← Moonshots with Peter Diamandis
Moonshots with Peter Diamandis
US vs. China: Why Trust Will Win the AI Race | GPT-5.2 & Anthropic IPO w/ Emad Mostaque, Salim Ismail, Dave Blundin & Alexander Wissner-Gross | EP #214
02:00:43
Loading summary
Transcript477 lines
- [00:00]
A
China is accelerating its push to become independent of Nvidia with CameraCon planning to triple output to a half a million accelerators in 2026.
- [00:12]
B
I fully expect, as I've mentioned on the pod in the past, that we're.
- [00:16]
C
Just going to see having an open source model that you can test as you're developing. It really closes the feedback loop a lot more aggressively here. And most of the Chinese models are optimizing around this very sparse MOE type structure with deep SEQ and similar arches with muon kind of acceleration. So you're getting towards one architecture that they can just engineer and output industrially and who's the best at industrial manufacture.
- [00:42]
D
The challenge with China is people don't trust it.
- [00:45]
B
I think we're going to see a Cambrian explosion, no pun intended, of architectures coming out of China now that China has been effectively decoupled from the US tech stack.
- [00:56]
A
So stepping up a level, is this good for humanity or not?
- [00:59]
B
I think the big question is, what does the finish line look like?
- [01:03]
A
Now, that's a moonshot, ladies and gentlemen.
- [01:08]
Anyway, I'll get us going in a second, but what a fun, fun day. Yesterday we were all in Seattle. Salim, we missed you. You were in Brazil. Good morning. You arrived from Brazil, what, at 6:00am this morning?
- [01:21]
D
Yes. And got an hour and a half of sleep. And so I'm foggy as f. All.
- [01:27]
A
Right, well, hey, that means we all have a shot at you, but I'm.
- [01:31]
E
Just back from Rome and Ahmad's in London. Let's kick off with the covering the world we've got going there.
- [01:37]
A
Yeah. Fantastic.
- [01:38]
E
And Alex is just back from San Diego.
- [01:40]
D
Yeah.
- [01:41]
A
And God knows.
- [01:42]
D
Yeah.
- [01:42]
C
I just got back from Vietnam and Japan yesterday.
- [01:45]
A
Look at us Globetrotting gentlemen. It's like, you know, no time for sleep. It really is. I mean, I feel like we're going 24 7. I don't know about you guys, but.
- [01:55]
D
Well, look, if you want to transform the world, you have to go out into the world, Right? And I think that's what we're all doing.
- [02:01]
E
I think that's a good opener, too, because, Peter, you just got back late last night. Last night from Seattle, and that'll come out in a couple days.
- [02:07]
A
Yeah.
- [02:07]
E
If we just mention the whole world that we've covered in the last week.
- [02:10]
A
That'S pretty globetrotting, but a lot going on. Shall we jump in?
- [02:17]
E
Yeah.
- [02:18]
A
Is that enthusiastic? Yes. Everybody.
- [02:20]
D
Yeah.
- [02:20]
E
Let's get in time.
- [02:22]
B
Make it happen. Engage.
- [02:23]
D
Trying to find my neocortex.
- [02:26]
A
It's there someplace. Don't worry about it. It'll show up. All right, everybody, welcome to Moonshots. This is the conversation that's changing the world. Hopefully we can help you get ready for the future. And this is the news that if you're not watching the Crisis News Network and you have time to watch Moonshots, we hope we'll deliver to you sort of what's going on. What's happened the last week in AI robotics, data centers, energy. It's a lot here with all of my incredible moonshot mates, we have a. Let's see a fivefold increase in capabilities here today because not only do we have AWG, we have EMOD as well. Salim DB2. It's going to be amazing. All right, going to jump into our first stories. We're going to start with where AWG was last week. Neurips 2025. So, Alex, this is you. Yeah, yeah.
- [03:22]
B
Neurips this year was a bonanza. I've called it in past Woodstock for AI. A few observations. It had more than 29,000 registrants this year, which is almost 50% increase over last year. It was enormous. The Alibaba, the Chinese lab had 146 papers accepted, including Best Paper award. Anecdotally, the language that I heard the most in the hall was Mandarin. There was a sense that the Frontier Labs have all the resources, the academic labs do not. And the Frontier Labs were at Neurips mostly to recruit academics. American. Again, this is sort of sense of the conference, if you will. The Frontier Labs and the Frontier Research coming out of the labs has largely gone dark. So what was being shown on the research end was largely coming from academic work resources or academically resourced labs that are lacking the resources. There was on the sidelines, even though Frontier Labs were there to recruit the publications and the oral presentations at this point, largely not coming from Frontier Labs, except for Chinese Frontier Labs, which is.
- [04:32]
A
So what does Neurips stand for, first of all? Let's start with that.
- [04:36]
B
So NeurIPS, formerly NIPS, stands for Neural Information Processing Systems. So it is the largest AI conference in the world. It's held once per year, every December. And it is where some of the most striking AI research historically has been published. It's also the place, the one time per year where all of the Frontier Labs are all under one roof. And you get really a sense of the pulse of the AI space just from being there. Some of the most interesting meetings happen in the hallways. Optimus Gen 3 humanoid robots there in force. You also see a sense of the. The Vibes the, the. The zeitgeist of AI right now. So you can see these are a bunch of photos and video I took from, from the conference. We've talked on the pod in past about how Chan Zuckerberg initiative is now pivoted to solving all disease with AI and that that solve everything mentality that math, science, engineering, medicine are just going to be solved imminently with AI was very much on the show floor to the point where it made banners. It's really, I think such a wonderful way to tap the zeitgeist of the space.
- [05:52]
A
Amazing. We should have.
- [05:53]
D
It's interesting to see that a lot of it is hardware and much more than you'd expect.
- [05:58]
B
There's a lot of hardware and the feeling of the moment is that robotics in particular is the next big thing after agents. I should also mention a lot of the attendees watch and are fans of Moonshots. And I think there would probably be a lot of interest if we were to do a recording in the future from Neurips or ICML or iclr.
- [06:19]
E
Well, it's a global thing. I don't know what those TLA is all over the world.
- [06:23]
It's kind of like the Olympics. It's in a different city all over the world every session. So it just happened to be in America this year. But where is it next year if we want to record?
- [06:32]
B
I don't think they've announced that yet.
- [06:34]
C
It's in San Diego again.
- [06:36]
E
Is it really back to back?
- [06:38]
A
That's easy. Neurips 2026 here we come. All right, let's move on.
- [06:42]
C
There's something interesting that I think I should point out. So iclr, which is another similar conference, the number of first author affiliated Chinese papers went from 9% in 2021 to 30% this year and the US number went from 52% down to 36%. I think we saw something similar with Neurips, but it'd be good to crunch those numbers.
- [07:05]
B
I think we can't emphasize that dynamic enough. And Imad, I'd be curious to hear if you have a different take on this. But my take on the floor was that American Frontier Labs have basically gone dark and are largely no longer publishing all of their internal results. Chinese labs continuing to publish in the same way that they're continuing to push open weight models. And so the gap that the research publication gap is being filled in part with Chinese labs.
- [07:36]
E
I got a question for you on that. So I know exactly why the US is going dark. Everybody working on these big frontier models is A former Googler. Not everyone, but almost all of them. And the billion dollar signing offers are a real problem for and we saw that yesterday at Microsoft too. It's just recruiting warfare. So Google has explicitly gone dark after being very open and publishing everything for years. So I know why that's happening. But why are the Chinese still being super open?
- [08:05]
C
Well, it's because the Chinese are backing open source now, aren't they? It's like deploy, use open source to make it more efficient and then that's how we'll win.
- [08:15]
D
That's how the models propagate.
- [08:17]
B
Yeah, strategically it's great differentiation if you have really strong American frontier models that are largely hidden behind APIs under the spirit of commodify or complement, release lots.
- [08:30]
A
Of open weight models and it's a land grab at this point.
- [08:34]
D
Right.
- [08:35]
A
If a lot of nations and entrepreneurs and companies are beginning to use the Chinese models, they have a foothold.
- [08:43]
B
It's also I think there's an integration angle. So if again, under the banner of commodify your complement a classic economic strategy, there is much more to AI than just the models. There's integration with society and the economy. So if there's strategically, if you're at a disadvantage on the model front, open weight, open, release all of the models and focus your attention, focus your economy on deeply integrating all of those open weight models and that becomes the competitive advantage.
