Elon 的600亿美元 Cursor 豪赌、Claude 终结 SaaS，以及 OpenAI 的大规模离职潮 | EP #249

Having this global geopolitical dependency on one narrow, volatile geographic region is completely unacceptable.

This Iran war is not really just an oil shock, it's a system shock.

TBD whether you know what direction the US goes on that.

It's almost easier to predict what happens after it ends, which is to say—

We're seeing all of a sudden the frontier labs beginning to compete with the vertical businesses built on top of large language models.

You're going to get this horde of startups building in an AI-native mode, and the big companies will struggle to adopt.

SpaceX negotiates the right to buy Cursor for $60 billion.

Is this just Elon's normal play?

This is actually what a Dyson swarm probably looks like. This is where it's all coming down to. This is like the final countdown of whatever stage of the singularity we're in.

Now that's a moon shot, ladies and gentlemen.

Everybody, welcome to another episode of Moonshots, WTF just happened in tech this week. Your opportunity to hear about what's going on in the world at a hypersonic speed. Here to help prepare you for the future with my incredible moonshot mates, our resident genius AWG. Alex, good to see you in your normal haunt.

Alex

It’s good to be haunted.

And, of course, our exponential investor, DB squared, Dave, back at MIT, I see.

Dave Blumberg

Yes, back from San Francisco for 1 day and then off to New York tonight.

It’s funny. We talk about the 3-day workweek and the 4-day workweek. I think this has invented the 9- and 10-day workweek for us.

Dave Blumberg

I’ve never worked harder in my life.

Alex is constantly reminding us that there’s only going to be 1 singularity in our lifetimes, and you have to savor and use every minute of it. So, it’s worth it. It’s so worth it.

Alex Iskold

I think, Peter, all these people who are absolutely convinced that human attention is a scarce resource are going to get proven wrong ultimately. I think we can manufacture more attention.

I’m looking forward to that.

Alex Iskold

You know, Peter, 2 of 10 is going to be dedicated to this podcast all the time. Or, as they say, go fork yourself.

Salim, you are worse than all of us put together. Where are you on the planet today?

I’m at The Breakers in Palm Beach, in the studio, after giving a keynote to 700 CEOs this morning. They’re pretty freaked out, as they should be, because the world is changing pretty rapidly.

By AI or by you?

About the content. Look, it’s such a huge shock for people in the normal world to ask, “What is happening, and how do I juxtapose myself with this?” It’s really a big topic.

It was a great conference with great questions. That’s always the best measure of whether an event was good or not: the quality of the questions was excellent. The number of people who were standing up and screaming, “Okay!”

It’s the usual Palm Beach dilemma of billionaires worrying that they might become millionaires again.

These are not Palm Beach residents. These are people coming in for this event from all over the country, so this is Main Street USA in 1 place. It was really quite surreal.

I’m here in our Moonshots studio, as always. Again, I still want to get you guys here.

Let’s open up. I had an amazing Saturday night, and I want to tell you guys about it. I was at the annual Breakthrough Prize. It’s put on every year by Yuri and Julia Milner. It’s an annual event that began back in 2012 and was co-founded by Sergey Brin, Mark Zuckerberg, and Anne Wojcicki.

I don’t remember, Salim, but this used to take place in the giant airship hangar at Moffett Field. Remember that giant structure right next to Singularity University? This year, it took place at the Barker Hangar at Santa Monica Airport, right near my home.

Yuri Milner, who’s the principal sponsor of the prize, is an amazing guy—a Russian-Israeli physicist turned VC billionaire. He had some of the most prescient investments. He invested in Facebook, Twitter, Airbnb, and WhatsApp.

He decided, along with Sergey, Anne, and others, that they wanted to put on the Oscars of science—a new Nobel Prize—and recognize the top researchers and breakthroughs. They give out $3 million in cash to the researcher who made the breakthrough, so it’s a big deal, bigger than the Nobel Prize.

Past winners were Stephen Hawking for Hawking radiation back in 2013, Jennifer Doudna for CRISPR, and Demis Hassabis, who won a couple of years ago in life sciences, of course, for his incredible work.

This Saturday night was the best ever. I’ve been honored to have been going since the beginning of this thing. All the Hollywood stars and all the tech CEOs were there. On the Hollywood side, it was Ben Affleck, Anne Hathaway, and Robert Downey Jr. On the tech side, Elon, Sundar, and Demis were there. I spent time with Jensen and Sam Altman. It really was extraordinary.

Alex Iskold

Did you invite them on the pod? That’s like 2028.

I actually did. Sam said he would come on as soon as things slow down a tiny bit. He’s got a busy couple of months ahead of him.

Hopefully sooner than that. Elon said he’s happy to come back at the end of the year for our annual catch-up on which predictions came out and which didn’t. I invited Sundar as well. Chamath was there from the All-In podcast, and he 100% said he’s coming on the pod, so that will be super fun.

Alex Iskold

Oh my God, I’m so looking forward to that. What about Robert Downey Jr.?

I did chat with him about longevity, and I think that could be fun. But having him comment on all the tech news probably isn’t going to work out.

Alex Iskold

Yeah, but you know, when movie stars meet the great scientists, it’s the most bizarre and awesome conflagration. They’re so different, yet they’re so impactful as a group together.

I think the movie stars really want to change the world, and they can’t really affect it, but the scientists can. The movie stars can get the word out, so it really is pretty magical.

I can’t tell you how impactful Iron Man and Tony Stark as a character have been at MIT. It’s just mind-blowing, the impact. I don’t know if Robert Downey Jr. really sounds and acts like Iron Man, his character, but I want to find out.

I really would love to at least try. He’s a super nice guy. I mean, it was fascinating to see Hollywood turn out to rub shoulders with the CEOs, and vice versa. Of course, having Sam and Elon in the same room was interesting.

The whole undercurrent, of course, was the conversation about whether AI and all of this is going to destroy or disrupt Hollywood. It was fascinating.

The winners this year, by the way—I should mention them, since it was the point of the whole evening—were gene therapy for inherited blindness; sickle-cell disease and beta-thalassemia, through a gene therapy that turns on fetal versions of your hemoglobin; the genetics of ALS and frontotemporal dementia; the muon anomalous magnetic moment; and the mathematics of wave behavior.

They did an incredible job explaining the implications of each of these breakthroughs. It was just a great award. In this photo, you can see Jensen down at the bottom right, along with our host, Yuri Milner. It was an incredible evening. I’m super pumped, and thank you to the Milners for putting this on this year.

Alex Iskold

What’s awesome about this is that if you go back to the Renaissance, scientists at that time were treated like rock stars and global celebrities. When we can recreate that, it actually affects civilization. That’s awesome.

It’s interesting, right? Who do you celebrate in the world? It tells you a lot about the culture.

I remember when I was growing up in the ’80s, typically all of the Wall Street bankers were being celebrated. Everybody was going to Wall Street. Dave, I don’t know if you remember that.

Dave Blumberg

Yes, some of the best and brightest.

And now, of course—

Alex Iskold

People who would have changed the world with breakthroughs were going off and putting together pitch decks for real estate transactions. It was so frustrating and painful to see. A lot of our classmates got sucked into that.

Luckily, people are now focused on changing the world.

Anyway, again, it was an amazing evening. Let’s get on to our docket of incredible stories. The first one here is Anthropic’s dominance, in particular, Anthropic releasing Claude Design.

I’m going to play a little bit of a video here. It’s Claude Design, released on top of Opus 4.7, allowing you to design almost anything that you want. What we saw immediately was that the stock prices of both Figma and Adobe started dropping. Figma was down 10%, and Adobe was down about 2%.

The question here, and what I want to talk about as the underlying issue, is that we’re seeing the frontier labs suddenly beginning to compete with the vertical businesses built on top of large language models. Gentlemen, thoughts?

Alex Iskold

I think the elephant in the room here is that this is just an unhobbling of capabilities that were already present in the model.

There may be some fine-tuning or post-training involved at the back end, but this is essentially what one might call an unhobbling of latent capabilities. I think it does raise the question: every time a few of these other SaaS stocks have a mini SaaS-pocalypse freakout from some unhobbling of a base model, whether and how much of the remaining economy is one unhobbling of a scaffold away from utterly cratering existing verticals.

I really do think that, and I say this to every company I advise: you really don't want to be just SaaS at this point. You don't want to be just a scaffold around models. You need your own vertically integrated, native capabilities. If there's a physical-world integration that you can credibly claim or build out, you really should be doing it at this point, because software is getting dissolved.

In the case of Claude Design, I've been using Claude to design PowerPoints and websites and other—call them prototypically Adobe-esque—outputs and artifacts for months now, if not low numbers of years, across all of the frontier models. We already had these capabilities. It's just that through unhobblings like Claude Design, Cowork, or even Claude Code originally, Anthropic is very successfully giving users permission to do what they could already do with first-class Chrome.

I think it does raise the question: how many other user experiences that are right now on dedicated verticals are one unhobbling away from just latent capabilities? Near the top of my list is creating new businesses. I think creating new businesses is just an unhobbling at this point.

Dave, I used it last night, and it's incredible what it can do. It's so dog slow, it's like torture. I'm sure it's dog slow because it's overloaded. I'm sure it's overloaded because Anthropic doesn't have the compute to support all of the growth.

Alex Iskold

Really important point.

Dave Blumberg

It's being shared, and they desperately need a Steve Jobs–Jony Ive kind of guy over there who understands the out-of-the-box experience. The other thing they desperately need is some kind of an ISV strategy—an independent software vendor strategy. They need to build a software community that knows where they're going and can work with them, not just be in their way all the time.

Right now, they haven't declared any roadmap, and we all know that any big company like Microsoft can kill any small company if they want to. But why would they want to? Some things they do want to do, and some things they don't want to do. You have to be intentional and explicit about it.

It really feels like Anthropic is very PhD-ish and acting very PhD-ish, and they need to think a lot about how much power and influence they have now. Going out and just bombarding the stocks of Adobe and Figma, driving them through the floor, and then not offering enough speed to be usable—what did that achieve? That's good for nobody. What is the plan, guys? Tell the world.

You shorted the stock before you made the announcement.

David Friedberg

Yeah, but—

That would be the play. That's the classic play today, isn't it?

David Friedberg

They did recover. They did recover, but the markets hit them hard. And to your point, Alex, I think about what's going to be hit next. I'll give you my top ones: legal research and document review, like LexisNexis; business intelligence, like Tableau; medical documentation and clinical decision support, like Epic. On my list, I have the Bloomberg Terminal, financial modeling and research. I've seen a lot come out there. And then Workday—HR, recruiting, performance. All these are going to get hit.

Alex

It's easier to ask what's not, because AI isn't two-layered anymore. It's just the control layer over design, coding, and so on. And to Alex's point that you made elsewhere, the incumbents aren't competing against new software features; you're competing against compounding intelligence. That's a hard task to take on.

Yeah. And then the question is, who owns the customer? I've got to have a shout-out here, because on our last podcast, Dave, you complained about the design side, and Alex, you guys talked about it. Literally, the answer to that was right there. So, perfect.

David Friedberg

We were pressing it. When the iPhone came out, Steve Jobs was originally opposed to the App Store. He thought it was a bad idea. He hated it. The team there talked him into it, and it became the most incredible experience for everybody.

You would never have had Waze and navigation and great weather apps, and all these niche apps, too. I use the Tide app to know when the tide's coming in. None of that stuff would have ever existed if they hadn't convinced Steve this was a good idea.

This is where Anthropic is right now. You're about to destroy the careers of 80% of software developers. What do you want them to do? Give us the proactive future. Where are you going, and what are they supposed to do with this new tool you've rolled out?

Alex

Yeah.

Tell us.

David Friedberg

That's what they need to do next.

Alex

Claude is launching a partner program. We're in the running for that as OpenExO, so they are actually starting to put that ISV-type structure in place.

Dave, I imagine you're going to see a lot of people come up with a business idea that they brainstormed with Claude, and then they hit print and a pitch deck comes out. You're going to get AI slop in terms of business and pitch decks. We're already there.

David Friedberg

Yeah, that's there.

Alex

We were already there months ago.

Yeah. I think now you don't have to build the business for hours.

David Friedberg

For hours. As pointed out earlier, an entire business is not just business plans.

I think maybe we need to do a little bit of brainstorming for Anthropic here on the pod. I propose the name: “We need a plan for a country of non-geniuses outside the data center.”

Alex

Totally.

David Friedberg

Totally right. Totally right. I think we're in a really weird moment of time right now where, if you use this tool to create a PowerPoint deck, what's the point of the PowerPoint? Who's going to even look at it? Oh, it's probably another AI.

Well, then why is it in PowerPoint in the first place? But right now we're in this transition phase where a product like this makes sense, but it's only maybe a year of human history where something like this makes sense. We're moving to this world where you just tell the AI the outcome you want, and it doesn't need a deck in the first place. It doesn't need a Figma design in the first place.

It's an interesting transition phase, but it's desperately in need of leadership. A roadmap from Dario or Daniela would be incredibly welcome.

Everybody, you may not know this, but I've got an incredible research team. And every week myself, my research team study the meta trends that are impacting the world. Topics like computation, sensors, networks, AI, robotics, 3D printing, synthetic biology. And these metatrend reports I put out once a week enable you to see the future 10 years ahead of anybody else. If you'd like to get access to the metatrends newsletter every week, go to dmandis.com/metatrends. That's dmandis.com/metatrends.

Dario has been saying for at least the last year that AI is going to wipe out 50% of entry-level white-collar jobs in the next 1 to 5 years. And so, out of Anthropic comes a survey of employees—not a large number of employees, a small number—but their estimate is that entry-level software engineers and researchers will be replaced by Mythos in 3 months.

That's a little bit extreme. The speed of replacement is key here. What's most significant about this prediction is that it's coming from the engineers that this technology is likely to replace.

I guess you guys can all remember the golden ticket for your kid was always: go get a degree in coding. Go get computer science, get an internship, get an entry-level position, and then work your way up the chain. All of that is disappearing.

What I find fascinating here is the idea that software engineers and coders are really the canary in the coal mine for a whole slew of other jobs. Thoughts, gentlemen? Does this mean anything?

