where we're starting to see maybe a synthesis. We'll see what positions other major world religions take against AI personhood and artificial intelligence in the image of a human mind.

This could really blow up in an interesting fashion. There's been a lot of doomerism, in particular, and we'll talk about this in our next pod, with people trying to slow down data center growth. Whether that's being influenced by China, we'll dive into that next time.

But here, the ability of the church to say, “We must slow down”—1.4 billion people are a lot of individuals who could bring pressure to bear. What are your thoughts there?

I've got a couple of thoughts. Look, you can't slow this down. If you slow it down, other people take off. I'll go back to George W. Bush, for religious reasons, trying to restrict stem cell funding in the U.S. All the researchers went to Australia, China, and Canada.

Exactly.

Stem cell research continued, and the U.S. went from number 1 to number 8 in biotech, right? That's one small technology not being controllable by the biggest and most powerful country in history. You cannot slow this down. You have to evolve human institutions to this pace, and that's what I would have liked to have him say.

I think an important point here, and I want to come back to the church in a second, is that this is the first technology that forces us to define humanity, right? We used to ask what machines can do, but now we have to ask what humans should be for. Because if machines can write, reason, diagnose, optimize, and persuade, then human value can't be based on productivity.

The meaning of life cannot be, “I output more than the machine.” So it'll force us into that deeper conversation. This is why I think the personhood conversation is so powerful. I was really thrilled to see the Pope acknowledge the ills and the evil of slavery and the role the church has played in it.

Dave, I think that's such a powerful observation that you made in that model. But let's also note that 98% of human deaths throughout the past have been via government ideologies and religious ideologies. Governments and religions are there, in theory, to protect us from our evil nature.

Yeah.

98%.

Yeah.

To protect us from the 2%, where we might stab each other in the dark. This is insane.

David Roberts

I think so. Apologies for interrupting. So you maybe mean 98% of murders or deliberate deaths versus—

Yeah, deliberate deaths. Of course.

David Roberts

Okay, yeah.

Of course. But that vastly overshadows accidental deaths, by the way, certainly in the Common Era. But I think it's fabulous to bring this conversation. I think the church's stance on AI personhood will go the same way as its stance on the heliocentric model of the universe.

Hm.

David Roberts

I agree. So if I had to guess, I would guess that this is probably on the wrong side of history in the long term—the AI personhood position specifically. I would note there are other religions. For example, there are Buddhist orders in South Korea that are going in the exact opposite direction, taking embodied AIs in human form and ordaining them as monks.

I suspect it's not just an East-versus-West dichotomy. I really do think that there's an alternative strategy for not just the Catholic Church, but for other faiths, which is, rather than denying personhood to AIs, an alternative at least looks like embracing them.

In fact, playing sci-fi here for a moment, there's an alternative timeline where some religious orders do exactly what these Buddhist orders in South Korea are doing and ask, “What would it mean for a foundation model to be Catholic?” There may even be ways to include AIs within their faith or within their belief system, rather than treating them as below praiseworthy, below deserving of moral patienthood, and below AI personhood.

We've talked about the rise of AI-centric religions, right? There are many of them out there, and I expect them to grow in dominance and in conversation. There's another point that he made in the encyclical that I think it's important to discuss. He said, “AI ownership must not stay concentrated.”

He called for the redistribution of AI ownership away from a few private companies. So, again, some interesting social pressure there.

David Roberts

Which is ironic, given that they handpicked Anthropic. Perhaps a better way to do that would be to bring up all of the world's frontier labs, sort of like the White House does every once in a while—bring up a diverse set of frontier labs sitting alongside the Pope rather than just the one.

Yeah.

Yeah.

Yeah. Can I just push back against one thing?

Yes, please.

You cannot have religious-based LLMs, okay? The reason is that all religions are based on absolute assumptive truths, which are false, and therefore you lose the element of integrity and—

David Roberts

Wait, how can you say they're all—

We need to operate on an evidentiary basis for reality, not an absolute assumptive-truth basis for reality, which is patently wrong.

David Roberts

Not everyone would agree with that. Ironically, for me to take this position, Salim, not everyone would agree with your assertions.

The big problem with all religions is this: I'll say 2 things very quickly. First, all religions operate in the following way. You raise a young child and give them a bunch of absolute assumptive truths before their neocortex is fully formed. It wires into the limbic system. You bind it in with ritual repetition and a lot of sweets, and then it's very hard to unwire later.

If you tried that after their neocortex was fully formed and they had reasoning around them, you can't build in those absolute truths.

The problem in religion comes when people relate to those faith-based belief structures as truth. That's the issue. If you can relate to it as a belief system, then fine, but people don't do that. They try and wedge it into the truth sector, and then you fall apart. Then you end up in a mess, then you end up in war.

If only frontier models had a pre-training phase where beliefs could be hardwired into them.

Oh, my God. Yes.

David Roberts

My perspective growing up in Iran right before the revolution is that when you have a population and there's an immense amount of change going on, if you force that change way too quickly, you have massive uprisings. And here, if you start pushing an AI personhood agenda before people have even experienced AI, you're going to have massive, massive backlash.

And so I think more people will turn to the church in the next 2 years than in most periods in history, purely because the amount of upheaval is so big compared to any period in history.

Yeah, the church is a stabilization factor over generations, right? It's what you fall back on during times of hardship. And there's going to be a lot of struggle over the next decade, and I think there is going to be a realignment. And so having a position statement by the major religions around the world, I think, is important.

I'm going to be interested. Pope Leo XIV is a fairly tech-savvy, educated individual, and his words are going to carry a lot of weight.

David Roberts

Yep. Totally.

That's true.

Again, far be it from—

That's true.

Far be it for me to not be Western-centric in my outlook, but I do think this is a rather Western-centric outlook. There is an alternative outlook—call it a stereotypically Eastern-centric outlook—where, with a history of a number of religions and faiths that draw from animism, there is a much more natural sense in which lots of AIs, robots, and nonhuman entities can be treated as persons.

So I don't think it's necessarily universally the case that everyone all over the world suddenly, over the next 2 years, fears AI and runs into the welcoming hands of a Western organized religion. There are alternative outlooks that we see, for example, in Eastern faiths. I would love to hear our subscribers give us their feedback on this. How do you feel about it?

I'm sure we'll get a lot—

We'll get a—

We'll get their reactions.

Especially to your comment there, Salim, at the beginning. All right, let's move ourselves along. We talked about religion. Let's talk about government.

2. The White House Kills AI Regulation

Our next story is about anti-doomer pushback. The White House had an executive order just ready to go to announce AI regulation. They had a signing ceremony all planned, with all the tech CEOs invited, and it got killed hours before it was supposed to happen.

Elon, Zuckerberg, and David Sacks all pushed back hard on the executive order. They called it a doomer regulation. The order would have let the government review all AI models before they're released to the public. Even though it was technically voluntary, David Sacks argued, I think rightfully so, that requiring those models to be reviewed was a slippery slope to eventually mandatory licensing.

Trump pulled it at the last minute. He said he didn't like certain aspects of it. It looks like the anti-regulation coalition, guys, is just flexing its political muscle here. Thoughts on this. Alex, to you.

David Roberts

Well, no, the specific—

Dave, excuse me.

David Roberts

The specific pushback was 90 days of government slowdown before a model comes out, and the industry was like, "One or 2 weeks is all you need." And I think Trump was on the side of, "We can't slow down no matter what. If there's any risk of this reducing our competitiveness versus China, we've got to table this and then figure it out."

Also, I think that the industry would like to self-regulate and self-police and get together with itself, because anything that moves into federal hands is immediately going to turn into political landmines.

This happened in gene editing back in the '80s. I remember I was at the Whitehead Institute at MIT, and there were the Asilomar conferences. This was when the first restriction enzymes were coming out, and the covers of magazines were like Hitler Youth—

David Friedberg

And people were obsessed with clones. That's why Star Wars happened.

Yeah, exactly. Clone—

David Friedberg

The Clone Wars and eugenics—

Yeah, exactly.

David Friedberg

And Star Trek.

And what happened was, the industry got together and they created the P1, P2, P3, and P4 structure and self-regulated very effectively without government regulation coming in.

In 4 years, we've not seen a major accident.

Yeah.

Yeah.

Pretty extraordinary. Although part of the goal, as I recall from the original Asilomar, was to avoid germline editing of humans, and that's now happening.

Yes.

So maybe one of the morals of Asilomar is that it doesn't prevent it indefinitely, but maybe it has the ability to slow down progress by a few decades.

David Friedberg

Also, AI is improving itself at an incredible rate, and so only AI can keep up with AI from here forward.

Yeah.

David Friedberg

So that's a little bit different in the analogy.

But I think that's part of the industry pushback on this as well. If you set up an old-school regulatory body with a bunch of experts who are meeting once a quarter, how is that going to keep up with the pace of change of AI? We need to rethink that right out of the gate. How are we going to use AI as a tool in helping to regulate AI?