- [09:13]
A
Every week, my team and I study the top 10 technology metatrends that will transform industries over the decade ahead. I cover trends ranging from humanoid robotics, AGI and quantum computing to transport energy, longevity and more. There's no flaw fluff, only the most important stuff that matters that impacts our lives, our companies and our careers. If you want me to share these meta trends with you, I write a newsletter twice a week, sending it out as a short 2 minute read via email. And if you want to discover the most important meta trends ten years before anyone else, this report's for you. Readers include founders and CEOs from the world's most disruptive companies and entrepreneurs building the world's most disruptive tech. It's not for you if you don't want to be informed about what's coming, why it matters, and how you can benefit from it. To subscribe for free, go to dashmandis.com metatrends to gain access to the trends 10 years before anyone else. All right, now back to this episode. All right, let's jump into Our first article is this one from Neurips. Google's titans and mirrors are helping AI have a long term memory. Google's Titan is a new architecture with deep neural long term memory that updates itself in in real time. Do you want to jump into this, Alex?
- [10:26]
C
Yeah.
- [10:27]
B
So.
- [10:29]
One of the many blockades to radical progress in terms of advancing AI models is obviously the context window size. It would be miraculous if we could have say a billion tokens in context or a trillion tokens in context. If we could have the entire web in context, or the entire human genome in context, imagine the reasoning powers we'd gain and all of the problems that we could solve. The problem is, as folks in the space have long used to motivate just about every recent paper on the archive, the quadratic complexity associated with increasing the number of tokens in the context. But it's really painful to increase a conventional vanilla transformer past a few million tokens in context. So we result to techniques like RAG retrieval, augmented generation and other techniques to try to effectively increase the amount of working memory, if you will, that an AI model can keep. So Titans and Miras, or Miras, these are Google's latest attempt to get past that bottleneck. And as we've talked about in the past on the pod, there are a variety of techniques, architectural techniques to try to break the context window limit, like recurrent neural networks, Popular example, they don't have any explicit context window limitations, but they forget. So the approach that Titans and Maurras propose is sort of biologically inspired, distinguishing between short term memory and long term memory. It's sort of ironic given that the attention mechanism itself that's powering basically the whole economy at this point was originally designed for differentiable long term memory. We found ourselves back there. The idea is to use surprise, some numerical metric of surprise, to decide what to commit to long term memory and what not to, and turns out scales really well without catastrophic forgetting.
- [12:29]
A
Just to give people a sense of this, the models historically, right, GPT 4 and 4.0 had about 128,000 tokens as a context window. Two million tokens, which we're talking about here, is about 3,000 pages of text, 16,000 novel. Sixteen novels, if you would. And you mentioned the human genome, which is 3.2 billion base pairs.
- [12:56]
That would fit 2 million tokens, only hits about 6% or 0.06% of the entire human genome. So will we actually get to a near infinite context window someday? Is there a strategy for getting there?
- [13:12]
B
Absolutely. And I think these papers and others, when Demis Hassabis talks about there being only one to two major AI advances at the level of transformers left before we achieve what he'd characterize as AGI. I think breaking the context window ceiling is probably at least half of one of those great advances and I think we will absolutely get there pretty soon.
- [13:36]
A
Imad, how do you think about this?
- [13:38]
C
Yeah, I mean I think I agree with Alex and also this fits with Google's hardware architecture. So the massive kind of toroidal TPUs that they're doing that can handle huge amounts of memory, gigabytes, probably terabytes of memory in one instance by making it more efficient, like from the graphs that we see on Titan and meras, it's everything else kind of just slopes straight down. This goes almost continuous as you scale without the also complexity overhead because you used to need almost exponential amounts of compute as well as memory to handle that increase. Being able to just capture everything in one go means that you don't need to store stuff to files anymore. Like you could have an entire picture of every email the organization has ever said just in the memory at once. So it can track everything and figure out the interconnections dynamically. And again, I think that's what helps you break through to that next level of performance and it's pretty exciting and I think a very logical approach that they've taken here with the surprisal element there as well right now before Transformers are a bit brute force to be honest.
- [14:43]
D
This reminds me of Ben Goertzel back around, I think 2010 or so, launched the OpenCog project, which was an open source effort at recreating a mind. So different modules of what constitutes a mind like memory, pattern recognition, sensory adaptation, etc. And they were trying to replicate each module in software and then improve the software. Typically like Ben, very, very early for the timescale. But as we look at this and we're building like memory and we've got the processing power, it feels like.
- [15:17]
This new generation will get to that point and essentially we're without realizing it, actually growing our mind.
- [15:25]
A
All right, I can't wait because frankly I would love to have perfect memory. And when we add augmented reality glasses and an always on version of Jarvis, having an assistant there that's able to constantly remind you of everything you've ever known and everyone you've ever met is going to be super handy. All right, let's go on to OpenAI is just, you know, we heard about in the last pod, we talked about their code red response to Google's, you know, growth and dominance. Well, some news is coming out. When I announced today that looks like GPT 5.2 will be coming online next week. And this chart we're showing, which shows the Benchmark and GPT 5.2 and Gemini 3 Pro. This is still hearsay. This was put up on X. Don't know in fact, if it's true. But just for fun, we're going to see once again, are we leapfrogging model on model on model? And we'll see Xai come out again with something, I'm sure, very shortly thereafter. Alex, thoughts?
- [16:35]
B
The catchphrase I heard over and over again at Neurips this week is that this is a rat race. There are so many employees at the Frontier Labs, so who are just grinding away, competing at what they view as a rat race to achieve the Frontier Max in this case. I fully expect, as I've mentioned on the pod in the past, that we're just going to see leapfrogging on a near weekly basis.
- [17:01]
A
And we're seeing it absolutely.
- [17:02]
B
Yeah. Until we get to the finish line. And I think the big question is, what does the finish line look like?
- [17:07]
A
It's interesting that Sam went out and made this code red announcement which got picked up by all the media everywhere. Right. And it's, it's kind of a. Interesting strategy because he's putting the organization on alert. He's letting the world know that he's put on alert. He got a lot of negative press from that, but he doesn't care. He just wants to refocus the organization. It really is a. An interesting management strategy.
- [17:36]
D
Iman commented, a good crisis is a terrible thing to waste.
- [17:39]
A
I love that. That's fantastic leverage that.
- [17:41]
D
Right. You're really. And I think that sends a message throughout. You know, look, Google did this exactly the same thing. Right. Sergey Brin said, I'm purse. You're doing back in. We're going into founder mode and we're going to solve this AI thing. And they've done it. And this is now the leapfrogging rat race, which is good for the consumer in so many ways. That's amazing.
- [18:00]
A
Yeah.
- [18:03]
We're going to be sharing the conversation we had last night on our next podC pod with Mustafa Suleiman, who's the CEO of Microsoft AI and just foreshadowing that. One of the conversations we talked about is safety. And if you're in a rat race and everyone's just trying to leapfrog everybody else, it feels like safety goes to the sideline and it's just an accelerationist point of view over and over again. Imad Any thoughts on this Code Red?
- [18:32]
C
Yeah, I think the code Red makes sense because the competition is intensifying and there's only so much attention that consumers have, as it were, for these kind of models. This benchmark chart I'd be very surprised at because for example it has video mmu and GPT 5. Can't understand video at the moment, so it'd probably be a brand new architecture underlying that. However, all these numbers will be hit in the next six months, probably let alone a year. That's why we're running out of benchmarks.
- [19:01]
E
Well, Peter asked me, should we put this out there because we have no idea if this is real or not. It's just, just a leak, an internal leak and we'll know next week. We could look really stupid if this is totally wrong, but I wanted to actually make sure we put it out there in case anyone wants to trade on polymarket because you know, right now, the end of year, the end of year, you know, you can buy ChatGPT or OpenAI at like $0.06 on the dollar. So if that, if that humanities last exam number is right, that's just mind blowing. So we'll find out next week but wanted at least give everybody a chance to, to see it and make their own guess on whether this is real or not. And like Ahmad said, we'll hit these numbers within six months no matter what. You know, some somewhere. You know the other thing that was really interesting last night at Microsoft is that we'll, we'll start adding a column to all these charts that has Microsoft's numbers independent of OpenAI. That's the mandate now I think, I think Mustafa was really clear that yeah, we're going to be another column on every one of these charts and another line on polymarket.
