NVIDIA 962亿美元季度营收、中国20万个假账号与OpenAI新芯片｜EP #284

Welcome to Moonshots everyone. Your number one podcast on all things AI and exponential. Your front row seat to the incredible accelerating singularity. I hope you can feel it like we all do. You know, guys, we just recorded two days ago and it feels like a lifetime ago. I can't believe it. Isn't it? I mean, every story we're going through today is new information in the last two days. It's crazy. It is like the podcasting singularity. Peter, it really is. Well, guys, I'm here with my incredible, my original four moonshot magnificent mates, Alex Weer Gross, our in-house ASI. Dave Blondon, our empressario of AI investing. And Salim is our global globe trotter. Our Where's Waldo? You're back from Brazil. I'm Peter Diamandis, your host and optimism provocator. So much happening in the last two days. We've reviewed 200 plus stories. I get a download from my AI and from Alex, my other AI, every day. We've narrowed it down to about 15 today that are fun. I would categorize these stories as provocative, massively impactful, and bizarre. Our mission here is to keep you optimistic and ready for the supersonic tsunami that is heading our way. If you're new to Moonshots, welcome. If you're a regular, a fellow moonshot, welcome back. We're going to be having a Moonshots AMA over Zoom. We're inviting all of you, our listeners, to join us. We're going to be doing it twice in September, once in the morning so that people in Europe and Asia can easily join, once in the evening so people on the East Coast and in the US can join. We want to meet you. You're going to have a shot at grilling Alex, Dave, Salim, and myself. And also, our new X handle is moonshots_pod. So, please follow us on X as well. We're putting up our clips and the full podcast. So, shall we get into it, gentlemen? Absolutely. All right, a lot to cover today. I'm going to open with a few AI stories, as we normally do.

Nvidia is on fire. Nvidia just booked $96.2 billion in revenue in a single quarter. I mean, this is more than the GDP of the majority of the countries on the planet. That's up 106% year over year, with guidance for the current quarter another 10% higher, at $108 billion. For context, that is more than $1 billion per day. Crazy.

Jensen's guidance for 2028 is 70% growth, against Wall Street's consensus of 44%. Jensen called the frontier labs “the first generation of startups that needed tens of billions of dollars of compute to get to their product.” Crazy.

In related news, Elon Musk is making some bold predictions about SpaceX revenues. We've heard him in the past say—and in the last earnings call, he said—that he's going to reach $1 trillion in revenue by 2030, potentially 2029. Now, yesterday, Elon tweeted that he may well reach $3.5 trillion in revenue in 2033. By reference, the company today with the highest revenue is Amazon, at $828 billion, just to give some context for Elon's $3.5 trillion.

Dave, let me go to you first. Is Nvidia's revenue growth sustainable, or is this just the peak here? What do you think about that?

Actually, it's a really interesting two forks in the road here. Nvidia has a lot of vulnerabilities, and it'll come up later in the podcast, but at the same time, they're going to be sold out for as far as I can see, no matter what. I think that everyone's got to use Nvidia for training neural nets. It's 90% of the revenue, 95% of the profits.

These revenues are not just revenues. These are 80% to 85% gross-margin revenues. The profitability of this company has never been seen before on the face of the earth.

Everybody's vulnerable too, you know, because all the major companies—all the Magnificent Seven—are starting to overlap so much in what they do.

The biggest vulnerability at Nvidia by far is that TSMC is still their 1 and only manufacturer of everything they sell.

I'm amazed they haven't vertically integrated, or at least tried to, like Elon has.

Well, it's tricky, because if they were to try and build their own fabs, they'd have to do it very quietly and sneakily, because they cannot piss off TSMC. About a third of TSMC's manufacturing goes to Nvidia, and another third, roughly, goes to Apple. The whole rest of the world gets the other third. But you can't rock that boat casually. You have to do it very sneakily.

As Alex has said on many podcasts, new physics and new technology are going to come through AI imminently, which could disrupt the entire manufacturing supply chain—how chips are made and what they're made of. There's just so much change coming.

Nvidia is making all kinds of moves to shore up its position while it's on top of the world, and you almost can't spend the money fast enough at the rate that it's coming in. I was talking to my kids about it. With so many of our portfolio companies, I meet with them weekly, and they'll say, “Oh, I had a meeting with a manufacturer in Vermont. I had a meeting with J.P. Morgan.”

Why are you meeting with anyone other than Jensen, one of the other Magnificent Seven companies, or one of their satellites? It's insane. The amount of money pouring through just that funnel dwarfs the entire rest of the economy, and your success and failure are intimately tied to just that one thing. Everything else is just a distraction.

Jensen is pouring out money in checks of $5 billion here, $12 billion there, $20 billion there. If you're not in that flow of conversations, you're not in the most relevant conversation in the world—in history.

One second. Let me call him. I need some more capital.

Yeah, exactly. Exactly. If you can't get Jensen, just get one of his executive staff. There's only, what, 20 or 25 people on that list. Talk to them.

Alex, what do you make of this?

I'd be the last person to suggest that this is one big wash-trade, circular-financing scheme, but I do wonder about the extent to which the financial markets have fully priced in Nvidia's financing of its customers.

I guess that, to the extent there is an elephant in this particular room, that would be it. I think it would sure be swell, as a participant in the financial markets, to have a clearer distinction between how much of this demand is Nvidia-financed demand and how much is not.

If it turns out that a disproportionate amount of this demand is directly or indirectly backstopped, financed, or credited by Nvidia through one or more intermediaries, then I think that starts to look a little bit too bubbly, which I’m not thrilled with at all. I’d like to see no hiccups in this singularity. I would not like to discover 2 to 3 years from now that some, or a large portion, of all of this demand at the infrastructure layer was being artificially propped up through financial engineering. I think that would be a highly suboptimal outcome.

I still think that even if it turned out that some quantum of the exploding demand we’re seeing is being artificially inflated through financial engineering, my forecast—this is not investment advice—is, at most, a mini-winter, as it were, because I think compute is fundamentally substituting for real estate, human labor, and other raw inputs as the fundamental substrate for human civilization. But when I see nosebleed growth numbers like this, and I see a bunch of other headlines about Nvidia financing a bunch of data center deployments, I have, in full transparency, a little tickle in the back of my mind about how much of this is financial engineering.

Nice. Hey, Salim, are you seeing the same videos on X that I am, from Ray Dalio and other financial leaders saying, “I’m out of the markets”? The P/E ratios are off the charts, and there’s this level of fearmongering in the financial markets. At the same time, you’re seeing this massive growth, and there’s the question of where you go. Do you stay fully in? Do you jump out? How do you think about it? This is a classic dilemma, right? You’ve got to hedge your bets in this model. I think Alex makes a really great point. The circular revenue is very concerning. I love the way you say it, Alex. It would be swell if it was all tickety-boo.

I don’t want this to be financial. I don’t want this to be a bubble. I really don’t need bubbles.

I totally understand it. It would be really bad if the whole thing collapsed and we were in a dead period, and that would slow civilization down dramatically. You’re better off having steady, regular growth where the market has time to self-correct in the new structures, and we don’t have that right now. Things are going vertical.

But there are a couple of things happening that I found interesting. One is that what’s clear is Nvidia is trying to move out of just being a chip company to trying to become literally an operating system for the whole of intelligence. I thought the Hugging Face acquisition was super smart.

And so that’s still just a rumor, right?

Do we know? Yes, it’s alleged, shall we say, right? But this reminds me of how clever Microsoft was to buy GitHub, right? And can I tell a quick story about this?

Of course.

Microsoft tried to acquire GitHub, which was less than 10 years old, for $7.5 billion. The CFO and the corporate development people started freaking out, saying, “The company has barely any assets. It has barely any workforce. It has barely any intellectual property. What the hell are we buying here? What do I put on the balance sheet?” They were literally trying to kill the transaction until Satya said, “Listen, we’re buying 30 million developers’ loyalty, because that’s how many people used GitHub at the time.” So we overrode that, and they got it done.

As we move into this AI world, GitHub turned out to be one of the smartest things they could ever have done because it gave them access to all of that open-source thinking and access to all of that stuff. The same thing is happening if this acquisition happens: you’re buying the community. That’s a very classic ExO play to do this.

To your broader point, Peter, about whether we’re in a bubble, I think there are some macro things happening that are much more concerning. For example, we’re printing money like it’s going out of style, and I think that has a bigger chance of collapsing the markets than anything else. What’s happening with the bond markets and Japan and stuff is absolutely insane in terms of the circular chaos in the fiat currency world, but that’s a different discussion.

Yeah. So let me jump to the SpaceX story of Elon projecting 3.5 trillion by 2033. Dave and Alex, I love your points on this.

Well, I mean, for starters, you said it up front: no company in the history of the world has ever gotten to a trillion.

Yeah.

That’s in revenue, and that includes massive companies like Amazon and Walmart—3.5 trillion. But Elon’s always been right before. He’s often been wrong by a couple of years on the timeline, but not because of anything other than barriers that didn’t need to exist: engineering mistakes, regulatory barriers, or whatever. He always says, “Look, this is the native size of the opportunity, and this is the best-case timeline,” and the size always turns out to be right, while the timeline is off by a couple of years because of unforeseen delays. So I have no reason to believe he’s wrong about the 3.5 trillion, but the scale of that relative to anything in history is just mind-boggling.

Yeah, I think Elon tends to like to move in 3-dimensional ways. Again, without this being misconstrued as any sort of investment advice, I think there is a path to that, but I think it’s a highly nonlinear, multidimensional path. The most obvious path to get there is for SpaceX and Tesla to merge sometime in the next year or two. That gives the new SpaceX, not the old SpaceX, Optimus revenue. I think Optimus alone—not financial advice—could be a multitrillion-dollar-per-year business.

And then there’s the cloud: SpaceX offering hyperscaler services, which, last I checked, were the majority of its revenue growth in the past few months. It could continue to grow its hyperscaler business both terrestrially, which is where it is now, and soon with the Dyson swarm. I think the Dyson swarm alone could probably generate trillions of dollars in revenue. In the face of Optimus and the Dyson swarm, I think Starlink probably ends up being a rounding error. I think cars are a rounding error.

I think applications—which, interestingly, in the SpaceX S-1, SpaceX went on and on and on about digital Optimuses, applications, and the service economy—maybe seem unlikely. I think the core strength of Elon’s ecosystem is less the application layer and more physical infrastructure. So if they just totally dominate robots and orbital data center infrastructure, I think that’s probably a few trillion dollars.

I mean, SpaceX is my largest financial bet right now, full disclosure. I think about it as a way of—if you’re investing in OpenAI or Anthropic, it’s a single-layer bet. You’re betting on the models. But SpaceX is everything: it’s ultimately chips, data centers, models, communications, infrastructure, and launch, and you get a chance to bet on all of them.

I think the biggest risk is just Elon being around long enough to implement his vision. I know he’s probably walking around with an army of guards. When I see him on occasion, it doesn’t look that way, so I hope he’s got the proper protection.

For our listeners here, Dave, we’re not a financial show, but we’re throwing around massive numbers here. How do you think about playing this market—staying in it, doubling down, or hedging against it?

You know, it’s actually, I think, in some senses getting easier to understand because everything that’s going on, you can start to understand it from just 2 things: AI training and AI inference. What Elon is putting into space is purely inference. All the satellites—all the revenue—is going to be from Starlink, which is about connecting AI-generated video and other AI-generated content to everybody on the planet. But the AI data centers are going to dwarf everything that Elon’s launched so far. It’s just inference-time data centers in space.

But that’s inference only. The training time is a very different beast. That’s where Nvidia is just going to kill it, because for training time, you need all the GPUs to be coherent in one location, like Tennessee. You need 100,000 and ultimately 1 million, and then probably several million Nvidia GPUs to be in this kind of very tight, all-to-all cluster that can train the next generation of models.

At inference time, it can be spread all over the world, all over space. Those are 2 very different emerging markets. Even the chips are going to be very different for those 2 markets. Then everything else kind of falls out from that. You’re like, “Well, what’s sustainable and investable?” Robotics is sustainable and investable, but only if you have great AI designing robots. The idea that you can build a better robotics company without great inference-time models that are specific to manufacturing and design—you’ll never sustain it.

Then you start to see all these great investments in vertical markets like drug discovery and robotics, but they’re all also built around proprietary inference-time models that are very good at that use case. Everything else looks like it’s going to be SaaS software and everything. It’s going to be undisrupted for a while, but that’s a choice that Elon and Dario have made not to disrupt all those verticals and kind of leave them alone.