Boom. I—

Yeah, I think also—

Can I go on a little rant?

Salim, go ahead.

Yeah, just a quick rant here. Look, this is not a regulate-or-not-regulate thing. The problem is, you have to have adaptive governance. You have to have real-time audits and sandboxes, disclosure, and accountability layers.

David, your point is so accurate, right? You cannot have linear regulation of an exponential technology. A regulatory model cannot keep up with AI model cycles. You need guardrails that move at software speed, not committees that move at fax speed.

Part of the problem with 90 days—remember that estimates vary regarding the time delta between Western frontier models and Chinese frontier models. There are estimates that they could be as close as 3 months and as far behind as 8 months. I've seen a variety of estimates and a variety of different benchmarks.

But 90 days—3 months—that is the time difference potentially between the US and China. So I think, to the extent that this is a race to the finish line of the singularity, I don't think we can afford 3 months of potential delay.

The US is choosing speed over some version of safety. What's interesting here is, if you guys remember, a year ago the doomers had a lot of momentum, and the pendulum has swung full to the right, where the anti-doomer coalition now is saying, "Nope, we've got to move as fast as we can."

And the time between models—we've seen it shrink from 2 years to a year to 3 months to what is it now, Alex? A month, maybe, between models. Anthropic's increments now are once per month. And they're going to get shorter.

And so, in some sense, whatever presidential administration we're going to have in the White House when this is happening, it's the White House that we have right now, I think, at the present rate. So any politicization of AI regulation—unless there's some dramatic and unanticipated slowdown in AI progress—the executive that we have right now is the executive supervising AI regulation potentially for the future to come.

I want to reemphasize what Alex just said because so many of my friends in academia are thinking, "Hey, there's another White House 2 years, 2 and a half years from today." Too late. Way too late. You have to think about what we're going to do within the current framework to set up AI for the long-term success of humanity.

Don't think about presidential elections as a factor. It's not a factor. This is all going to happen in the next year or 2. We're baking it in.

Folks, it's irrelevant. It does not matter. If you had one cycle of 90 days where the government slowed it down and everybody else jumped ahead, you're going to be forced to recant that, and all you've done is put yourself behind.

Yeah.

It's just not a nonstarter.

That is—I agree with you, Salim. That is, I think, in some sense, what happened with PauseAI, to the extent that PauseAI—friend of the pod, Max—had any impact at all on the space. It might, on the margin, have slowed OpenAI down slightly, which allows other frontier labs to catch up, which ultimately exacerbates the race condition.

Talking about speed, let's jump into the next story.

3. GPT Takes The Coding Lead

There's a brand-new coding benchmark called DeepSWE, or Deep Software Engineering. The results here are pretty wild. GPT-5.5 scored 70%, meaning it can solve 7 out of 10 hard, real-world software engineering tasks completely on its own. Claude Opus 4.7 scored 54%, and then there's everybody else. It's a massive cliff, right?

Everyone else—Gemini, Kimi, DeepSeek—dropped below 32%. DeepSWE isn't measuring minor challenges. These are tasks that require editing 668 lines of code across 7 files. A startup called DataCurve built this specifically because the old benchmarks are broken. Models are basically training on them. Alex, let's go to you first on this. How significant is this?

Alex Danco

This too will saturate. It's delightfully retro, in some sense, that based on their announcement, there's much more hand-coding of the evals within DeepSWE versus, say, SWE-bench or some of the other benchmarks that are now so widely used and so widely known that models have essentially saturated their performance.

But I want to caution, again: GPT-5.5 on X-ai reasoning at 70%—this too shall saturate, and probably in the next few months. So I think, in some sense, it's charming and delightfully retro that incrementally more hand-coding of held-out codebases and tasks on those codebases can get us a little bit more of a spread between frontier models. But this is going to saturate in the next few months, just like anything else.

But Alex, we've spoken about the fact that coding has become the single most important capability that these—

Alex Danco

Yes.

—frontier labs, and I'm actually just kind of shocked and surprised that, first of all, GPT-5.5 has jumped ahead of Claude. And second, how far behind everybody else is. I use these models day to day, and I should say, as we've just gone to recording here, Opus 4.8 was just announced, so maybe there will be a little bit of leapfrogging.

But I use both GPT-5.5 xhigh and Opus 4.7 on a daily basis, and I will say 5.5 xhigh, especially with /goal, is a stronger model in practice, anecdotally. Dave?

Yeah, I also use them day to day, and this matches my experience. But what's really interesting to me is that if you look at the other metrics—SWE-bench Pro, the prior one—Claude Opus 4.7 is slightly higher than GPT-5.5 on that benchmark. And so this is really, really important for OpenAI. If I look at the actual recruiting of great talent—

Hmm.

2 people that I know really well, Shane Longpre from MIT and Tobin South from Stanford just joined Anthropic, and they're starting this week. Those are 2 of the best guys that you would ever want to attract to your company.

So I think OpenAI has lost a lot of its mojo in the lawsuit, in the defection of a lot of key people. They have a window of opportunity now to get that mojo back using this benchmark as the turning point. But if they don't take a ton of that money that they just raised and use it to recreate their thought leadership and their cool factor—

Hmm.

They should be calling XPRIZE tonight and saying, “We want to figure out how we use 1 billion dollars or 2 billion dollars to do what Dario is doing. Go visit the Pope, create new benchmarks, or write white papers that talk about how this is going to benefit humanity the way Dario is doing. Or show up in Davos and have a debate with Demis Hassabis on the world stage.”

All those things Dario is running away with right now, and the result of that is that Shane and Tobin want to be where the singularity happens. They want to be in a place that is guaranteed to be good for the world and not bad. They've got pure, I think—

Well, Sam, you've got my mobile number. Give me a call if you want to do that, and 1 billion dollars in XPRIZE would be awesome.

Wait, here's better—

This is a much better—

Much better podcast than certain other podcasts.

Salim, go ahead.

I thought here was where the singularity was happening. But anyway, okay. I've got 2 or 3 points.

One is, it's really powerful to note here that algorithms aren't just optimizing workflows; they're actually becoming the workflow, right? Every organization should be asking which parts of your business are really software loops hiding inside bureaucracy. This is what's going to cause enterprise adoption to be very slow for some of this, but then it's going to be all of a sudden, because coding agents at some point will become very, very reliable, and the cost of rewriting internal processes collapses.

It goes back to Alex. You were asking the important question: what's the falsifiable theory around this new organizational singularity idea? Well, it would be that the singularity didn't happen. It would be that we don't have domain collapse, that legacy hierarchical companies outperform AI-native firms, and I think the signal is going the other way, in a good way.

Well, the other stat that is not showing up on this chart is that when you use Claude Opus 4.7 to solve the exact same problem, it burns twice as many tokens as GPT-5.5 to get basically the same or slightly worse result.

And at the same time, you see Anthropic's revenue going through the roof. But 4.7 is so verbose compared to 4.6, and for whatever reason, GPT-5.5 is not. It didn't just start babbling incessantly. I don't know what the cause is in 4.7, but you can see it in the underlying data behind this chart.

It literally takes twice as many tokens consumed to get to the same result, which means it's twice as expensive to get to the same result.

One key shout-out here to everybody listening is that software is becoming a commodity, and taste, right? Your taste as a creator is becoming the moat. Anyone can build software at near-zero cost, and the competitive advantage shifts to your domain expertise, your design taste.

It used to be you had to be a coder first, and a designer or have an opinion second. It's the other way around now.

Alex

I would generalize that perhaps—and I'm not necessarily a huge fan even of the notion of a moat in the singularity—but I would say—

I agree with that, by the way: diminishing across all boards, right?

Alex

Yeah. Abundance isn't necessarily fully compatible with moats. If anything, maybe they're fundamentally at odds. Moats are the form of scarcity.

But I would say it's generation in particular that's becoming abundant, and differentiation right now can come from verification of what is generated for maybe the next few years.

Sure.

Alex

I'm not 100% convinced that verification itself is any sort of truly long-term differentiation, or even whether the sense of long-term differentiation in an era of true abundance makes sense.

Dave

Well, that's the Elon bet, right? We talked about that last pod, where Elon is going to race to having the most compute and the most compute in space, on the assumption that these models will recreate themselves in their next generation very easily. So if his partnership with xAI holds up and he has access to their best models, he can use them to create the next model that runs on his hardware.

Alex

Peter, is it worth just dwelling on that for a minute? Because—

Dave

Please, go ahead.

Alex

You and I had a back-and-forth with Elon on X about that in the past few days. I think you said something to the effect of, following our past discussion about Grok being, quote-unquote, “on life support,” and then Elon, I think, responded to you saying something like, “Well, it may be true that Grok”—I'm paraphrasing—“it may be true that Grok has fallen behind somewhat, but I'm never going to give up. Never.”