- [20:03]
A
Yeah, well, I guess one other point just to, just to note, Sam has got a lot of capital to raise.
- [20:13]
In order to implement the build out that he's announced. And I think this kind of like I'm willing to do whatever it takes to stay out front is part of the strategy being able to raise capital and get ready for an opening.
- [20:30]
D
That's a great point. That's a great point, Peter. This is as much for investors as it is for the general public and for the employees.
- [20:37]
A
But I think the bottom line for all of our subscribers listening is expect this on a week by week basis. Which is what makes our conversations on this pod so interesting. It's like it is watching. I don't know what the equivalent race would be. It is a horse race, but it's continuous with a billion dollars per day plus going in to fuel this crazy competition.
- [21:04]
B
I think the closest analog I can think of is this is like a world war with multiple campaigns and multiple fronts and multiple thrusts and initiatives.
- [21:12]
A
Yeah, yeah.
- [21:13]
E
If you only have time to look at two numbers on this chart, look at the first and the last, because the first one is the one that's most correlated with self improvement and accelerating AI.
- [21:23]
A
Well, just to read them, for those listening here.
- [21:26]
E
Well, I mean, it's speculation, but the high bar on humanities last exam is without tools, 37.5%, which is really good.
- [21:33]
A
Which is Gemini 3 Pro, and that's Gemini 3 Pro.
- [21:36]
E
Then with tools, it's higher than that. It's closer to 50%. Here the speculation is that they've leapt all the way up to 67.4%, which would be crazy. I mean, again, only Alex can answer those questions. As far as we can tell.
- [21:50]
A
I think Imam would do a good job as well. But what is.
- [21:54]
E
I mean, seriously, they're damn near impossible.
- [21:56]
A
What does with tools mean for those who don't know?
- [21:59]
E
Well, so then the AI, it doesn't just answer in one pass. It's allowed to actually use a whole variety of calculators. And really any kind of a tool that doesn't give it the answer, it's allowed to use in its chain of thought. And it can iterate many, many, many times. And so it's not just a standalone LLM, it's the LLM accelerated or enhanced with other software, which is perfectly fair if you're trying to solve a world problem, cure a disease or whatever. That's why they use that benchmark in addition to the raw benchmark. And then the last line is the one that Alex loves for good reason. It's self improving is one thing, but then self financing is another. And Alex, you talk about that better than anyone. I'll hand it to you for that.
- [22:41]
B
Vending Bench two and vending Bench arena. I love to the extent that we have any sort of economic Turing test or economic benchmark for agents ability to autonomy deliver a return on capital. That is what we have right now. And.
- [22:57]
As I've mentioned in the past, maybe just project this into the ether. I would love far better benchmarks for measuring economic autonomy than vending bench. But vending bench is what we have.
- [23:07]
A
Right now because it's coming.
- [23:09]
C
Actually, there's another thing that just crept in, didn't make it into the slide deck, which is there's a trading competition with the AI. Bots, I think, on the crypto side of things. And there was a mysterious model that actually made a profit reliably all the way through. And Elon Musk just announced that was GROK 4.20.
- [23:30]
A
Nice.
- [23:31]
C
So.
- [23:34]
Leapfrogging. And so he said that's how you pay for all the GPUs. Just going to let Grok5wild on the stock market. So you have to compete against Elon and his million GPUs dollars are the best benchmark.
- [23:45]
A
All right, Anthropic is making news once again. There's a lot of excitement and energy and right now it's still a bit of rumor, but that they will have an IPO as early as 2026. So anthropic is negotiating a new funding round that could value them at $300 billion as revenue is projected to reach 26 billion next year. They're aligned alongside OpenAI, which is also exploring a future IPO. So if we've got OpenAI going public and to access capital and Anthropic going public, I have to imagine XAI is also going to go public sometime in the near term. Thoughts on this one, gentlemen?
- [24:29]
E
Well, I mean.
- [24:29]
B
Comment? Go ahead, sir.
- [24:31]
E
No, go ahead.
- [24:33]
B
I think if anything, XAI is relishing not being public, given its history. But I think more broadly, the worst case economic scenario for superintelligence intelligence is that it maybe not the worst, the next worst case is that it remains decoupled from the human economy and that an intelligence explosion happens not on the publicly traded markets, and that insiders and early employees and the machines themselves sort of skyrocket in terms of real wealth, but are largely decoupled from retail investors and the rest of the human economy, that would be, I think, a highly suboptimal economic outcome. Whereas if we get enough IPOs from Anthropic, from OpenAI, from SpaceX, from some of these other firms that are achieving hyperscale on land in space. I think that is probably the best case from a macroeconomic perspective for the economy. And once those happen, assuming they happen, I think it's far clearer to see how the economy grows past the so called debt crisis, how we achieve hypergrowth, real hypergrowth, over the next three years.
- [25:42]
A
Dave?
- [25:43]
E
Well, having founded and taken a company public, this is a big, big step. And when Dario, the CEO of Anthropic, Dario Amadei, gets interviewed, he says, I actually never visualize myself being a CEO at all. I'm surprised I'm in this position. Then when you become a public CEO, It's a whole nother level, so I'm guessing he'll grow into the role, but it's a big deal. And so why do it? And why would Elon be relishing not doing it? Well, once you're in the public limelight, all your dirty laundry and your code Reds become crystal clear in your stock price every day. It's very hard to do what Sam does right now. Roam the world selling the story when your dirty laundry is right there on every stock ticker. So it's another level, but they have to do it because you can raise 10, $20 billion privately, but you have to tap into the public markets. Yeah, you gotta be talking about much bigger numbers, and the only way to do that is through the public markets.
- [26:39]
A
And it gives you currency for acquisitions.
- [26:42]
Which is important.
- [26:45]
I think it also increases trust in a company if it's a public company versus being a private company.
- [26:52]
E
Sam was at Davos last year, and I was following him around on the streets of Davos. And I mean, just the.
- [26:59]
A
Were you stalking him, Dave?
- [27:01]
E
He actually had this. You couldn't miss him. He had an entourage as big as the president of a nation.
- [27:07]
But he's just doing meeting after meeting after meeting, saying, give me another billion dollars. Give me another billion dollars. And, you know, you can see how there's no way that can scale to what's happening next in.
- [27:18]
A
I mean. And part of what's coming out in the news right now is a lot of the deals that Sam and OpenAI have announced are options and not actual deals, which is fascinating.
- [27:29]
E
Yeah.
- [27:30]
C
I think you're at this fascinating time now, though, whereby the dollar benchmarks are starting to accelerate, the revenues ramped up like anthropics. At 10 billion of revenue just a few years in, it's amazing. And they're actually catching up to OpenAI. But then the competition is going to get intense. Like Opus 4.5 is $25 per million tokens. Grok 4.1 is $0.50. Will you see substitution occurring? Or will you see these models actually just being used to literally make money? Like I said, like, I could easily see Elon in particular, just say, Grok 5 is going to pay for itself and all the GPUs by being the best hedge fund in the world and just let it loose on the stock market? And MacroCard is going to replace all the SaaS.
- [28:15]
E
Yeah.
- [28:15]
C
So it's an interesting.
- [28:17]
E
You know, the other thing that happened this week concurrent with this, it's not in the deck, but the cost of HBM memory, the memory that drives AI skyrocketed. It went way, way up. And OpenAI in the news said we've reserved 40% of the world's supply of memory for our own data center work in Abilene, Texas at Stargate, which is crazy. Normally memory comes down at a rate of almost half a year in cost. To see it go the other direction is the first bellwether that, hey, there's going to be a huge shortage of computer and if you don't go public and raise the capital and lock up the supply, you're going to be left compute starved.
- [28:55]
A
Yeah, I mean something's going on in fundamentals, right. I didn't put it in the deck. But there's a copper crisis right now. The price of copper is going through the roof because of wiring these data centers. All right, let's move on to our next story here.
- [29:11]
D
Wait, quick point.
- [29:12]
A
Yeah, please.