And so, like Alex was saying earlier, it looks a lot like the AI economy is a circular economy, where Jensen is putting money into data centers. The data centers are buying the chips that provide inference to the vertical-use-case companies. Isn’t that a closed-loop, circular economy? Couldn’t that collapse?

But to me, it looks more like this: Imagine there’s a civilization on Alpha Centauri, and they’ve discovered superintelligence. What are they going to do? They’re going to build this massive, valuable economy within themselves. Are they going to trade with Earth? They’re not going to give a rat’s ass about Earth. It’s too far away. It’s too slow.

And that’s exactly what’s happening in the new AI economy. Every economy is circular in nature. Dollars just move in a circle; euros move in a circle. So now the circle is tightening up into this new economy that only cares about itself and cares very little about the legacy economy.

I talked to my brother over at Vestmark, who works with Fidelity, Ameriprise, and UBS, and his initial reaction a year ago was, “AI will not really have hit until I see it show up at Fidelity.” Now he’s like, “You know what? It doesn’t care. It literally doesn’t care when it’s discovering new medicines and new physics.”

We already mentioned that Ozempic and Mounjaro are bigger than all of AI inference combined—just 1 drug discovery. So if it comes up with 5 or 10 more drug discoveries, which it inevitably will very soon, the amount of revenue flowing back into the AI economy from just that will dwarf the legacy world.

Well, there are these meta-signals—these high-level, orthogonal signals. If I ask an entrepreneur, “Do you know how to train a foundation model? Do you know how to fine-tune one? Do you listen to the Moonshots podcast?” every single one of them goes up 10×, 20×, or 100× in value in some cases—up to 1,000× returns on those deals. No losers yet. Zero. 0.00 failures.

So I don’t need to be a rocket scientist to say, “And then somebody says, ‘Well, I’m more of an application-layer guy and a vibe-coder guy.’” There are lots of losers in that bucket. So do you really understand what’s going on? Do you listen to Alex and what he’s about to say? If yes, you will succeed.

Close us out on this one, Alex.

I think there’s probably an important distinction, just on the financial-engineering and circular-economy side. I think there are perhaps 2 different forms of circular economy one can have.

In 1 form of circular economy, you have Company A selling goods to Company B, and Company B buying or selling goods back to Company A. Call it an income-statement-based circular economy. In balance, you don’t want circular or wash trades that are overly engineered to inflate revenue either, but maybe call that, from a bubble perspective, a slightly more healthful form of circular economy.

Then maybe there’s a slightly less healthful form of circularity in this context, which would be Company A using not its income statement but its balance sheet—in the form of private-credit loans, et cetera—to finance Company B’s purchase of Company A’s goods.

1. China’s AI Video Boom and the World Model Race

Right now, as a shareholder of the entire market via low-cost, broad-index funds, I don’t feel like I have tremendous visibility into which of those 2 forms of circularity we’re seeing at the very heart of the innermost loop. I would love more clarity.

This episode is sponsored by Google for Startups. Think about this for a second: you now have access to the same generative AI models that cost hundreds of millions of dollars to train. Google's startup technical guide for generative media gives you a complete blueprint for deploying Google DeepMind's models in production: images, video, audio, all of it. Real architecture, real results. Find the link in the show notes below. All right, I’m going to move us on to our next 2 stories coming out of China in the world of video production.

The first story is the rollout of Wan 3.0 by Alibaba. Wan 3.0 is capable of generating 30-second, single-pass videos from documents, spreadsheets, slides, and web pages. I wonder what kind of video you get out of a spreadsheet.

Its API runs between 5 cents and 20 cents per second of generated video. That’s something like $20,000 to $60,000 for a 90-minute film, right? These prices are beginning to collapse. This is 1 day after announcing their largest-ever private follow-on offering by a Hong Kong-listed company, at $10.2 billion. Clearly, they’re engineering their stock price.

That’s Wan 3.0, the first story. The second story comes from Runway’s co-founder, who reports that video generation now represents roughly 70% of all AI token consumption in China, driven by short-form content and robotics. It’s growing faster than Claude Code grew in the United States.

The framing here is that America is LLM-pilled and China is world-model-pilled. The implications are that the 2 superpowers fundamentally have different priorities. America is building large language models; China is building world models. Language models predict the next token. World models predict the next state of reality.

These world models are going to be particularly important not only for video generation, but also for robotics, autonomous driving, physics simulations, and video games. Alex, I’ll go to you first on this. Your thoughts on Wan 3.0 and the 70% video-token consumption. Crazy.

I agree with the sentiment. I think it’s striking, and I have a unified field theory that would explain it. My unified field theory is that American AI labs are revenue-maxing and Chinese AI labs are not.

When OpenAI tried to be in the video-generation business, they ended up abandoning it because Anthropic ran away with their lunch by revenue-per-token-maxing with code generation, which is, call it, a more LLM-shaped use case for their compute.

The Chinese labs really—I mean, yes, they’re generating revenue, yes, they’re IPOing, and yes, they’re growing quickly—but they’re not revenue-maxing. They’re giving away model weights for free. You don’t revenue-max by giving your model weights away for free.

So if you’re not in the business of maximizing revenue per token, you might as well be spending tokens on video versus more economically productive applications. That’s my unified field theory for why you see so much world-model inference in China and relatively little, by comparison, from American labs. American labs are too busy doing more economically productive applications.

Alex, which one gets us to AGI faster?

It’s a trick question, because I think we hit AGI no later than the summer of 2020 with LLMs. With LLMs, we didn’t have good video models.

Which progresses us from AGI toward ASI faster: world models or text-based models? I’m moved to ask that question.

Okay, so I’ll ask myself that question. The answer is that, in the end, omnimodal models are the best.

I see, like, Fable 5 is so strong, and yet its visual capabilities and visual-reasoning capabilities are so weak. In the medium term—which is 6 to 12 months—I think that’s an impairment that hopefully Anthropic is busy repairing with M&A activities.

I think they’re on a collision trajectory, and whether we get diffusion transformers at the end of the day or some sort of hybrid model that combines the best of vision, video, and audio with autoregressive, text-based generation, I think they have to combine one way or another.

Mhm. Dave, any thoughts?

Yeah, big time. You’re going to have to shut me up on this topic, so just fair warning. Everything Alex said is dead right, of course.

But I think there’s a story within the story there. The Chinese don’t have the ability to sell Kimi or Qwen into the enterprise use case because nobody knows if they can trust it. Video is perfectly profitable. It’s a fantastic way to take your maybe-not-trustworthy model, get it to market, and generate huge amounts of revenue and huge value.

But I don’t think the question of world models versus language models determines which is going to recursively self-improve first. I think it’s purely about smarter engineers with plenty of cash and lots of training chips working on the problem.

That’s the race. The Chinese are brilliant by going after video because they can just launch it on the world. It’s a global product. No one’s worried about security in that use case, and it just generates tons of revenue.

But remember, OpenAI was ahead of them and decided not to invest in it anymore because the enterprise use case is so much more valuable. So America’s taking kind of the highbrow road: let’s solve real business problems and cure real diseases. It’s more profitable. It maximizes revenue per token, so it’s better in that sense, anyway.

But at the end of the day, it’s all about that revenue pouring back into more and more training chips that turn into better and bigger—10 trillion and then 20 trillion parameter models—that then get distilled. And so that’s what’s kicking off RSI.

2. Pancreatic Cancer Survival Nearly Doubles

I’ll tell you what else: Demis Hassabis was the first guy to really think this through, but he couldn’t act on it at Google. It’s too bureaucratic. Dario Amodei took it, and he’s the first guy who said, “We don’t care about consumer movies, videos, whatever. We care about a model that can improve itself through better code and better training ideas. That’s all we’re going to focus on. If we can productize that at the same time, great. But at the end of the day, that’s going to recursively self-improve first, and that’s what’s going to get ahead of everybody else.” And the Chinese believe the same exact thing.

I miss Sora, by the way. It was so much fun with my kids.

It was fun.

I think Sora will be back, Peter. I don't think it's gone for long.

What do you think about the idea of 70% of the tokens being spent on video out of China? Is there an implication there for us?

Well, I think there are a couple of things going on here. I think Dave nailed it with what he said in terms of: are you going for RSI, or are you going for world models? What’s happening right now is you’ve got the U.S. optimizing around software engineers and knowledge work, and China is optimizing and heading toward manufacturing, commerce, media, and, to some extent, machines like robots.

Eventually, you’re going to converge. The winning systems are going to have language and vision and all of this memory strung together. The future, if I think about the future of AI, AGI, and ASI, is where you hit the physical world—that’s where you’re going to make the biggest difference, right? Humanoid robots or other forms of robots.

I think the intelligence that can model and then manipulate the world and act inside the world is going to win. In that case, I think where China’s heading may have a slightly better advantage. But if we get Anthropic or one of these guys to RSI, then I think Dave is correct. That kind of trumps everything, and that’ll be the inner loop that just swallows a whole bunch of other stuff. So you could go either way here.

You know, we had a conversation last week with Emad. By the way, to all our listeners, Emad sent his regrets. He was supposed to be here, but he had last-minute travel.

He made the point, I think very importantly, that AI is one of the most persuasive agents out there. The ability to understand a person and persuade a person, combined with video generation and AI, is a very powerful force. It’s a force for spreading culture.

So the question becomes: with China about to become, if they’re not already, the dominant world producer of video content—in any language, on any person, on any topic—what are the implications of that?

Oh my God, they’re huge. They’re crazy. My youngest son, Jack, actually told me this morning that he turned off Instagram. He just deleted it from his phone.

It’s so good, and it just sucks you in. Before you know it, an hour of doom-scrolling has gone by, and you’re like, “Oh my God, what am I doing here?”

But you heard Alvin Graylin on the podcast say that in China they banned everything. You can’t show blood in a video. You can’t show blood in a video game. The blood comes out green.

If they’re allowed to ban that level of detail, they’re going to control all of these actions and try to prevent their children from being sucked into the vortex. Meanwhile, they’re perfectly happy to sell it to every other country in the world and generate that revenue.

But I’ve got to imagine the CCP is sitting there saying, “Wow, these guys are just suckers. We can launch any AI-generated influence machine on any country, completely change the national opinion, and turn the dials here from Beijing, and they don’t seem to be able to do anything about it. They’ve all got free speech and whatever, and they can’t protect themselves at all.”

But these AI agents are so good. They’re going to get smarter and smarter and smarter, and become user-specific. That’s the new thing.

3. China Cracks Down on AI Companions

Dave, that brings us to the next story outside of China. It’s an extremely important story for America to hear loud and clear, and especially for any parents listening.

China is tightening restrictions on AI companions amid concerns that users are developing emotional dependence on virtual partners. They’re going to prohibit AI companion services for minors and restrict their use for adults.

The regulations come as policymakers grapple with the loneliness epidemic there, alongside falling marriage rates. The one-child policy really devastated China. The concern here is that people are forming emotional bonds with AIs that substitute for human relationships, and the substitution is accelerating. It’s a demographic crisis.

We’ve already pointed out on the pod that much of the world is now below the 2.1 children per family required to sustain a country’s population. Among the worst is South Korea, which is down to 0.75 children per family. South Korea is evaporating. Singapore is at roughly 1 child per family, as is Ukraine, and China is still at 1 child per family, even after they removed that ban.

China is becoming the first country to regulate against AI companions. My bet is Japan, Singapore, and Korea are going to follow closely. This raises the question for every country, not just China: what happens to society when the easiest emotional connection you can make is to your AI and not to another human?

Exactly.

Yeah.

It’s crazy. You know, Peter, do you remember Boris Zelman from MIT?

Biology. He was actually—his dad was in charge of food science at General Foods. He invented the coffee crystal.

Okay.

But he was like, “You know what? We all know that smoking is addictive now.” Smoking took over the entire country, then took over the world, and then we said, “Oh my God, this is killing everybody.” Then we stopped, and then we did the same with junk food.

He could develop a food at General Foods that was so delicious—a French-fry coating that was so delicious that you just cannot physically resist it—and then the whole country is diabetic.

We do this habitually. We react to it 10, 20, or 30 years later, and then we correct it reactively.

But if we start rolling out addictive AI, and then we say, “Well, the government will wake up when there’s a crisis,” that crisis is when a whole generation of kids is wrecked.

Yeah, but you cannot afford the 20-year delay with this particular topic.

I mean, the shout-out to parents is: please be careful. You form normal social relationships in your teens. Salim, you’re a father of a 15-year-old. How do you think about this?