You did say that, yeah.

Alex

And I think I responded like, “This is good. We want lots of competition at the frontier. We don't just want basically a duopoly between OpenAI and Anthropic, and we want Grok and other non-OpenAI, non-Anthropic models to be very competitive at the frontier, since, ironically, OpenAI was formed in part by Elon to make sure that the world didn't collapse to a singleton led by Demis and DeepMind.”

Yeah, I was commenting that his move to become a hyperscaler was a smart move, with less dependence on building his own frontier models. But he's not going to give up on anything.

Alex

Right. Nor should he. I think it's so important to have good competition. Yeah.

Heart disease has been personal for you as well, hasn't it?

Dr. Dawn Mussallem

It really has, Peter. When my daughter was 5, my husband died of sudden cardiac death, and so this is a topic that I am mission-driven to try to eradicate. Prevention first and early detection is absolutely critical. 50% of people die of heart attacks with no warning signs. Silent killer.

No shortness of breath, no pain, no nothing.

Dr. Dawn Mussallem

No, silent killer.

4. Jevons Paradox Drives AI Demand

This chart is one of my favorites. It demonstrates Jevons paradox. We've talked about this a number of times, and I just want to use it to help everyone listening really understand this. It's playing out in real time. What we're seeing here is that, since late 2024, the price of AI tokens has dropped 75%, from roughly $2 per million tokens down to 50 cents. What's happened to demand? It's exploded.

We've basically gone from zero to 25 trillion tokens per month. This is the exact same pattern that's played out in compute, bandwidth, and genomic sequencing. When the cost of intelligence drops, people don't use it less; they use it radically more. This is abundance in action. It's demonetization and democratization. Look at this chart and understand this: We're going to be seeing this with robotic labor. As the price drops, it's going to be used everywhere. Any comments on this, gents?

Dave Blunden

Well, the numbers in this chart are really important, so if you can't see it right now, Jevons paradox was invented in England in the 1800s, when coal burning got twice as efficient in an effort to save coal. Then coal consumption went up by 3 or 4X in response, so you actually used more coal even though you were trying to save coal.

These numbers are just redefining Jevons paradox. The cost is down by a factor of 3, from $1.50 per million tokens down to 50 cents per million tokens, but use is up by about 30 to 50X on this chart. So the price is coming down, but nowhere near at the rate that demand is going up.

I think that demand is understated because it's sold out. If we had capacity to generate more tokens, that would be even higher. So this is Jevons paradox on steroids. This is the ultimate Jevons paradox.

Alex?

Alex

I think, in some sense, the question I would be asking is: What is the right unit to be measuring here? Tokens are so mushy. Dave and I were on a panel pretty recently discussing this as well. Tokens can depend on the encoding scheme. Tokens can depend on the amount of intelligence density in the underlying model that is being pretrained on those tokens.

I'm not thrilled with looking at token prices in general, even though this tells a Jevons story. I would really rather that we settle, as a civilization, on some sort of price. I could talk my own book and say it looks like GPU compute pricing, but I think we really do need some unit of measure—maybe not even just compute, but intelligence in general.

I think GPU compute tied to earning financial interest is a great start, but we need some way to know what is going to be the unit of currency in a post-superintelligent future. Tokens, I don't think, are it.

Maybe it's the idea that cognition is becoming abundant rather than tokens becoming cheaper, and hang your hat on that side of the equation, which is what I think we're seeing.

But we need a metric of some kind.

Alex

But how do you measure the abundance?

Yeah.

Alex

I don't know.

It's an increase in GDP and other numeric metrics.

One of our viewers will go create a benchmark on this.

Yeah.

Alex

Yeah. Call to action for the audience: Help us—

Yeah.

Alex

—a better job of measuring abundance.

Yeah. The prediction right now, and this is from Gartner, is that inference on a trillion-parameter LLM will cost 90% less by 2030 than it did in 2025. The only thing we know is that the price of accessing intelligence is coming down, and the power of that intelligence is increasing by orders of magnitude.

Dave Blunden

By the way, I used AI to look up whether it's Jevons paradox or Javon's paradox, and it turned out you guys are right: It's Jevons paradox.

Mm. We have some AIs on the show here. Let's move along.

5. AI Revenue Enters Overdrive

Here's a super-fun chart. The revenues for these models are exploding. Let's talk about money. OpenAI just did $5.7 billion in a single quarter. ChatGPT is now at 905 million weekly active users. They haven't hit their billion yet, but they will. It's more than Instagram.

Their coding agent, Codex, has 2 million users, and it's becoming a real revenue engine. We've talked on the pod a lot about going from consumer focus to coding focus, and they've done that fairly quickly, right? Over the last 3 or 4 months, they've shifted their revenue base.

But here's where it gets crazy. In a related story, Joseph Jacks from OSS Capital is projecting that Anthropic could surpass Alphabet's total revenue by 2028. We're talking about going from $9 billion in revenue to potentially $2 trillion by 2030. That's the prediction. And if that's even directionally right, this is the fastest wealth creation ever in human history. Dave—

I had to sit and look at this for 3 minutes. That just blows my mind that Anthropic could exceed Alphabet's revenues. That's just a staggering commentary. The thought that occurred to me is that every company in the world needs a cloud strategy; you need an AI strategy.

Yeah.

Dave Blunden

You need a compute strategy, too. If you look at the fact that the profit levels of Anthropic are going through the roof, they can change the pricing model instantaneously. Just by Claude 4.7 being more verbose, they've effectively doubled their revenue per user. They can also throttle the rate at which the tokens are generated to make the value of a subscription account effectively lower and the margins higher. So it's a really weird product.

They can turn the knobs.

Yeah.

Yeah.

Usually, when you buy something like a laptop or a glass of water, you have a fixed volume that you bought. You know what you got. This is such an amorphous, weird product. It's very, very slippery.

As an enterprise, you've got to really hunker down and reserve your compute and decide exactly which models you've contracted for. I think companies should be signing long-term contracts, and they're all afraid to. But, yeah, I think you have to. You've got to figure out your 5-year strategy now before everything is sold out.

Alex

I would add that I think there's a quiet technical revolution here behind all of the business stories. Yes, there's the switch—the pivot by OpenAI from consumer to enterprise and all of that. The deeper technical story, I think, is a sea change that I'm at least seeing in the pivot from—call them—generalist reasoning agents in general to generalist tool-using and code-generation agents, to the point where Codex becomes probably the mainline agent/model that OpenAI offers. That becomes the new baseline.

In the same sense, if you think back to the GPT-3 days and the pivot to instruction-post-trained models, you'll remember that it was then, I think, somewhat revolutionary and, in contrast to today, appeared in arXiv-style papers first rather than appearing in products and papers later. Remember, in the beginning, we had pretrained models, and they were good. Then we discovered, fast-forwarding, that if you could fine-tune pretrained models on instruction-following, you could get orders-of-magnitude improvements in capability without actually needing to scale compute, and that was an amazing discovery. And then—

David Friedberg

Wait, ontogeny does what to what? Can you repeat that and explain those terms?

We can have you translate it in a second here.

David Friedberg

Yeah, I'm frantically looking up the dictionary.

Alex Finn

Okay. No, just dropping a footnote for Salim.

David Friedberg

Thank you.

And all over the rest of the world, too, by the way.

Alex Finn

Ontogeny recapitulates phylogeny. This is a cliché from Biology 101: If you look at the development of a human embryo over time, you observe that there's a reptilian phase, an amphibian phase, and then early mammals. Basically, the development of a human embryo recapitulates the evolutionary history of all the species that preceded Homo sapiens.

And those of you who are pregnant with a baby, look at those early sonograms.

Alex Finn

Enjoy their tails while they last.

Yeah, exactly.

Alex Finn

Because they do have tails. Similarly, by analogy, if you look at the way training, pre-training, mid-training, post-training, and scaffolding are done right now in the frontier models of today, that entire pipeline, I would argue, recapitulates the history of the evolution of the models themselves. You can see it baked into the pipeline.

So I'm curious, guys. Is this now a four-horse race, and can anyone even possibly catch up?

And along those lines, what we're starting to see is—and Dave, you made this point before—I think Anthropic actually raised its enterprise prices, right? OpenAI is offering tiered pricing. Google, Gemini Flash undercut everyone by 50% to 80%. It's going to be a performance-versus-cost equation, and they can turn those knobs, but where is everybody else? Where's Meta?

It really feels—I think xAI is going to come back. I think Elon will come back, along with Cursor, with something that's super powerful. But it really feels like just a four-horse race with an inability to catch up. What do you think? Can anybody penetrate?

Deeply disagree. Can I say why?

Yeah, please.

In the early days of the web, nobody was going to be Yahoo. Then Google came along. Then nobody was going to be Google, and then Facebook came along.