- [29:13]
D
If they're next year revenue projections are 26 and the valuation is 30,300,000,000. That's only 10 times revenues. Palantir is trading at 111 times revenues. So that's cheap in that sense.
- [29:27]
A
It's a good point. It's a good point.
- [29:29]
E
No, it is reasonable. It's crazy big numbers, but it's perfectly reasonable.
- [29:34]
A
I think Slim is saying it's overly reasonable. It should be at a higher valuation and it will be. It'll probably spike after an IPO. All right, here's our story. OpenAI finds confessions can keep language models honest. So new confessions method trains models to openly admit when they're hallucinating or when they've broke instructions. The method encourages models to self report mistakes instead of hiding them. So I am completely curious here. It's like, okay, so you use this methodology. Do you actually get two reports? Here's my answer and here's my confessions. Imad, what's going on here?
- [30:16]
C
Yeah, I mean this whole thing with next token prediction, right, is that the models kind of go along and then they can't have the self reflection and more things like that. I think that what you find is when you've actually got the right prompts, the right planning and then the right loops, you get very interesting things occurring. Particularly as the hallucination rates are dropping now as well because the models used to jump a lot and skip these behaviors because they didn't have the self reflection, they didn't have some of these other things. So I'm not really surprised by this. And again, I think what we'll eventually see is what we saw with The Deep Seq v3.2 math paper, a concept called a meta verifier, where models learn from their mistakes. So rather than just checking against a very small baseline, are you being honest? Where have you made mistakes? Having that as a verification loop is very similar to how humans learn. And that's what causes big leaps in some of these more frontier areas of thought as well.
- [31:17]
A
You know what I found shocking? I asked one of the models, how often are you hallucinating or providing wrong answers on average? And I found a couple of studies and one of them was that 25% wrong answers on everyday user questions. Another one said GPT 4.0 and Claude 3.7 sonnet hallucinate on an average of 15 to 16% of the time. And I never expect that when I'm asking my questions, I'm assuming, and I think the majority of all of everyone, perhaps not you, Imad and Alex, I'm assuming that they're correct. If it's really 10 to 25% hallucination, that's scary.
- [32:01]
C
It's basically the same as a human doctor. Right. And the interesting thing though is that it's dropped. So GPT5 dropped it from 18% down to 3%. So the last generation of models.
- [32:15]
A
Yeah. Yes, go ahead.
- [32:16]
D
Can I give you a dramatic example of this? I think the. It's worse than a human doctor because it's. They're actually trying to please you, the model. So therefore they're saying, whatever. I had a TV where the power went bad and ChatGPT said, oh, if it's this model here and it's making a buzzing sound.
- [32:34]
A
Yeah, you mentioned that in our last pod.
- [32:35]
D
Yeah. So I asked it then, okay, who do you know that can fix this? And it gave me names of three local TV repair shops that were completely hallucinated with phone numbers, with phone numbers, websites, names, addresses, and I started calling them all the phone numbers didn't work and then I went and looked up none of them existed. So this is a big problem. If I lift up a level for this confession thing, I think goes back to the earlier point that it's great to have a feedback loop. It gave me somewhat chills because I went to Catholic high school and the idea of confessions is somewhat chilling. Who's the priest? Is my question when you do this type of model. But I think the feedback loop is very powerful to have. Wow.
- [33:18]
A
Wow.
- [33:19]
B
I think it's also worth mentioning the 50 year old notion from economics of Goodhart's Law, which is that when a measure becomes a target, it ceases to be an good measure. And the way that these models are trained certainly and superficially rewards various sorts of behaviors that might be construed as dishonesty and being able to avoid Goodhart's law. Clever ways to avoid Goodheart I think are admirable on the part of OpenAI. And maybe the final solution looks a little bit less like naively optimizing just next token prediction objectives or reward maxing on RL objectives to solve math problems or programming problems. Looks a little bit more like some sort of multi objective optimization problem where maybe there's some blend of a good heart avoiding honesty reward and an accuracy award and an ethics reward that maybe almost start to look a little bit like separation of powers in what we see in in government systems there's an executive in some systems, a legislative, judiciary, et cetera.
- [34:31]
D
I'm going to have to bookmark that comment, Alex, because you lost me in multi objective optimization problem.
- [34:37]
A
On that note, I'm going to move us forward.
- [34:39]
B
But you have a good heart, Saleem.
- [34:42]
A
All right, so again, it's like the next Release, Google releases Gemini 3 Deep Think, which uses parallel reasoning testing multiple solution paths at once. That makes sense to me. An upgrade from 2.5 deepthink and it's excelled once again. Humanities last exam and GPQA diamond arc AGI 2. Going to our benchmark expert, Alex.
- [35:08]
B
This is a template for how revenue is going to scale to justify the trillions of dollars of capex. It won't just be faster models or better models or stronger models. It's going to be lots of agents, fleets of agents, millions of agents, many millions, billions of agents all running in parallel to solve problems. That is in my mind and certainly based on the architecture of Gemini 3 Deepthink, which isn't just a faster, better, singular model, but it's also scaffolding to have fleets of agents, fleets of Gemini 3 agents that are all running in parallel to solve a given problem. That's the deep think part of Gemini.
- [35:48]
A
Sounds like quantum computing to me. You know, it's like I'm going to. No, I know that, but there's the concept. I'm going to run this, this problem in multiple parallel universes and bring back the answer. I'm going to run this problem in, you know, a billion or a trillion agents and bring back the best answer.
- [36:05]
B
Parallel. Yes, but quantum computing wishes that it had the economic utility of Gemini three Deep Think.
- [36:13]
D
Peter on this one.
- [36:15]
A
Go ahead, go ahead, Salim.
- [36:17]
D
I'm with Peter. I'm with Peter on this One. It feels like that kind of parallels. But I think the broader point you're making, Alex, is that when you have millions of agents, each specialized, if you take something like the Manhattan Project, you have thousands of people, each with a deep specialty, connecting together and the hive mind then solve the problem, and we're going to see the same with agents. Is that a metaphor that works?
- [36:38]
B
Yeah. When Dario speaks of countries of geniuses in a data center, this is what it looks like. We're going to have billions of agents that are all going to be independently, probably pretty expensive, even though the cost of intelligence is going to zero. But collectively, yes, this is going to generate trillions of revenue if we have so many agents. And this is how we pay for all those data centers.
- [36:57]
E
Well, yeah, I think that's a really important point, Alex, is, you know, everybody talks about the cost of intelligence going to zero, but it's not actually going to zero. It's going down to a low number. But concurrently, the fleets of agents are getting so much bigger so quickly. And, you know, we saw earlier in this pod too, the. The process of iterative expand the context window to entire hundreds of millions of tokens and run many, many iterations to get rid of the hallucinations. Those forces are going in the other direction, and it's working far better than anyone thought it would. It's very unlikely that the cost of intelligence is going to go to anywhere near zero. Everybody's going to want more intelligence and more intelligence and more. So there's going to be an acute shortage. I only mention that because many business leaders are out there saying, let me just wait and see what happens. You're going to be starved of access. And you can see this already when the new models come out of Gemini and they add another level of deep thinking, it works incredibly well. But you wait three, four, five minutes to get your answer. And very often it says, we're experiencing unexpected loads right now. Sorry, we're offline. Like, how's that possible if the cost of intelligence is going to zero? Well, it's not. The cost per token is going way, way down, but the use cases are expanding on at least three different dimensions on this ridiculous curve the other direction. And everybody's going to want it because it works so well.
- [38:25]
A
So what does this actually mean other than we have yet another faster model able to hit the benchmarks a little bit better, and next week we'll be announcing the next better model. Imod.
- [38:42]
C
I think the models are getting to the point now where ensembles of these models doing different things are just genuinely useful for real world, advanced, complicated tasks. I'm really looking forward to using a Gemini 3 DeepThink 2.5. DeepThink was pretty good, but right now I think the only one usable for real frontier stuff is GPT 5.1 Pro, which again, does something similar. Ultimately, what you want is you don't want a task that's really complicated to be done in seconds, nor minutes. You want to be iterating on a task for a period of time, giving it input and feedback and the model just not making mistakes like Gemini 3 Pro still makes mistakes in math when I'm using it. And so Gemini GPT 5.1 Pro doesn't the model usage. Again, I think the Deep SEQ Math paper is fascinating for this. It's the first open source model that gets a gold on the IMO. And to each of the problems they use 2 billion tokens, so about 2 billion words. That gives you an idea of how many more tokens you can use.