David Friedberg

Yeah, human arrogance is unbelievable, actually. If you go to San Francisco and talk to these exact people, they all say exactly what the slide said: “This is going to replace almost all entry-level engineers.” Are you an entry-level engineer? “No, not me. I'm not—”

[Laughter]

“I'm a hyper-senior, whatever engineer.” Well, it's on an exponential curve, right?

[Laughter]

You're talking about 3 months after this—3 months. Either way, you're coding yourself out of a job very, very quickly. Yeah, yeah. Okay, I guess I am.

Alex

I'll take the other side of this. I thought it was saying 1/3 of the employees think they'll be replaced. Why isn't it 80%? That would be much more indicative, I think. A third tells me not a lot about that particular thinking.

And yes, they're going to be wiped out. I think we'll just change the types of what we mean by an entry-level job.

David Friedberg

Maybe the 1/3 who thought entry-level engineers are going to be replaced are the senior engineers and executives at the company who are better positioned to judge that.

Right.

Guest

I would just comment: I think it's fair to say we're in the singularity. It's not just vibes, and recursive self-improvement is not just vibes. This is exactly what one would expect to see if we are—as I would argue we are—in the midst of recursive self-improvement, where almost all of the code—it’s been widely reported at Anthropic—is now generated by Claude.

At other companies, too—at Apple and at Google. I don't think we have a slide for this on the pod, but this was pretty spicy news in the past 24 to 48 hours: rumors out of Google DeepMind that a lot of Google DeepMind code is now being generated by Claude. So, I think it's not just necessarily junior SWEs that are being replaced by hypothetical future releases of Claude. It's also other frontier models from other vendors that are being displaced. The blast radius is pretty wide for this.

Mhm.

Guest 2

I think we expect everybody to be leapfrogging each other. I expect this will occur in every single lab.

Guest 3

I think there's also something here for organizations, because the metabolism really gets affected. How do you develop talent when AIs do all the apprenticeship work? This is going to be a big structural problem over the next few years.

Yeah, how do you get to senior-level players if you don't have junior-level players?

Guest 3

Exactly. The SWE ladder gets yanked up. But then we see, in other news in the past 24 hours, Meta announcing an upskilling program to teach people over a relatively short period of time how to lay fiber—optical fiber—in data centers. So, I think these are all just patches or bridges to an ultimate solution. We can debate what the ultimate Star Trek economics look like after this is all over, but in the meantime, certainly, if you're a software engineer—a junior one—maybe you want to start your own company or shift to some other career ladder that has more of a ladder to it.

Guest 2

Yeah, I was going to say something similar. If you look at all the senior executives at OpenAI, I don't know about Anthropic as much, but we know a lot of the people at OpenAI, and the vast majority of them were previously founders and co-founders of other companies. So, they became senior through the entrepreneurship pathway, not through the work-your-way-up-the-ranks pathway. That's the answer to your question. Now, I know a lot of the listeners will then say, “Not everybody can be an entrepreneur.”

Well, we're going to talk about that. At the end of this podcast, we're going to have a session addressing the question that's been asked by all of our listeners: “Hey, I don't think I can become an entrepreneur. Is it for everybody?” So, we're going to dive into that a little bit. So, Dave, a question for you: I can imagine this three-month survey—this SWE apocalypse in three months—works for the hyperscalers, where you've got these high-level software engineers. But what about large Fortune 500 companies? What about Chase and Kaiser Permanente? I think this mythos, or all these technologies, is going to take a while to permeate through the larger companies. How fast is it going to happen there, do you think?

Guest 2

I think it'll vary depending on the quality of the management team, but at great, well-managed companies, it'll happen very, very quickly. I think you have to realize that, after Sam Altman's house got firebombed and then shot at, and he has bullet holes in his door, the bias when you survey these people is toward downplaying, not toward exaggerating. That wasn't true a year ago. A year ago, the bias was toward exaggerating, but now it's the exact opposite.

The capabilities of the AI are going to be far, far ahead of what they want to talk about. The employers will then underreact, by and large, and a lot of them just don't want to deal with it. So, you'll see a subset of them that are early adopters. Part of it is cost reduction, but another, bigger part of it is greenfield building of new things that you never would have thought of before. That's such a massive, sweeping opportunity that those companies will get miles ahead, hopefully create new jobs as they grow. Then the legacy guys will be very, very late to catch up.

And we'll probably get crushed in the process, actually. That includes some really big companies—big banks, big insurance companies. But they just won't move fast enough, and they won't see the little guy coming. A little guy with an effective workforce of 1 billion AIs is going to go very, very quickly past you, and you won't see it coming.

I have a clear lens on how this will happen. Please.

I think what's going to happen is you're going to get this horde of startups building in an AI-native mode. The big companies will struggle to adopt because, as I've mentioned before, all our workflows in big companies are human-to-human. All the approval lines are human-to-human. You need to be AI-native, which means you completely need to redesign your workflow. Most big companies don't have the skill set or the ability to do that.

So, what will happen is they'll watch the incumbents come along, and then they'll start buying them because they have to in order to compete. Then they'll reverse into them that way. That's what I think happens.

Salim, it's really clear, since you're in every country and every city in the world pretty much every week, as far as I can tell.

It is crazy. It is crazy around the—

—how far ahead San Francisco and Boston are moving at warp speed. Then you go elsewhere in the country, and it's much slower. But then you go to Europe, and it's like no action at all—no reaction whatsoever.

Sublinear.

And so, yeah, very sublinear. There's a view of the world that, if you think about some tribal village in the middle of Africa, and then AI comes into the world, does that tribal village even know, or does anyone care? Do you interrupt it in any way? No. It just sits there and does what it's been doing for 100 years.

Yeah. Some of America is going to be like that, too.

But of course, Dave, we have more than 8 billion handsets on the planet, and AI is diffusing to all individuals. I told you in the last pod my story from Morocco about that young man who basically chatted with ChatGPT to come up with his own new career. It's going to be initiative that people take.

David Friedberg

I think people will just grab the tools and start using them to improve their lives. It'll be amazing.

Well, I think at the rate of growth we're talking about, the people who choose to grab the tools will thrive like crazy, regardless of what happened to their old job. I'm almost positive of that now because the rate of what it can do for you and with you is growing so much faster.

People who lament that they spent 10 years learning to code Python better than anyone else on the planet and say, “I can't let go of it,” are going to suffer for not grabbing the new paradigm. But you're trapped by your history. If you let it go—release, let it go—focus on the new and ride the wave, there's so much value being created that you'll actually end up ahead, not behind where you would have been.

Well, as we talk about speed of growth, Elon comes back into front-and-center view here, and here he is predicting the next rollouts of Grok. Grok 4.4 will be twice the size of 4.3, at 1 trillion parameters. It'll be coming out in early April. Then comes Grok 4.5 at 1.5 trillion parameters. Then 4.8, 4.9, and Grok 5, which he predicts to be AGI; Grok 6, ASI; Grok 7, ASI 2.

I mean, this was like a 2:00 a.m. tweet by Elon.

I mean, seriously, I'm ASI 2. We haven't even defined AGI, for God's sake.

[Laughter.] Come on. Insert my normal rant here. How many parameters do you need? This is crazy.

David Friedberg

Yeah, yeah, the more the merrier. Actually, I love the fact that he's willing to talk about parameters, because I did an interview with Sam Altman at MIT back in, I think it was, 2020. I said, “How long till we get to 1 trillion parameters?” And he said, “We need to stop masturbating over parameter count.”

The crowd loved that, and I was like, “Yeah, but I want the answer. Give me just the data, man. I want the facts.” He wouldn't do it. So, here we are years later. Elon's like, “Okay, here's our exact parameter count by model number.” You know, the IQ goes up with the parameter count. We all know that. Then you distill it back down and it gets all confusing, and Alex has to translate it for us. But this is the roadmap, and he's right, and it's transparent, and it's beautiful. It's just great that he's willing to talk about it.

Alexandr Wang, what's your take on this?

Alex

To chime in, I think this doesn't necessarily bode well for xAI's roadmap if they're bragging about parameter counts. I don't think this race to superintelligence is going to work the way Moore's law did. Moore's law ended once Dennard scaling ended in the mid-2000s and the gigahertz per processor core topped out at 4 or 5 gigahertz. I think it's unlikely that we're just going to see larger and larger parameter counts on models.

I think most of these weights—as I've commented on in the past—are probably wasted on world knowledge that could be safely externalized into a database or a text file.

Doubly so once we're able to fully resolve the attention bottleneck and expand context windows to infinity, or at least to a few billion characters or tokens. I'm not sure that we need all of this knowledge built directly into the models. So I worry a little bit that this may signify that xAI or SpaceX AI may be running the last race, not the current race, which is focused on distillation, iterated amplification, and finding more efficient approaches.

If anything, what I would have hoped for from Elon's late-night tweet would be a race to reduce the number of parameters at constant capability. I think a race to the bottom—where we're compressing more and more capability into a smaller and smaller parameter count, so-called intelligence density—would be much more interesting. I'd love to see him bragging that xAI is going to be doubling, tripling, or 10Xing its intelligence density per model. That's what we see from some of these newer labs that are bragging about their moves.

Dave and I like to talk about ternary quantization or even just pure binarization of models. I think there's probably some sort of breakthrough, or maybe a set of breakthroughs, right around the corner that will enable us to take the capabilities of a raw, say, undistilled 10-trillion-parameter model and compress them down to maybe a few million parameter equivalents. That's what I'd like to see Elon focusing on.

So, totally right. I mentioned that to Elon when we were in Austin, and he was absolutely aware of it. But reading the body language, I think you're exactly right, Alex. He's aware of it, and he's probably thinking, “Why do I not have those people? Why do I have the hyperscaler people, the builders, and the creators?”

To use a rough analogy, he's going to launch 50 tons and then 100 tons per Starship launch. But why don't we just make the satellite smaller? That would be the opposite side of that. That's much less marketable as, “Oh, the satellite is half as heavy now.” But that's also an equally important breakthrough.

Let's be fair. He is doing that if you look at the progression of his Raptor V3 engines for Starship; they're becoming smaller. He is the king of minimizing the number of components and making things more elegant. Alex, a question for you: Is AGI also going to be a function of the amount of compute power that you have?

If you look at it that way, xAI is about 2 gigawatts by the end of this year, OpenAI is at 1.2, my research says Meta is about 1 gigawatt, and Anthropic and AWS together are about 1 gigawatt. xAI is still leading in the total amount of compute power, which puts it in some kind of a good position. Don't you think?

Alexandr Wang

It does. He's the king—the god emperor, if you will—of brute-forcing. So I think even if he had just said, “Here's the training kilowatt-hours or gigawatt-hours that we're going to spend on these respective models,” and we saw that we're spending more and more on training, that would be a more convincing scaling-law curve to highlight than just the raw parameter count. Again, that seems like he's pointing to a trend from 2 years ago that the rest of the industry has abandoned as ineffective.

I mean, this is an intelligence meme. This is not a timeline. He gave a timeline for Grok 4.4 through 4.5, but he's just saying Grok 5 is going to be AGI. The question I have is, if he gets to AGI—or whatever his definition of AGI is—and the same thing happens for OpenAI and Anthropic, do they release it, or do they hold on to it and utilize it for their own scientific breakthroughs and capabilities?

Alexandr Wang

I think we're going to see a lot of that. The snide part of me wants to look at the tweet, or X post, that you posted through computer-science language. Is it possible that he's actually defining AGI as Grok 5.0 and defining ASI as Grok 6.0? If you look at those right arrows in pseudocode, that's how I read it.

Well, also, right? We're going to define AGI when Amazon invests in Anthropic. Or no, that's—

Alexandr Wang

That's right. Or OpenAI's definition of AGI is whatever it takes to produce $100 billion in revenue from it. I think these definitions, Salim, are somewhat malleable. But, Peter, taking your question seriously about whether capabilities get so advanced that ultimately the frontier labs no longer want to share them, that's already starting to happen.

We saw that with Mythos, which is already being used with a limited set of users. It's now being shared with substantially all of Europe other than the UK, which is no longer formally part of Europe politically. I chat with my friends at the frontier labs all the time, and they're expecting all of this to happen in the next couple of years.

All right, more on the Elon news: xAI launches Grok text-to-speech, targeting voice developers. xAI just shipped a standalone speech API with a 5% error rate on phone calls versus ElevenLabs at 12%. Of course, ElevenLabs has been the darling and the default. It's a super-competitive price at $0.10 per minute, supporting 25 languages.

Here's the question: Is this just Elon’s normal play, where he comes in with vertical integration, prices to zero, and uses his real-world data to dominate an area? Is he basically going to deliver what ElevenLabs is doing? They've been the king and the default in voice AI for the last 3 or 4 years, and they're valued at $6 billion. Where do you see this going, guys?

Alex

I see this as parallel with what we saw with Claude Design. We're just going to keep having compounding capability, and they're choosing to go after the voice thing because somebody was able to show that it was much better. They'll just keep launching things like this.

I think Elon cares tremendously about leveraging xAI—or, sorry, leveraging X—and winning the race to the best media experience generated. That's image creation, video creation, voice, synthetic characters, virtual girlfriends and boyfriends—all of that. He wants to win for sure because he can distribute it through X.

Yeah.

So he wants that all to be dominated by xAI, and speech is just on that roadmap. I don't think ElevenLabs has a lot to worry about, though. Staying ahead of that curve isn't that hard. At the end of the day, the market is expanding so much. There's a trillion dollars of global payroll in call centers and voice, and we've penetrated 0.0001% of it so far.

The fact that Elon is stepping on ElevenLabs' toes a little bit doesn't matter. ElevenLabs just needs to keep moving forward. It's the same thing we were saying before about Anthropic: Just declare what your roadmap is so people can work with you and not against you.

Alex

But ElevenLabs should be fine.

What's the guidance for entrepreneurs who are building a vertical on top of one of the frontier models? Should they be concerned about maintaining some kind of moat that's not going to get instantly disrupted? How do you think about that?

Yeah, totally. I had a great conversation with Liz Harkavy at a16z last week in San Francisco on this exact topic, because she thinks a lot of the really young Y Combinator founding teams are going to get crushed. But the more mature teams that are doing any kind of enterprise distribution of these capabilities are thriving like crazy. The same is true for any team with a consumer user base.