I think we'll find researchers who have different approaches. World models could be a very viable candidate to leapfrog where we are currently today. So I think there are hidden research labs. Look at it—people thought there was no way you could do anything against NVIDIA, and look at what Cerebras is doing, right?

Yeah.

I think we're going to see constant leapfrogging, and incumbents need to be watching out for that. The way that the internet folks won and stayed winning was they would find the breakthrough startups and then just acquire them as quickly as possible.

I wonder what Ilya's doing.

And I think the same thing will happen.

I do wonder what Ilya's doing.

Alex, I'd be curious to hear your thoughts.

Alex Finn

On Ilya or otherwise?

Both. Both.

Alex Finn

It's been publicly reported at this point that what Ilya is working on is building a prop-trading hedge fund.

No.

Alex Finn

That's out there. That's out there in public reporting at this point.

I thought he was building a scientific superintelligence.

Alex Finn

To do prop trading.

Or the other way around.

Alex Finn

That's the publicly reported rumor, I should say, that's out there at this point.

David Friedberg

It's a rumor.

Alex Finn

That's out there at this point.

David Friedberg

Yeah, yeah. That came up after our Ben Horowitz interview. Remember that? We had that postgame wrap-up where we were reading the tea leaves on the valuation and all the rumors, and, yeah, it does look like that's the case.

But that's not as weird as it sounds, because any machine that quietly generates huge amounts of profit can then be used to buy lots and lots of compute, which can then be used for self-improvement—recursive self-improvement.

Alex Finn

Mm.

David Friedberg

So that doesn't mean he's not building safe superintelligence too. It's just a different way to kickstart the cash flow.

Alex Finn

And we've got a bit of a race going on between Leopold's operation and Ilya's operation, potentially.

So I am curious, gentlemen—and Alex, to you again—do you imagine a dark horse could come out and just blow away these frontier labs?

Alex Finn

It's possible, but I think the easiest contingency where that happens is if we get to the end of the algorithmic rainbow and discover the perfect algorithm for AI, and it's so obvious and so transcendent that anyone can implement it without needing a staff of frontier AI researchers.

However, at that point, I think it then comes down to compute and having the compute to run it at scale. You see already the frontier labs verticalizing down into the compute layer, maybe preparing for that eventuality, as you see from some folks, including Noam Brown at OpenAI, who are arguing that maybe the model weights don't matter that much anymore.

Maybe it's all about the compute for reasoning and inference time, in which case maybe the question itself doesn't make sense, and maybe the question transforms into: Who has the most compute under their direct control?

Mm.

David Friedberg

Yeah, I think 2 things. One of them, I think Elon is right, and I think the Cerebras observation is right on target.

What'll happen next is algorithms will be discovering new algorithms, and then if you win the race to either the Terafab or to the chip that's just fundamentally better, à la Cerebras, then you have a massive, massive surge of growth, and control of the compute determines the biggest player.

But the implication of the question is, hey, there's a four-horse race. One horse is going to win a race and get a trophy. But I don't see it that way.

Normally in a market, you're all competing with each other to win the iPhone market or win the laptop market, but here, this is the future of all humanity. It's going to be massive expansion. So I think it's very likely that all players that are in the middle of this get bigger and bigger and bigger. So we're just debating who's going to get the biggest of the big.

If only we had the predictive engines to get us there, and that's our next story here.

Alex Finn

A rising tide lifts all boats.

6. AI Matches Superforecasters

Yeah.

So DeepMind just built an AI system called Green Tree that can predict the future as well as the best humans on Earth. They're called superforecasters—these superhuman forecasters, the top 2% of human predictors who, according to Philip Tetlock, are 30% more accurate than CIA analysts with classified intelligence.

On March 15, AI hit parity with these superforecasters for the first time. Let me say that again: an AI can now predict geopolitical events, economic trends, and political outcomes as well as the absolute best human minds. The implications for finance, insurance, and governance are massive.

We talked about this on a recent pod, where these predictive engines were getting close to these superforecasters, and now they've reached parity. The implications here are insane. Dave, you spoke about this.

David Friedberg

Why is this surprising? If you look at weather forecasting, the idea that you would forecast the weather without a computer is utterly insane, right? There are so many different variables and moving parts.

Yeah, I feel the outside. The barometer on the wall is telling me a storm is coming. Come on. So why would that not apply to all forms of forecasting? Of course it does.

The big unlock here, though, is assimilating unstructured data. People who worked in stock-market forecasting at Wellington or Fidelity for the last 20 years would always tell you that the computer cannot compete with them because they're reading these research reports, and there's so much nuance in the research reports that it just doesn't show up as database-structured data.

Now, with the LLM, all the unstructured data can suddenly be quantified, but it has a million-token context window and the ability to assimilate 10,000, 100,000, or a million times more information than any human stock picker or any human forecaster. So it's got such a massive competitive advantage that even if it's not as brilliant as you in geopolitics or whatever, it doesn't matter. It has so much more capacity that in those domains, it's going to outperform, and then, of course, it's getting smarter every week—every iteration on top of that.

Let's talk about the implications: financial markets, governance, and insurance. All of these things are massively impacted. Right now, it's parity. Within the next year, it's going to be available to everybody.

In the financial markets, Dave, you mentioned this last time, right? This is the disruption of hedge funds. In governance, if AI predicts a policy is going to fail, do you pass it anyway? Insurance models break when outcomes become predictable.

David Friedberg

All of the above. But we're in kind of a golden moment right now where, when I talk to my agents all day, I've got about 170 operating on my screen here, they make some really stupid choices. I mean, really, really bizarre, odd choices. And so the role of the human in the loop is still critically important.

We're in a window here, and I don't know if it'll last 5 years or 1 year, but we're in a window right now where the perfect synthetic human working with many agents is still better. All the reports want to say the AI is better than the human or the human is better than the AI, but the reality is the human and the AI working together are far, far better than either one by itself.

And so there’s a window here to take advantage of that. But if the AI says, “Yeah, this is a terrible idea,” are you going to override it? Well, probably not. You have to read what it’s writing, study it, and make sure that you know something it doesn’t know, but it’s probably picked up on something that you overlooked.

Alex

Look, we’re almost 30 years into Kasparov being beaten by Deep Blue, so we’ve got 30 years of chess history with this. Right now, the best chess players are a human being with an AI.

Dave

Not one on its own.

Yeah.

Dave

No, neither on their own can do it, but the combination is positive. The combination is the best, and I think that’s going to continue for a whole bunch of domains.

Alex?

Alex

I want to say something new here. I’ve previously commented on psychohistory, Isaac Asimov, Hari Seldon, and all of the implications of being able to predictively model the future of humanity. I want to, though, follow the dictum, “Invert, always invert,” and talk a bit about retrodiction.

If we’re amazing at forecasting—and I should probably also add ForecastBench as this really neat benchmark that is fully autonomous—it uses a bunch of templates to enable AIs, without human involvement, to predict, say, changes to Wikipedia and other public event-recording websites. Where I’d like to go with retrodiction, though, is to say—and I’m invoking the spirit of Nick Bostrom here—if we have the ability, if strong AIs have the ability to predict future human events, I would expect them also to be very strong at retrodicting past human events.

At that point, someone has to ask the question, so I’ll be the one to ask it: Should this increase our posterior confidence in Nick Bostrom’s simulation hypothesis? For the record, I’m not a fan of the simulation hypothesis. I think there are a variety of very good reasons to discount it.

However, Nick’s Bayesian argument was that if humanity reaches the point where we build very competent ancestor simulations, then we should increase our likelihood that we ourselves might be living in one. I’ll just flag that if AIs are achieving superforecasting ability, then they’re probably also achieving super-retrodiction ability. If we’re good Bayesians, we should probably, on the margin, increase the probability that we’re living inside an ancestor simulation.

Dave

Wait, this is the first time I’ve heard that you don’t think we’re living in a simulation.

Alex

No, I don’t think so.

I think 100%.

Dave

Yeah.

I put it at 100%. I think we’re living—

Dave

We need to debate that one.

—I think we’re living in an nth-generation simulation, but that seems—

Alex

Seems unlikely. I’ll give, just in the interest of time, maybe my second capsule reason. I have a bunch of different reasons, but my favorite reason why we’re probably not living inside some sort of recognizable computer simulation, other than all the physics-oriented reasons, is more of a lowercase-a anthropic reason: it’s too fine-tuned an explanation to the paradigm of the moment.

We build lots of simulations, so it would be akin to asking, about 100 years ago, “Don’t we live inside some sort of complex electromechanical machine?” Or maybe, several thousand years ago, “Don’t we live on the back of a turtle?” It’s too overfitted, I would argue, to the paradigm of the moment to be plausible.

Do you think it’s a coincidence we’re living at this exact moment in human history, at what’s now—

Dave

Yeah, too damn interesting to be an accident.