If you step back, the right regulatory structure is probably some level of transparency and consent, figuring out what types of harm could be done, and then mitigating them, right?

Trying to have a government decide which relationships are authentic or not is going to be very tricky, because this is what’s happening right now. We’ve gone from regulating what AI says in the China case to saying we’re regulating how attached you become to it, right? This is a totally new regulatory frontier.

Social media was very, very addictive because an algorithm, as Dave points out, can learn your intimate habits and, without you realizing it, totally capture you. That’s really, really bad. All the kids are completely addicted to social media today, and banning it outright is pretty much the only way, because all their neocortexes are forming at a time when they’re getting totally addicted to stuff. It’s going to ruin them structurally for a long time, and then you need a lot of fixes for that.

The broader picture, though, I don’t know how to think about. We are dropping so dramatically on the childbearing side that this is a huge structural challenge for civilization, right? How do you navigate that? It’s going to be really, really tough.

There are also other issues in all this. The privacy issue, right? Your companion AI is going to know so much more intimate information about you than you do. Then there’s the benefit side. There are some really solid benefits around loneliness, disability, elderly companions, therapeutics, and coaching social skills, right?

You shouldn’t just categorize the whole thing as pathological. It’s the standard problem we have with any technology: how do you extract the promise without the peril?

Yeah.

And that's the challenge for regulation across the board.

Yeah. I have a bit of perhaps a contrarian perspective on this. I think the Chinese Communist Party is actually going to do an about-face sometime in the next year or 2 or 3 and decide that they actually love AI companions. The reason is that it's another method for party control. If you can make sure that all of these AI companions go through all of your ideological filters and become a little bit more like a CCP version of Neal Stephenson's A Young Lady's Illustrated Primer and a little bit less like some sort of distractant, I think the party—this is a prediction—decides that actually having everyone, or maybe not everyone but a sizable fraction of the youth of China, deeply engaged with deeply party-controlled AI companions is, from the party's perspective, net helpful, not net harmful.

And I think so, my forecast here would be that once the CCP has decided that it has a way to steer the country's youth via AI companions, it will decide, “No, actually, we like these AI companions after all.”

But bring it home, Alex. What do you think about the US and AI companions?

There are US states that are already trying to regulate these things out of existence and trying to ban romantic relationships between America's youth and AI. I'd like to think that in this country we enjoy enough freedoms that they won't succeed. I'm just reminded again of one of my operational definitions of the singularity: every sci-fi trope everywhere all at once. And I can't shake out of my head the famous line from Futurama: “Don't date robots.”

I think different countries will react in different ways, and my forecast for the US is that you may see individual states. Maybe it'll be a federalist issue where some states ban them and some states don't—question mark as to whether we end up with a national policy on romance with AI companions. But if I had to bet, my bet is that at least in some places in the US, AI companionship, including romantic companionship, including robotic romantic companionship, will be legal. That's my bet.

Yeah, we're going to talk at the end of this pod about robotic sex partners just to spice it up.

The place to watch on this is Japan because they tend to be much more tolerant of this type of stuff. There's already a case where a Japanese woman is marrying an AI, and they're like, “Okay, go.” So we'll see this play out, I think, there first. Well, South Korea is super high-tech.

And a very early adopter of everything without a lot of regulation, and they have all kinds of issues: the lowest birth rate in the world, one of the highest suicide rates in the world, just all kinds of issues.

Yeah. Maybe 2 points. One, I have a difficult time getting myself worked up over the collapse of human population. I think we're so many decades at this point past The Limits to Growth, which I view as nonsensical propaganda at this point. Humanity would be fine, barring some total disaster. We have a carrying capacity in the tens of billions if we use just the Earth that we have correctly.

We can also survive with an enormous amount of AI and automation with fewer people. So, one area where I perhaps differ from Elon's sort of virulently pro-natalist policy: I think the world is going to be fine regardless of whether we have more people on the margin or fewer people on the margin. We're going to have so much AI in the solar system. I just don't think it's going to be a major X-risk at all.

Yeah.

I've got to say something here.

Please go ahead.

In a rare moment, I'm 100% aligned with you, Alex.

This is a rare moment.

I think that here and there you and I have opposing points, but this one I'm absolutely 100% with you on. This is trying to worry about how many kids we're going to have. It's going to end up however it ends up.

Listen, the challenge to a country like South Korea is that it's sublimating. It's evaporating. If all of a sudden the population is dropping smaller and smaller, you're losing the culture there. And that's the issue. And I hate it when people—I'm giving a presentation on longevity and people say, “Oh my God, can the world sustain enough people when people are growing to 150 years old?” It's like, look at the actual numbers.

4. Flock Safety and 20 Billion License Plate Scans

I don't think we lose culture if population, on the margin—not like catastrophic nuclear war, but on the margin—goes down. If these cultures are centuries old, when populations were smaller, I think it follows reasonably that if the population on the margin halves, the culture will still be there in one form or another. But the reality is we're going to have so much, again, AI. The AI is being trained off of our culture anyway.

All right, moving us along. This is a story, Alex, that you fed me, and it's a fun one. Our next story is about a company called Flock Safety. It's one of the fastest-growing public-safety tech companies in the US. Flock was founded back in 2017. What is that, 9 years ago?

Its core product is a network of AI-powered license-plate-reading cameras used by police departments, neighborhoods, schools, and businesses to identify vehicles associated with crimes and missing persons. The company has further expanded its platform to include video cameras, gunshot and audio detection, mobile surveillance trailers, drones, and software that ties all these feeds together. Flock says its systems are used by more than 6,000 communities across 49 states, generating over 20 billion vehicle scans per month.

So, let me show a video here that Alex provided me. This is out of San Diego, where Darth Vader is testifying on behalf of Flock. Take a look at this:

“The Emperor is a fan of Flock, and we must continue utilizing Flock technologies so that we can follow and surveil the rebel scum as they move from playground to playground, from playground to pool, from pool to gymnasium. Because we all know that the Flock cameras are not only following the license-plate readers. They are following children. They are following children in parks and gymnasiums. And we need this. I need this so I can stalk my ex-girlfriend.”

Okay, so, Alex, let's go to you first on this one.

I think this, honestly, is the face of AI at the municipal level at this point. We talk all the time about the implications of AI for enterprise use cases and for transformative scientific and technological discoveries, but I think many people, when they start to think about how AI is transforming everyday life in many communities, I worry that some of their earliest interactions won't be with driverless cars. They'll be with cameras that are tracking license plates.

In many communities around the country, I read stories about people who are chopping down AI cameras from Flock or other companies because they feel that somehow they're being surveilled by their police department. You read stories about police departments federating their surveillance data to enable new forms of tracking. In the odd case—I think this is what Darth Vader was gesturing at, perhaps—you read stories of police and other first responders using their access to these AI cameras to stalk ex-partners.

And so I think Flock, in some ways, has become the petri dish for either surveillance or what an old friend, Dave Brin, might call sousveillance: who gets to see whom inside the panopticon that AI enables. In China, it's pretty obvious that the party gets to look through the cameras at everyone else. In the US, we're grappling with this right now. If we put smart cameras everywhere that can look at every public space, and then they're all federated so you can track movements of cars, who gets to look at the data?

So, I think Darth Vader testifying before the city of San Diego is a very American moment, when we have the Empire testifying on behalf of surveillance, not sousveillance. Salim, you and I have talked about the loss of privacy a lot, and the fact that people assume they have privacy, they want privacy, but I think that's pretty much a myth at this point. What are your thoughts on Flock?

I have a lot of thoughts here, and I think this is a very, very dangerous moment. I've gone to see a bunch of Rush concerts lately, but let me quote Pink Floyd here: “Another Brick in the Wall,” right? That's Pink Floyd flipping over to show some level of thing. But there's a huge structural pattern because now we've transitioned from “Find this suspect” to “Find me people who behave like suspects,” and that's a very, very different animal.

Pattern matching now becomes essentially free, meaning surveillance stops being constrained by manpower at all, right? Civil liberties were kind of protected to some level by the friction, because following everybody was just too expensive. And now that's gone. The cost of following everybody has gone to 0.

So this is a big structural problem. You already have cases where police officers are using this technology to track their partners, to track ex-girlfriends and ex-boyfriends and stuff. You've already had people literally dying because they got tagged with the wrong license plates because Flock read it wrong. This is a massive issue that completely wipes out our civil liberties in some ways. You've got to be really careful, and the problem is that technology has eliminated the cost of mass surveillance now.

Okay. The Constitution does not update when transaction costs go to zero. This is a very, very big moment. I think I like the way Alex frames it: this puts the whole thing in a Petri dish and brings it up to this level of conversation. We need to have this kind of conversation at a constitutional level and say, “How do we want to be? How do we want to operate in this environment?”

Now we are very close to essentially where the Chinese have the CCP mandating it, and they just operate that way because that’s what it is. But essentially, we’re in the same place here. It’s just done by corporations and by unwarranted, fairly rogue police forces and surveillance systems. So it’s a very, very delicate moment. I don’t think it ends well.

I want to just steelman this for a moment. This is not just the CCP. This is also the Emirates, for example. I was just having a conversation with Jamie Justice, who heads our life sciences at XPRIZE. She was saying she went out for a run—she’s a prolific runner, oh my God, getting up at 6:00 a.m.—but she went out for a run at 10:00 p.m. in the streets of Dubai, and she felt completely safe. She said she would never do that in the U.S.

I mean, you’re trading some level of civil liberties. There’s always a trade. When you go through customs at the airport and you’re no longer waiting in giant lines because facial recognition puts you through instantly, I’ll make that trade. I will happily make that trade to save myself time, and I’ll make the trade as well for increased safety for my kids. So the question is: do we put this at an individual vote, or is society saying, “We value safety over privacy?”

I just have to remind you of the Ben Franklin quote: “They who can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety.”

Yes, I’m with you. This is very, very dangerous. I have a personal experience here. I once came into the U.S., and they took me aside and said, “Sorry, you have to miss your flight because there’s an Afghan warlord by the name Salim Ismail wanted by the FBI for poppy trading.”

And I’m like, “What?” And they’re like, “We have good news and bad news.” I’m like, “Okay.” They said, “Good news: you’re, you know, Vice President at Yahoo. It’s pretty clear you speak at all these conferences. We’re pretty clear you’re not an Afghan warlord.” I’m like, “Phew, glad to hear that. What’s the bad news?” The bad news is, “We’re not allowed to make that judgment call. We have to check the watch list. It’ll take a few hours. Have a seat.”

As an Afghan warlord from now on.

You can’t call me an alleged Afghan warlord; nothing was ever proven. But this is a real problem. And I have to say something: the folks I’ve encountered at U.S. immigration have never been anything short of beautifully professional and constructive. That’s been my experience—unbelievably proper and appropriate.

But a few times I’m like, “Guys—” and they’re like, “Yeah, we know you’re not the Afghan warlord, but have a seat. We have to check. We just have to go through this process.”

Might have shaved your hair.

Yeah. It took a few years to get that cleared up. So I said to them, “Does that mean every Salim gets stopped like this?” And they said, “Sir, you have no idea. When Juan Rodriguez commits a misdemeanor, it means we have to take aside every Juan Rodriguez.”

This is where you can get the benefits of this technology, because it can disambiguate, and names resolve at the name level very quickly with technology. On the bad side, the negative consequences and accidental consequences—like innocent until proven guilty and habeas corpus—are essentially gone now in the U.S. This is a really, really bad place.

So I don’t know how this resolves. But I think the Ben Franklin quote should be sitting on the front desk, tattooed in front of everybody, for the next 10 years to solve this. It’s a very big problem.

It can go very badly for quite a while. We can try to fix the systems, et cetera. It’s a really, really difficult problem. And we can use technology in very powerful, constructive ways, but we really risk losing a lot of civil liberties here, especially if you have an environment or a government that’s not friendly toward you.

In the UAE, they have a benevolent dictatorship. As long as it stays benevolent—and they’re about as benevolent as you could possibly imagine—that’s why there’s so much incredible safety. Nobody steps out of line. The minute you step out of line, they take you out very quickly, and therefore everybody feels incredibly safe. It’s a fabulous thing, but the problem with benevolent dictatorships is they typically don’t stay benevolent very long. That’s a structural problem, right? The solution—

We have to really figure this out.

Sorry, we have to figure out human governance now at a level that we’ve never had to deal with before. This is, for me, one of the singularities: how do you construct regulatory frameworks that can have any hope of keeping up with the pace of technology?