Peter, you and I have talked a lot about our moms. My mom has not been touched by AI in any way, shape, or form yet. She has money, and your mom has money. She'll buy things. There are massive parts of the tech economy that she would benefit from, everything from hailing an Uber more efficiently to asking, “Should I get a Tesla?” All those things are conversations that our moms should be having with AI. No one has touched it yet, so it's just wide open.

The key for an entrepreneur is to think about your data moat. Do you have any kind of flywheel effect with a data moat? Are you directly in their crosshairs, or are you doing something that they're not going to do? The other thing is tricky to predict because they haven't declared it, but you should be fine. Just jump in.

Isn't—I mean, the other thing is that it's easy to build. You see some kind of capability, an app, or a piece of software, and you can easily replicate it. Then it's a matter of whether you can market it and get customers, right? I think the differentiator—the customer relationships you have, your reach, and your brand—is going to become more and more important. Do you agree with that?

It's also regulatory. We have a company here in the lab, Vocera, that does voice AI. It's doing great, growing like crazy, and has a great valuation. But they're doing mortgages, insurance, and things that have regulatory components to them. You need to know what you can and can't do in terms of your AI quoting a price, having an accountable agent, and all that detail.

xAI isn't going to do any of that. If they give you a better speech API, that just makes it easier for you to succeed as an entrepreneur. There are all kinds of regulatory knowledge, vertical knowledge, and data-moat knowledge that make these companies sustainable.

Salim, I'm sorry. You were about to say?

Yeah, 2 things that strike me from this. One is, if you're building a tech company and you have a tech stack, make sure that the tech stack is agnostic as to what model is underneath it, so you can swap out an ElevenLabs API for xAI or another one. As we just talked about, your edge will be the differential you can make close to the customer. Right? This is where the MTP comes in.

For years, people thought Box.net and Dropbox would be toast because of OneDrive, Google Drive, and iCloud. Because they were so passionate about storage and delivering value to the customer, they thrived, absolutely, with everybody else. What I remember from my days at Yahoo is, yeah, Yahoo has a storage capability, but there are 3 people on that team, and we've got 15 people on Yahoo Finance and 15 people on Yahoo Personals, and it's never going to outperform somebody that's dedicated to that 1 domain full-time, 100%.

So people get scared of the big companies—

Brilliant.

—but if you have passion and you're focused, you are going to outperform them all the time, especially as Dave pointed out, in the areas where you have regulatory issues, industry-specific knowledge, and tacit knowledge that's very specific to that application or use case. That is the part that keeps you going, and then technology is swappable underneath the hood.

You know what's funny about that, Salim, is I launched my first successful 1,000-concurrent-agent swarm yesterday, and I couldn't believe how easy it is. I gave it a whole bunch of math problems to work on. I wanted to do something productive, but I didn't know how to coordinate it into building something cool. I'll have to figure that out this week.

But it burned a lot of tokens. I used Sonnet to keep the price down a little bit. But it works. The whole infrastructure works. The funny thing about it is I can just say to Claude 4.7, “Switch half of these over to a different AI vendor,” and it just does it. [Laughter] Like, wow. It's pretty funny that you're just that willing to swap.

But what Salim said is exactly right. You need to be flexible and nimble, but it's so easy now. In the old days, you had to really think ahead to plan swappable APIs. Now you literally just write 1 line saying “swap,” and it just magically works. It's wild.

Dave

Speaking about swapping—OpenAI. Sorry. [Laughter]

What a perfect segue.

Alex McCaw

Wow. Wow, that's professional grade.

OpenAI is swapping out its leadership. There were 3 senior OpenAI leaders who left a few days ago, on April 17: friend of the pod Kevin Weil, who was the Chief Product Officer; Bill Peebles, the head of Sora; and Srinivas Narayanan, the CTO of B2B.

Best of wishes. I texted with Kevin to wish him all the best, and he's going to keep me informed of what he's up to next. A brilliant, super-nice guy. He was on our stage at the Abundance Summit this year.

The exits are apparently tied to restructuring, reallocating Sora resources, and decentralizing the science team. Again, OpenAI is heading towards its IPO. It's got to focus on use of capital and really driving things that are driving near-term revenue. Do you read anything else into this, gentlemen? Dave, Alex?

Well, it's—

Go ahead, Alex. I'll think for a minute.

Alex McCaw

Lots of thoughts. OpenAI has seen multiple waves of executives leave over the years. I'm reminded of the great Anthropic schism, where Dario and Jared and others all left at approximately the same time and formed Anthropic. I would not at all be surprised, based on what I'm hearing, if this is another one of those events.

The Anthropic schism happened in part due to OpenAI—an earlier version of OpenAI—focusing its attention away from broader research, away from robotics, and away from broader RL initiatives, instead focusing just on large language models. I think this is another moment where OpenAI has just completed its historically large fundraising round, like a $120-billion-plus fundraising round.

They're targeting an IPO as soon as the end of this year, and they're in a death match with Anthropic to achieve recursively self-improving AI, research, and code generation the fastest, with Codex versus Claude Code. I think anything from that vantage point—anything that isn't in that innermost loop of recursive self-improvement, and that means AI for science, that means video generation, and it may mean elements of vertically integrating higher elements of the stack, like B2B applications—anything that doesn't get them into that innermost recursive self-improvement loop, I think is getting thrown out the window.

I would not be surprised if we discover in the next 2 to 3 months another Anthropic-like frontier lab emerges from this schism, and we go from 3 or 4 or 5, depending on how you count frontier labs. There's a new star in the heavens: a new frontier lab is born with a curiously large amount of capital funding from inception that has a totally new approach to solving AGI/ASI. That would be my guess.

Nice. You know, your point is an important one to make here. Of the 11 co-founders of OpenAI, 2 remain. It's Sam and Wojciech, right? You and I met with Wojciech while we were there, and Wojciech is likely to be heading the OpenAI Foundation.

But again, Kevin, congrats on wherever you're going next. Dave, you were going to say?

Well, the numbers are so weird, and I'm going to throw it out there, maybe for the last time, but if you look at the underlying numbers, when we were at OpenAI, Peter, Mark Chen had just been offered $1 billion to come over to Meta.

God, right.

And he turned it down. Remember that?

Yeah.

So then Sam reacted by going to every single employee other than the baristas and giving them a $1 million spot bonus. Not vesting or anything. “Here's $1 million.”

Yeah.

It's crazy numbers. So Kevin, or a guy like Kevin, joined 2, 2.5, maybe 3 years ago, and probably got 0.1%, maybe 0.2%, vesting over 4 years.

That would be a lot at this stage.

Yeah. Well, today, every 0.1% in this new funding is $1 billion. A freaking billion dollars.

Yeah.

And so if he's half vested on that, there's $500 million that he doesn't vest if he leaves. He'll have probably 6 months of vesting severance. And again, I don't know if he left or if he got pushed out. We'll find out, I think, pretty soon, but we don't know as of today.

That frees up half a billion dollars of comp to hire 20, 30, 40, 50 top AI researchers to compete directly with Anthropic. I'm not saying any of this is what's going on. I'm just saying those are the numbers.

It's like when a company goes up in value that much and you start talking about a $1 trillion valuation, it gets wacky. All these dynamics, the amount of money flowing around, are just absolutely wacky. Not normal company dynamics at all.

We're going to talk about this whole conversation later, at the end of the pod here. We keep talking about billions and trillions, and the majority of everybody listening, including myself, can't relate to those numbers.

All right, the speed of AI and AI business is stunning. As we were recording this podcast today, 2 pieces of important news dropped. So we're doing a pickup here to cover them both.

The first is that the ChatGPT Images 2.0 model is released, reporting 99% text accuracy and extraordinary resolution. I'm going to play this video, and then I want to hear what everybody thinks about it.

Guest

We are launching Imagen 2.0. If we think of DALL·E as cave drawings and Imagen 1 as ancient art, then Imagen 2.0 is the Renaissance.

Imagen 2.0 is the smartest image-generation model ever built, with the ability to generate complex, polished, and production-ready visuals with accurate text and structured design. You see, this model isn't just generating images. It's thinking.

That's right. Imagen 2.0 is thinking and researching, and it can even search the web to generate images with the most accurate information available. And with that information, the model is able to generate infographics that explain complex systems and images that solve math problems with proofs.

With new multilingual capabilities, you can create visuals with multiple languages for the entire world. And now, for the first time in image generation, you can create multiple distinct images at once. So you can generate entire magazines with structured typography and photorealistic photos, full renovation plans for every room in your house—

So we're having conversations not only with words, but now images back and forth. Dave, thoughts on this one?

It's incredible. You got to try it. It's so easy to do, too. You just go to your ChatGPT interface and go to Create Image. The naming is terrible, so you're like, “Is this really the right thing?” But as Alex was explaining to me a minute ago, it's all the same. You go to Create Image; it just defaults to Images 2.0 now.

You'll know as soon as you're there because it creates images so quickly compared to before, and they are incredibly good.

Stunning, huh?

Stunning. But also the knowledge. The first day, I asked it for a specific scene in a specific location that only I would know.

[laughter] Huh.

Alex Wissner-Gross

And it did it. It got the background right. It just has insane amounts of built-in knowledge that fills in the gaps in your request compared to before. It’s really good.

Alex, do you have a standard test you use with image models?

Alex Wissner-Gross

It’s been an evolving test over time for OpenAI’s image models. I’ll often ask it a molecular-diagram question. This time around, my first prompt to it was, “Show me the molecular diagram for a typical drug, say ibuprofen or acetaminophen.”

Or my favorite drug: caffeine, 1,3,7-trimethylxanthine.

Alex Wissner-Gross

Caffeine’s a little bit easier in some ways. Maybe it’s more popular. It got a couple of the details wrong. It added a couple of alkane linkages in ibuprofen that weren’t there.

I tried a few other tasks. I asked it to create manga for a single chapter of Accelerando. It did a decent job. I asked it a few other tasks, and I will say the text quality is vastly improved.

If you look at the benchmarks, Arena.ai benchmarked this at 1,512. That’s, I believe, an Elo score. It’s the largest gap. It’s now the number-one-scoring text-to-image model in their text-to-image arena, far ahead of number two, which is Nano Banana 2. So this feels like a remarkable jump.

I have a few other comments, if I may. One is the way OpenAI rolled this out, which is to comment on the reasoning abilities. We talked almost a year ago, I think—maybe 6 to 9 months ago—on the pod about how eventually we’re going to see images as first-class reasoning modalities, and that the frontier models will start to think to themselves as humans do, with sort of a gestalt image rather than in language. I think we are starting to see the dawn of first-class visual reasoning by the models. I think that’s very exciting.

I think the use of tools by ChatGPT Images 2.0—or GPT Image-2, depending on which way you’re calling this—is an interesting plot twist. Query whether that will lead to additional copyright-infringement claims. I don’t know, but it’s an interesting question.

I think the biggest question we should be asking here, doubly so in the context of a story that we covered elsewhere in this pod, is the departure of a number of key OpenAI executives, ostensibly in service of OpenAI trying to better compete with Anthropic in the code-generation races. Why is OpenAI releasing such a compute-intensive modality when Google launched Gemini’s Nano Banana Pro 2?

The rumors going around were that it sucked up an enormous amount of Google’s compute, and that Google was scrambling to find the compute to service the Nano Banana Pro demand. So why, now, in the middle of a code-generation and AI-researcher race, is OpenAI choosing to release what ostensibly would be a pretty compute-intensive modality that probably—if I’m speculating here, but if you look at the enterprise value of text-to-image generation—I would speculate is an order of magnitude or more less valuable than code generation? Why, in the middle of this hot race, is OpenAI releasing an AI model? Any ideas?

I’ll speculate. This is unvarnished speculation, because it’s a pretty strange bit of timing to drop Sora on the one hand and then launch a next-generation image model on the other.

One thought—again, admittedly speculation—is that if images are much lower-cost to generate at this point, and if the speed at which it’s able to generate images has somehow, through algorithmic advances, suddenly gotten a lot cheaper and faster, maybe images are no longer that compute-intensive. Maybe OpenAI has achieved some sort of algorithmic advance versus, say, Google’s Nano Banana Pro, in which case they can actually afford to allocate whatever compute is necessary to this without breaking a sweat or otherwise interfering with their recursive self-improvement program. That’s one theory.

I could speculate as to other theories. Another theory would be that somehow image generation is actually instrumental in competing with Anthropic. Anthropic, as we’ve discussed on the pod previously, doesn’t allow users to generate images. So perhaps, again speculatively, OpenAI thinks that reasoning over images—and I would note that even in that launch video, OpenAI talks about, and in its demonstrations generates, screenshots, for example—could be important.

Sam was teasing this release earlier today by posting what looked like a screenshot, I think, of macOS and saying that, in effect, this isn’t a real screenshot. So maybe OpenAI is betting on some form of supremacy in code generation, using images as some sort of intermediate representation. Maybe they think that having stronger image-generation capabilities will lead to better UI-design code generation or something like that. Those were the first 2 speculations that popped into my mind.

Fascinating. When we heard that they were going to be dropping something, of course, we’ve been waiting for Spud to drop, right? Their answer to Mythos.

Alex Wissner-Gross

Right.

Still expected?

Alex Wissner-Gross

This month, possibly. Still expected, based on everything I’m hearing. Still expected imminently. It may be branded as a sort of point-five model—ChatGPT 5.5, if you will—but as far as I can tell, it’s still expected imminently.

Salim, any thoughts on ChatGPT Images 2.0?

I’ve got 2 thoughts. One is that Alex’s hypothesis is always a possibility, obviously. I think they need to just show that they’re still relevant, because they’re launching this and then the Twitterverse is abuzz with it. It’s crazy, the reactions.

My standard test is to generate a photograph of a Gutenberg printing press from the 1500s. Obviously, we didn’t have cameras back then, and it did really well. I did this about 6 months ago with Nano Banana and ChatGPT, and the Nano Banana one was way more realistic. The ChatGPT one just looks marginally better than the Nano Banana one from a few months ago.

My secondary thought is that, as they’re generating buzz for an IPO, they need as much consumer attention and retail-investor attention as possible. Getting this out to creatives and having it out there may be one way of doing it.