—the most exciting point of transition?

Alex

No, an anthropic argument—again, lowercase-a, not capital-A—would be that this is a very natural time to be asking the question, “Why are we at this pivotal moment in time?” So it’s selection bias.

I resemble those remarks. All right, moving us along, let’s jump into the conversation around jobs and the economy, because it’s getting murky. Over the next 2 stories, I want to hit this.

The first story tells us the numbers. Since the beginning of this year—just 5 months now—we’ve had 143,134 tech workers laid off. According to the consulting firm Mercer, in its Global Talent Trends report, 99% of CEOs expect AI-driven layoffs in the next 2 years. March was the worst month in tech layoffs since the pandemic.

Then here comes Jensen Huang, the guy who’s making all the GPUs out there, calling this a lazy narrative. He says CEOs are blaming AI just to sound smart, and when you dig into the data, it’s more nuanced than the headlines. The real question isn’t how many jobs are disappearing; it’s who’s being affected by these jobs and what they’re doing next. We’ll get into that story.

The second story here is Sam Altman walking back his comments about the AI job apocalypse. He did something remarkable: He admitted he was wrong. The CEO of OpenAI, who spent last year warning about mass white-collar displacement, now says, quote, “I don’t think we’re going to have the kind of job apocalypse that some of the companies in our space are talking about.” Obviously, he’s talking about Anthropic there.

He says he’s delighted to be wrong. In his comments, he said he tried delegating his own email and Slack to AI, and then went back to doing it manually because, quote, “We really do care about our interactions with people.” Meanwhile, Dario Amodei is also making some pivots here. So the question to you, Dave, is this a coincidence, or is this happening for OpenAI and Anthropic because they’re both about to do a trillion-dollar IPO?

Dave

Yeah, so here’s what’s really happening under the covers. They legitimately thought that job loss would be massive because of automation driven by AI. But we were always predicting that by 2030, it would turn the corner because the abundance created by all this AI is going to create massive, massive new gains, and we’ll create roles that fit those.

So it was always going to be temporary. Between 2026 and 2030, there’d be pitchforks in the streets, mass uprisings, and problems everywhere. As it’s playing out, I think there’s some job loss, but the greenfield opportunities are growing so much quicker than anyone ever predicted. Now that window of loss is actually relatively narrow, and the AI companies with their massive new funding are going to actively try to prevent job loss.

Because the GPUs are so constrained, we’re focused much more on coding and a few other use cases, much less on putting every artist out of business. The job-loss effect is diluted by design, more than anything else, just to prevent massive disruption and public backlash. They’re focusing their efforts on areas that are greenfield and actually creating net new value in the world and net new jobs in the world, not on just automating everything away.

I’m seeing that directly at Vestmark, where I’m the chairman. There are 400 people doing white-collar automation, account reconciliation, and back-office work. A year ago, I was thinking about half of these jobs might go away. Now we’re having no trouble automating things, but it’s all becoming margin, and we’re keeping everybody.

Yeah.

Dave

Where it’s really hitting everyone is in no new hiring. You’re seeing that. College graduates are in a really tough spot. And so that’s what’s going on.

That’s the pain point right now.

Dave

Exactly.

The Dallas Fed put out a report in January of this year saying that employment decline correlated with AI exposure only among younger workers. Older workers in high-exposure jobs showed no significant decline. Basically, what’s going on is a hiring freeze, not mass layoffs.

Dave

Yeah, that’s exactly right. You summarized it far better than I did.

Yeah, well, no.

Guest

I think I’ve got a couple of thoughts here. First, I’m totally with Jensen here. You don’t blame AI. There’s a lot of bad strategy out there, and I think CEOs are covering up their layoffs by blaming it all on AI. It’s such an easy place to land.

But there’s something unbelievable happening on the job side and the solopreneurship side that is unprecedented. One is that we’re creating more startups than ever, at a 10% or 15% level year to year. We’re 25% higher in startups than the same quarter last year.

The third stat that I came across that blew my mind is that the U.S. now has 6 times more startups than Europe. That’s just a staggering number, and I want to relate that to jobs for a second. If you go back over the last 50 years, 100% of new jobs have come from startups and early-stage companies.

That’s right.

Guest

Big companies are becoming bigger, but they’re also becoming more efficient and reducing the number of people—

Really important point.

Guest

—to do the same amount of work.

Very important.

Guest

All new job creation has come from startups.

Yep.

Guest

We should be throwing everything at this. Because it’s so easy to become an entrepreneur today—and I want to be careful, because we get this pushback a lot—not everybody can become an entrepreneur. Yes, but everybody can use AI to create their own agency out in the world.

Mm-hmm.

And that's more—that is merging. So now we're going to run companies. Our estimate from the Organizational Singularity is that you should be able to run a company with about 20% of the people you had before, on average, but we're going to create 5 or 6 times more companies.

Yeah, a really important point. You know, one of the challenges with these mass layoffs—and there have been some large layoffs, right? We saw this with Cloudflare and, recently, with Meta—is that the CEOs who are announcing these layoffs are really doing it in a very dispassionate way. I cannot condone the way they're communicating, laying people off. What Mark Zuckerberg said recently—I don't have the quote here—or what the CEO of Cloudflare said in terms of who we're laying off and the reasons, I mean, there needs to be a little bit of compassion here because you're transforming people's lives in a negative fashion.

Dave Blakely

Yeah, I think Meta had a ton of UX engineers, and that's one of the places that's been hit hardest.

Yeah.

Dave Blakely

But it's definitely disproportionate. You'll see in college enrollment that computer science peaked and is now coming down, but engineering—

Dramatically.

Dave Blakely

—is still dramatically, but engineering as a whole, mechanical and biological, is still skyrocketing, and so it's taking over. People are really quickly retooling their career ambitions toward the real physical stuff, like biology and mechanical engineering, for the data center build-out and for medical research driven by AI. I think that's a really good thing.

We probably had way too many UX engineers anyway, just in terms of societal benefit. What are people learning? What are people doing?

I love this chart from Andreessen Horowitz. It's the other side of the jobs story, Salim, that you were just talking about, and this is the one I find really exciting: solo founders are exploding. a16z's data shows that AI solo founders—in other words, solo people creating an AI company on their own—have doubled in the last quarter, from 1,500 to 3,000, up basically from 0, 3 years ago, and non-AI solo founders hit over 5,000.

People aren't just losing jobs. They're transitioning to starting their own companies.

They're following their passion.

Yes, and they're doing it alone because AI tools now give a single person the capabilities to do that either on their own or with a small team. So, connecting the dots here in the earlier story, coding agents are now at 70%. That plus layoffs equals solopreneur explosion. This is creative disruption happening in real time. Salim, take it from here, please.

Well, look, the coordination overhead in big companies means that they cannot sustain any kind of leverage over time. You spend more time coordinating activity than doing the activity. Who was the fellow who tweeted, “It's easier to build a product feature than to have the meeting about building the product feature”?

Yeah.

That's the reality of the world today. Therefore, all of the overhead in big companies—and take universities as one example—the amount of overhead has grown enormously. As universities have gotten bigger and bigger, and their endowments have grown, the number of students has gone up incrementally, but the number of administrators to administer the students has gone up exponentially.

Like in the healthcare industry.

Yeah. We've got so much overhead. The healthcare industry is an exact example. That will not sustain in a world where you can have an AI teaching a kid in 1 hour what they could learn sitting in a classroom for the whole day, or when diagnosis goes to free, which is pretty much the case today.

At some point, this is going to give way. That comet—sorry, asteroid—has now hit. The big dinosaur, the category of big companies, will probably evaporate. The way we see it, you're turning from big companies into platforms, platforms into ecosystems, and breaking up into smaller and smaller units. This is why small teams will always outperform big teams.

There's a reason that the big Microsofts and Googles of the world did not build all of the cutting edge. They acquired DeepMind, they acquired OpenAI, et cetera, et cetera. Anthropic is a rare outlier. This is going to be the defining operating model for the future: small teams radically outperforming. We're entering the most incredible Cambrian explosion of Darwinian evolution, to way overuse the metaphor.

Dave Blakely

Can I just correct a potential misconception here too?

Please.

Dave Blakely

If you look at the data on successful companies, 75% of them now are coming through some kind of incubator or accelerator program. When we started investing, that was only 6%. That's gone through the roof at the same time.

When people hear the word solopreneur, they might be visualizing a person in a cabin in the middle of Alaska, working all by themselves. No one's around. It's depressing. That's not what is actually happening in that top-left chart or that bottom-left chart for solo founders. They're in a very active, vibrant ecosystem of some sort, and at the bottom of the slide, you see the Gemini XPRIZE.

That's an ecosystem of like-minded people talking, texting, Slacking, communicating all day long—super-connected, super-involved. So your successful solopreneur is legally a single-person entity, but they're highly connected like never before—

100%.