- [39:44]
E
Yes, and it works.
- [39:46]
C
And it works. It got killed.
- [39:47]
E
I mean, if you want to experience this firsthand, just go to GPT 5.1, soon to be 5.2, ask it to do something complicated for you that it can't quite do, and they just keep asking it to try harder about a thousand times back to back, and it will eventually get it right. You're like, why does that work, Dave?
- [40:05]
D
What's an example of a hard thing to do?
- [40:07]
E
Oh, I do this like all day long. In fact, right after this pod, I'm back. I have a fleet of Kimmy K2 agents scouring the world right now, working on hard problems. But if I ask it to translate legacy C code into Python and then it comes back slow, make it faster, improve the algorithm, or, you know, more, more down to earth.
- [40:26]
A
Do you do that with your students? Do you do that with your students too? Work harder. Try, try it again.
- [40:33]
E
Well, that's the, that's the difference. That's where everybody's analogy breaks. Because there are a limited number of humans and students, but there are an unlimited number of AI agents. And so deploying a billion in parallel to work on something is out of your normal range of intuition, but it just flat out works. You need to expand your intuition to this new world we're moving into.
- [40:51]
A
I think that's one of the most important things that we can say coming out of this, which is we're about to enter a new world where there is a near infinite amount of intelligence to be thrown at things.
- [41:07]
C
Interesting things.
- [41:08]
D
Sorry, Please go ahead, Iman.
- [41:10]
C
Yeah, I think one of the super interesting things on the context when just realizing it is again, the stuff that we get wrong when we're trying to solve problems. Usually the only stuff that survives is the stuff that we get right. Like at Neurips, it's papers full of all the stuff people got right. Being able to actually have scientific method, strategy and things like that, where the context window includes everything you got wrong for all the different models.
- [41:33]
A
Fascinating.
- [41:35]
C
We know that will do better because often it's what we got wrong that.
- [41:38]
A
Actually guides learn from your mistakes. Yeah, sure.
- [41:41]
E
I mean, if you had asked everybody in the community three years ago, is that going to work? 90% of people probably would have said, yeah, I really doubt it. But it does now. Everybody agrees, everybody at Neurips, I'm sure it just flat out works. But that means you need massive, massive numbers of parallel agents and even any given agent needs many iterations. It's just a huge amount of compute, but it just solves problems. It's incredible.
- [42:08]
A
All right, our next story here is the reasoning. AI became so efficient. So AI got more efficient by mostly for huge LLMs with training becoming 22,000 times more efficient, while smaller LLMs only improve by 10x to 100x. So what's the story here, Alex?
- [42:27]
B
Yeah, this is sort of finger in the eye to those armchair theorists who say that the small guys, that the small labs are going to benefit from algorithmic advances. So this is a study that found that 91%, this is a study out of MIT. 91% of algorithmic efficiency gains between 2012 and 2023 were the result of only two things. One, the switch from LSTMs to transformers, and two, the switch from kaplan scaling, named after my former office mate in the Harvard physics department, Jared Kaplan at Anthropic and switch from Kaplan scaling to chinchilla scaling. Those two things. So LSTMs to transform LSTMs were recurrent to transformers.
- [43:14]
A
Not what are LSTMs?
- [43:18]
B
Long short term memory. So LSTMS were. Prior to the transformer revolution, LSTMs were the favored language model. I remember the old days prior to transformer, prior to GPT, when Andrej Karpathy had his char RNN language model that stunned people by being able to generate code. There was a, there was a life prior to GPT, but just those two algorithmic transitions. LSTMs to transformers and Kaplan scaling to Chinchilla scaling. Those two were 91% of the efficiency gains. And what that says is that this story that, well, we're just stacking small wins on top of each other and that eventually somehow algorithmic efficiency gains are going to enable smaller labs to have some sort of advantage relative to larger labs. This suggests that's just not true and that most of the algorithmic efficiency gains are actually accruing to the large labs that are able to scale out the most.
- [44:19]
E
There's one other thing in this paper that's noteworthy, and please don't read it. Alex summarized it perfectly. That's everything you need to know. It's way longer than it needs to be. Classic MIT work, but it's a very, very good summary at the beginning of the document of why this work is so important, because we're putting an immense amount of societal energy into scaling the hardware. And Elon Musk talks about Tennessee all the time and Stargate and huge amount of thought and research and discussion on our podcast about these massive data centers because they're so visual and they're so expensive. But the software side of it is very under researched and very underanalyzed. So they're taking a first shot at trying to give us better insight into the future rate of improvement of the software side of it, because that's where it's not as expensive, but the lift could be enormous. And so I think this is a really, really good focus area and I'm really glad MIT is on top of it. But, you know, Alex's summary is all you need to know about the work so far.
- [45:17]
A
Amazing.
- [45:17]
D
So the inner loop here then is. Is energy going to gpu, to agents going to intelligence, and therefore all of that scales. And there's the demand is so infinite in terms of adding intelligence to everything. That'll be a long time before you run out of that. That's the, that's what we're traveling the world with data centers.
- [45:39]
B
So I think there's some YouTube viewers somewhere with a drinking game or a bingo game for how many times we can say tile the earth with compute or disassemble the moon or whatever it is. So drink your whatever or cross incredible bingo game. To this one.
- [45:58]
Robots are the part of the loop.
- [46:00]
D
We're robots in the loop.
- [46:02]
C
Here we go.
- [46:03]
A
This next article here I find really important. This is about visual chain of thought. It's the notion that chain of visual thought methods are now able to give us a better understanding of images. This visual thought delivers 3 to 6% gains in continuous reasoning performance. The image here on this, on this is asking question, is the wall behind the bed empty or is there a painting hanging on the wall. And what you see then is the analysis. The ability for an AI to understand what it's seeing at the same time that we're bringing about augmented reality glasses and we have humanoid robots coming online is going to be fundamental. I want my AI to understand what I'm seeing. I want it to be able to remember during the course of the day where I left the keys or who I ran into, or recognize a face and give me their name. How fast is this accelerating?
- [47:10]
B
I think this is, if I may, even more profound than just garden variety acceleration. If I ask all of you, or I request, don't think about pink elephants, what's the first thing that happens? You start thinking about pink elephants, not in terms of text tokens. In your mind, you're not probably thinking in terms of language. You're using your visual cortex probably to create a mental image of pink elephants. And the ability to visually reason is something we've talked about on the pod in the past. We're finally, a few weeks later, a few weeks after this was predicted to happen, we're starting to see major gains in reasoning performance by models that can include visual tokens in their chain of thought, not just text tokens. And we're going to see a lot more of this.
- [47:55]
A
Iman, how excited about this are you?
- [47:58]
C
Yeah, I'm not surprised at all by this. It's very exciting. I think it's actually something fundamental to reality. Models are the things with the best math that approximates reality. And we've seen some interesting things before. Like originally we built stable diffusion, and then from stable diffusion, we extended it to 3D using the same knowledge. We found out, actually Harvard did a study that a image model understood 3D, then we extended that out to video doing the same thing. It somehow actually had a concept of physics in there. In fact, if you look at the latest image model that's top of the charts now, Flux by the Black Forest lab team, my former colleagues. It started with a language model that then got a video model that then became the best image model in the world. And you see now, for example, Luma recently raised 900 million from humane and others to build world models where you input all this data. Image, video, text, etc. Because text is low dimensionality. If you actually want to understand and reason, you need to have all the different types of data. But the latent spaces are actually very, very similar to them all in terms of your understanding of the universe. So you can go from a text model to a video model. Actually, just by Adding the right types of data, but the underlying structure doesn't change. Which I think has big implications for, again, the actual nature of reality itself, because each of those is modeling a different part of reality.
- [49:21]
A
Can you imagine going back to Alexnet when they were putting this together and showing them this capability when all of it, you know, we're trying to recognize the number seven. That was a conversation we had yesterday. I mean, truly extraordinary it is.
- [49:38]
E
And to experience it firsthand, take a screenshot of something you're doing on your computer, dump it right into Gemini and say, help. What's going on here? It's incredible that that works. It would have shocked anyone 10 years ago. Nobody would have believed you at all.