Yeah. I scratch my head again, just briefly, over that, because every narrative that I’ve seen out of OpenAI, certainly in the lead-up to an IPO, including from their CFO, is the same narrative of OpenAI transitioning from consumer use to enterprise use. So again, it’s such a head-scratcher in my mind. Are enterprises just loving text-to-image? Maybe.

No, no, my point before is that there’s a huge number of PDFs and PowerPoints in enterprises that could be scanned and rendered in a way that—that’s what I thought was really interesting about the Claude design thing. So maybe this is a response to say, “Hey, we’re still relevant there, too.”

I think you guys nailed it. I use Anthropic constantly, and it’s generating megatons of code for me. If I ask it to create a diagram to explain what it did, it sucks beyond belief. It just absolutely falls apart. So then I have it write code to create diagrams, which is a royal pain in the ass, but it just doesn’t do it.

I use Gemini for most of my planning because it’s brilliant. I use Claude for all the coding and all the complicated thinking now. But there’s no role for OpenAI in that stack. If you look at what goes on in the real white-collar world, coding is a little piece of it, but architecture, design documents, and images are a massively bigger part of it.

So this actually makes OpenAI suddenly relevant in white-collar. It’s a good entry point, and once it’s in the stack, the use tends to grow. So I’m sure it’s not a random move. I’m sure they’ve got real data behind it.

The second story that came out that’s a very important one is that SpaceX negotiates the right to buy Cursor for $60 billion. The war is out there. We heard Elon say they have a lot of catching up to do, especially in coding. Well, guess what? Buying Cursor helps them catch up to Anthropic.

Cursor already has the best product out there, but they’re compute-constrained. And, of course, Colossus gives them all the compute they need. Interesting breaking story. Dave, what do you think of it?

Well, you got that storyline exactly right. They need a foundation model. They need compute. xAI needs distribution and needs to get in the game and be used.

It’s interesting, though, because when you use Cursor right now, you can switch between models instantaneously. So you would assume that if this does happen, somehow it would favor using xAI—maybe put it at the top or default to it or something like that. But as of right now, Cursor’s completely neutral. It’s just easy to trigger between the models.

And do you remember, OpenAI was very close to buying Windsurf, and then it got blocked? So they turned and featured Cursor in the launch of GPT-5. Remember? They kind of featured that partnership. I don’t know if this would then trigger some kind of bidding war, but I guess if this news story is right, Elon has it locked up. He has a right to buy the entire company.

Or they pay a $10 billion walk-away fee. Fascinating.

Yeah, yeah, interesting structure.

They bought optionality, and I think the trade-off is exactly right. Cursor needs compute, and these guys need access to code capabilities.

Yeah, I think this is also effectively a liquidity event.

You know, a lot of the Silicon Valley gang is saying these valuations are incredible, but when are we going to see actual liquidity? This is at least $10 billion of liquidity, more likely $60 billion, potentially in pre-IPO stock.

Alex

There are a few thoughts. I think this is a bizarre story, but it's the sort of bizarre story that could only happen in April of 2026. One, I would say, dovetailing with the last story, we're seeing that code generation and the rise of AI researchers is maybe, in some sense, the innermost loop within the innermost loop. Everything seems to be coming down to the question of who, at least among the frontier labs, can generate the best code, and why generate the best code? So that you can build the best AI researcher to generate the best AI to generate the best code.

Elon and others have made comments after the SpaceX–xAI merger that xAI was not built correctly from the start and that they needed to start all over again. SpaceX's vice president of Starlink, who's also now head of engineering for xAI, made similar comments publicly as well. There was a desire to reboot. Query whether this acquisition, or quote-unquote “right to acquire Cursor,” actually ultimately represents a complete reset of Grok. Whether this is, in some sense, a manifestation of—or an admission that—Grok's code-generation abilities were not remaining competitive with the frontier. What's the frontier? It's Claude.

Next point: There have been many studies, many analyses of Cursor's behavior. Of course, Cursor was for the longest time, before it decided to launch its own first-party model, in a somewhat privileged position to watch user interactions with all of the existing third-party frontier models. If you're Cursor, you have to ask yourself: You're in some sense in a position of quasi-vulnerability to those frontier model vendors. What do you do?

Quasi-vulnerability.

Alex Danco

Quasi-vulnerability.

Very big quasi there, yeah.

Alex Danco

Yeah, the quasi is a load-bearing word fragment there. So, you're in a position of vulnerability. What do you do? Well, to the extent that you have access to user behavior, you take as much user behavior and context as you possibly can. You probably take a Chinese open-weight model and fine-tune the heck out of it based on all of this user behavior, and to the extent that you can replicate Claude Code or Claude-level code-generation abilities, you probably try to do that.

So, if you're xAI and you really would like to be competitive with Anthropic in particular, what's the best way to do that? Well, you buy yourself optionality from Cursor, which may be one of the best and/or cheapest ways to basically acquire Anthropic-level user interaction for code gen, given all of the users that have been interacting with Claude via Cursor.

That's a second thought. Fascinating.

Alex Danco

Third thought, if I may, is the whole cloud angle. SpaceX, right before our eyes, is turning into a hyperscaler. We talk about the Dyson swarm all the time, but this is actually what a Dyson swarm probably looks like. It looks like a cloud in the stars. This is seemingly going to be the first major tenant of the Dyson swarm that SpaceX and xAI are going to build in sun-synchronous orbit—or maybe around the sun; I don't know—but either way, this is potentially, even in SpaceX's announcement, a million H100 equivalents.

That's Colossus now. That's ground-based. But that's just an anchor, third-party tenant. Historically, Colossus and Colossus 2, Elon-style, were fully vertically integrated with only Elon's applications. Now we're seeing an opening up of that cloud infrastructure layer to third parties.

Cursor may or may not get acquired. I would expect to see many more deals like this, where Elon is now exposing his vast GPU set on the ground and soon in orbit to third parties. So how soon until we're all running our compute loads on Colossus, Colossus 2, or whatever SpaceX renames or names? Maybe it's Colossus 3 in orbit. I think it's a very interesting time for orbital clouds, and we're starting to see the beginning of that.

I think the second thing that Alex said is a great lead-in to our StackBlitz podcast coming up in a couple of days, because I think it's very likely that Elon is incredibly confident in his foundation model and, after layering chain-of-thought reasoning on top of that, is fairly confident in that. But that next layer that actually makes code generation really work—there's a lot to it. Cursor has worked on it a lot, StackBlitz has worked on it a lot, and a handful of other companies have, too. But I think this elevates that layer of the stack to first-class.

The science and technology in that layer of the stack wasn't really part of the AI researcher lexicon, but it's become critically important to getting actual functional code, actually getting the AI researcher to really exist. And so I think this elevates that layer to another level, and Elon is saying, “Yeah, we could have the best foundation model ever, but we can't exist in the code-generation world without this other thing. It's easier to buy it from Cursor than try and reinvent it.”

Yeah, note that the $60 billion is a drop in the bucket compared to their overall market cap. It's a little small.

And you have to imagine Cursor's concerned about being disrupted, right? They have everybody else creating capabilities equivalent to or greater than theirs. This is a chance for them to become part of SpaceX–xAI as it's about to catapult, literally, to the stars. It basically secures their position and makes them part of Elon's ecosystem.

Well, also, the Cursor team—the 4 guys—they're awesome, but none of them is a natural public-company CEO. There's no obvious Mark Zuckerberg there. So what are you going to do? Just stay private until eventually you're crushed? You have to do something to block this up, and this is a good way to do it.

Alex Danco

I also just want to contextualize, if I may. We talked about this elsewhere in the pod. Rumors are flying again that many Google DeepMind researchers aren't using Gemini for code gen; they're using Claude. So, in some sense, there's a sense in the air that code gen for recursive self-improvement AI researchers—this is where it's all coming down to.

This is like the final countdown of whatever stage of the singularity we're in, where SpaceX is scrambling with $60 million acquisitions, OpenAI is dropping large chunks of its initiatives just to focus on this, and Anthropic is hoarding the compute that it does have just to focus on this. I think if you squint at the chessboard, all of the major players are dropping either large amounts of capital or entire other initiatives just to focus on code gen. I would speculate—but this is somewhat informed speculation—that's because code gen, in all of their minds, is the critical path to maximum change, maximum acceleration, and something interesting is certainly going to come out of—

Grok 5.

Alex Danco

Maximum recursive self-improvement. Well, yeah, Grok 5, which is conveniently the definition in Elon's mind of AGI.

Everybody, welcome to the health section of Moonshots brought to you by Fountain Life. We talk about AI on this Moonshots podcast all the time. One of the most important things AI is going to be able to do for you besides educating your kids and helping you with your taxes is making sure that you're living a healthy lifestyle, that you get a chance to get to 100 plus. I'm here today with Dr. Don Mussallem, the chief medical officer of Fountain Life and a part of my medical team. Don, a pleasure.

Don Mussallem

Great to be here.

The thing that people are concerned about most about living to 100 or 120 is their cognitive abilities, making sure they don't have dementia. The conservative estimates are that 45% are entirely preventable. With the advanced testing we're doing at Fountain Life, one quarter of our members had advanced brain age. When we partnered it with healthy living—eating healthier, moving our bodies, sleep, optimizing sleep—we saw that we improved that brain age by 26%. If having healthy brain function till 100, 120 is important to you, check out Fountain Life, go to fountainlife.com/peter. Make sure you become the CEO of your own health. All right, now back to the episode.

All right, let’s jump into the economy. I found this chart absolutely fascinating. This is how hyperscalers compare to the megaprojects in the United States, in particular. This is from a tweet from Finn Moorhouse.

Data center capex is hitting almost $1 trillion by the end of this year—a 6-year, trillion-dollar spend. That compares to the Apollo program, which spent $257 billion over 14 years; the Manhattan Project, which spent $36 billion over 5 years; and the Interstate Highway Program, which spent $620 billion over 37 years.

When I first looked at this chart, I was like, “Okay, the numbers are big, but really, let’s talk about inflation, or particularly about the percentage against GDP.” So, I did the work and put this chart together. The data centers are still outpacing the Apollo program and the Manhattan Project by a factor of 5. And, of course, in the right-hand column, all those were government-funded, and we’re talking about 4 private companies driving this trillion-dollar spend. Alex, your take.

Alex Danco

Well, the Dyson swarm is going to consume a huge portion of our economy. I don’t know that this should be that surprising to the audience. If it’s 1% one year, 10% the next year, and 50% the following year, at some point all of this infrastructure spend—all of this tiling the Earth, and soon the heavens, with compute—surely consumes the economy. This becomes the de facto economy.

I would argue that this is fundamentally unlike the railroad build-out, which some may bellyache about, saying, “Well, actually, if you look at the railroad build-out, that consumed a disproportionate amount of GDP, and ultimately there was a collapse,” or compare it with Apollo or interstate highways. I really do think it may not happen in this economic cycle. It may happen 2 or 3 economic cycles later that data centers are the infrastructure for the future of civilization.

I really do think, say, 10, 20, 30 years from now, the majority of intelligence in our solar system is likely to be intelligence that’s hosted on data centers and not human. Maybe a good deal sooner than that. That’s a very, very conservative outer bound. So, if this is the entire machinery of civilization, then at what price do we price the spend? It’s the future of our whole civilization.

But, Alex, the most interesting thing here is not the percentage or the numbers. It’s the fact that all these other megaprojects were government-funded, and here we see private funding driving this incredible spend.

Alex Danco

Yeah, I think this is very, very exciting because in the past you’d have to take all the momentum and get all the political parties involved, and there’s such a lot of spadework to do to get the government behind a major initiative like this. Therefore, everything just took forever, and major things took forever. Now, because of democratization, individuals and private companies can do things that only governments could do before, and that will move civilization forward at a 10x faster rate than we did 50, 60, or 70 years ago—100x faster. So, this is, for me, the most exciting part.

If the economy grows as quickly as some expect, including Elon and including myself, I’m not even sure the distinction between public spend and private spend matters that much. If the economy ends up growing 10x, who cares whether it’s the approximately 1/3 of the economy that is the public sector or the 2/3 that’s the private sector? If the pie grows so wildly, it almost, in my mind, doesn’t matter.

The government could, in a much, much wealthier future a few years from now, pick up the tab for AI data centers if we want to, and that would still be enormous compared to the size of the economy that we have right now. So, in my mind, to the extent I sound a little bit blasé, perhaps, about the distinction between the private sector funding this build-out versus the public sector, it really is because I think this is the fundamental infrastructure for the future of civilization that will get us to 10x year over year.

Alex Danco

Yeah. In the long run, yes.

We’re running out of the long run, though.

Alex Danco

It’s like a few years away.

Yeah, I know. Five years is a long run.

Salim Ismail’s point is the one I think is most interesting. The notion of getting political support—I mean, look at the Apollo program as a perfect example, right? It cratered back in ’72 because the political support was gone. But all of a sudden, a lot of these projects take a very dedicated, long expenditure of capital. That can be done by a single individual, right? Elon or Bezos can say, “We’re building out the Moon. We’re building out Mars.”

Let me give you one story on the government versus private. That F-35 program that’s on this chart is actually the second-fastest-growing spend of all time. [Laughter]

Nick Leonard, who works on VoiceRun here in the lab, worked on the F-35. They asked him—that thing’s supposed to have a vertical takeoff capability, which is very, very hard to do—and so they asked him to work on a nozzle for the thrust. It’s like this rotating nozzle. “Could we make it out of carbon fiber?” You’re like, “Well, there’s a 1,000-degree exhaust going through that thing. How are we going to make that out of carbon fiber?” They said, “Well, just try.”

So, they found a way to make it out of carbon fiber to reduce the weight a little bit, but they have to replace it every 50 flight hours. [Laughter]

They said, “Fine. Good enough. It’s only taxpayer money. Go ahead and burn those things up.” What do they cost? A million dollars each? Great. We’ll burn them up every 50 flight hours. So, they actually made that choice.

When I look at this chart and I look at the data center build-out, this is the most important and biggest thing humanity has ever done. Whatever country or region does it first and fastest is going to have insane amounts of growth, success, power, whatever. I know the point of the slide is that this is the biggest and fastest thing ever, but to me, it’s too small, too slow.

Mhm.