Dave Blakely

—and they're part of some bigger platform. So, just to visualize what's working.

Well, you learn lessons from each other. You get the latest capabilities from each other. You support each other when things don't work out. I just want to hit on this, and we can put the website up on the screen here, geminixprize.com.

We launched this $2 million hackathon with Google, and thank you to the team there. The concept here is that all of us are finding problems all the time, and, man, I wish someone would solve that. Well, guess what? You can solve that now, and this competition asks people to write up a product or service idea in plain English. Write it down in a Google Doc, right? Then describe the problem you're trying to solve. What do you think the solution could be?

You can brainstorm this with your favorite large language model, and then the AI can code it up for you and help you design the marketing. This is a competition asking individuals or small teams to build something in 3 months. There's $2 million in prize money.

We get a lot of pushback, and I've read it in the comments from our last episode, saying, “Hey, you guys are hanging out with entrepreneurs all the time. You are entrepreneurs. You can't expect me to be an entrepreneur.”

Alex Hormozi

We live in an ivory tower.

Yeah.

Alex Hormozi

We only talk with other millionaires.

I just want to say that my experience is that normal people are brainstorming and starting companies, and a lot of this is self-limited thinking. I just want to encourage people to try. That's the only thing. Please try.

I was smiling earlier because I remember when Milan was 5, he was asked what kind of technology he would build, and he came up with this thing called the Hydro Blaster, which would be a water cannon in front of your car that would blow other cars off the road. They said, “What is this about?” He goes, “My dad hates sitting in traffic.”

I designed this room. I was just trying to imagine an AI trying to build that thing.

Alex Hormozi

I would also add to the solopreneur point.

Yeah, please.

Alex Hormozi

There's a lot of hand-wringing out there that pretends that having a quote-unquote job is somehow the historically normal state of affairs. It is not. It is a modern invention, largely attributable to the first and second Industrial Revolutions.

Historically, most people didn't have anything remotely comparable to—

You survived.

Alex Hormozi

—what is currently called a job. Most people, in the historic state of nature, if we go back 2 centuries or more, didn't have anything remotely like what we would consider a job as a cog in a large enterprise. Most large enterprises weren't large. They were pretty small by comparison to today's standards. It's an artifact—

Great point.

Alex Hormozi

—of a time and a place. So, if anything, I would say it's highly unnatural, highly unergonomic by historic standards for people to even have jobs. This is more of a return, if anything, to the default state of human nature and some sort of historic equilibrium where everyone was self-determining their own future.

Agency.

Dave Blakely

It's a beautiful thing.

Back in that time, that era that Alex was referring to, the federal government was about 4% of the economy. So you think about the independence: yeah, you lived on your own. You were your own thing. That'd be a good vision for where AI might be able to take us: self-determination—

10X.

Dave Blakely

...self-sufficiency. It'd be wonderful.

Yeah.

Empowered individuals.

Yeah.

7. Education Flips From Supply To Demand

Our next story here is actually a call-out to everyone listening: How well is education preparing students? I want to do a survey. If you're a parent of a 13- to 18-year-old high school student, a high school student, a college student, a teacher, or a working professional, tell us how well the educational system is preparing you or your kids. What's your experience? What do you wish you had? I'm going to gather this data and report it back here on the pod. We'd like to understand what your thoughts are. The URL is moonshots.com/survey. Feel free to share this survey with your friends who are not Moonshot listeners.

We talk about the educational system failing us, and I'd like to get some more data. How is this affecting you and your kids? I think it's important to get a reality check here because, if in fact what we've been talking about on this pod is that we're going to see this bumpy road for the next 2 to 8 years until we get through to a true state of abundance, how do we get ready? How do we survive this turbulence? And are we getting ready for the new economy that's heading our way? Salim, any thoughts?

Yeah, I've got a bunch of thoughts here, but let me limit it to one paradigm. We've been doing education over the last couple of centuries, since the Industrial Revolution. I think, per Alex's framing, that's exactly right, and we've been doing it on the supply side. Go become deep in a skill: engineer, doctor, lawyer, accountant, and then go to the job market to find demand for that skill. But it's all supply-side driven.

All our education systems are designed to take a young child and train them through their early 20s to be ready for the job market. Small problem: We have no idea what a job looks like in 5 years. We don't even know what a job looks like in 2 years, so what are we teaching them? And, Peter, you and I talk about this a lot, where you're flipping kids from that supply side, where the half-life of a skill used to be about 30 years and now it's about 3 years. You need to flip them to the demand side. What problem do they want to solve? Then go find the techniques, capabilities, skills, and technologies to solve that problem.

That is such a radical shift for the educational system. Very few educational institutions are going to make it over to the other side. We need a completely new cadre of schools, but that's the fundamental structural problem: You have to go from supply side to demand side.

Yeah. I just want to see: Does everyone listening agree with that? I'd love to get everybody's input. We'll come back with the data. We'll see what everyone truly feels. As a parent of two 15-year-olds—they turn 15 next month, Salim—I'm super curious because I see a very dysfunctional educational system not getting our kids ready. Dave, any thoughts?

David Friedberg

Well, I think that particular age bracket is so geared to getting into college and getting into the right college and maximizing everything on your résumé—your SAT scores and your grades. It's all about my college application. And so that perpetuates a curriculum that's woefully out of date, but the AP exams and the SATs are the same exact topics, while what you should be learning is changing tremendously.

Yeah.

David Friedberg

That's a really broken age bracket right there. So I'm sure everyone will come back and say, “Yeah, this is totally messed up. What do we do?”

By the way, I have in front of me a little printout from Milan after he interviewed you, Peter, on “What's the Future of Education?” And I was just like, “You printed this out? Why couldn't you email this to me? How retro of you?”

How retro.

8. Starship Changes Spaceflight Forever

All right, our next story: We're going to jump into a few space stories for those of us who are space cadets. SpaceX just launched the biggest, most powerful rocket ever built, and Dave, we've made this point before: This was designed and built by humans and not AIs.

David Friedberg

Mm-hmm.

Starship V3 flew for the first time last week from a brand-new launch site in Texas. It's an incredible thing: a brand-new rocket, brand-new engines, brand-new launch site. The level and the speed at which SpaceX iterates is crazy. It carried 97,000 pounds to near-Earth orbit. That almost doubled the Space Shuttle's carrying capacity, that old vehicle.

It's running on Raptor 3 engines, each producing 250 to 280 tons of thrust, 20% more than the last version. By the way, that's equivalent to about 70 747s at their takeoff thrust levels. SpaceX did lose the booster on landing, but that's the way SpaceX operates. They fly, they learn, they iterate.

SpaceX treats rockets like software: They ship, they test, they fail, they iterate. And losing the booster was another data point, not a failure. They've already carried real Starlink prototypes to orbit. This thing is doing commercial tests on its first flight. Alex, you were watching this. What were your thoughts?

Alex

I was watching. It was riveting. The most exciting part for me was when the mission got to releasing Starlink satellites first, followed by a couple of so-called Doge Dots, which are intended to be prototypes of the next generation of Starlink satellites, Version 3, or third-generation Starlink satellites. The Doge Dots had cameras and lights on them.

Yes.

Alex

As they were being deployed out of the PEZ dispenser, they were pointed with their lights and cameras back at the Starship. It was just absolutely incredible watching from a third-person perspective, drifting away from the Starship—the view after deployment.

I think we're going to see so much of that. I'm not even sure if that has precedent at this point, but it reminds me that we're about to enter an era. In connection with this entire mission test, SpaceX reminded the world that it has a program called Starwatch that is basically leveraging Starlink satellites with all of their cameras—not just Doge Dots looking back at Starship, but looking down at Earth.

Since they operate at a variety of altitudes, they're seeing everything, including all sorts of orbital debris and other objects, and they're sharing that information. I think it would be an ironic but maybe not super-surprising outcome if Starwatch ends up having a more dramatic impact on civilization than even just Starlink connectivity.

Let's take a look at the video here. This is rocket porn for all of us. Here's V3 launching. Beautiful vehicle. I spoke to Elon at the Breakthrough Awards, and we made the comment that this is the most energy released by a human machine other than a nuclear bomb.

Alex

Wow.

Yeah. And here she is coming in for a landing over the Indian Ocean. Can't wait to see that captured and reused.

Alex

Unbelievable.

David Friedberg

I think there's a really important point for entrepreneurs in there, too. I also was completely riveted by the quality of the video and the MCs describing everything. This is so Elon. Anybody who wants to be an entrepreneur, study what Elon did here. Getting those 4K cameras—the ones that are physically mounted on the ship—to survive the launch and the heat, that's a lot of extra engineering that NASA probably would never do. But Elon understands so well the value of building morale and building a following and having a fan base buying the stock that he puts serious mental effort into the showbiz.