- [49:57]
A
Go ahead, Iman.
- [49:58]
C
Well, the crazy thing I think it's always worth coming back to this is that if you told someone 10, 20 years ago, they would have thought it'd be like this massive logic tree, right? We have to remember, is it this.
- [50:08]
A
Or is it that? Right?
- [50:10]
C
They're just ones and zeros. They're literally like a movie file. And you push words in or images in one side and it squeezes out this stuff on the other side. The reasoning isn't actually reasoning at all in the way that we think about it. And again, I think that says something profound about the way our brains work in the universe works. But the static group of ones and zeros can do that.
- [50:30]
A
I think what's important to realize here is where we're going, all of us are going to have a AI with visual capability always on helping you, supporting you, right? And I think that's a vision of the future. People say, well, I don't want to lose my privacy and so forth. But it's going to be watching what you eat. If you want to turn on health mode, it'll tell you, eat more of that, don't eat that. Or there's a staircase over there, go take the stairs instead of taking the elevator. I mean, the ability for an AI to be your always on visual Jarvis assistant is going to be profound throughout our lives, increasing our efficiency of what we do and what our objectives are. Yeah. Salim, do you want to add on that?
- [51:16]
D
Just to build on that point, Peter, I'm expecting in a year or 18 months some sensor that in your stomach saying, hey, you're about to eat that donut, wait 10 minutes because I'm still metabolizing the coffee.
- [51:27]
A
Right, okay.
- [51:28]
D
And creating radical efficiency and all these very little things that we never thought about much is going to be one of those areas that we're going to add a ton of compute against.
- [51:39]
E
So I took a screenshot of our podcast as we're speaking and gave it to Gemini just to prove the point. And I said, hey, are these guys having fun? And it completely interprets the scene, it knows exactly what we're talking about, and it says, yeah, it looks like it's fun. If you're a tech enthusiast, you like futurism, or you enjoy brain food, it's probably not fun. If you dislike technical jargon or you want casual entertainment, okay, that's probably true. The point is it completely knows what we're doing from just that screenshot. And, you know, this is going two different directions, too. It's making the AI more in touch with humans and the way we live, but the data is not specific to that. It can also go the other direction, where you feed in genetics data, you feed in satellite image data, and it can then get intuition in those domains where nobody that you know has intuition. So it's going in both directions at the same time. So if you, if you study what's going on with vision on this slide, you can develop some intuition about what it's very soon going to be capable of with medical imaging, with satellite imaging, with other types of sensors that we're not familiar with.
- [52:40]
D
I'm still reeling over last week's comment from Alex that we're taking brain scans and running them through AI. I mean, that's going to just generate some unreal insights.
- [52:50]
B
Well, Imad has thrown some cycles at that previously. I know, Imad. We were catching up with some of your former colleagues from the MEDARC days at Europe. FMRI wants to be its own modality.
- [53:02]
C
It does indeed. I think all the modalities, again, we tell the world for some reason.
- [53:06]
A
Right. All right, we're going back to one of the stories we opened up with. Which is the response China is having or the leadership it's providing? So China is accelerating its push to become independent of Nvidia, with Cambrican planning to triple output to a half a million accelerators in 2026. So this is a response to US policy. It's always that way. As soon as we restrict a country from buying a product or service that we're providing, especially if it's fundamental to the lifeblood, they will develop competition. And without question, I think the competition will. There's a huge amount of intelligence resident in China. Don't forget.
- [53:54]
The Chinese sort of educational system excels at math and compute. So I would not expect that they would deliver anything sub Nvidia Iman, do you want to kick us off here?
- [54:08]
C
Yeah. I think necessity is the mother of invention. Right. We also saw more Threads IPO this week in China. They raised just over a billion dollars. They're again another competitor GPU, it was 4,000 times oversubscribed. So $4 trillion of demand. Now obviously that's a bit much, but again you're going to see more and more of this stuff ramping, particularly for the specific Chinese models, because having an open source model that you can test as you're developing, it really closes the feedback loop a lot more aggressively here. You don't need just to build for one vendor, you can build for everyone. And most of the Chinese models are optimizing around this very sparse MOE type structure with deep SEQ and similar arches with muon kind of acceleration. So you're getting towards one architecture that they can just engineer and output industrially and who's the best at industrial manufacture.
- [55:03]
A
Yep, China has been, you know, it's important to remember Nvidia used to supply 95% of China's advanced AI chips. And when that, when that supply got cut, you know, there was a red alert going on in China and I'm sure the government orchestrates and supports and says, okay, we need to, we need our own Nvidia or multiple Nvidia companies in China. Alex, your thoughts?
- [55:29]
B
Yeah, we don't have a slide for this, but I would definitely encourage the audience to read the National Security Strategy that was just released in the past 48 hours. It's most certainly eye opening and I think spells out a pathway for tech decoupling between the US AI tech stack and the Chinese tech stack. I'm reminded during the Cold War the Soviet Union had what for those years would have amounted to an independent tech stack and was experimenting with all sorts of crazy architectures like ternary computing and other to westernize unconventional choices. I think we're going to see a Cambrian explosion, no pun intended, of architectures coming out of China now that China has been effectively decoupled from the US tech stack. And maybe many of those innovations will end up one way or another, benefiting the overall world, benefiting the US tech stack. I think we'll see a lot more experimentation coming out of China post decoupling.
- [56:27]
A
So what's the implication of this? I'd like to spend an extra couple minutes on this because China is going to be going as rapidly as possible developing its, its models. Its, its developed fully its energy ecosystem, you know, 10x further than the US has. We're going to talk a little bit about China's desire to put data centers in space.
- [56:50]
I mean, any thoughts on the long term implications of this complete parallel development between the US and China?
- [56:58]
B
I think we see an intelligence race and that will lead to diversity. We're going to see so many different architectures that are all competing in the us in the west, we have a whole handful at this point of Frontier Labs that are all vertically integrating with their own chip architectures, many of them in partnership with Broadcom or other lower level infra providers. Now we're starting to see the same happen in China. I think in the end, this heterogeneity that we're seeing in terms of tech stacks is only going to further accelerate the race that we're already in to the finish line. And again, I would pose the question, what is the finish line that we're racing toward? Because we're going to go much more quickly with this level of integration.
- [57:39]
A
So stepping up a level, is this good for humanity or not?
- [57:42]
B
I think all other things being equal, more experimentation can probably be better for humanity. Query whether it's good for the US or not. Query whether it's good for interoperability or not. But all other things being equal, more experimentation is probably better.
- [57:59]
A
Imad and Salim, I'd love to hear your thoughts here.
- [58:02]
D
I think the good news is that more technology development is generally better for the world. If I think about the counter position between the US and China, I think a lot of the future will depend on where you end up with trust.
- [58:17]
And the challenge with China is people don't trust it. Now people are losing trust in the US on a week by week basis. So there's that to be considered. But I think over time.
- [58:30]
The concept of do you trust Google or do you trust ChatGPT? In terms of what the future of AI is going to be, a lot of it's going to come down to where do we place our trust?
- [58:40]
A
It's really important.
- [58:42]
D
If you're an African nation over time, where will you put your trust?
- [58:46]
A
Right, Salim? Really important. I saw a tweet today to entrepreneurs saying if you're building something that increases trust, double down. If you're not, then stop doing it. I think trust as a scarce asset is a really important thing to look at.
- [59:02]
D
And I got a shout out to Jerry Mikulski here who made that phenomenal comment that scarcity equals abundance minus trust. It's just like amazing.
- [59:13]
C
Yeah, I think that you'll see just like China flooded Africa with smartphones With TCL and others, these chips will be very aggressively priced. So let's to put Cambricon in context, they raised about $2 billion. Their market cap is $100 billion right now. And the 5090 is equivalent to an Nvidia A100, the 6090, about a H100, but it's about half the cost and it's much more power efficient as well.
- [59:40]
A
Does this hit Nvidia's bottom line?
- [59:44]
C
Not for a while. For a while they'll all be used locally. But as they ramp from 500,000 accelerators to 5 million to more, and again, China has the full end to end supply chain as well. Then you'll see it flooding in a few years time. Again, this is in the acceleration phase here. To put the 500,000 in context, I think there were about 4 million hoppers sold and about 10 million Blackwells coming. So in a few years you can expect even just Camerocon, someone that no one's really heard about. They already get 80% of a generation back Nvidia chips in a few years. You probably expect them just like Tesla and byd, to actually be fully competitive.