Sorry, Alex. One of the solutions that has been proposed—I know David Brin and others have advocated for this—is to give everyone access to these surveillance capabilities. Allow civilians to do the same license-plate tracking that first responders and police have. Let everyone track everyone. If it’s a public space, make it a public resource.

That’s right. And I’m a huge fan of the EFF and the work that they do there. They’ve highlighted some of these things. For example, in Russia, which is a bit of a Wild West, a policeman will come and break your headlight, then give you a ticket and extract corrupt bribes from you.

The Russians were the first to adopt dashboard cameras. Do you remember that meteorite that hit a decade ago? There were something like 200 live videos of this meteorite hitting. Somebody asked the question, “How the hell do we have 200? How many people had time to whip out their cell phone cameras and track this thing in real time?”

It wasn’t that. It’s just that the Russians had learned they better have a dashboard camera to protect themselves against dodgy police extortion. So they all have dashboard cameras for their own safety, and that captured all those things. You can see, in that case, the citizenry is starting to point the camera the other way, and we need a ton more of that. I think Alex’s point is really important: everybody should have access to these systems so that you can see what’s going on.

Amazing. Well, as you guys may know, we put out a call for optimistic, hopeful visions of the future. This is the Future of Vision XPRIZE. We had over 5,000 teams registered and 2,550 video trailer submissions. It was amazing. It blew us away. Largest film competition ever. And we're down to the final 50, and we're asking for you to help us vote. You can go to my tweet on this at Peter Diamandis, my XPRIZE handle, or you can go to vote.futurevisionexprize.com. You'll be served up two videos at a time, and the question is, which future do you prefer? You vote on one, then you'll be served up the next two, and we're doing essentially an ELO bubble sort here. We're going to be bringing the top five of these to the Moonshots Live summit. On September the 25th, you're going to have a chance to meet all the Moonshot mates. Alex, Emad, Dave, and Salim will all be there. We'll be doing a Moonshots Live broadcast from there. We have incredible speakers. Palmer Luckey, Astro Teller, Ben Lamm, Kathy Wood, Neil deGrasse Tyson, and Neal Stephenson. Salim, what are you going to be talking about at Moonshots?

I’ll be talking about my normal stuff, which is how humanity copes with all of this. Specifically, I’ll be talking about the organizational singularity: when a world is changing this fast, how do you build organizational structures that are resilient for this new world?

Or daily, or in real time, as Alex would talk about. Just get in one big room and to hell with everything else.

Yeah, this is live. Live.

Yes.

Live, 24/7. And Alex, you're going to be on—we have, by the way, through the day an incredible program with the Mates, and then in the evening each of the Mates is going to be spending an hour with you. So, Alex, what are you going to be doing during your hour with all our guests?

Well, I think the question is: What are the questions? I think the agenda calls for an AMA. So, I'm extremely curious to hear what the questions are and do my best to dispel any myths that I might be propagating.

And Dave, you're going to be doing a deep dive in AI investing. Talk more, please.

Well, I'm anticipating we're moving into this red-pill, blue-pill world where you can choose to be oblivious to all of this and be happy living it out without being aware of it, or you can choose to be in the middle of it. I'm anticipating every attendee of this event has decided to take the red pill and be in the middle of it.

And so then they're going to be asking, “Okay, what do I do career-wise? What do I invest in? What are the scarce constraints of the future? How do I interact with these 11 megacap companies? Are they going to crush me, or are they going to be my friend?” All of those questions are going to come up with this gang, and I want to be prepared to answer every single one of those life-changing questions. I have a specific question I want to ask Neal Stephenson.

Yeah.

Which is, if you went back 10 or 20 years and he was writing all these amazing novels, you had some sense—a glimmer—of what the future might hold, right? In a world that we live in today where it's so hard to predict where things are going in 3 months, forget 3 years or 30 years, how do you build the models that give you viable future narratives? And that's the question I've got for you.

You know, I spent the day with Neal Stephenson, one of the most extraordinary science fiction writers. He wrote The Diamond Age, The Young Lady's Illustrated Primer, and Snow Crash. Wow.

Snow Crash, Seveneves. He wrote that over 30 years ago, and I just reread it, and it's perfect. I mean, it's so good.

It's so good. It has held up over 30 years, right? Predicting where we're going.

Yeah, because we should ask him to be a guest, Peter, because it would be really great to get him as a guest. He's brilliant. He's not really a live-stage kind of guy, but I'd love to have him as a guest. If anyone could open him up in a live environment, it would be us. That would be fantastic.

You know, one of his other themes that I think is incredibly relevant over 30 years is, if you look at the world leaders today, they're overwhelmingly really, really old. And if you survey young people and say, “Do you aspire to be president someday?” “No. Absolutely, positively not.”

And so, in Neal's books, society worldwide cuts the other direction. It's not like, “I'm in this country. I'm trapped in this country. I live in this country.” People actually, in the age of AI, cut in the other direction, where people all over the world are like-minded on a topic and they bond through global communication, through Starlink and through AI, across countries, which could be a great sign for future global peace. So I'd love to talk to Neal about that.

We're also going to have Rod Roddenberry there, the son of Gene Roddenberry. Star Trek is an underlying theme at Moonshots Live 2026.

I think, in some ways, just on the sci-fi note and Neal Stephenson, I think, in some ways, it was easier, ironically, 30 years ago to write sci-fi that was predictive than it was in the decade or 2 after that.

Peter Thiel and others have made the point, or suggested, that the 1990s might have been a local optimum in terms of clarity on the future because we had the economy booming, we had the dot-com boom, and we had the internet hitting consumers. So, if you imagine that history isn't sort of a monotonic exponential—or at least consumer and broad public perception of history isn't like a smooth exponential, but ebbs and flows and has booms and busts—then a moment of peak clarity would come during a boom, when you're able to see above the tree line and see where we're going.

And then, when you're in a local bust, you lose the forest for the trees and you lose clarity. So, by that metric, I think the happy side effect would be explaining why Golden Age science fiction from, call it, the 1950s or so—the post–World War II era of sci-fi—was also a local moment of clarity, when you could see rockets and you could see fission-based energy too cheap to meter.

So, by that metric, then, I would say it's not that surprising that so much amazing sci-fi was written in the 1990s, from Neal, from Charlie Stross, and from others.

Robert Heinlein. Yeah. Yes, amazing.

When we did our week-long Singularity University executive programs, the most fun session ever for everybody, universally, was the unconference.

Right. And for folks that don't know, what you do is you put up a set of rooms and a bit of a timeline. People get up and say, “I'm going to talk about AI and biotech,” and they put up a thing. Somebody else gets up and says, “I'm going to talk about this,” and they put up a little poster, and you go wherever you feel like going.

It turns out to be the most incredibly engaging session you can ever have. And so, if you ever get a chance to participate in an unconference, go do it. An unconference of this crowd will be absolutely epic.

Listen, because if you look at the comments, we have some really, really smart followers and commentators. It'll be unbelievable. Unbelievable. It'll be amazing.

Yes, for sure.

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 code bases with millions of lines of code. Engineers start every development sprint with the Blitzy platform, bringing in their development requirements. The Blitzy platform provides a plan, then generates and pre-compiles code for each task. Blitzy delivers 80% or more of the development work autonomously while providing a guide for the final 20% of human development work required to complete the sprint. Enterprises are achieving a 5x engineering velocity increase when incorporating Blitzy as their pre-IDE development tool. Ready to 5x your engineering velocity? Visit blitzy.com to schedule a demo and start building with Blitzy today.

5. Only 4% of Companies Expect AI Job Cuts

All right, moving us along. We're going to jump into the stories around jobs. I actually have 3 stories around jobs. The first story is an article in The Washington Post this week that reported an analysis predicting that the job apocalypse is unlikely to happen.

Again, I've been saying for a while that this has been really murky. You have half the community out there saying we're going to lose all the jobs, and the other half saying, no, we have job growth. So, The Washington Post argues that historically, technology transitions—from the loom to the automobile to the internet—consistently created more jobs than they destroyed. We've had that conversation on the pod here many times.

It displaces specific tasks rather than entire occupations, while creating new categories of work that were never previously even conceived of. The thesis of The Washington Post article is that AI is not replacing jobs. It is replacing tasks within jobs, and the people who learn to use AI become more productive, not less employed.

So that's the first story. Let me go on to the second one. We'll stop there, and then, Salim, I know you have a particular story you're going to cover here.

The second story comes out of Goldman Sachs, which this week warned that professional-services firms—consulting, law, and accounting—face an existential risk as AI capabilities improve faster than their partners can adapt, creating a dangerous skill gap where junior staff trained on AI outperform senior partners who refuse to use it.

The skill gap is not between the educated and uneducated. It's between the AI-fluent and the AI-resistant, and the AI-fluent are increasingly the juniors, not the partners.

Yeah, 2 or 3 things. First, let's all urge everybody to please be evidentiary and data-driven in your orientation for how you view the world, right? There are people that doomsday—Bill Gates, 2 days ago, came out and said, “Oh my God, all the jobs.”

And let me just call out here: I'm a massive fan of Bill Gates and his philanthropy, but as a futurist, he has somewhat of a dismal track record. He missed the internet, he missed mobile, and commenting about the future of work, I don't think, is the right thing to do in an area like this.

And by the way, this is true of, I find, Ray Dalio, Larry Fink, and the other folks that do commentary—Yuval Noah Harari. They tend to be unbelievably great at framing the past, somewhat questionable on the future, right? Because many times they don't really understand exponential, which is a whole thing, and so you really have to be careful around all of this.

If you look at, say, the Goldman Sachs or the professional-services-firm conversation, I deeply disagree. Why? Because as the world gets more volatile, companies are going to need more help than less, because that volatility will require a lot of help, which is what we're seeing.

You’ll have to change your business model from selling hours to selling outcomes, which is starting to happen, but I don’t think that’s that big of a deal. The bigger thing, I think, is the jobs question, and I would love if we could throw up the slide right here.

Principal Financial Group, a friend of the pod—their people watch us on a regular basis—did a survey. To give you a sense, Principal Financial Group helps small businesses with health plans, benefit plans, retirement, and so on at the small-business level. For SMEs, they have 130,000 customers, right? This is a big company with a great cross-section across the entire country and, frankly, around the world.

Here is what they are seeing. They did a survey across a very meaningful, representative sample, and what they found was that only 4% of the companies anticipate that AI will reduce staffing and wages. Thirty-one percent of small-to-medium-sized enterprises expect an increase in staffing and wages.

Which is the majority of the workforce, by the way.

Let’s note a very important fact. Over the last 50 or 60 years, 100% of job growth has come from small-to-medium-sized companies. Big companies are getting bigger but also getting more efficient. Net job creation: zero. One hundred percent of job growth has come from small-to-medium-sized companies.

So this is the demographic to track if you’re interested in the future of work. Thirty-one percent expect increasing staffing and increasing wages. Ten percent say they don’t use AI. Last year, it was 19%, and now it’s down to 10%.

Of the people that have reduced staff, only 1.4% attributed those reductions to AI or automation.

This is so fantastic.

It’s so inverted from the current BS tropes of radical unemployment that we see. It’s unbelievable. Please, people, look at the data. Look at this stuff. We can get you the full report if you’re interested, and look at the staffing trends just in the last few months.

Fifty-two percent have increased staff, 30% have maintained it, and only 12% over the last 3 months have reduced staff. So this is very recent and a great predictor of where things are going. It’s completely opposite to the “Oh my God, jobs—everything is going to take all the jobs” narrative. This is where to track the future in actual reality.

Love it. A great optimism story. Dave or Alex, do you want to jump in?

I think we’ve come full circle, actually, in the sense that, remember when we were talking to Elon? He was predicting great social unrest with unprecedented abundance going on concurrently.

Yes.

What the labs have now decided is, “Look, if I automate everybody’s job and they’re all unemployed, I need to give them a new sense of purpose.” What’s their new sense of purpose? Well, that job I used to have gave me a sense of purpose. Why did we take it away in the first place?

So now what they’re doing is saying, “Look, we don’t need to displace everybody’s job just for the sake of automating everything. We’re going to leave everybody alone while we work on incredible new breakthroughs, and we have infinite sold-out tokens anyway. So we don’t need to disrupt all these voters.” That’s the way it’s going to play out.

When I was a kid, we were the first people we knew to get a PC—an Apple II. My brother and I begged and pleaded with our parents to get this Apple II, and it was life-changing for me. But I showed my dad, “Hey, there’s a word processor called SuperScribe, and you don’t have to handwrite all your notes anymore.”