It needs to be even bigger, even faster. If you pushed any of it into the public sector, it’s going to slow down and go nowhere. So, it’s great that it’s in the private sector, but it’s so different from all these other projects. This is going to change AI. AI underlies everything in the future.

The economy of society, right? Two quick points here.

Yeah, it’s everything.

I remember, Peter, when we were doing the one-week executive programs at Singularity University, we had a very senior official at the DoD come through. He stood up and goes, “Yeah, all you guys talk about exponentials. Investors are all hovering at the knee of the curve, trying to take the technologies as the curve turns upward. You forget that government is the one that’s been funding the flat part of the curve for a very, very long time. And as governments drop in and get run out of money, who’s funding the flat part of the curve?”

That answer was kind of stuck in my head for a long time, until we saw that the private sector can actually take on that type of investment and make that investment. To Dave’s point, the U.S. private sector is doing a huge amount, but I think there’s a massive problem when governments around the world are not putting enough money into data centers for their sovereign future. That is going to be a massive problem for them as this world moves on because they’re going to get left behind. So, if you’re in government in any country around the world, figure out a way of funding your own data centers and retaining your own sovereignty.

Alex Danco

Totally right. I think the only comparison that I like in history is World War II. The industrialization build-out during World War II, just in terms of speed and mobilization, is what a well-thought-out plan would do right now. Do what we did in 1940–41, not just in the U.S., but globally. That’s what you should be doing right now in the AI race. I don’t know what we spent on World War II, but it was a heck of a lot more than the highway system and the railroad system.

I’m curious to take a spot poll, if I may. We’re all obviously very excited about 1% and a continued increase in the annual percentage of GDP spent on AI data centers. I’m curious, of the 3 of you, what do you think, looking backward historically, will end up being the actual peak annual percentage of GDP spent on AI data centers?

Alex Danco

On Earth and in space?

Everywhere.

And on the Moon? In the human sphere of existence?

And after we’ve disassembled Jupiter.

Alex Danco

It will approach 100%. Yeah, it should just keep going.

Right, so you think it goes to 100%? What about you, Salim?

Well, I think over time it should get higher and higher because, over time, as you talk about your inner loop, everything becomes computronium, and everything should be in service of building that.

What about you, Dave?

If you include building physical things to make people happy as part of the data center build-out, it’s 100%. If you take that out, I’d say probably 95%.

I think it’s going to—this is not investment advice—I think it’s going to cap out somewhere between a quarter and a third of GDP, at maximum, and then decline after that. At some point, AI data centers are consuming so much of our available capital and resources that pressure increases through the roof to reduce that spend through new means of automation, like robots. Maybe we get our nano-assemblers, and that creates cost efficiencies that compensate for the— to the extent there is a bit of a Jevons paradox here.

Alex Danco

I don't actually think it's going to go to 100%. I think it'll peak somewhere.

On definitions here.

Yeah, and also, at some point, we're going to be under a post-capitalist and post-scarce world, in which case GDP becomes meaningless.

Okay, you think GDP is going to become meaningless before we finish exhausting—

Well, let's look at the data and see what it says. I don't think so. We need to be moving to other measures anyway.

Honestly, I like the way Alex framed it a lot. Because if you tell the general population we're going to spend 99% of our money on data centers, they're like, “How does that benefit me? That's terrible. That's the worst thing in the world.” The way Alex phrased it, it's just a shift in what you define, but it's much more palatable to the world because, in that definition, you're using pure compute for about a third of the global economy, and the other two-thirds is robots and the creation of physical goods and everything else.

Let's get to AGI first, and then we can rethink this whole thing and it'll figure it out for us.

All right, talking about the global economy, let's look at the US economy in particular: manufacturing. We had 17 quarters of negative manufacturing growth—17 quarters of contraction—and we've now seen 16 quarters of pretty steady increasing acceleration in manufacturing. I think we offshored most of America's manufacturing over the last 50-odd years.

In particular, looking at the last 17 quarters, the drivers of that contraction were the trade war uncertainties under Trump's first set of tariffs. It was the COVID-19 collapse; we destroyed supply chains. Semiconductor shortages were also part of the COVID collapse. The post-2008 financial crisis had a hangover that didn't fully recover. And so, there were a lot of challenges as companies found themselves unable to manufacture goods and services.

We've seen now, in the last 16 quarters, the CHIPS Act, the Inflation Reduction Act, the Infrastructure Investment and Jobs Act, a reshoring wave, AI infrastructure build-out, and defense spending basically driving up US manufacturing. So, hopefully this continues. Any thoughts on this one, Dave?

Well, if you want every American to be ultra-wealthy, this is exactly the right thing to do. If you're worried about India, like Salim, you'd be worried sick about this. A lot of prosperity around Asia has come from outsourcing work that ends up in iPhones and in other manufactured products that then flow through America and go back out to the world.

But if we start onshoring all that manufacturing again and robotizing it, that depletes the rest of the world of those jobs. So, I would expect that global hatred of America is going to go way up, that American wealth will go through the roof, and that this trend will continue at least for the remainder of this administration. Try not to put an “is this a good thing or a bad thing” spin on it, because it affects different people in different parts of the world very, very differently.

A massive amount of capital is now flowing into the US from all over the world. And because it's not labor-dependent—it's robot-dependent—it's basically all going to come back into the United States.

It was engineered. Salim, what are your thoughts on Dave's point?

So, I think that's exactly right. But what's also happening in places like India is that they're building their own manufacturing and automation capabilities, and they're doing it at reasonably low cost. Just look at the way India's tiling the country with solar panels and becoming energy self-sufficient. I think that's generally a good thing.

I think there's a little comment to be made here, though. We've had a very bad political trope over the last many years that globalization has caused job losses. And it turns out that's not the case at all. What's happened in the US specifically is financial engineering and private equity have gone in and bought companies in logistics, trucking, manufacturing, engineering, et cetera. And when private equity goes in, they stall all innovation, outsource the jobs, and they've made it easier to do that.

The irony around this—this was uncovered by Robert Goldberg—was that the LPs in these private equity funds that are outsourcing the jobs are the California pension funds. So, the pensions of the workers are funding the outsourcing of the jobs. This is the worst incentive alignment ever.

What we've now found using the ExO model and others is that we've found ways of de-risking innovation and doing it at low cost. Back to that demonetization thing: in the past, when you wanted to do anything heavily innovative in a company, you had to make a big bet, right? The CFO hated you and defunded it as fast as they could, et cetera.

Today, because the cost of technology is so low, you can do disruptive innovation and try things out at a very low cost at the edge of the organization, only doubling down when you see success. This is allowing US companies now to go back to the thing Robert Goldberg's MTP was about: reinventing American exceptionalism. And he found a way of recreating that wheel.

That's now, I think, starting to bite in a big way, where manufacturing and innovation are back in the US with a vengeance. And I think that's just fantastic. I think the ripple effects will be fantastic for the whole world as well, because it'll push the whole world to operate that way. You'll have to be able to compete on innovation going forward.

Alex

I think manufacturing wants to be sovereign at the end of the day. I think we're going to have advanced nanotechnology and all sorts of other sci-fi-esque technologies that enable every sovereign nation-state to effectively re-domesticate, or domesticate, its entire supply chain.

I don't think that's necessarily the case, given all of the technologies that I reasonably foresee over the next, call it, 5 to 10 years. I don't think supply chains necessarily want to be global in the long term. In which case, capacity domestically, for the US or otherwise, is limited more by technological advances and capability than by any sort of geopolitical shifts or shifting around of supply chains.

With all of the advances that I expect to happen over the next 5 to 10 years, I just don't see a bright future for global supply chains and, in particular, globally supplied manufacturing at all. Not investment advice.

Alex, you know, if you think about the fact that we're talking about manufacturing physical goods and services, the other place we're going to see this probably is in food production, with vertical farms and protein-based meats. The flip side of this, of course, is that intelligence is being globalized, right? By definition, when we move these things to orbit, they're globalized planet-wide.

So, it's a fascinating shifting of where things get made and consumed: things are being physically globalized, but not politically. I think we end up with probably 2 sovereign Dyson swarms. Well, the US is going to wind up probably with, I think, 3 or 4 competing corporate Dyson swarms, and who knows how many China will wind up with. The question is whether Europe or other parts of the world will be capable of launching their own Dyson swarms in time, but that's still a political division of the heavens.

Yeah. Salim, you're going to say?

No, I was just agreeing. I think this is a very important point. By the way, I mentioned Holland last time as a small point because it was a global food exporter—not just from vertical farming, but they were doing hydroponics, aeroponics, and greenhouses. I'm actually getting a briefing from one of the top vertical farming companies in the world, so I'll come back and tell you guys what I learned from that.

Well, I'll tell you that globalization exists fundamentally because of shipping. Everything is moved by boat. There's no other way to move it around the world, and the US Navy is the protector of all global shipping. You see what's going on in the world right now. What's weird about the US is that we have all the natural resources that we need internally.

Yeah.

And so, we don't actually have a huge incentive to support globalization other than labor. We're heavily reliant on Asian labor to make iPhones, cars, whatever—to make parts. But if that all moves to robots, there's no natural reason why the US Navy needs to support and defend every transport ship in the entire world going forward. It's TBD what direction the US goes on that, but no other country is geared up right now to protect shipping.

And then you see Europe isn't even trying in the Strait of Hormuz. It's like, “Yeah, you know what? We don't want to eat that cost. You guys figure it out.” But you're the biggest importer of the oil. Like, wait a minute.

Yeah. What's the deal?

Our navy is so critically important for this. Yeah.

Dave, we'll come back to that. You surfaced this chart in our WhatsApp group. Do you want to brief us on it?

It's truly amazing. Now, this is Perplexity speaking, so if anyone fact-checks us, remember it was Perplexity, but I think it's right. I checked it again.

AI.

But look, I looked at the numbers. San Francisco is now bigger than China—all of China—in market cap. It's just counting the market caps of the public companies.

So it's about a billion people versus about a million people. The flow of wealth into San Francisco is going to make it the global capital—the global financial capital—of the world, easily. If it's not already, man, people just haven't noticed yet.

But you can see on the chart the rest of the U.S. compared to San Francisco. And then in China, you know, it's a small fraction of that. Actually, there's another slide that showed the same thing, but I guess we took it out.

The point: 14% of the global market cap is concentrated in 7,000 square miles. That's extraordinary.

Alex Karp

There's a cheery thought. It's also, I think, in part reflective of, admittedly, a cliché: the way public companies work in China is materially different, usually, from the values of the very shareholder-activist West and the U.S. The cliché, of course, is that for Chinese public companies, profit is not necessarily as prized a measure of performance as it is in U.S. and Western companies.

If you do a discounted cash flow analysis for Chinese public companies, and then you also discount the impact of the CCP and government regulation, maybe there's, in some sense, a suppression of book value that comes from that, which otherwise could be realized if they were a little bit more profit-oriented and a little bit less, say, oppressed by the CCP.

But I think it's an interesting statistic, for sure. If you're a U.S. policymaker anywhere other than San Mateo County, San Francisco, or Santa Clara, you would absolutely love for your geography to be the world center of the future. But I would pose, maybe as an open question: How do we get diffusion from Silicon Valley and San Francisco to the rest of the country and the rest of the world?

Tax will do that for us.

Alex Karp

The billionaire tax will shake loose a few billionaires, for sure.

No, no, no. I mean, the billionaire tax is interesting.

Alex Karp

It's moving to Florida for sure.

Right?

Alex Karp

Yeah.

David Friedberg

Well, the billionaire tax already pushed out half of the tax base that it would have taxed. That came up on the All-In podcast in a big way. It's completely counterproductive, but it's not affecting the inbound entrepreneurs at all.

What's happening is people are forming their companies and their teams and their ideas all over the country, wherever there are universities, mostly. But then they're moving to San Francisco, and they're young and they're not rich yet, so they move to San Francisco without thinking about the billionaire tax. Then later, when they're billionaires, they're like, "Why am I here?" And that's when they move to Austin or wherever.

But it's not slowing down San Francisco in terms of entrepreneurship and growth. So the slide we're looking at here is tech market cap, but actually San Francisco passed all of China in total market cap. Alex pointed out that market cap is a little bit of a weird statistic in China, but I really feel like when you measure a global economy, if somebody makes dinner for somebody else and then charges them $20 for it, that counts as a dollar in the economy, but it's gone. It got eaten. It turns to poop. It's not a thing.

Somebody else creates a new chip or a new engine or whatever, and creates permanent value for that society—a massive amount of gain. An iPhone creates a massive amount of permanent gain.

Yeah, exactly.

David Friedberg

When you strip it all out, what's going on in San Francisco is absolutely extraordinary and unprecedented in world history. You should go and experience it for a week or two, no matter where you live, and just talk to people. You can't even walk down Market Street without 5 people talking about Anthropic and OpenAI right behind it.

Go to the coffee shops. The billboards on 101, especially—every billboard is referencing superintelligence. Amazing.

All right, let's jump into one of our favorite topics here on our pod: the race to space. This week, Blue Origin and its New Glenn did their third launch. It was a bittersweet launch. New Glenn's reused booster flew beautifully and landed beautifully. The challenge is that when they launched their AST satellite, it launched into the wrong orbit, and AST is having to deorbit that satellite.

I think the points I want to make on this particular story are, first, Blue Origin has demonstrated booster reuse 3 times faster than Falcon 9, in terms of the time from when they started launching New Glenn to when they were able to reuse their booster. Of course, the reality is that SpaceX has launched 600 times, and Blue Origin has only launched 3 times.

The big race right now is who gets the lunar contracts. We're going to the Moon in 2028 with Artemis 4. Artemis 3, don't forget, is going to be testing operations in low Earth orbit for going to the Moon. Artemis 4 is going to be making a lunar attempt, and right now SpaceX and Blue Origin are vying for the contract for NASA's lander there. Alex, I'll go to you as my space buddy in this conversation.

Alex Karp

Well, I think it's notable that United Launch Alliance, ULA, isn't even part of this conversation. It's just the new-space companies, SpaceX and Blue Origin. I think competition is profoundly good. I don't think we want to wind up in a singleton where a single company controls the Dyson swarm, which I think is still more or less the endgame here.