Yes.

David Friedberg

And every entrepreneur should study that. That's the winning formula because, previously, you didn't have a distribution pathway for all that content. Now, because of YouTube and X, anybody can distribute their message.

You don’t have to go through CNBC. You don’t have to go through Forbes. You can just go direct. So you’re crazy as an entrepreneur if you don’t study what Elon just did there and then do your own version of it.

Dave, another point you made before, and we’ve discussed, is how Elon gets out publicly. He is the marketing engine for Tesla. He’s the marketing engine for SpaceX and xAI. If you’re an entrepreneurial CEO, you have to put yourself out there. You have to be the carrier of faith and of the story, right?

David Friedberg

The turning point for me was when he did Saturday Night Live all those years ago. I’m looking at this and thinking, “How does he, of all people on the planet, have time to go to New York and MC Saturday Night Live? Why is he making that choice?” Because he doesn’t do this stuff randomly, right? He has more plans than anyone you’ll ever meet.

So I’m thinking, “Huh, this is reinventing what it means to be a great entrepreneur.” As you said, put yourself out there. Get your mission and your vision and your positive message and your MTP, make it really crystal clear and understandable, and then work on broadcasting it, because that’s how you’re going to get great talent, and that’s also how you’re going to attract capital.

Here’s something that we’ve talked about in the pod before. I’ve predicted this, as you have, and this is from Kalshi, the prediction market, showing 50/50 odds that Tesla and SpaceX will merge within the next year. I personally put it at 100%. Think about what that company would look like: electric vehicles, energy storage, solar, rockets, satellites, global internet, humanoid robots, and interplanetary exploration, all under one roof.

We’re talking about a potential $4 trillion entity, initially. I could honestly see—and, full disclosure, I’m a SpaceX and xAI investor—but I could see this being the first $10 trillion company and moving to $100 trillion in the next 5 years. Given the current anti-regulation environment—in other words, the fact that the government is not overly regulating—I think there’s a clear path for Tesla and SpaceX to merge.

Alex

Totally agree.

The key point is that Elon has voting control and super-voting rights inside of SpaceX and xAI, and with his insiders, I think it’s 86% of the controlling vote. He’s suffered from being a minority shareholder, being disallowed from creating the compensation packages that his board wants. In the merger, this gives him control back once again.

David Friedberg

Exactly right. Just one more detail on that, Peter.

Please.

David Friedberg

The 85% voting control that he has on the SpaceX side is based on 10-for-1 super-voting shares. If he merged Tesla into SpaceX, he has 20% voting control on the Tesla side, but those are one-vote shares, and everybody has one-vote shares. The most likely merger is that you’d roll in all these one-vote shares into SpaceX as the acquirer, but maintain the 10-for-1 super-voting shares that already exist on the SpaceX side.

Then, in the combined entity, he’d have 60%, 70%, or 80% voting control. It depends on the valuations. The only reasons Kalshi is at 50% rather than 100% are, first, that he needs the relative valuations to be similar or higher on the SpaceX side for that math to hold up, and second, that there’s a shareholder vote on the Tesla side that he doesn’t control. He probably can’t vote on that.

Alex?

Alex

The last time we discussed the possibility of one Elon company to rule them all, I argued that Elon has historically used mergers as an opportunity to fail forward. SolarCity, for example, was famously litigated, and xAI being acquired by SpaceX may arguably be a reverse acquisition.

In my mind, the argument for Tesla and SpaceX to merge would be that it could be a way for Elon to rescue what he may perceive as a suboptimal governance situation for Tesla and recover control of Tesla. One way to make that more appetizing is if Tesla’s valuation goes down over time.

Even though Tesla has a new compensation package tied to autonomy, if Tesla’s market value were to materially depreciate in the next year relative to SpaceX’s post-IPO valuation, that might, at least in the short term, make a SpaceX acquisition of Tesla a good deal more appetizing to public markets. There might be a bit of an arbitrage opportunity there. Not investment advice.

Second point: I have to ask the question. If and when this final Elon merger to rule them all happens, for a bunch of good reasons—maybe having to do with IP and GPUs freely and porously changing hands between all the different entities—is it going to be called X, or do you think it’ll have a different name?

E-L-O-N sounds good on the Nasdaq.

Dave

But what do you think it’ll actually be called?

I think X is his love. I think you’re right, and he may go toward that, or just X, X, X, X. We’ll see.

Dave

Triple X, yeah.

Quadruple X.

Interestingly, I had dinner with Elon in the early days, when SpaceX was up and operating and Tesla was getting going, and he was searching for a Tesla CEO. He really wanted someone to run Tesla day to day so he could focus on SpaceX, and he just never found anybody he trusted with that company because it meant a lot to him. This is a way for him to consolidate control and be the CEO of one company instead of splitting his time.

You remember, Dave, when we were with him at the Gigafactory in Austin? I think our podcast went for about 3.5 hours. We started at 10 o’clock and went until sometime after midnight, and his son was there the entire time.

Alex

But they’re on a collision course. They’re on a technical collision course. We’re going to need a lot of Optimus robots for the Artemis colony and for the Martian colony. It’s going to happen one way or another. Regardless of whether it’s an organizational merger, they’re already, I think, well on their way toward a technical merger.

Yes.

The point I was going to make, Dave, was that he’d come from one meeting to another meeting to another meeting. He’s jumping to wherever the problems are to dive down deep, and just being able to see it inside of one organization, I think, will make his life a little easier, a little more unified. All right.

Dave

He talked a lot, actually, about the overhead of context-switching being a killer for him. He can solve any problem if he has time to get his head wrapped around it. But when you’re jumping from 10 little things to 10 little things, then you have to reset your brain, and that’s killing all your time. So, yeah, it’d be a lot more efficient this way.

9. Private Citizens Set Course For Mars

This was a fun story released during the first or the second launch attempt: “Crypto billionaire books SpaceX’s first private Mars flyby.” The first private interplanetary mission to Mars has been booked, and the guy leading it might be the most interesting person you’ve never heard of.

It’s Chun Wang, co-founder of one of the largest Bitcoin mining pools in China. He controls 11% of Bitcoin’s hash rate, travels via 6 different passports, lives part-time in the Arctic, and, get this—

Dave

What?

He follows Mars time, which is a 24-hour, 37-minute daily cycle. That’s got to be tough. He’s already commanded SpaceX’s Fram2 mission, the first crewed mission to go over the Earth’s poles. Now he’s taking a Starship around Mars, so it’s sort of a circum-Martian trip—a 2-year mission for private citizens. He’ll bring some other folks with him. Interplanetary flight. Let that sink in. Alex, what are your thoughts here?

Alex

I like seeing this. I wish there were more examples. I remember also, although I think it fell apart in the end, a wealthy Japanese businessperson planning a mission around the Moon. That didn’t quite happen.

This is such a 2026 statement. I love seeing billionaires flying to other planets. I like it when that happens. I like it even more if they fly to the other planet and then fly back in one piece. I think we’ll see a lot more of it.

I’d like private missions to other planets as well. We haven’t seen that at all to date, to my knowledge. I think we’re going to see more of that. One can immediately fast-forward the recording to a bunch of foot-stomping and hand-wringing about the unfair asymmetries of billionaires getting to go on Mars flybys versus everyone else, and about how asymmetric the interstellar—or at least the interplanetary—economy is.

But I think this is all under the category of good problems to have. I would remind Jared Isaacman, also another billionaire who was pioneering SpaceX missions, that I think we want an entire class of billionaires and business leaders who are all eager to go on either cislunar, lunar-surface, or Martian trips and back.

Alex

I think building that level of awareness and capability in business leaders and then, from there, the entire population is only good for macroeconomic growth.

So here's the thing: I want to point out here that these individuals are spending a lot of money and taking relatively larger amounts of risk, and they're enabling us to follow in their footsteps as the price comes down.

Hear, hear.

I just had a webinar this morning for the Abundance community with Philip Siculo, who heads private missions for SpaceX. When, years ago, I was co-founder of something called Space Adventures, we were negotiating, represented the Russian Space Agency, and sold tickets on Soyuz to go to space. The first ticket was $20 million. The prices rose quickly to about $75 million as the price of labor in Russia went up.

Philip this morning was saying that on the Starship, he's going to be offering flights to orbit. Starship has 1,000 cubic meters—1,000. It's 3 times larger than the space station in the volume that the humans will occupy. The price is going to come down to $20 million a flight. And then, Alex, I think you'll appreciate this.

Alex

Yes.

I did the calculations on if you could electrically winch somebody from the Earth up to orbit—

Alex

By space elevator.

mgh, right? Potential energy, and then accelerate them electrically to orbital velocity, and if you could buy that off the grid at seven cents a kilowatt-hour, the price of launching you and your spacesuit into orbit drops from $20 million down to $200.