He was like, “You know what? I’m in my 40s. I’ve been doing it this way my whole life. I don’t need to be a computer person.” So he finished out his entire career at GE without being a computer person.

I can kind of forgive him for that, because the PCs would break all the time. The printer was a nightmare. This is not hard to do, relatively speaking, and it rewards builders, creators, and visionaries regardless of their technical skills.

You are absolutely empowered by this. You’re crazy not to get on the bandwagon, but AI is going to allow you not to do it. We’re actually making the door open to not get on the bandwagon.

I get the question a lot: “As AI is automating lots of stuff, what will human beings be doing?” The best metaphor I’ve come up with is this: If you went back 100 years and looked at accounting, people were doing double-entry bookkeeping in big ledgers—credit over here, debit over there—in pencil, with a big eraser and a whole set of tabular columns and lots of ledgers.

As we moved to automation, we had calculators and slide rules that helped you do the tabulation faster, but we were still doing the entries by hand. Today, we have accounting software that reads your bank account and does all the entries automatically.

The human being has been lifted above the loop, not in the loop. The human being is above the loop, categorizing the transactions, looking for problems, categorizing vendors and types of income coming in, figuring out reconciliation gaps, and looking at the process flows. It’s much more value-added to be there than doing the rote double-entry.

This is what we see happening. Eric Brynjolfsson calls this white-collar drudgery. There’s a huge amount of stuff happening in big companies. It’s a lot of massaging of information—taking stuff from this sales report and putting it over here, and so on.

All of that will be taken out, so you can do really curatorial, judgment-thinking work and apply experience to it rather than doing the actual work itself. People miss that.

I’ll go back to my specific experience. Writing the first book was 3 years of hell, just horrible, because you had no help.

Well, no. The first book, I was partly with you, but you had to write every line and check every line yourself. The second book, Peter, that you and I did together fully, was 2 and 1/2 years, but even more hell, because you kept having to think, “How much of the first book do you bring over?” and so on.

By the way, it was an absolute joy working with you, so don’t you dare say anything like that. But this third book has been 6 months of unadulterated joy, because I can say, “Bill Gates just said this,” or “Satya Nadella said this.” I can ask, “Look through the book. Tell me, as a developmental editor, where we should put inserts,” and have it scan the thing.

It’s absolute pure fun. People forget how much more fun it is when something is doing all of that cognitive BS stuff that everybody had to think about. Now you can do pure thinking and judgment. You can be much, much more creative.

This is the heart of what it comes down to. All of this automation, AI enablement, and cognitive abundance allow us to be deeply, deeply, deeply creative. That is the most fulfilling place you can be as a human being.

Beautifully said. Let me hop on Dave’s point. I’m sure everybody listening here at the pod is using AI and enjoying these conversations, but if you have someone in your life who isn’t, I think the most important thing to realize is that you have access to the most patient, most capable teacher on the planet.

Pick your favorite model and ask it to begin at zero: “I don’t understand this. I just heard this term. Help me create a curriculum to go from not understanding AI to fundamentally being able to use it.” At a minimum, spend 30 minutes a day in conversation with your favorite model.

It’s that easy. It’s free. It’s everywhere, on your phone. There’s no reason not to dive in. Human cognitive challenges typically prevent a person from jumping into something they don’t understand, but it has never been easier in the world.

Alex or Dave, do you want to close us out on this one?

Yeah, I’ll just comment. I think the original impetus for the story was actually from our friend of the pod, Eric Brynjolfsson, as you mentioned, Salim.

I think there are many people in this economy who don’t realize that not only do they have permission to be high-agency, but we’re now in a regime where high agency—and agency in general—is one of the few human traits that is actively rewarded in an era of superintelligence.

I suspect this is my theory of the case: Too much of the human economy, at least in the American economy, the one I’m most familiar with, thinks agency is something exhibited by superheroes in Hollywood movies. People don’t realize that the shackles are off. Anyone can exhibit high agency, empowered by superintelligence now—not just in the cinema.

To the contrary, this is one of the few things left that can actually yield extreme vertical mobility. So if you’re listening and, to Dave’s point, you’re not fully availing yourself of superintelligence to accomplish superhuman feats, I encourage you to start now, because I think there’s a window of time, and I don’t know how long it’s going to last—maybe only a handful of years.

Do it now. Otherwise, do it never.

Yeah. Wow. I want you to reflect on that. Society will reward agency in this world of AI automation. That’s such a great point, because the more you are a self-starter and have the gumption to go ahead and do this and build something with it, the more you’ll be rewarded. Absolutely love it.

Alex said “extreme upward mobility.” We’ll use that as a theme at the Moonshot Summit, but the data behind what he just said is like nothing you’ve ever seen before. The gap is getting wider and wider, but the rate of upward mobility that’s possible has never been seen before.

How's it possible to be depressed if anybody watches this? That's my question.

Yeah, I mean, it's such an amazing time, and I love your story. Thank you for sharing that data. It's data-driven optimism.

Do you mean that or the Afghan warlord thing?

You mean this one? Okay. All right.

6. China’s 200,000 Fake Accounts and the Data Center Fight

You, too, can be an Afghan warlord. Everyone can be an Afghan warlord. There's agency of a different kind. There's hope for everyone.

Let's jump into the world of data centers, chips, and compute. There are 2 breaking stories this week. The first comes out of the Global Affairs team at X. Let me share a slide, and then we'll talk about what it means. Is this data absolutely 100%?

This is what was posted on X by the Global Affairs team at X. It says, “The X safety team conducted an investigation into suspected Chinese inauthentic accounts involved in influence operations. We identified a bot farm of approximately 200,000 accounts. Within this farm, we found 200 accounts posting in a manner that could manipulate a legitimate debate about American AI and energy policy. These posts contained claims that AI data centers are driving up household electricity prices and straining the grid. Others included AI-generated cartoons that depicted data center operations enriching operators themselves at the public expense.”

Here we're showing one of those cartoons. We've talked about this. What's causing this 75% “not in my backyard” sentiment, where people would rather have a nuclear power plant in their backyard than a data center? It seems completely illogical.

The second story is one I want to put front and center. It comes out of Quincy, Washington, where the data centers have had an extraordinarily positive civic impact. This story, reported on CNN, says, “Data centers helped resurrect Quincy, Washington, where the poverty rate fell from 29.4% to 6.2%. Tech tax revenue funded a new high school, hospital, library, police and fire stations, while residents' property tax rates decreased.”

This is a story about the tech industry succeeding, and bluntly, this is the story that the tech industry needs to be telling and needs to be making happen. So, gentlemen, your thoughts?

Shocked. Shocked, I say, that there are foreign influence operations attempting to suppress American AI data center deployment. Shocked.

Salim or Dave?

Standard ops. You'd expect to see this happen. The difficult part is that we have no easy defense against it. The problem, as I mentioned a couple of podcasts ago, is that people in the U.S. are very, very narrative-driven rather than evidence-driven. That's the problem.

I don't know what the U.S. is doing in China, but the numbers here are crazy, right? We've talked about this before. If you or someone you know is against data centers, please dig and look at the numbers. Water is not an issue.

With energy, the right policy is that the county says, “Okay, you can build here, but generate your own power or subsidize community power so it's cheaper.” And for God's sake, hyperscalers, invest in the schools, invest in the libraries. I don't know if libraries are a thing anymore, but invest in the community. That's the right way to do it. Take a fraction of the cost and make everybody's life there better.

But, Peter, you're gesturing at something interesting. Libraries—remember the digital divide and concerns that some areas might be left behind with less compute? Now we're seeing, whether as a result of foreign influence operations or other political factors, the exact opposite: all of these regions that could, in principle, be economically uplifted, energetically uplifted, and compute-uplifted are worried, hand-wringing that they might be suffering from too much compute. How much the tide has turned.

Amazing. Well, every now and then there's a story where we absolutely, positively need to get to the truth. We saw this during that COVID outbreak, where it's like, “Look, this is a medical thing. What's the actual truth? Do the masks work? Does ivermectin work?” We need absolute truth, and instead we got politics. We got a Republican opinion and a Democratic opinion, and now Fauci's on trial. Come on, man. We need the absolute truth on this story.

Are the Chinese—is it the policy of the CCP to disrupt American opinion on data centers, slow down our construction of AI, bypass us in AI, and then control the world by winning the race to RSI? Is that their plan? Or, if you talk to Alvin Graylin, he's like, “They don't even think vaguely that way. They're just trying to have a better life for themselves and improve their own progress. They couldn't care less about trying to invade the U.S. with false stories.”

Which is the true story? I don't actually know. But we absolutely don't need politics on this. We need to know what is actually going on. Dave Blundin

So I just think, hypothetically, they may come up with some sort of infinitely scrollable streaming video app that will become insanely popular on mobile phones—

Addictive to all the kids.

Just speaking hypothetically.

7. OpenAI’s New Chip and Apple’s Local AI Push

I think American companies are doing a great job at that already. But that's a different story. All right, let's move on.

OpenAI released the first performances of their chip Jalapino. It's a custom inference chip developed with Broadcom, reporting 1.5 to 1.9 times more AI work per watt and up to 3.6 times lower end-to-end latency compared with NVIDIA's GB200 and GB300. Jalapeño runs on 700 watts, against the GB300, which runs at 1,400 watts: half the power and 1.5 times the peak token rate per kilowatt.

The significance here is that OpenAI is no longer just a customer of NVIDIA. They're becoming a chip designer. I saw a great interview where Jensen was being asked by the reporter, “How do you feel about your largest customer competing with you now?” I think this is what we're going to see everybody doing. Everybody's moving up and down the stack. They want independence.

Yeah, there's no doubt inference is moving off NVIDIA, for sure. The 100-times performance gain—this is the first of 2 100-times performance gains. That does not mean NVIDIA has a problem, though, because training is staying on NVIDIA, and training is infinitely sold out, just like inference.

But there's an enormous—90% of the compute is inference time—so there's an enormous new industry for inference-only computing emerging. The other side of the story is that this was just an idea at OpenAI a few months ago, and now it's a production chip. That shows you how short the cycle time on innovation is becoming with AI assistance.

Yeah, a few thoughts on this. Maybe the most startling statistic that OpenAI put out regarding Jalapeño was the throughput per second per user—the tokens-per-second-per-user increase that they got on Jalapeño, not just on their frontier models but on their open-source models. Remember GPT-OSS? They claim almost a 54-times throughput increase hosting GPT-OSS versus what they obliquely refer to, I guess, as the existing best—presumably a reference to some NVIDIA architecture.

That's so startling. OpenAI—and where is Broadcom in this? Obviously, it was public information that they worked with Broadcom on this, but presumably the only way OpenAI, using its own AI, could achieve a tape-out so quickly is in conjunction with a Broadcom closed PN.

This is so startling and potentially such a huge throughput increase. It makes me scratch my head and wonder whether OpenAI might actually do this to themselves. Even though they are competitive with Anthropic, they're one of the 2 top-tier frontier firms left, and they might actually choose to become a hyperscaler themselves with their own chips in-house. It's like a crazy future. It would be a heck of a plot twist.

This has me wondering: if they have such an amazing balance of prefill of the prompt with otherwise memory-bandwidth-limited decoding for LLMs, does OpenAI want the optionality in a year or 2 of saying, “Hey, market, in addition to our own models, we're going to respond to the competitive threat of Chinese open-weight models and other open-weight models by—surprise—offering its own cloud”? OpenAI offers its own cloud. We'll call it OpenAI Compute. It'll be hosted on OpenAI chips, and we'll still make money just like Elon is. He's making money hand over fist hosting everyone else's models, including Anthropic.

I think we're in such a crazy future where I could imagine a world where, say, an OpenAI Compute cloud hosts an Anthropic model, and OpenAI and Anthropic win at the same time.

Love it. Wow. Okay, I have a question for you guys.

Yeah.

One of the things that was mentioned was that NVIDIA's CUDA moat may be dead in the next few years. How big of a deal is that? CUDA seemed to be their moat, right?

Yeah. No, CUDA is still the dominant moat for AI researchers who are working on new training.

Take a second and explain what CUDA is for those who don't know.

Yeah. So, all the AI researchers in the world moved to Python and PyTorch because you can iterate on your ideas so much more quickly in that environment. Then, to run it at scale, you're like, "Well, here's my code. Just run it." The only way to say "just run it" is through the CUDA layer, which translates your PyTorch directly into NVIDIA microkernels and runs them at blazing-fast speeds.