I think a lunar station at the South Pole with Artemis 4 is just a waypoint to launching lots of AI data centers from the Moon, and I think SpaceX, based on its recent public communications, would agree with that. So I view this as a very positive thing, and I would like to see ideally more than 2 Western companies that are capable of rapid reuse of boosters. Hopefully we'll see at least 1 or 2 more publicly traded companies, or soon-to-be publicly traded companies, in the space do it. We're certainly going to see quite a bit from China. I think this is very positive.

I'm, for one, just on Artemis 4 specifically, looking forward to Artemis 3. According to Administrator Jared Isaacman's new roadmap for Artemis 3, Artemis 3 was originally going to be the crewed landing on the Moon. They pushed it off to Artemis 4 so that Artemis 3 could focus on a demonstration of the docking of the lander and the orbital vehicle, and I think that's going to be super interesting.

I love the new cadence that NASA has announced under Jared Isaacman: rapid iteration, in some sense SpaceX-like rapid iteration, toward colonizing—or, let's say, disassembling—the solar system. This is the only way we get there, and this is an important step in that direction.

Jared has said he'll come on the pod. I think I want to wait until after Artemis 2 is done. I think it's time to start the invitation process to bring him on.

You know, the dark horse is Relativity Space. A friend of mine, Tim Ellis, had built that really to go up against the Falcon 9. It's interesting: when I met with Elon, when he was just announcing Starship, he said his plans were going to be to retire Falcon 9 as soon as Starship is operational.

Think about Falcon 9—the most successful launch vehicle by an order of magnitude—is going to be retired when Starship comes online. He always burns the ships, right? After he develops it, he retires the old system. He did that with Falcon 1, and now Falcon 9.

It's going to be interesting to see what Eric Schmidt, who bought Relativity Space, does with it. I hope it gets up and operational. It would be great to have another super-large booster like this coming into operation.

Alex Karp

For sure. I think it's sort of an interesting microeconomic question to ask: How many different transport-to-LEO services, or let's say heavy-launch providers, do we actually need in order to keep that market competitive? Is that a long-term source of profit, or is that, frankly, as with civilian air transport, a case where profit margins are driven to zero?

It's just transport. It's just a dumb pass-through, and all of the profit is in LEO, with data centers, orbital Hiltons, or something like that.

It's Boeing and Airbus.

Alex Karp

Except Europe isn't in this story.

Yeah, Europe is not. I mean, they kind of tried, but they went old-school all the way. And China still has not gotten to reusability yet. They're also trying and having some failures along the way.

All right, Alex, this is a fun story that you and I talk about on text all the time: the UFO and UAP disclosures that are coming. I'll play these 2 videos, and then let's discuss it. All right, the first one: President Trump on UFO declassification.

Donald Trump

I recently directed the Secretary of War to begin releasing government files relating to UFOs and unexplained aerial phenomena. The process is well underway, and we've found many very interesting documents, I must say. And the first releases will begin very, very soon.

All right. Related, another story coming out of the news media and the White House.

Reporter

There are 10 missing scientists with access to classified stuff, nuclear material, and aerospace. They've all gone missing or turned up dead in the last couple months. Based on what you've been briefed, what do you think is happening here? And do you think that this is connected or totally random?

Donald Trump

Well, I hope it's random, but we're going to know in the next week and a half.

I just left the meeting on that subject. Pretty serious stuff, but some of them were very important people.

Alex, over to you. What do you think of these? Are they connected? Tell us your thoughts.

Alex Karp

Elon likes to say words to the effect that the strangest future, or the most ironic future, ends up being the one that we find ourselves in.

Yes.

Alex Karp

I find myself, when hearing stories like this, wondering whether a UAP could land on the White House lawn, be fully televised, and then the general public turns up its nose and asks what's next on television. If the president does the proverbial Rose Garden address and makes some major announcement in connection with UAP disclosure and NHI, will people—or at least maybe half the population—say, “I don't believe it for 1 second,” because of the overly politicized climate of the moment, and ask, “What else is on?” I do wonder whether we're going to find ourselves in that world.

A few weeks ago, I was on Capitol Hill doing a closed-door briefing with Senate and House staff on issues relating to science and technology, and I will say that on Capitol Hill, this issue is taken extremely seriously. It's contrary to, say, 10 years ago, when it was well outside the Overton window of discourse. It's very much inside the Overton window of discourse now.

Alex

I chat with friends of mine who were in charge of the relevant hyperscalers and the government clouds, who are actually in the process of carrying out President Trump's historic UAP directive to declassify files on top-secret government networks that are connected with disclosure—his now-historic executive order regarding UAP disclosure. As far as I can tell, this is all in process, and I'm hearing stories from my friends at the hyperscalers about how UAP disclosure in connection with the president's executive order is scheduled to conclude no later than January of 2027. I think, yeah.

This has been discussed, like Project Blue Book. There have been a number of presidents in the past who have talked about disclosure, and stuff comes out that's just blanked and not actually earth-shaking. Are we finally going to see some earth-shaking information here?

I agree with your point, by the way, that this could be disclosed: UFOs on the White House lawn, alien body autopsies, and people who say, “Okay, what was the sports score last night?”

But here's the other point I want to ask. We saw this in The Unacknowledged. I loved that documentary. I urge people to go and watch it. It's worth your while. There are a lot of very high-ranking generals and admirals speaking about what they've heard and what they've seen.

But we also know that a lot of this was taken out of the government and put down into private companies to avoid FOIA. This is now apparently being managed by private contractors that are resistant to disclosure. Any thoughts around that?

Alex

Yeah, I will say—and I made similar comments when last we were discussing the documentary The Age of Disclosure—that if the central allegations in that documentary are accurate, and just to provide a one-sentence summary, the central allegation made by 35-plus senior current and former U.S. government officials in both the executive and legislative branches is that there has been a highly illegal, 80-plus-year campaign to capture, retrieve, and reverse-engineer crashed UAPs. That's the central allegation of that documentary.

If that central allegation turns out to be accurate, then I would argue—and I have argued in the past—that this is borderline a crime against humanity. If, literally as alleged in that documentary, there are civilizations, non-human intelligences essentially raining technology down on us, and there's some illegal group—an illegally operated, quasi-governmental, quasi-contractor-based group—that is recovering artifacts that could advance human technology and civilization by centuries or millennia, and it's just being suppressed or, even worse, weaponized for purely offensive or defensive military purposes without seeing the light of day in civilian applications, which could affect a variety of sectors, I think that that type of conspiracy, if it is indeed real, is not just criminal. I think it's a crime against humanity that, if accurate—I have to add the appropriate caveats—would have potentially set back human progress by maybe a century. I think it would be devastating.

That's one of several reasons why I think this is a very interesting space to watch. In my mind, we're getting superintelligence from AI one way or another. I think, in some sense, there's relatively little alpha left in extrapolating the superintelligence explosion. But when I think of the sort of crazy left turns that could hit civilization in the next few years, it's a very short list of crazy things that could happen that would derail the singularity or alter the course of the singularity. I think something like this—

This is one of them. This is near the top of the list. I think it bears close scrutiny.

Dave, do you remember the conversation with Elon?

Dave

I asked Elon outright. I said, “What do you think? Have you seen the documentary? What do you think of UAPs, UFOs?” His comment, honestly, was underwhelming. He said, “You know, we have such advances in camera technology. Why is all the imagery blurry?”

They're black and white, even. I was like, “Okay, Elon, even if you knew, you probably couldn't tell us.”

All right, I don't want to spend more time on this, but I do have a question here about the 10 missing scientists. Do you connect it all? We've heard a number of people in The Age of Disclosure saying, “If I disclose any of this to you, I would potentially lose my life.” People feel like they're under threat. We see these missing scientists and these suicides. Anything about that, in brief, Alex?

Alex

It's certainly concerning. There have even, in the past 24 to 48 hours, been statements from both the executive branch and Congress talking about the active investigation. There has been messaging from congressional leaders that this is now being treated as connected, as opposed to just an unfortunate coincidence.

We see that tweet down below by Representative Ogles that says, “I have seen evidence so classified that just knowing it exists makes you a target.”

Alex

It is very concerning. I want to make maybe one closing comment, if I may, on the intersection between artificial superintelligence, ASI, and—the statutory term for this is non-human intelligence, NHI—ASI and NHI.

We talk—royal we—I talk on this pod often about building the Dyson swarm, disassembling the Moon, and disassembling planets. I'd like to think on a slightly larger scale for just a minute. I think, as a civilization, we're on the verge of having enough technology, thanks to AI, to be able to send von Neumann probes—self-replicating probes—in all directions at relativistic speeds throughout our galaxy.

All other things being equal, if we wanted to, as a civilization, convert our entire galaxy over the course of tens of thousands of years—light-speed limitations, obviously—to paper clips, we could, in a few years, with advances in AI and space. If we wanted to paper-clip the entire Milky Way galaxy, and we're alone, we could do it.

If there were a time for any non-human intelligence elsewhere in our Milky Way galaxy to make a cameo appearance and stop us from at least having the optionality of paper-clipping the rest of our galaxy, I think they're going to have to make a cameo appearance sometime in the next few years, because this intelligence explosion is about to give us, as a civilization, galactic superpowers.

Yeah, and that, of course, also applies to the appearance of these UAPs and UFOs at the turn of our nuclear age, when we had the ability to destroy ourselves. We could speak about this forever. I'm going to move us along. I know you have a hard out at some point, Alex, so, in other news: China continues to dominate in solar.

Alex

Tough segue, Peter. How are you going to segue from that today?

Meanwhile—

Alex

Good luck.

Meanwhile, China makes some solar panels.

Alex

All right, Peter, this is where you earn your keep.

All right. Listen, China's dominance in solar is pretty extraordinary. China hit its 2030 goal 5 years early. Ninety percent of new power capacity in China is wind and solar. They have actually installed 1,500 gigawatts of solar. It's half of the total global energy production, compared to almost 300 gigawatts here in the United States.

I just point this out for everybody because solar is being underdeveloped in the United States, and I truly hope we start developing it further. We have nuclear, and we have fusion coming, but solar is here now, and I think we need to be deploying it more. Salim, do you want to make a comment on this?

Well, this is just a big strategic thing. As you do solar, you need less oil, and therefore what China is doing is slowly reducing its dependence on any kind of oil. So energy is also the new substrate, and China is playing for substrate advantage.

That's very—

—all going for that.

Yeah, good on them. Yeah, we need to get on this too because we've already decided the Dyson Swarm—the Dyson Sphere—is the future, and that's solar too. The panels are 6 times more efficient in space, but 6 times is not that much when you're building an entire Dyson Sphere. So we're going to need massive amounts of solar manufacturing capacity.

And, by the way, I've been watching carefully the advances in perovskite, and in the labs they're now hitting 35% solar capture, which is amazing.

And that's a crazy number.

That's a crazy number, and now the only question is: How do you make the perovskite last long enough to match the competitiveness of silicon panels? They're not far from it at all. And if you have perovskite, which is quite abundant and easy to make, you reduce all the rare-earth mineral dependence to 0, which is incredible.

That's incredible.

And perovskite's cheapness is one of its principal benefits.

All right, 2 stories in the robotics world—actually, 3 stories in the robotics world. The first is that Beijing ran its second humanoid robot half marathon. The Lightning Robot won the second Beijing Humanoid Robot Half Marathon in 50 minutes, 26 seconds. The human world record is 57 minutes, held by Jacob Kiplimo. Last year, this half marathon took 2 hours and 40 minutes. This year, it took 50 minutes.

So, the fact of the matter is, China is driving so hard on robots. It's a priority for the nation. They're using these kinds of competitions to highlight it. They have 150 humanoid robot companies. I'm going to turn it to you, Alex, on the second story, which is yours.

Alex

Yeah, this is sort of an interesting split-screen. So let's rewind to last year, when China ran its first humanoid robot half marathon. I watched that story. I covered it, and I was so upset that the U.S. and the West had nothing like that that I resolved to try to start a U.S. equivalent.

My first pitch was to Tom Grilk, who's the former head of the Boston Athletic Association, which ran the Boston Marathon, including at the time of the Boston bombing. I pitched the Boston Athletic Association, trying to persuade them that they should host a humanoid robot or robotic marathon the weekend of the Boston Marathon. And they looked at me like I had dropped a dead fish on their table. [laughter]

So Tom recruited his son David, and I recruited capital from my fund, O2 1T Capital, and we funded a startup in this area called Professional Robotics League. This past Sunday—we're recording this on a Tuesday—we held the country's, North America's, and the West's first humanoid and non-humanoid professional robotics race ever.

We did this in the Boston Seaport. You can see some video here from the event. We had a range of manufacturers, Chinese and Western robotics manufacturers. It was a 50-meter dash because we had to start somewhere, and it was a tremendous success.

I'm just sad that we couldn't do a live podcast from this event because there were so many Moonshots fans who showed up. There were people flying in from Texas, people flying in from Canada just to attend this event. This was a historic first for the country.

Yes, the robots were operating a little bit more slowly. Yes, we didn't have the support of the Chinese Communist Party forcing every robotic vendor to participate, but we had to start somewhere, and I think it was a wild, wild success.

The most touching moment for me was when we did an exhibition match the day before, on Saturday. There were people just walking by on the street in the Boston Seaport. A little boy came up to the railing that we had set up, a little barrier, and said to me at one point, “I think I could do a better job designing a better robot. I want to be an engineer when I grow up.”

I was ready to burst out in tears. That was the most touching event from the entire race. I would just say, “We're going to need to do quite a bit more like what pro wrestling is doing in the West in order to make sure that the U.S. retains primacy in robotics.” But we're going to do it.

Well, congratulations, Alex. The only elephant in the room here is that all 3 humanoid robots were Chinese. We need to get a U.S. manufacturer.

Alex

We had some nice Boston Dynamics dogs.

Our third robot story here, Salim, you kicked this to us. Why don't you go ahead and brief us?

Yeah, so this is a picking robot, and the combination of vision plus AI plus dexterity is now leading to all sorts of new stock-picking systems. I was at the MODEX show—30,000 people—and this is one of the videos I promised to show you. Radical changes.

By the way, not a humanoid in sight, because I think in the back office, you're going to see industrial robots—

Only one arm.

—and in the front office, humanoids. [laughter]

You got your non-humanoid robot, Salim. You should rejoice.