- [60:25]
A
And now you're seeing BYD displacing Tesla and over again.
- [60:29]
D
Yeah. It also comes down to the engineering versus legalistic thing, right? The US is lawyers managing immigrant and engineers and China is all engineers with kind of an authoritarian state. And where will this play out?
- [60:45]
E
This episode is brought to you by Blitzi Autonomous software development with infinite code context Blitzi 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 Blitzi platform bringing in their development requirements. The Blitzi platform provides a plan, then generates and pre compiles code for each task. Blitzi 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 Blitzi as their pre IDE development tool, pairing it with their client coding copilot of choice to bring an AI native SDLC into their org. Ready to 5x your engineering velocity? Visit blitzi.com to schedule a demo and start building with Blitzi today.
- [61:49]
A
Iman, you're kind of in Europe or you're in a previous European nation and we've had this conversation on this pod over and over again about Europe has been in sort of a AI.
- [62:03]
Winter or ice age as the case might be. So Here we see EU to open bidding for an AI Gigafactory in early 2026. Europe is finally making a serious move to close its compute gap with the US and China by greenlighting an AI gigafactory bidding in early 2026. Iman, what does this mean?
- [62:23]
C
I mean every nation needs sovereign compute because their intelligence of their nation will be dependent on number of GPUs. Right. The EU with their regulatory acts has kind of held AI behind. But I mean recently we saw Yann Lecun now hiring for teams in Paris. You know, we see teams in Germany. We had the stable diffusion team. There's a lot of talent there. It's just they got to cut the red tape. And we're seeing, I think a change in that. And now the uk, Europe and others are like, well, this is the future, we have to cut the red tape. But the US is still far, far ahead.
- [62:56]
A
Can they move fast enough? I saw, you know, they're going to relax GDPR to give access to data finally. I mean GDPR was just a chokehold on entrepreneurs. I mean, how are you feeling in the UK right now?
- [63:09]
C
So I think again you've seen a step change just in the last few months. And as the agents hit next year, proper agents, not these stochastic parity agents. Everyone has to change. No, no country has an option but to change and to go all in on this. Because if you don't, then you're going to be left behind, you'll be out competed by your peers.
- [63:32]
E
Well, I'll tell you, just, just observing without judging. There's a square mile in Palo Alto, a square mile in San Francisco and a square mile in Cambridge. And the gap between those three.
- [63:43]
A
Cambridge, Massachusetts.
- [63:45]
E
Cambridge, Cambridge, Mass. Yeah, yeah, yeah. Not the other, not the original Cambridge. The gap between those three square miles and the rest of the world is getting wider and wider at an incredible rate. And I have always the same observations that Ahmad has. There's amazing talent all over Europe and all over the world and shouldn't this proliferate out to all that talent? But when I observe, you know that Peter, that meeting we had with Richard Socher, I heard you go holy shit like 12 times during that meeting.
- [64:14]
A
Yeah.
- [64:14]
E
And there is nothing on the planet. And we will talk about that.
- [64:17]
A
Not that we can say anything about.
- [64:18]
E
It, but hey, we can, we can't right now. But, but the gap between what's going on in those three square miles of the earth and the rest of the world is mind blowingly big and accelerating very, very quickly. So I think you know, building data centers around Europe is, is way too little, way too late. Unless it's done with some other, in combination with, with some other force that I don't know about yet. But just observing it, that gap is accelerating very quickly.
- [64:45]
D
Europe is best at public private partnerships. The problem is that in this world, speed is the ultimate higher bit and speed is not the strength there.
- [64:55]
A
I mean, you and I have had these so many meetings throughout many of the European nations, Salim. And the energy isn't there, right? The drive, the absolute. My favorite Joseph Campbell quote is like a man whose hair is on fire seeks water. Right? I mean, that's what we're seeing right now in the hyperscalers here. It's like, you know, this code red. It's like everybody jumping in. It's 24 7. It's not. What was it? It's not 996.
- [65:32]
I don't know what it is. It's 6 12, 7. That really is that way.
- [65:37]
E
Yeah. Not just that, though. I tried to make a point of this with Mustafa yesterday and that'll be on our next podcast. You'll see it. But he just name dropped. Well, when I was talking to Sam the other day, and then I saw Demis last night, and then Sam and I were thinking about Dario. It's all first names. All of these are just first names to him. And so it's not corporate, it's not data center investments made by the government. It's this very small group of people that are on a first name basis that have now, no joke, fifty hundred billion dollars budgets to build this out. So that's what's really happening.
- [66:16]
A
We saw the same thing in the space industry, right? In the United States, we gave birth to Blue Origin and SpaceX and Virgin Galactic and a whole bunch of entrepreneurial space companies. And in Europe, it's the industrial military complex creating Ariane 5 and a few other smaller rockets. But they can't compete with the current entrepreneurial space industry. The only way to compete is because the government buys local and that simply makes the entire space based services that are launched out of Europe more expensive. It just can't be supported well.
- [66:53]
E
And it's changing so quickly, it has to be on a first name, informal basis. At the rate of change in Massachusetts, this will drive Alex crazy, but in Massachusetts we have a very good relationship with our amazing governor. And she said, you know what I need to do? Put together an AI task force.
- [67:09]
And so the timeline on that kind of action is like three orders of magnitude slower than the Evolution of AI.
- [67:16]
D
So in Europe, in Europe, they would say sometime in Q3 of 2026, we'll start the discussion to create that task force.
- [67:27]
C
Actually, there's a very interesting thing. There's an initiative called Next Frontier AI to build Frontier AI Labs. And again, it's well intentioned. It's like they'll give 12 teams 25 million euros over the next two years to see if they can accelerate up and get there.
- [67:42]
A
25 billion euros per second.
- [67:46]
D
Can I double down on this just for a second? We are working with one of the biggest companies.
- [67:52]
A
And listen, I have to apologize to our European listeners. I don't. We don't want to make fun of the situation there. It's serious.
- [68:01]
E
Just spend two years in the middle of it and then go back home. That's the way to solve the problem.
- [68:05]
D
Look, I spent all the 90s living all across Europe in like five different countries. So I've got some kind of personal thing here. In terms of the ability to live and have a great life, Europe is amazing. But in terms of technological progress, it's not really the place to be. You have to move to the west coast and others. We are working with one of the biggest.
- [68:29]
European companies, it's one of the biggest global companies on transforming their metabolism. And we finished one of our major sprints with them. And they said, well, we need to start another one. Right. Right away. And this was back in February, and they said, let's have the first meeting about it in October.
- [68:46]
B
Yeah.
- [68:46]
D
And this is the problem. The people are seeing that the metabolism of everything that is happening needs to accelerate by 100 times across Europe for them to join.
- [68:56]
A
And it's culturally, it's culturally blocked. You know, there's a old adage. In Europe, you work to live. In the US you live to work.
- [69:06]
D
Yeah, yeah.
- [69:08]
A
Which may not be the very best thing for your life.
- [69:10]
D
Now we're going to live to compute. Yes.
- [69:13]
A
All right, let's jump into jobs and economy. A few interesting articles this week. The first one is Michael Dell's 6.25 billion billion investment in America's kids. So it's called Invest America and it'll give every child born after January 1st of 2025 an investment account of $1,000, and it'll be deposited to build financial security. Let's take a listen to Michael and Susan Dell describe what they're doing.
- [69:41]
D
We're making a $6.25 billion investment in America's kids through our charitable fund.
- [69:48]
C
Next year, every American child will be.
- [69:50]
A
Able to get an investment account powered by Invest America.
- [69:54]
D
We've seen what happens when a child gets even a small financial head start. Their world expands.
- [70:00]
A
The real power of these accounts is.
- [70:01]
E
That anyone can contribute. Parents, relatives, friends, everyone can help shape a child's future.
- [70:09]
D
To philanthropists, companies, community leaders. If you want to be part of something truly meaningful for our kids, for our communities, for our country, join us.
- [70:19]
A
So I.
- [70:22]
Celebrate them for that effort. There's been a number of players who have talked about similar situation. Of course, being able to invest versus just save is a part of what's made America great. Now the question becomes, is this too little, too late? Right.