NVIDIA invested a lot of time and human effort, before AI, coding that up. It became so pervasive that when AI took off, only NVIDIA chips could be used for all this training and research because it would have been a nightmare to try to port all these model algorithms to AMD or some other vendor. That's why NVIDIA is the most powerful—or most valuable—company in the world today: specifically because of that brilliant early support of AI researchers when nobody else cared about them.

Now everyone's like, "Well, I can vibe-code anything in a microsecond. Why is the CUDA thing still a moat?" At inference time, it's not. Your inference algorithms are so easy to port to AMD or Intel that you just don't care about CUDA. You can vibe-code your own solution.

But for advanced researchers, they still overwhelmingly use CUDA. NVIDIA realized that CUDA can't last forever, so they went and bought Mellanox to create a massive, high-speed interconnect. Now their big moat is: if you want to run 100,000 or 1 million GPUs in one coherent cluster, you still have to go with NVIDIA.

I think Jensen is smart enough to know that your moat is only good until the next moat. He's moving the moat very intelligently, and he has a huge amount of capital to keep moving with it. CUDA as the inference-time moat is already dead. CUDA as the training-time moat probably has a limited lifespan. But that's okay, because the interconnect is the new moat.

All right, let's turn to Apple. Apple's new Mac Studio pairs the M5 Ultra with up to 512 GB of unified memory, giving a desktop Mac enough memory to run some of the largest open AI models available locally on your desktop. Apple also unveiled the M6, its first chip manufactured on the 2-nanometer process.

So, finally, Apple is in the AI game, pushing local AI as an alternative to AI in the cloud. Salim, you wanted to talk about this one. What are your thoughts?

Well, it's just such an easy and obvious thing for Apple to do, right? If they hadn't done this, they would have been the stupidest company in history, and they're not that dumb. The M5 Ultra's 512 GB of unified memory puts a massive amount of intelligence under your desk.

This really changes the economics from paying per token perpetually to buying a capital asset and then using it continuously, especially since Apple has gone to the extra level of starting to lease its equipment to people. This is really a big deal.

I think the big winners here are companies, law firms, and healthcare companies that have HIPAA issues and can't put their data on the cloud. Much more can now be done on-premises. I think this is going to be a big deal for medium-sized healthcare firms that can't afford a private cloud and all the work involved in running one.

This is such a natural advantage for Apple to pursue. You get 4 of these Mac Studios clustered together, and now you've got a small private data center. That's a pretty big deal. Even Dave would go all-out for that.

Rather than agents. I mean, look, I'm at the end of my rope with Apple.

Nobody has changed my life more—someone I've never met—than Steve Jobs. The Apple II that I got when I was a kid totally changed my life. Then, when they came out with Mac OS based on Linux, it changed my life again. It's absolutely the greatest company I've ever seen.

But they've totally missed the boat, and it's just embarrassing that a company with that much cash flow has no AI strategy. They don't even deserve to be a Mag 7 company anymore. It's just pathetic.

If the best they can do in the age of AI is add a bunch more RAM to a machine they already had, I mean, what the hell, man? I'm not saying this is a great thing that they came out with, but it's like 0.01% of what they should have done by now in AI.

I totally agree with all of that, but it's obvious they should have done it. At least they did this, right?

The irony is even deeper. If you look at the ancestry of how unified memory architecture came about and the Neural Engine inside the M series, you can trace the lineage back to the Apple Car. The Apple Car that never happened was the reason Apple first introduced the Neural Engine with access to this huge, unified, addressable memory space.

I almost think Apple's interaction with AI is a history of hardware that's too good for the software—or, conversely, software that just isn't worthy of the hardware. Apple had early unified matrix-multiply acceleration built into a unified addressing space. Apple had the TSMC connection. Apple had the raw transistor throughput to basically be NVIDIA and then some. And they fumbled it.

They had Siri before everyone else. They had audio interaction. All of these various forms of interaction, and Apple's software fumbled the ball. I can only hope that under John Ternus, Apple is going to decide it really does want what Tim Cook—and before Tim, Steve Jobs—used to always say: hardware plus software plus services.

It seems to me that Apple's hardware, at least on the consumer side, is incredible. Its services are also not that bad. But they're still fumbling and infamously underutilizing their amazing hardware with their software.

John, if you're listening and you're looking for guidance on what Apple should be doing in this new era under your leadership, please, please, please just take your amazing hardware and unleash it with much better software that integrates AI natively.

So, yes, hire Alex as a consultant first and foremost.

I don't need it.

Well, do it. You'll do it because you need it, not because of Alex. Maybe for charity.

But also, the world needs it. Apple is the one brand that you can trust with your information. It's the one and only brand. And here we are in the age of AI. Everyone's confused. It's incredibly powerful. It would change the world.

Such a different world if Steve Jobs had not made his stupid health decisions and not treated his cancer. You know, Dave, I met Steve once. I had a one-word conversation with him.

So, I was having an XPRIZE meeting. Larry Page, co-founder of Google, was on my board, and he hosted us for an evening dinner. Laurene Powell Jobs, his wife, was there, and Steve was in the back of the room.

We were talking about the latest XPRIZEs and the ones that had just been won. Afterward, I had to go meet my hero over here. Steve was in the back of the room with his arms folded the entire time like this, looking pissed off.

I walked over, introduced myself, and said, "Steve, it's a real pleasure to meet you." He barely nodded his head. Then I said, "You don't want to be here, do you?" And he goes, "Nope." That was the entire conversation.

I've got a story. A couple of months before the big iPhone announcement in 2007, Jerry Yang put together an off-site—a full-day off-site—with all of Yahoo's vice presidents. Steve Jobs came in to give a keynote, right?

At a handshake hello, Steve Jobs looked at the audience. There were about 300 people in the room, right? He said, "You have 300 vice presidents. I think I see the problem with Yahoo right there."

The fact that he had that much organizational overhead—he just couldn't cope with that. My claim to fame is that I have a whole bunch of MacBooks signed by Wozniak.

Oh, wow.

Because I did a 90-minute debate with him on stage a few years ago. I had him sign about 4 or 5 of them, guarding them jealously.

How old are they?

Yeah, that one was a few years old. I had him sign about 4 or 5 of them, guarding them jealously.

Yeah, you've got to preserve those.

You've got to preserve those. I gave a couple away to my community because they were so thrilled with that episode. It was a great conversation.

Steve Wozniak is famous for saying that all of this technology gives us the opportunity to tinker and just play around. That's the heart of all creativity: just playing around with technology to see what comes up. He was really doubling down on that. It was a great conversation, but that memory of Steve Jobs looking at 300 vice presidents and saying, "I think I see what's wrong with your company," has never left me.

Oh, my God.

It's funny. The very first share of stock I ever bought was Apple. We tend to overthink things. All I had to do was just hold it.

8. Nuclear, Solar, and the AI Energy Race

I should still have that stock today. I'm going to move us into the world of abundance.

Two stories on energy and one on water. Again, our mission here is to deliver to you data-driven optimism about why the world is getting better. Let's hit energy first. Again, 2 stories here. The first story is about uranium enrichment to meet the growing needs of our advanced Gen IV fission plants and the coming SMR, or small modular reactor, power plants.

Historically in the US, conventional reactor fuel in the form of uranium 238 has been enriched to about 3% to 5%. This week, a commercial startup called Actonide said it had demonstrated the ability to enrich natural uranium into high-assay low-enriched uranium, or HALEU, up to a purity of 15.38%—5 times the traditional purity previously achieved by the government labs. So, a startup is enriching uranium to this level, and that's, for me, a remarkable story.

Every SMR company—and we had that incredible podcast with Ramez Naam on the future of energy. If you haven't seen it, please go see it. Every SMR company says its reactors will be coming online in 2030, but none of them have the fuel, and this could well be the solution. Alex, any thoughts on this story?

Yeah, a number of thoughts. I think something went wrong right after World War II. That's my best guess. In the immediate aftermath of World War II, so 80 years ago, the Atomic Energy Commission was eventually formed in the wake of the Manhattan Project.

I think something just got screwed up with regard to how nuclear physics and advanced physics in general were handled in the post-World War II era. The government, probably because of the bomb, granted itself, under the guise of mixed civilian-military leadership, the Atomic Energy Commission. I think something governance-wise got fumbled, and 80 years after the end of World War II, we're finally starting to shake off the quasi-governmental monopoly on nuclear physics.

I suspect there has been so much progress in an alternative history, sort of a For All Mankind-style history, where, if in the aftermath of World War II the Atomic Energy Commission had either been organized differently or we had just had a totally different governance structure for advanced physics after the war, we would have advanced so much more quickly. I think startups like Actonide that are producing HALEU are the tip of the iceberg for all of the economic growth and technological growth that can now be unleashed, 80 years after World War II. All of these shackles that we put on our own society, perhaps with the best of intentions, are finally slowly atrophying away.

And just a narrow point on the technology here—

Please.

This is nonobvious in many cases: Uranium enrichment in this country, to the extent we do it at all and aren't just importing enriched uranium from, say, Russia, often uses Manhattan Project-level technology that has not been materially improved since World War II. It's using uranium hexafluoride gas and centrifuges, when we have much better technology.

There are alternative ways to enrich uranium. For example, the term is a calutron. This was technology that we've known in principle how to make since the Manhattan Project. Rather than using centrifuges, spinning uranium compounds in gaseous form and then later having to compress the gaseous uranium molecules down to solid pellets or other solid forms that could be used in reactors, we could instead be using magnets, electromagnetics, vacuum systems, and power-electronics controls.

We have 80 years of advances in power electronics that, by and large, are not being used to enrich uranium, either because we're scared of doing it or because we've handcuffed ourselves in terms of our ability to do it. I think we should expect to see, now that the shackles are starting to come off in a limited fashion from the post-World War II era, 80 years of chip advances and Moore's law advances finally being applied back to this basically World War II-era technology. I think it'll create an energy boom.

Well, I hope so. I think, from a venture capitalist point of view, you would never in a million years have invested in this when the regulatory cost of getting a nuclear reactor up and running is going to be 1 billion dollars anyway. So you can see how the dysfunction ripples through. You start with government dysfunction, and then the venture capitalists are like, "Well, given that that's dysfunctional, I won't invest in these innovations." So the whole value chain just doesn't go anywhere.

But now the SMR companies are being funded by venture capitalists, and they're going public. Let's jump a little bit to solar. I'm going to show this chart here. This is a chart that comes from Ember, the global energy think tank. They shared this data showing China's increasing energy generation, both in solar and in nuclear.

Monthly solar generation in China has exploded 8-fold—800%—over the past 6 years, growing from 20 terowatts per month in 2020 to about 160 terowatts per month in mid 2026. At the same time, nuclear power in China has increased by at least 50%, while US nuclear power generation has basically remained flat.

We need to remember that the AI race is not just about chips. It's also about the cost of energy to run those chips. The US is building its whole data-center power generation system based on natural gas, which is like $6 per million British thermal units, while China is basically getting this for near-zero marginal cost from solar.

The abundance story here is that it is possible to create a massive amount of energy from solar. The numbers, if you talk to your local model, are that there is 8,000 times more energy hitting the surface of the Earth than we consume as a species. So being able to double, triple, or 10 times that is well within reach. Salim, do you want to jump in first?

A bunch of things to say here. First of all, if you look at the difference between that red and blue curve, it's completely a copy of our old linear-versus-exponential curves, right? Linear is like this, and exponential curves go up, and we're on the wrong side of that equation. This is linear intuition confronting an exponential curve.

Solar keeps getting underestimated because people look at installed capacity today rather than the vertical slope of the deployment curve. And that's the nightmare. People can't get their heads around this. This is such an old, tired story that we should just get past it, because China's advantage is not a magical solar panel. It's the whole industrial system optimized around manufacturing, the supply chain, permitting, and deployment.

In the US, the big challenge is interconnection with the grid, transmission, and permitting speed, right? So you need all these different things—you need nuclear, you need solar, you need all of this. But battery technology makes this a killer opportunity because you can add a solar panel today rather than wait 15 years for some gigawatt-scale energy project.

This is a massive opportunity that's being left behind. If I were the US, I'd put a Manhattan-type project into perovskites and new solar capabilities. Photovoltaics have some huge opportunities there, and when you have energy abundance, as I think we've all pointed out, that becomes a multiplier in every other exponential technology. So this is such a no-brainer, and it really is very dismaying to see the GDP challenge here.