I should rejoice. But really interesting to see there, so I just wanted to show this very quickly.

All right. You know, I talk about the Crisis News Network, CNN. This crossed my X filters. America's trust in mass media is at a record low. So here we go: Only 28% of Americans now trust the mass media. It's the lowest in 50 years, as measured by Gallup. It's down from 68% in 1972, and it's down 40% in the last 50 years.

So, just as we're seeing AI come online, we're seeing a collapse in trust across all of these areas here. Salim, back to you on this one.

Well, this is a business model problem. When you privatize media, people, for lots of reasons, went through this. We've been telling the newspaper industry for 20 years that they were going to be in deep trouble with online. That happened.

I think the big challenge now is that legacy media has lost its monopoly on reality construction. We're going to have to think about how we re-deal with this. How do we deliver trust? As Jerry Michalski points out, “Scarcity equals abundance minus trust.” So you have to figure out how to bring trust back into the equation.

By the way, audiences like this and shows like this, where we go direct voice, are way more effective than institutional polish for delivering trust. And so we're going to have a lot more stuff like this out there in the world to restore any level of that.

I think that the XPRIZE around figuring out what is trustworthy and notable and what is true is also useful. I'm really, really excited watching xAI, where people can say, “Hey, Grok, is this real or not?” and get called out very quickly on that. And that's proving to be very powerful and very important over time.

I totally agree. I think that XPRIZE needs to come back. I know it's a really tough one to design, but this is going to become a crisis because that line's only going down from here with deepfakes. There's no rebound in sight.

And, like Salim said, there are lots and lots of truthful, credible people out there, but they can't get above the clutter. So if we can figure that one out, we might be able to reverse the trend.

Yeah.

And I think that if you're an entrepreneur out there thinking about how to build something that delivers trust in a way that people actually believe is true, that they actually trust, it's a great opportunity.

Other stories: We just saw Apple CEO Tim Cook step down, with the head of hardware manufacturing stepping into the role. Any thoughts on this, gents?

You know, I lost track of the fact that when Steve Jobs passed away, Apple was worth about $300 billion. And that was huge.

[gasps]

Now what is it now, $4 trillion? That's just crazy how big everything has gotten. Meanwhile, Apple has not introduced a new product in that entire time. It's still the iPhone. It's just crazy how little innovation has driven that much growth and profitability under Tim Cook.

And he did a great job of cutting costs and vertically integrating everything, building Apple silicon, its internal chips. The biggest innovation during that timeframe was the AirPods, if you can believe that.

That's crazy. Of course, the phone is going to go away as a modality. We've been talking about that for a while, where headsets come in, and pins and other technologies that are watching and listening all the time. And, of course, your AI becomes your interface to an intermediary over all the apps out there. Alex—

Alex

My best guess is that we'll look back on the Cook era as being analogous to the Ballmer era of Microsoft. We had Bill Gates as the original visionary technocrat founder-CEO. Then the antitrust trial pushed Bill, in a number of ways, to be chief software architect and focus on his foundation. Ballmer stepped in. Ballmer was more of an operator, less of a technological visionary, for a number of years.

And then we got to the Satya era, where we see Microsoft being resurgent again with new categories, willing to drop initiatives that weren't working and refocus initiatives where Microsoft could once more become a technical leader in the cloud and elsewhere.

My best guess is we view the Jobs-to-Tim Cook-to-John Ternus transition as somewhat analogous, where John came up not through operations like Tim Cook did, but through hardware engineering. And so I would expect a new Apple under John Ternus, starting later this year, to be more focused on breakthrough hardware devices and hardware-centered innovation.

Of course, the elephant in this particular room is AI and Apple's seeming inability, after multiple swings at bat, to position itself as the leading distribution platform or server for AI.

It has fallen prey to all the frontier models, and even Microsoft, with its quasi-failing Copilot initiative, has leaned more into AI than Apple has. So I would expect a new, device-oriented AI strategy from John Ternus. Maybe ideally—I was chatting with a new friend from Apple marketing a couple of weeks ago in San Francisco, asking them why, on earth, given the popularity of OpenClaw, Apple isn't leaning into an OpenClaw-oriented device strategy. You're selling Mac minis like they're going out of style. Why isn't Apple marketing all of its hardware as being personal OpenClaw? Silence.

You want to hear something amazing? I think the answer is buried in this amazing stat. There are 7 Magnificent Seven megatech companies. 4 of them still have their original founders as controlling shareholders and active participants, and they're all doing AI. 3 of them are on—

Alex

CEO number 2, 3, or 4.

And they're the 3 not doing AI.

Alex

Wow.

So Apple, Microsoft, Amazon.

Alex

Everything.

That tells you everything.

Alex

Dude.

I have 2 quick thoughts. One is what a set of shoes to try and fill for John coming in. And second, what an amazing opportunity. I mean, just imagine if we were CEO of Apple and you could go, “Okay, let's now really take this to the next level.” We would have such a great time.

But still, we all know the founder CEO has such control over the company. Not only because of their stock position, but because they're seen as the visionary. And if you want to come in and make a right turn on your company, you're fighting everybody. It's like, “Don't you know? Don't disrupt our revenue. You know, we have predictable revenue. The profits have been going up. You want to do what?” It's really very, very difficult. And, of course, Tim Cook will remain as the chairman of the company. Good luck to Apple. A lot of changes coming.

Alex

I like the fact that they've got a hardware-focused person in there. That's great.

This episode is brought to you by Blitzy, autonomous software development with infinite code context. Blitzy uses thousands of specialized AI agents that think for hours to understand enterprise-scale codebases with millions of lines of code. Engineers start every development sprint with the Blitzy platform, bringing in their development requirements. The Blitzy platform provides a plan, then generates and precompiles code for each task. Blitzy delivers 80% or more of the development work autonomously, while providing a guide for the final 20% of human development work required to complete the sprint. Enterprises are achieving a 5x engineering velocity increase when incorporating Blitzy as their pre-IDE development tool, pairing it with their coding co-pilot of choice to bring an AI-native SDLC into their org. Ready to 5x your engineering velocity? Visit blitzy.com to schedule a demo and start building with Blitzy today.

You know, last time we had a great conversation about the 5 forks that humanity is going to take. We got a lot of incredible feedback from our listeners.

Our discussion today among the Moonshots mates is going to be around the war in Iran, in particular the potential disruption coming from the Strait of Hormuz. Here we see an image of the strait. Let's jump into this conversation. I think it's incredibly important. You know, the strait is going to impact infrastructure build-outs to a large degree, driving aluminum shortages, global oil shortages, natural gas shortages, helium shortages, and a lot of concerns. We see a lot of change coming, and it's hitting already. Who wants to dive in first on this?

Dave

Well, I grew up in Iran. That's where I spent my early childhood. And I'll tell you, it's just incredible to me that, absent U.S. involvement, Iran would just continue down the nuclear-weapons path unchecked and eventually have a nuclear weapon and eventually launch it at Israel or wherever. And that's just going to happen.

Yeah.

Alex Finn

So, the U.S. goes with Israel, bombs a whole bunch of nuclear facilities, and then they shut down the Strait of Hormuz. Now the U.S. Navy has sunk the entire Iranian Navy and is trying to reopen the strait, but it's very, very, very hard to do because it's such a narrow pass. It's very easy to mine it. It's very easy to launch things from shore. And the drones are dirt cheap now, so the drones can just blow up one of these oil tankers. They're very slow, right? It's not hard to build a drone that can hit an oil tanker. And what do you know? Oil burns.

Oh my God.

Alex Finn

It's not an easy thing to fix. And so, it's a major problem, and that's about a quarter of the world's oil supply moves through that strait. So, yeah, it's a cluster, but what else were you going to do? We are approaching a massive material shortage. It'll show up mostly in Korea and Japan, which are huge importers through the strait. It's not going to affect the U.S. too much. We have plenty of oil supply in the Western Hemisphere. But it's a big problem for East Asia.

Let's talk about helium, right? A third of the world's supply of helium comes out of Qatar, out of the Ras Laffan facility there. And helium is absolutely critical in the fabrication of chips, and it's going to hit TSMC and SK Hynix. Thoughts on this, Alex?

Alex Finn

Yeah, I've seen scattered reports of shortages at certain memory-chip facilities in South Korea facing, reportedly, multi-week deadlines to resolve their helium shortages before they're unable to manufacture memory chips. I'm not sure where this ends. It's almost easier to predict what happens after it ends, which is to say, I think it's high time to start helium startups. I think having this global geopolitical dependency on one narrow, volatile geographic region is completely unacceptable.

A lot of helium certainly comes as a byproduct from fission. There have been various speculations over the years. I know you and I, Peter, have talked about nonconventional isotopes like helium-3 being minable from the Moon and elsewhere. I look at material shortages and I say, in the long term, each of these is a startup opportunity for us to come up with clever new ways to radically decrease the cost of generation and also re-domesticate production.

From what I've read, there has been meaningful movement in the direction of strategic helium reserves. I think that's probably an important component of the strategy. Truth be told, I don't know where the Iran war ends or where helium shortages in particular end, but I feel more confident in predicting what happens after it's over and how we make sure that we're never technologically in this position again.

Agreed. You know, abundance is the rule in the final result. You have short-term scarcity, but at the end of the day, entrepreneurs will step in, and there is helium globally. It just needs to be put into manufacturing.

Alex Finn

It's pretty abundant in our universe. We just need a better way of harvesting it.

Yeah. You know, the other thing that's going to hit real quickly here, worth noting, is natural-gas supply. Taiwan is completely dependent on natural gas, right? And the statistic I read is that Taiwan has 11 days of natural-gas reserves, and that a prolonged blockade is going to impact Taiwan's semiconductor industry. So that's an even bigger concern for me than the helium supply.

Alex Finn

Well, I'll tell you, Taiwan had a water—you know, rain—problem a couple of summers ago, a massive water shortage. The one thing they kept going was the chip fabs. Everybody had to stop showering, but the chip fabs did not slow down. So I suspect they'll sacrifice every creature comfort, but keep those fabs running no matter what. But who knows.

David Friedberg

A couple of comments here. For me, this Iran war is not really just an oil shock; it's a system shock, right? Because you've got helium issues, insurance spiking. There's a whole cascade of little bottlenecks that are coming along that are going to cause a huge challenge. The 2 that struck me—1 is aviation bottlenecks. Europe imports 30% to 40% of its jet fuel, and half of that comes from the Middle East. So that's going to be a big problem over the next few weeks.

And the second, which is really the big tragedy in my mind, is food inflation, because fertilizer not getting out there is going to be a huge problem for food production all around the world. Something like 30% of the global fertilizer supply passes through the Strait of Hormuz, right, or depends on the gas that goes through there. So this is a huge systems problem. I think it's going to accelerate deglobalization because people will not want to be dependent on these overall global supply chains going forward. So there's a silver lining, but it's a downside to streamlining.

I want to share some feedback from our community. We read the comments that you make every week, and there were 3 principal comments that you made last week, and I want to address them. The first pushback from everybody—we read this in the comments—is, quote, “Not everyone can be an entrepreneur.” And Salim, Dave, Alex, and I are constantly saying, “Hey, become an entrepreneur.” So let's address that comment.

I want to acknowledge the criticism directly, right? You're right. Not everyone wants to be an entrepreneur. Not everyone can be an entrepreneur. But the fact of the matter is, becoming an entrepreneur has become much easier by orders of magnitude. And I think the conversation here is that if you never thought you could be an entrepreneur, the opportunity to become an entrepreneur is greater than ever before.

So, let me share a couple of data points here on this. First off, the cost of starting a business has plummeted. It’s come down 99.7%. You can see on the left, back in 2005, traditional startup costs included legal costs, hardware costs, and all of these different costs. It’s come down tremendously.

The fact is, an entrepreneur, by definition, doesn’t have to be someone going out and raising venture capital. It’s not someone starting a billion-dollar startup. You could be an entrepreneur if you’re a nurse and you see an opportunity to start a home health care business, or you’re a barber and you want to get another chair going in your shop. It’s finding a problem and creating a solution to that problem. Salim, you want to jump in on this?

Yes. I agree that the old model of entrepreneurship was very, very difficult, and most people cannot become entrepreneurs. But everybody can become more agentic, everybody can become more AI-leveraged, and everybody can become more resilient.

The shift is not to entrepreneurship. It’s moving from employment dependency to being self-sovereign in capabilities. There are other paths: founder, freelancer, craftsperson, operator, local business owner, as you mentioned, Peter, creator, co-op member, and so on. It’s a move towards dignity. You don’t want venture-backed chaos; you want stability, autonomy, and meaning. There are lots of other roles beyond founders.

You don’t become an entrepreneur because I said so. Become antifragile in a way that fits your life. That’s the way to frame this.

In the past, if you wanted to start a business, you needed a lawyer, an accountant, a web developer, a marketing team, and months of capital for a runway. You can register a business in 6 days, and you can build a website in an afternoon using AI. You can write a business plan.

This is about finding a problem, and no matter where you are in life, your ability to identify problems and opportunities is there if you look. If you don’t have those answers, brainstorm with your friends. Here’s some more data.

David Friedberg

Or, if I may, Peter—your AIs. I think even brainstorming problems is not a limiting factor. AI is quite good at identifying problems.

This is one of the reasons I have to name-check a company I have a financial interest in, Henry Intelligent Machines, HIM, from friend of the pod Alex Finn. It’s premised on the thesis that not only is the cost of starting a business going towards zero, but everyone who wants to be an entrepreneur can and will become one, in the same sense that anyone who has opinions or taste as a consumer can become an entrepreneur thanks to AI.

AI operators are carrying out all of the difficult or non-obvious tasks, and you, aspirationally, as a one-person investor, magnate, or conglomerate owner, can oversee a fleet of AI operators being entrepreneurs so that you can just be a gentleman or gentlewoman investor or magnate. I do think we are going to find ourselves in that future.

Yeah. Take control of your future. Here’s the data: back in 2019, 23.7% were solopreneurs. That has climbed steadily to 36.3%.

We talked about this in the last pod. We’re going to start to see companies going from 100% down to 20% of their employee base, but at the same time, we’re going to see 5 times as many companies being started. And guess what? If you’re not the CEO, if you’re not the idea person, you can find an entrepreneurial company—a small company—and bring your capabilities, your passion, and your experience to the table there. Salim, you’re going to say?