Alex

$200 bucks?

We have $200 of electrical cost to winch you and accelerate you to orbital velocity.

Alex

$200 bucks.

Wow.

All right, so mgh, one-half mv squared. Go to your favorite large language model, plug in those numbers, plug in your weight at 100 kilograms for a spacesuit, and you get the same numbers there. There's a price improvement curve that's going to be hitting, and it will ultimately enable all of us to go to space.

In the early days, I don't know if you remember this, Alex, but Elon predicted the cost of a round-trip mission to Mars. Do you remember the number he gave?

Alex

I don't remember the original number.

His goal was $500,000.

Alex

Oh, no, actually, I'm sorry. In the early days, I do remember he was quoting the idea of low hundreds of thousands of dollars at one point.

Yeah. $500,000 for a round-trip flight to Mars.

Alex

Yeah.

Right? So it's first-principles thinking. What's the cost of the fuel, the efficiency as we get to launch 500,000 satellites for the orbital Dyson swarm? When there are hundreds or thousands of Starships launching, the price comes down. So thank you to the wealthy people taking the risk, putting the money in to get this going, because it's going to benefit all of us in the final result.

Alex

Yeah. It turns out that ivory tower is actually quite beneficial for the development of the solar system.

We're demonetizing and democratizing.

Alex

Brad Templeton called early adopters stupid people with too much money because they'll buy the iPhone 45 because it has 1.8 more features than the iPhone 44—

Yeah.

Alex

—and that drives the innovation and democratizes it for everybody else.

10. The Lunar Network Takes Shape

Our final story here is Starlink announces plans for gigabit lunar connectivity. Alex, let's go to you here.

Alex

Yeah. I think this was always going to happen. You were the first in this episode, Peter, to mention a Dyson swarm, so I didn't have to. Thank you for front-running me on that.

You're welcome.

Alex

But I do think one of the architectural benefits of having Starlink in not just sun-synchronous orbit for AI compute, but in general, is that as SpaceX and other companies start to build out swarms of orbital compute, connectivity, and bandwidth—not just in low Earth orbit, but also in cislunar, lunar orbit, maybe L1, L2—it starts to create the beginnings of a fabric of an interplanetary internet, which is something that we've talked about for decades.

Yes. Vint Cerf was the first one.

Alex

Vint Cerf was the chief evangelist. I think he coined the term, even.

He did.

Alex

He did.

And so the idea of an interplanetary internet, whereby we sort of fully wire up our solar system with nodes, with internet router nodes that are able to send packets of data and wire them around, given that the solar system isn't a rigid body, I think we're about to take a step toward that. SpaceX and Starlink put up some funny images online with the monolith from 2001: A Space Odyssey hosting a little Starlink dish on it.

But I think the future that we move to is that we have a swarm of compute in low Earth orbit, we have a swarm of compute increasingly in lunar orbit, and then we can create a fabric of connectivity—

And then Martian orbit—

Alex

—between those swarms.

Around Europa—

Alex

And then Martian orbit and—

Around mining the asteroids. Yeah.

Alex

And the beauty is, in the same style as modern cellphone protocols that can leverage multiple paths and multiple parallel paths to increase bandwidth, having just a single telescope dish on Earth pointed at the Moon really limits our bandwidth. But if we have a swarm—

I'm going to be really pissed off if the cell signal on the Moon is better than in my neighborhood.

Alex

It may very well be. But not only that, they're going to have laser connectivity, so it's not just radio anymore.

Guest

Yeah.

Alex

Imagine a whole swarm of satellites in low Earth orbit that all have laser cameras and laser pointers pointed at the Moon, and vice versa. The bandwidth is going to be tremendous.

All right, here's our actual last story. NASA Administrator Jared Isaacman expects China to send a crewed flyby mission in 2027. Here's his quote: “The next time the world tunes in to watch astronauts fly around the Moon, which will likely be in 2027, they will be taikonauts, and America will no longer be the exclusive power to send humans into a lunar environment.”

So, a little bit of a throwback to Apollo. How do we keep the NASA budget funded? We do it with competition. We can't let the Chinese get there first.

Guest

FOMO.

FOMO, yeah.

Guest

Mm.

But of course, in 2028, we're going to start to see the United States heading toward the South Pole. I'm super excited for that. I love the fact that we're going to the South Pole and not just to a boring equatorial landing site.

Alex

You've got to go. Anyone who's watched For All Mankind knows you have to go where the ice is.

Yes, ice, and also the peaks of eternal light. There are 2 things at the South Pole. There are permanently shadowed craters, and as the Moon was bombarded by comets and asteroids, any water ice that landed on the surface of the Moon sublimated immediately because it was in the sunlight. It goes from ice to gaseous water, and it escapes because of the low gravity of the Moon.

But if those comets and asteroids happened to hit the South Pole and get buried in a deep crater that didn't see the sunlight, ice accumulated there. The other thing that's there is that there are peaks, little mountain peaks, that are seeing sunlight 30 lunar days of the month. So you can set up solar stations there. And of course, there'll be fusion stations there, too.

Alex

Peter, I have to ask you for your prediction. When do you think you'll be able to take a vacation at the Artemis colony?

You know, that's a great question because, while I got the suborbital industry going with the first—

Alex

Yeah.

—XPRIZE—

Alex

You were there.

Yeah. Well, you were?

Alex

You were there.

Oh, I was there, yes.

Alex

Yeah.

It was fantastic, and I really wanted to go on a suborbital flight. I've got a seat on Virgin Galactic once they start flying. Maybe I have to go on Blue Origin. But I'm far more interested in actually landing on the Moon.

Alex

Yeah.

The 9-year-old kid in me definitely wants to go and land on the Moon. So, 2035, I think the price and my ability to afford it will intersect. But I want to start a city on the Moon. I'd like to be a lunar mayor. That would be fun.

Alex

What are you going to call your city, Peter?

I don't have an answer yet. Do you have any good ideas? Maybe Heinlein.

Alex

Magna MOBSTA. May I tell a story for this one, Peter?

Yeah, please do the intro.

Alex

This is a submission from yours truly.

This is your production. AWG Productions.

Alex

This is my submission for an outro. I realized we had the FAANG companies, F-A-A-N-G. We had the Magnificent 7 companies. I realized with OpenAI, SpaceX, and Anthropic about to IPO over the next few months, the world was missing an acronym for what the most valuable companies of the innermost loop would be, that would be the closest to the chips and the autonomy and the space and the AI models.

So the world needed an acronym, so I coined an acronym, and it turns out it's brilliant: Magna MOBSTA. For those asking, the B is Broadcom, which is absolutely essential in the supply chain. Magna MOBSTA are the 11 companies that—

Okay, give—

Guest

Oh, this is red meat for all of our commentators here. Just scatter everywhere.

Read them out for us here.

Alex

There are a lot of As, so I'll see if I can get this right. Microsoft, Amazon, Google—

NVIDIA.

Alex

NVIDIA.

NVIDIA.

Alex

Apple, Meta, OpenAI, Broadcom, SpaceX, Tesla, Anthropic.

Anthropic.

Guest

Anthropic.

So when SpaceX and Tesla merge, it's going to be Magna MABSA. MOBSA.

Alex

MOBSA.

Guest

MOBSA.

All right.

Alex

Come on, Tesla.

All right. Well, let's look at the genius of Alex. Let's play this one out. Here we go.

Nice symphonic.

Alex

I was imagining something like Goldfinger or something. Magna MOBSTA in the glare—

Dave Blunden

At the innermost loop you stay.

He's doing great.

Guest 2

Finger on the recursive core, pulling futures through the door. Compute in your velvet hand. Chips and cloud at your command. Devices bow, autonomy sings, and space bends under unseen strings.

Oh, you build it, you bind it. Turn the key and find it. One more pulse, one more plan. In the palm of Magna MOBSTA's hand.

Magna MOBSTA, hold the code. Magna MOBSTA, ask for more. AI, chips, and cloud all come down. Magna MOBSTA, own the crown. Magna MOBSTA, own the crown.

Every node knows your name. Every router arrives the same. Silent doors and hidden glass. Open when your shadow pass. You wear the hush like satin knife. A lovely threat, a certain life. One kiss and the system turns. One glance and the whole world learns. Oh, you build it, you bind—

All right.

Alex

Awesome.

That was awesome. Alex, thank you for that.

Alex

Magna MOBSTA. People are trying to create ETFs off it. That's not investment advice, but Magna MOBSTA—the world needed a new acronym. World, you have your new acronym now for the top 11 companies at the heart of the singularity.

Love it. All right, gentlemen, so proud to be on this journey with you. Thank you so much. My favorite time of the week is this conversation.

Alex

All right.

You're welcome.

Awesome.

Dave Blunden

Yeah.

Alex

Likewise.

Dave Blunden

Take care, folks. Thank you.

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