The GDP of a country is directly proportional to the amount of energy it creates. And, Alex, the point you made about nuclear being frozen—it's very clear here. We have not changed nuclear in decades. We've still been producing—

A half century, something. Again, I think in the fullness of time, maybe 10 years from now, we'll have a better understanding of what happened during the 20th century that got screwed up after World War II. But I think several things on the governance side, especially as they relate to nuclear physics, went wrong just in the wake of World War II. We went on the wrong path, and my hope is that it gets corrected.

Yeah. Dave, do you want to make a comment on this one?

Well, I'll only make one comment, which is, on that chart, the US is way behind in solar. All those US panels are made in China.

9. Rainmaker Drones and the Future of Water

It's much worse than it looks.

Yeah. All right. Our next story is about water. The weather-modification startup called Rainmaker says that 10 drones that flew over Alaska's Kenai Peninsula generated an estimated 19 million gallons of additional rainfall in just 3 hours, using what they call glaciogenic cloud seeding.

Ultimately, weather modification is no longer theoretical. It's here. Applications for this kind of technology include drought relief, agriculture, wildfire prevention, and water security. Just for reference, you can look it up: There are 3.4 quadrillion gallons of water in the atmosphere. Atmospheric water capture is now a thing, whether it's making it in the clouds or capturing it on the ground.

So this is the abundance thesis that predicts that water becomes massively abundant when the technology to create it becomes cheap. And 10 drones creating 19 million gallons of water in 3 hours is pretty damn cheap. Alex—

And the drone part—the UAV part—I think is especially novel. So imagine a near-term future where we achieve our weather-control grid. Not necessarily. When I go through the mental exercise of how I would build a global weather-control system, I think the easiest way to do it is probably with reflectors, either terrestrially—cheaper—or in LEO, a little bit more expensive.

But this points the way, I think, to a third path for global weather control, which is cloud seeding. Imagine a swarm, a fleet of drones in the style of Rainmaker drones that are continuously in flight, probably solar-powered so they can be permanently aloft, and are just dispensing, on an as-needed basis but under the control of some centralized AI algorithm, the particulates—the nuclei needed for either snow or rain—all informed by a global weather AI model. In principle, I think, with some further development of Rainmaker technology and similar technologies, we could have total weather control, and that would be, from a natural-disaster perspective, incredible. Imagine being able to divert hurricanes—

Or disperse hurricanes before they form.

Yeah. Disperse them with precipitation on the water instead of on land, or address tornadoes and other extreme weather events. Not to mention giving good weather to the places that want it and are willing to trade it. We could have a global weather trade where regions trade with each other for precipitation. There are so many possibilities that are unlocked by the confluence of AI, drones, and weather modification. And now, finally, we're about to start this era of geoengineering, I think.

Love it. I can see the business model for Rainmaker. It's like different cities bid on where the drones go. Rain arbitrage. We could have a global market for rain. Salim—

Rain becomes programmable. For me, the big breakthrough is not that they made rain—we've been able to do that for a while now—but the fact that they could prove how much additional rain they made. I think that is really interesting because it then unlocks a huge amount of stuff.

Weather—the challenge is that weather typically doesn't respect political boundaries. Trying to figure out if you're downstream of somebody that seeded a bunch of clouds, and now you're not getting the rain while they're getting it, is going to create some interesting tensions around that. But this is exactly the type of abundance technology, right? It allows you to scale something really powerful. And I think there's a wonderful future. The business models around geoengineering, as Alex puts it, are pretty rampant.

You could do some really interesting stuff. I mean, crop management—all sorts of things become capable. It's reminding me of one of the more creative Singularity solutions that we ever saw at Singularity University. One of the summer students, I think his name was Simon Daniels, had this idea of putting platforms in the ocean that would spray up water, take ocean water, and spray it up to block the sun from hitting Earth and slow down global warming that way. His business model, too, was to do it along flight paths, where you could spray up water in the shape of a Nike logo and sell that advertising.

You'd do it in the shape of certain patterns that pay for the whole system; at the same time, you lower the temperature in a controllable way. There are lots of people who say, “Oh, my God, geoengineering is a bad thing. We don't know the aftereffects.” The counterpoint is, well, we've already been geoengineering the world for 100 years in a horrible, uncontrolled, unmeasurable way, and it's causing massive damage. Look at the disaster in Nepal that's happening right now. It's a complete tragedy. My heart goes out to all—

—and nothing anybody could do in that sense. You could avoid a lot of that, and so I think it's incumbent upon us to use technology to navigate this stuff because you do it consciously rather than unconsciously, and that's the powerful point.

All right, let's move into the world of health. A couple of stories here. First off, this week the FDA approved a drug called daraxonrasib. It's the first RAS inhibitor for metastatic pancreatic adenocarcinoma, attacking the RAS family of proteins that drives tumor growth in most patients with the disease.

So pancreatic cancer is really considered almost a death sentence. It's the most lethal, least treatable cancer for humans in history. In previously treated metastatic cancer, daraxonrasib nearly doubled median survival from 6.7 to 13.2 months. It also tripled tumor response rates from 11% to 32%.

And while this isn't a cure, and just getting an extra 7 months doesn't sound like a lot, that's an additional 7 months for science to continue to make more breakthroughs. I mean, that's the way I think about this: longevity escape velocity. Your goal is to live long enough to get a breakthrough to the other side. It's also worth noting that RAS mutations drive about 30% of all human cancers, so a drug that inhibits these proteins may also be useful for other cancers as well. We've heard this said so many times by Dario, by Demis, and by many others: the goal is to cure all these diseases, including cancer, in this next decade. Alex, do you have a take on this one?

Yeah. A couple of items. One, I point out that, remarkably, this is a daily oral tablet. Pretty incredible. Secondly, I'd point out that I was doing some research on the full lineage, as it were, involved in the development of this drug.

I think this is actually a finger to the wind. I think this is probably, hopefully, one of the last-generation pancreatic cancer medications that almost precedes the AI wave. I would expect lots of new targets and lots of new approaches to emerge from both the AlphaFold 3-style protein-folding and structural-biology breakthroughs, as well as virtual cell models. As far as I can tell, daraxonrasib used neither of these in its design. It was good old-fashioned drug design, but I think we're about to see a whole wave of AI-driven drugs that take it from its current response rate to near 100% using AI. I think that's to be celebrated.

10. Regrowing Tooth Enamel

I also want to celebrate the FDA for approving this relatively quickly. The FDA noted that they approved this, I think, 60 or 90 days—something like this—prior to the due date, which, given historic FDA speeds, is actually pretty remarkable speed. We're seeing a much more responsive FDA that now starts to care about Bayesian versus frequentist statistics, that cares about INDs quite a bit more, it seems, and that's interested in lowering the number of clinical trials in exemplary cases. Hopefully, this is, I think, a baby step towards a future where AI solves 5,000 diseases and then we have a regulatory apparatus in the form of an FDA that is able to metabolize all of those innovations and turn them into cures.

Amen. Amen to that. Let me move to our second story in the health corner. This was published in Nature Communications, and it comes from an international research team at the University of Nottingham, who developed a biomimetic gel capable of repairing and regrowing damaged tooth enamel. This fluoride-free treatment uses lab-engineered proteins that mimic the natural processes of enamel formation during infancy.

As you might know, tooth enamel is the hardest substance in the human body, and it doesn't regenerate. Until now, the only thing you could do was replace enamel with a synthetic filling that would eventually fail. The biomimetic gel is an elastin-like protein matrix designed to mimic the natural protein scaffolding that organizes enamel during tooth development.

The treatment didn't just make something that looks like enamel. It regenerated layers and restored hardness, stiffness, water resistance, friction properties, and resistance to brushing, chewing, and acid exposure in labs. The gel acts like a scaffolding that pulls calcium and phosphate from the surrounding environment and directs them into properly organized minerals rather than merely depositing an amorphous coating.

It’s true enamel. For me, this is huge. Oral disease affects half the population. It’s estimated at $500 billion per year in associated costs.

To put this in perspective, this is still ex vivo. It’s still done using extracted human teeth, effectively in the test tube or petri dish. But it’s going to be moving to human testing very quickly. As someone who’s had poor dental health and poor dental genetics all my life, I couldn’t be more excited about this, especially for my kids. Salim, do you want to jump on this?

Just fantastic. For me, the big thing is showing that the future of medicine is not repairing broken bits. It’s convincing the body to rebuild them and finding the mechanisms and pathways. This is really powerful stuff. It’ll completely change dentistry.

Yeah.

People undervalue their oral health so much, right? The mouth is a direct corridor to the brain and to the heart. You need to be careful about your oral health. Alex—

Regenerative medicine has been the future. We’ve been waiting for stuff like this to happen for a long time. It’s great to see it finally come into being.

Alex, did you dig into this story, too?

I did. I think these results actually started bubbling up last year. I’m incredibly interested in materials engineering of the human body, including human dental enamel.

I think this was an AMA question that we got in the last podcast episode. People were asking, “Where’s all the innovation?” I think the story—which, again, I first started seeing last year—is just one data point in a larger manifold of applying good old-fashioned materials science to the problem of human teeth.

If you can help shape and engineer the pathways through which tooth enamel grows, I have to think that similar biomaterial discoveries and inventions will also be helpful for bone regrowth and for a variety of other key biomaterials in regenerative medicine.

Again, looking at the research behind this, I saw approximately no AI. That’s always my sniff test: how and if AI was used in this research. If it wasn’t used, as appears to be the case here, and there was approximately no AI that I could see, then take this and multiply it by many orders of magnitude. That’s the sort of progress I’d expect to see in a few years in this area.

Dave, how’s your oral health?

It’s fine. But I get really excited about that Moderna story we had recently, where they had a melanoma vaccine that works really well. The stock went way up, but what was exciting is that it’s a platform that applies to virtually any other cancer.

When Alex says, “Yeah, these weren’t done with AI,” I’m like, “Well, they’re really cool, but they’re one-offs.” I can’t wait for the repeatable version: I can grow a tooth, I can grow an arm, I can grow an ear, I can grow anything now. Then you get into true scalable regenerative medicine. I feel like that’s very, very soon.

And by the way, work on limb regrowth is making amazing progress.

Yeah. All right.

I think, Dave, you’re also touching on an interesting, almost singularity-flavored technological deflation story, which is: if there’s no AI in it, to first order, what’s the point? Either infuse it with AI and make it scalable so that you can solve everything—scale it up by 1,000×—or just don’t do it and do something else that’s higher leverage.

Not to discourage people from regenerative medicine—

Sell your data to Dennis, and Dennis will AI what you just invented.

That’s perfect.

11. The Trillion-Dollar Lunar Economy

All right, finally, let’s jump into space and robotics stories. The first 2 stories are from the Elon universe. Deloitte reported this week about the potential lunar economy. They put a financial figure of $566 billion in cumulative economic value by 2050. I couldn’t expect a lowball estimate than that number from Deloitte.

The biggest early markets are infrastructure costs, transportation, power, comms, mobility, construction, and life support. They cite SpaceX as the biggest winner because of the central role that Starship will play. Clearly, they’re not taking into account Elon manufacturing Starlink on the moon to fuel Earth’s—

Disaphab pedophab is MIA.

Yeah. To put the numbers on it, that $500 billion lunar economy may sound large, but it’s a minuscule estimate. Elon previously stated that his plans for Starlink started with about 100 to 200 gigawatts per year of orbital AI compute launched from Earth, and then moved to manufacturing AI satellites on the moon for launch with a mass driver.

He estimates that we could yield as much as 100 terawatts per year of orbital compute based on lunar production, which would generate—and wait for it—an absurdly large amount of lunar economy. The estimate, if you’re building 100 terawatts per year of orbital compute, is that the economy would be $2 quadrillion per year of gross revenue.

I mean, I don’t know. Maybe planet Earth is too small.

Two quadrillion dollars. Now we’re talking about a good economy.

But, Salim, you deal with the Deloittes and McKinseys, and these guys are so conservative in their numbers all the time.

Crazy. Yeah, because a lot of them are accountants, so they’re scared of being called out for anything vaguely approaching a real estimate. They want to make sure their numbers are defensible to the nth degree, so there’s no speculation at all. By definition, you’re speculating, so you might as well speculate to some level of honesty.

This is the same type of pattern we’ve seen before. One of the Big Four—I won’t say which one, to protect the innocent—put out an agentic AI report saying this is the future of AI agents, et cetera. I excitedly read it, thinking I’d get some glimmerings that would give me a sense of it. But when I looked deeper into it, the data they were using was 2.5 years old.

How, with 2.5-year-old data, are you going to come up with anything about the agentic world when everything happened in the last 6 months? They have so many checks and balances and internal review systems that dilute everybody’s thinking: “Careful, you can’t say that.”