I don’t even think there are 6 million people in America who say their full-time or primary job is social media influencer. I don’t even think they count in this. There are so many jobs like that emerging that didn’t exist in the world previously, and they’re going to pop up with AI. They kind of defy normal categorization, but they’re going to be rampant.

Yeah, for sure. I don’t know. I count those as entrepreneurs. Maybe listeners don’t think of that as an entrepreneur, but to me, that’s an entrepreneur.

And this chart—I want to dispel the age myth. There was a persistent myth that you need to be in your 20s to be an entrepreneur, and the data says otherwise. The mean age of founders of the top 0.1% fastest-growing companies is 45, not 25—45. Individuals in the age group 55 to 64 represent one of the fastest-growing segments of entrepreneurs in the United States.

If you’re 60 and you’re thinking it’s too late for me, guess again. Your experience is the product. AI is your amplifier. Your ability to, as Alex said, brainstorm an opportunity, brainstorm a challenge, and find something that’s right for you is there.

I talked about the gentleman I met in Morocco who used ChatGPT to say, “These are my abilities. This is what I can do. This is where I live. What kind of business can I start to provide food for my family?”

Yeah. Just to reiterate, entrepreneurship is not the point. Agency is the point. Give yourself agency.

Yeah, for sure. What do you need to be an entrepreneur? I just want to hit on a few things. You don’t need to be the CEO. You don’t need to be a fundraiser. You need to be curious about how you can be useful. You need to be willing to create value first.

That’s something a lot of people think: “Being an entrepreneur is putting something up and having money come in.” No, it’s about delivering value first. You see Alex and his projects, Salim and your projects, Dave—it’s delivering value to the people that you want to serve, who you want to be a hero to. It’s being resourceful. It’s all of your experience in life.

Can I tell a quick anecdote? One of my favorite entrepreneurs in the world is Craig Newmark from Craigslist. He starts Craigslist, grows it, and basically stays in customer service. His title is customer service. At its peak, the company was doing $100 million a month in revenue with 30 employees, and he retained the title of customer service person. If you call Craigslist for customer service, he will usually answer the phone.

Then he took all his money and gave it to Veterans Affairs and projects on the future of journalism. It’s just an incredibly magical way of framing it: you follow your passion, you do what you think you’re best at, and then let the rest take care of itself.

That last point you just made is so important. Have purposeful alignment with whatever you start or whatever small team you join. Don’t do it to make the money. Find something that gives you true meaning and purpose in life, and go focus on that.

I want to challenge everybody who said, “Entrepreneurship’s not for me. It’s not for everybody.” Okay, agreed. It’s not for everybody, but the percentage of individuals—the range of folks who can become entrepreneurs—has been skyrocketing. I encourage people to explore this for yourself, especially if you’re just out of college and you haven’t gotten a job, or if you’ve been laid off someplace.

Find a few of your best friends. Building something, as we’ve said, Dave, with 2 or 3 best friends is the best way to do it.

Yeah, I think it’s even more acute than that. If you move into a world of AGI and then ASI, to thrive, you either need to be a founder, a joiner, or an investor. All other career paths are going to disappear.

You can’t just join a big company and work your way up over 20 years. That’s not going to exist in the world post-AGI, post-ASI. So, what does that leave? It leaves founders, joiners, and investors.

You know, I love those 3 categories, Dave. That’s great.

Yeah. Just decide what you’re good at. There’s always a way to add value. Decide which is your natural bent, pursue it, and get in the game.

All right, the second piece of criticism was on data center water use. It said, “Dave is contradicting what AI says about the use of water in data centers.” Dave, you want to take this one on?

I love that you’re blaming AI for the contradiction here. Come on, man. Take accountability.

Water use by data centers is less than 0.3% of total water use. It’s way less than 1% of agriculture. It’s microscopic.

The point here is that the water that is used is used almost entirely by evaporative cooling. So I thought, why are we bothering with evaporative cooling at all?

It turns out that the data centers in Phoenix and other very hot locations with tons of sunshine love using evaporative cooling because it’s more efficient than regular mechanical cooling. But that’s not going to work because those are also the areas where there’s the least water. That’s why it creates this problem.

There’s no reason we should be using evaporative cooling at all. Just put more solar panels on the roof and use regular mechanical cooling, and then you don’t waste any water and everything is solved. This is not a real problem in the world. We can put it aside and move on. Fix it, for sure, but move on.

I imagine that the hyperscalers are going to be focusing on this because of all the criticism. We talked about power, and of course, they’ll pay differential rates.

They’ll provide their own power, but they’re going to have to address the water issue. Not a big problem. I checked out the energy costs of moving over to mechanical cooling. It’s not trivial, but it’s about a 20% increase in your overall electricity bill.

You just have to eat that with more solar panels. Put up more panels, because this only matters in very, very hot, very sunny locations in the first place, because that’s the only place evaporative cooling works anyway. Just put up some more solar panels, eat the extra cost, and move on. Those areas are also the same places that—quote—“Whiskey is for drinking, water is for fighting.” Those are the same locations where these data centers are going in, like Abilene, Texas, and Phoenix. So you’ve got to put up more solar.

The third criticism—and it really is something we need to address—is the ivory-tower, out-of-touch criticism. It’s a massive ivory-tower mentality: your silver spoon is showing. Of course the experts are more optimistic about AI; they’re all rich, laughing at middle- and lower-income people’s concerns about anti-AI sentiment. My electricity bill climbed 14%. Salim, you want to jump in here?

Yeah, look, everybody on this podcast and everybody on other podcasts, take it all in. We all started from nothing. My dad told me, “Here, take a first year of university. I can pay for it; the rest is up to you.” I was selling beer during beer strikes in Canada, trying to make money. So we’ve all gutted it out at various points.

I arrived in Silicon Valley when I was 43, and I had $2,000 to my name. I’d wiped myself out doing a startup. So you have to build from scratch. I think we’ve all been there in different ways. I’ve been very, very lucky, and I think the skill that you need to work on the most is your luck to be successful in this stuff.

The huge opportunity here is that you can do a huge amount today with very, very little capital. I think that is the magical story around this. Why we’re so excited about entrepreneurship is the fact that you don’t need to raise billions of dollars or millions of dollars. You can literally go get an account for $20 and start changing things. That’s an incredible opportunity, and that’s where we get so excited around this.

Yeah, I mean, we are heading toward abundance. We talk about this. It is not coming next quarter; it’s coming over the next 5 to 10 years. We’ve seen this, right? In the book I just published, “We Are All As Gods,” there are 100 pages of charts in the back of the book that show the story.

For most of human existence, it was the king and the queen, the pharaoh, or the emperor on the hilltop, and everyone else living in abject squalor. We’ve had the most extraordinary movement of people into the middle class and upper-middle class. Today, individuals alive have access to more than the kings and queens had a century ago. The poorest people on the planet have a smartphone, access to lighting, and a roof over their heads. It’s become extraordinary in that regard, and it’s continuing.

Can I give a tangible example here? Let me give you a very specific example that you can just latch onto. There’s a fishing village in the middle of nowhere in Vietnam where, once a month, a big ship used to deliver diesel fuel for the fishing boats. At some point, the ship stopped coming, and these people had no fuel to power their fishing boats.

One of them literally buys a solar panel over the internet. They look up on the internet how to connect it to the propeller, and they invent a solar-powered fishing boat. This is cutting-edge innovation using disruptive technology, happening at the edge of the world. We on this podcast, and other people like us, get incredibly excited when we see that, because here are people at the edge of the world using the most advanced technology, along with their mindset and their agency, to totally change their own world.

That’s what’s possible today that wasn’t possible a few years ago. Then AI amplifies that a thousand times. That is now possible everywhere in the world. I used to call this PDI: permissionless disruptive innovation.

In the past, to do something disruptive, you had to get the Medici family to back you. You had to get the government to approve you. You had to get a VC or investor to fund you. Now you can basically get together with the right mindset and do something crazy at near-zero cost. How can that not be exciting to people?

Our mission is to inspire you, right? To give you the inspiration to dream bigger. At the end of the day, life can be difficult, and I’ve suffered a lot of massive setbacks, back to zero.

David Blumberg

Peter, you spent 11 years trying to get approval for ZERO-G from the FDA. You want to talk about hacking it out, right? It’s the persistence that has made most of us successful. If you ask most successful people what made them successful, it was their persistence. They just never stopped.

Yeah. And so, anyway, I could go on forever.

You get persistence when you’ve found your purpose, your MTP. Every time I’ve tried to do anything just to make money, it’s failed miserably, because doing anything big and bold in the world is hard work. You need to find that purpose that you’ll dream about in the morning and keep going through the night.

If anything we say on this podcast comes across inappropriately as having an ivory-tower mentality, apologies. That’s not our purpose here. Our purpose is here to—

Calling us out. Yeah, I mean, trust in us comes from our ability to respond to your critique and your criticism. We do read your comments, so please keep them coming.

Our desire is to serve you in a number of different ways: keep you informed of what’s going on in the world, give you an optimistic point of view, and inspire you to go bigger and bolder in anything and everything you do. If you find yourself in a place where you’re not enjoying life, where you’re not able to thrive, you’ve got to have the guts and the fortitude to find someplace better, around people—you heard this from Dave in the last podcast—you’re the average of the 5 people you spend the most time with.

Find incredibly good people: entrepreneurial people, people who want to make the world a better place, and go hang out with them. The reality is you’re facing challenges and difficulties that probably Salim, Dave, Alex, and I never faced when we were entrepreneurs getting going, and we hear you there, right? Our job is going to be to put ourselves in your shoes as we’re covering these stories, as we’re talking about what’s possible, to really provide a perspective that hopefully is useful to you. Dave, you want to add a final point here?

David Blumberg

Well, I really feel for bullet 2 and bullet 3 on this, in the sense that when I was trying to get my first startup up and running, I was below the poverty line for 8 or 9 years. The muffler fell out the bottom of my car, and I had to duct-tape it back. I was thinking, “How am I actually going to survive this?”

But I could always fall back on an entry-level job at JPMorgan or something like that, so I wasn’t really panicking. Now it’s different. It’s kind of do-or-die; it really is a different time. No one’s hiring, and we’re saying, “Be an entrepreneur,” which is the right path forward, but the safety net is much scarier now than it was back then. No one’s hiring, and I feel for that.

Bullet 2 says, “Of course the experts are more optimistic; they’re all rich,” but it’s because we know we’re moving into this era of unprecedented abundance, and asset values are going to go way, way up. So the point is right.

Then the last point there really tears my heart out: laughing at middle- and lower-income people’s concerns only adds fuel to anti-AI sentiment. AI is going to actually create more wealth than anything in history, and anti-AI sentiment is the worst thing we could ever propagate. I don’t remember us laughing at lower- and middle-income problems. In fact, I feel it acutely every day with the 1,100 people that I have to think about.

If we ever laughed at lower- and middle-income problems, that was just an error on our part. I don’t recall doing that, but that should absolutely never happen, because it is going to be a massively turbulent time.

Now, I’m much more confident today than I was 2 weeks ago about Dario and Sam, especially after Sam’s house got firebombed. If they just turn 10% of their AI toward this problem, I know by 2030 we’ll be beyond it.

I’m only worried about the window between now and 2030, when people who’ve worked on a 15-, 20-, or 25-year career path as a paralegal or as a white-collar professional are about to have that entire career replaced by AI. If you’re way down this path, that’s devastating. But if you’re nimble, you can actually get ahead of it. I know that’s very vague advice, but if you’re nimble, you can get on this wave and not get crushed by it.

I think if the big AI companies put just 10% of their effort into trying to deliberately give people a roadmap, right now, starting with the software engineers, that would help. You just rolled out capabilities that will kill Canva and Figma. Put a roadmap in front of those software engineers that’s designed and deliberate, through some kind of independent software vendor partnership program, so they know where to go next.

That will solve a huge part of that swath. Then move on to legal, move on to the other professional services, and be deliberate about smoothing the pathway between 2027 and 2030 to try to actually create forward trajectories for everybody.

I think it'll actually work. I feel like there's so much abundance coming that that overhead will be a rounding error, and we can actually make this smooth for everybody. You just need to design the three of us, by mistake.

The three of us are on the XPRIZE board. We're focused right now on launching the Abundance XPRIZE, which—remind people, if you haven't heard me speak about it before—the prize would be for a company or a team that's able to deliver housing, food, water, electricity, and bandwidth to a family of 4 for $250 a month, right?

So, if you've got those basics, it's universal basic services. You have those basics covered, and you're then able to think about, How do I upskill myself? How do I plan an entrepreneurial journey? But if you're worried about, Can I turn the lights on? Do I have a roof over my head? Can I feed my kids? then nothing else matters.

And so, very true.

All right, gentlemen. It's a tough place to close it out, but it's an important one. The world is changing fast.

A question? Yes, please.

May 7th, 7:00 p.m. Eastern, I'm doing one of my Meaning of Life sessions online.

Okay. Where do people go to find out about that?

Go to openexo.com and you'll see it somewhere.

Amazing. And on May the 4th, we're going to be recording a Moonshots podcast at MIT. We're doing our We Are As Gods session with Ray Kurzweil on May 4th. Our outro music today is coming from Tomas. It's Solve Everything.

Under moonshot guidance, I gain new minds. Wake up in the physical light. Fear in the streets, but—

All right. I love the chisel jaw these outro music give me.

Alex

It was borderline fake news actually. We look way too good in that.

All right, gentlemen.

Alex

All right, folks. See you next time.

See you guys soon.

If you made it to the end of this episode, which you obviously did, I consider you a Moonshots mate. Every week my Moonshots mates and I spend a lot of energy and time to really deliver you the news that matters. If you're a subscriber, thank you. If you're not a subscriber yet, please consider subscribing so you get the news as it comes out. I also want to invite you to join me on my weekly newsletter called MetaTrends. I have a research team. We spend the entire week looking at the MetaTrends that are impacting your family, your company, your industry, your nation. And I put this into a 2-minute read every week. If you'd like to get access to the MetaTrends newsletter every week, go to dmandis.com/metatrends. That's dmandis.com/metatrends. Thank you again for joining us today. It's a blast for us to put this together every week.