By the time it comes out, it’s been whitewashed so many times that it’s meaningless. It’s lawyers and cross-reference checkers of even more cautious people, and then nothing realistic gets put out. In general, I say take any of their estimates and multiply them by 10 or 20 times.

Dave, what’s your thought?

We had the Deloitte AI team in the office the other day. They’re very, very good, actually. But they were also saying the same thing about the accounting side. So, Salim, you’re confirmed from inside Deloitte.

Yeah, quadrillions of dollars. That’s what Elon was talking about. I believe it because the gravity well is the key. Peter, you talk about this a lot, but everyone’s like, “Why would I bother manufacturing that on the moon? What’s the point? I can just make it here.” No, no—it’s the gravity well.

You start building power plants based just on lunar minerals, and all of a sudden you’re like, “This is so much more efficient.” Plus, there’s no worry about pollution and stuff like that.

Yeah, lunar regolith is silicon. What’s that used for? Oh, okay: power generation, aluminum, nickel, iron, and oxygen. It’s like the perfect materials for building Dyson swarms. Alex, I’ve got to feed it to you to uplevel this vision here.

Well, the moon may not be a harsh mistress. She may be a wealthy mistress at this point. I do think the pedophab is the killer app for the moon. I don’t think it’s tourism. I don’t think it’s transport infrastructure or comms.

The Apollo program was, infamously, arguably a successor to the Manhattan Project. To the extent that the full history of the Apollo program is already fully understood and in the public record, I think it was a relative failure.

I would love to see the moon reach its full economic potential, and I just don’t think that’s going to happen any way other than through full private development. Right now, it does appear that the critical path is mining chips and fabbing data centers on the moon.

It’s not rods from God. It’s not deploying NASA astronauts to the moon. That’s all well and good. Friend of the pod Jared is doing an amazing job of reinvigorating the case on the government side for fission reactors on the moon and all of that.

But I think, in the fullness of time, even that ends up being a rounding error compared to data center deployment and mining on the moon. Unless 2001: A Space Odyssey happens and we discover some monoliths on the moon, in which case that’s the killer app.

It’s really interesting to game it out, because when you look at the fab construction process, a lot of the components in there are just routine aluminum and routine construction. There are a few really critical pieces that you can still make on Earth and launch, and it’ll be just a tiny fraction of the total cost.

Once you start manufacturing chips and solar panels on the moon—

I mean, it’s just exponential takeoff from there. The technology is not that far away.

The problem is that these guys are not looking at what the lunar economy is today. You can’t ask what the lunar economy is today; you have to say what industries become unlocked when the cost of moving mass off Earth drops by 1 or 2 orders of magnitude. That’s the imagination gap.

Yeah. I just want to give a shout-out to Jared Isaacman, who’s a friend of the pod. What an extraordinary administrator. I could not be more proud. I said it during our podcast with him. I’ve known, I think, all the administrators pretty well over the last 40 years, and he is by far the best, most eloquent, most visionary, and most business-minded. He is an accelerationist, and if anybody’s going to support humanity moving beyond the bounds of Earth, it’s him.

12. Tesla Robotaxis and the Future of Transportation

Peter, every time the CCP sponsors an influence op to deter American terrestrial data centers, the lunar pedophab gets its wings.

Yeah. All right. Our second Elon story comes from Tesla. The company announced that it has expanded its rollout of Cybercabs in Texas. It just registered 79 Model Ys for the robotaxi service, and this is starting to look less like an experiment and more like a real fleet. In Las Vegas alone, Tesla recently said it’s going to deploy 2,500 vehicles in the next year. Tesla’s official long-term milestone is to reach 1 million robotaxis in commercial operation as soon as possible. If you do the estimates, that’s probably about $100 billion in rideshare revenue for Tesla.

Let’s take a look at a quick video here to remind people how cool the Cybercab is. I have to say, I love Google, but Waymo’s new vehicle is kind of clunky. This is pretty beautiful. This is Elon at the original Cybercab rollout. I love those doors.

Way cooler than I thought.

Yeah, those are 3 components or 4 components. Pretty cool.

Yeah, such a great design.

It’s a beautiful design. In gold, it’s perfect.

It really is. At the Gigafactory, remember they had that one that was all exploded out so you could see all the parts inside? It’s just a beautiful design, and it doesn’t have that many moving parts, so you can see how they’re going to manufacture them at an incredible scale.

Yeah.

Just from a consumer point of view, all my friends who are driving the new full self-driving Model Ys are like, “Oh my God, this is life-changing. I literally go anywhere with FSD.” The waitlist right now is about 3 months for a Model Y, but 3 months from today, the fully engaging, you-want-to-talk-to-it-all-the-time AI will be everywhere. You might want to get on that waitlist because you’re going to want to be talking to your AI and continuing your dialogue with your

Robot.

You know, it’s going to be amazing. And if there are entrepreneurs out there, when the Cybercab comes out, buy 100 of them, put them out to work, and we’ll have a fleet of Cybercabs generating revenue for you. I think that’s going to be an incredible business.

I will say, I love my Tesla Model Y 2026 Juniper Launch Series Edition. It is life-changing. The FSD is incredible, and the manufacturing quality of the car is incredible. I can’t wait for Cybercabs and robotaxis in general to come to Boston. I’ll beat the drum once more for my broken Waymo theory: if a city can’t accommodate robotaxis, then it is not equipped for the singularity. Hopefully, someone from the city of Boston is listening to this, scratching their head, and wondering what it is doing with Boston and why Boston doesn’t have proper Waymo support. If you’re scratching your head and you’re in a position of authority, call Dave or call me. We’d love to figure this out.

Yeah. The same thing happened when Uber was made illegal in France. I was like, “Okay, if you’re an entrepreneur, don’t move to France.” These are telltale signs of which cities are on the singularity curve and which are not.

Yeah.

See.

I’m actually having to sell my beloved 10-year-old Tesla Model S to get a Model Y because I need the FSD. I’ll send a photograph of it for our next pod, just so everybody can see. I totally love my car.

This is a big deal.

I think there are 3 things here. One is that this changes from purchasing a car to autonomous revenue miles. What’s the cost per autonomous revenue mile? If you can do it at a decent, safety-adjusted cost, this is actually massive.

The fleet growth will lower wait times, which will improve the user experience, which will increase rides. It generates more data, which will create that inner loop that Alex is talking about. I think there are some really amazing downstream benefits from all of this. It totally changes insurance, cleaning, and charging. You’ll turn gas stations into autonomous car-cleaning depots. That kind of thing will start to happen.

For me, the big one is that I made a bet. I got up on stage 10 years ago and said, “My kid will never get a driver’s license.” I’ve got just about a year for this Cybercab to roll out before he may get one anyway, just to get away from his parents, but he wouldn’t have to. Then I’ll be proven correct. It’s important not to lose your credibility as a futurist.

Yeah. The abundance story here is that the poorest people on the planet will be driven around by electric autonomous vehicles.

Yes. It will be the cheapest mode of transportation.

And think about the elderly, right, and people who can’t get around easily. The market suddenly becomes so magical, delivering huge capability to segments of the population that can’t do it today. Huge.

Yeah. I’m going to put up this image here. I love it, just comparing Cybercab and Waymo. I have to say, I don’t know how Waymo is going to compete when Cybercab rolls out. First of all, Cybercab has 8 cameras. Elon made the point that if humans can drive with 1 or 2 eyes, you should be able to have a car do it with just cameras, compared to the lidar, radar, and 13 cameras that Waymo has. The point with the brains on this image is that Cybercab is massively AI-driven. I could not agree more.

It’s going to be a big fight over this. By the way, didn’t Tesla acknowledge that cameras aren’t enough in this?

No, I don’t think so. Tesla does use ultrasound for short range, but everyone uses ultrasound. It does not use lidar.

I think there’s an interesting, maybe speaking prospectively—not financial advice—bet, insofar as Luminar, the lidar provider, declared bankruptcy half a year ago at this point. If I had to bet, again, not investment advice, I think the market abhors monopolies, and there’s probably room for at least 5 or 10 competing providers. There will be many of these around, competing, beating themselves up, and lowering the price across the board, and we all win.

Yeah. I introduced Austin Russell, the CEO of Luminar, to Elon at a party before it went out of business, and I watched them duke it out over whether lidar was important. Anyway, it was a fun show, and Elon was like, “No way. Not interested. Not going to go there.”

13. AI Companions Get a Physical Form

All right, our final story. It’s a little bit of a bizarre one, but expected in the world of robotics. A company called Somnia Lab has built a robot for intimacy, and the internet has lost its mind. It’s called Model L—maybe L for love. It’s a 5-foot-9-inch, 44-pound robot with 24 degrees of freedom, warmth simulation, touch response, and preference memory, designed for human-robot sexual encounters.

Okay, it’s finally here, guys. Real humans wear motion-capture suits to record 165 intimate actions that were fed into the model. Somnia Lab is taking $300 deposits for delivery in late 2027. What makes Model L more interesting than a sophisticated sex doll is the attempt to combine physical embodiment, persistent AI memory, and personalization. Somnia says the robot is intended to remember conversations and preferences across encounters, learn a user’s rhythms, and learn their communication style.

The body is made from lightweight carbon fiber, with modular skin and structural components designed to be serviced and upgraded. Customers will be able to customize the face, makeup, eyes, skin appearance, and body silhouette. I guess this is a story around the abundance of intimacy. Let’s take a look at this.

This was inevitable. Peter, I’m waiting for you to say, “Fully functional, programmed in multiple techniques.”

No, I’m not going to go there.

Not yet.

Let’s watch this video. This is a video of the motion capture being done to train the AI models here, with cheesy music to boot.

We’ve hit a new high here. This is incredible.

One has to admire the low degree of freedom. That was a pretty simple face. I think what we were seeing was a human in a motion-capture suit, right?

It was.

That was low quality in every way, from the music to the video to everything.

A robot porn video. Oh my God. You combine this with the conversation we had earlier about AI girlfriends and boyfriends, and the human relationship is cooked.

You think? I mean, I go back and forth. Is there going to be an app store for mocap movements, or is this just going to be pre-training or post-training data for a foundation model that knows about everything?

I want to see the job ad for that one.

I have a couple of comments on this one.

Please.

The first thing is, this type of embodied AI has a pretty difficult data problem, right? Because you can scrape the internet for text, but there’s no internet-scale data set of physical human interaction at all.

Are you talking about the internet? Are you looking at—hold on, hold on, hold on—

In terms of physical movement, what I love more than anything here is the economic approach, right? You do crowdfunding of the early thing, get early customers, and then use customer demand to finance the hardware and generate the interaction data as you scale the product. This is like a pretty classic ExO flywheel. You got a new career out of it. You got it all inside.

Now I’ll give a hot take on the hot sheets: I don’t actually think this is that promising as an approach. There is an abundance of evidence at this point that you can do a great job either pre-training or post-training robots for—call them economically productive applications—just from watching videos, without any need for human mocap at all. So, if I had to guess, my guess is this is like a very 2026 story, with people mocapping sex maneuvers manually, but it doesn’t scale nearly as well as just pre-training off of internet video, is my guess.

Okay, circle of trust here. How many have you ordered, guys?

I won’t tell anybody. Oh my God.

We need guests and panelists who are in their 20s for that.

Oh my God. Dave, any closing thoughts on this topic?

None whatsoever, actually, as it turns out. First time in my life. Nothing to add. Sorry. Speechless.

My God. All right, everybody. I want to say that’s a wrap, but we do have a closing video.

And we’ll see you guys in 4 weeks to the day.

If you have outro music videos, please send them to mediadmandis.com. We love sharing them. We love the creativity of our community. We'll see you guys in four weeks to the day at Moonshots Live in LA, the evening of September 24th and all day on September 25th. You can go to moonshots.com to apply. It is by application, and it's going to be an amazing community for you to network with and spend the day with all five of us. The Moonshots quintet will be there. You get a chance to grill Alex, Salim, and of course Dave. It's going to be awesome. All right, let’s play this video by Ellen Vamman. Thank you, Ellen, for your submission. Hey, nice heroic music, gentlemen.

It’s the Lord of the Rings in space mashup I never knew I wanted.

That was fun, guys. Really loved it. A lot of fascinating stories and definitely one bizarre one.

Until we see you again twice next week for another episode recording.

Be well.

Can’t wait. We’ll try and sleep in the middle. God help us all.

Live long and prosper, as they say.

Likewise. All right. Take care, folks.