- [70:42]
E
Or just flat out irrelevant given that all education is moving to AI? You know this better than anyone, Peter.
- [70:47]
A
Well, I mean, one of the questions became, you know, maybe what's going to happen is every single kid will have an AI agent that is out there generating revenue for them. Right. This was a con. Where was the conversation we had about this? No, it was. It was the conversation that Ilya had on his recent pod saying, one of the problems is if you've got an AI agent that's doing all this work for you, generating revenue, supporting you, representing you, the question is, does the human fall out of the loop and becomes irrelevant and is it important instead then to ultimately merge with AI? But that's a different conversation.
- [71:27]
E
That's a different conversation. But I think I'm 100% sure of this topic. Teaching at MIT, Harvard, and a little bit at Stanford.
- [71:35]
There was a huge push to move all educational materials online when the Internet exploded. And it kind of worked and kind of didn't work. You know, it made all the materials available. Yeah. Surprising. I'm 100% sure that now that there's an AI face and voice that matches your personality on top of that, that it's going to absolutely take off.
- [71:52]
A
Sure.
- [71:53]
E
And traditional education will be completely irrelevant imminently because it can match your. It matches your accent, your favorite voice, your favorite star, your favorite.
- [72:03]
A
Well, I think this is more than just for education. Right. I think this is more about how do you provide financial stability. We've talked about this in the pod before that, you know, this was a few episodes ago on the data from Fi9, that the majority of the world is absolutely concerned about not being able to be employed and the cost of living. And if you have a, you know, a seed kernel of capital when you're born that is growing by the time you're 18, does that give you some additional stability? Salim, you were going to say, I.
- [72:38]
D
Have two Thoughts about this one is I really, really love the fact that it has every child born and it's kind of essentially universal. It's creating a wealth floor for every kid, which means every kid from day one will be thinking about how investment and how do I think about it, etc. I didn't come across the concept of investment until I was like 6, 16 or 17.
- [72:57]
A
Right.
- [72:57]
D
If I'd had that way earlier, I'd be a way richer person today than anything else. And I think that by adding the employer add ons and encouraging people to contribute, you're creating a community effort here. So I think that it may be late to be doing this, but at least it's being done and I got to applaud them completely for doing it.
- [73:15]
E
Well, I think also it's just money so it'll be pivoted over to access to compute and say look, education, oh it doesn't have to be tuition. You can also get your GPUs that.
- [73:26]
A
You otherwise ultimately that's the currency that matters. Ultimately it's just hard for folks to acknowledge that and understand it. Imad well yeah, I mean it used.
- [73:35]
C
To be that capital is what compounded, right. You give people money earlier, that thousand dollars become $20,000. Now it's computing cognition that compounds as you move to self learning systems like your capital almost becomes irrelevant as the compute can get capital quicker than anyone. It's all about again how do you build that whole cognition architecture around you. So I think this is great. And then the other thing is we've got to give people access to frontier compute as young as possible as well in a way that makes them able to compound the benefits from that.
- [74:10]
B
I would just add this to my eye. Looks like the beginning of universal basic equity. We've spoken on the POD about ubi, ube, ubs. This looks like universal basic equity where every person in the economy will have an equity stake in the economy. And if we do see hypergrowth macro hyper growth over the next few years, then $1,000 in a 530A account which is again what this Invest America. Thank you Brad Gerstner for helping to conceive this idea. What a thousand dollars in an account now in a 530A account a few years from now if we account experience hyper hypergrowth could be quite material to a person's living circumstances.
- [74:49]
A
Yeah, and Brad's brilliant and he's agreed to come on the POD and talk about his moonshot. So we'll make that happen probably in early 2026 along these lines. Our next story here. College students flock to a new major, AI. Okay, not very new for us, but AI majors are exploding with popularity with schools like MIT and UC San Diego and launching AI branded degrees. So, Dave, I'm going to go to you first. Everything is AI at MIT these days, right?
- [75:17]
E
It sure is, yeah. So this is at MIT lingo. This is course six four, which was only added just a minute ago, basically. And it's already almost caught up to 6.3, which is core computer science in terms of people who are majoring in it. And I'll tell you, when you talk to students, they say the curriculum sucks. There are two or three great classes. Well, because you're trying to build an entire major and there's only two or three classes so far. You know, it takes the school too long to build the material because they're used to this much slower timescale. I'm sure it'll fill in because the demand is so high. But as of right now, there's just a couple of classes and then a whole bunch of garbage, which is frustrating the heck out of the students, by the way. Everybody wants to move to this and for good reason.
- [76:00]
No matter what you're trying to achieve in life, whether it's biotechnology or space travel or whatever, the way to achieve it is via AI. So if you get a good grounding in AI, you're actually then empowered to do virtually anything. So it's the perfect thing to study anyway, especially when you're young and you have time on your hands and you can really grind through these complexities. I'm really glad this is happening. I just want the curriculum to move much faster to catch up.
- [76:25]
A
The interesting note here to add is that AI related job postings in the US was up 50% year over year from 2024. So continue that. And that's going to continue. Any other thoughts on this one? I mean, it feels kind of obvious and I think the biggest challenge I've got is it shouldn't just be in college. I mean, we should be seeing this in high school as well. We're going to see a lot of people that are going to skip college. I think that's a debate that we've had. So if we can get you started in high school to think about AI, think about the world you're going to be inhabiting and inheriting, and how do you use this technology to create your vision and your passion? Imod?
- [77:06]
E
Well, I'll tell you, if there's one actionable thing too. Talking to every School administrator, every high school principal, every college administrator approve the applications. When the students say I want to study this on my own, I don't want to study that, just say yes, just approve it. Let them carry themselves forward. Don't hold them back.
- [77:28]
A
Intrinsic motivation. Intrinsic motivation.
- [77:30]
E
Unleash them. Unleash them.
- [77:32]
D
My favorite Joseph Campbell quote is follow your bliss. Let them follow their bliss.
- [77:36]
A
Yeah, that's a good one too. Yeah Imad.
- [77:38]
C
Yeah, I think it's fascinating because the fundamentals of AI are not actually that hard. Like it's not easy math, but it's not like that hard mathematics. My take is like if you have a semi vocational course where you do fast AI, which is a fantastic intro into the math and the basics for programmers. Andre Karpathy's video series on YouTube and then you just vibe, code and build and the entire class like implements some latest research every month that will put you way ahead of everyone. I think that CVs and qualifications move to show me what you have built and done with AI.
- [78:19]
E
Yeah, so if you're listening to what Ahmad just said and you're a student, take exactly what he said. Take the Karpathy material. Karpathy is the one guy from that OpenAI original crew who is not a self made billionaire because he's building education for the world right now. He's given up. He could be a billionaire tomorrow if he just signed some documents. He probably is anyway actually from his OpenAI stock. But putting that aside, he's building out the best educational platform you could ever imagine and just go find him online. Then tell your high school teacher or your college professor, I want to study this instead. Can you allow me to do that in replacement for this class I would have taken?
- [78:56]
A
Brilliant.
- [78:57]
E
That's the solution.
- [78:58]
D
Well, what you just said is antithetical to the concept of a university structure which is.
- [79:03]
A
Or a high school or a high school program.
- [79:05]
D
Yeah.
- [79:08]
C
Just quickly. I think that if you implement the stuff together and discuss it and again that fits with it, it's way better than doing it by yourself.
- [79:19]
A
For sure. But I'm just saying you have to literally hijack a high school curriculum or university curriculum to do that because it's not being offered today.
- [79:27]
D
To Dave's point earlier Peter Lilly, my wife has been pressured by all the local parents to have a day of just AI mind shift for all the teenagers. So we're going to do that and pilot that out and see how it goes.
- [79:39]
A
I heard that you've stolen Max Song, my strike Force member, to join for.
- [79:43]
B
A day Yeah, I should also point out, Peter, I mean, just want to look at this for a minute. From the perspective of economics right now, AI engineers are complementary good or complementary service to AI Compute. The cost of intelligence is going to zero. And so right now, pursuing careers and majors in AI, highly complementary. But, but, but, but recursive self improvement is also potentially imminent. And to the extent that recursive self improvement gives us soon AI engineers, we might start to see AI itself become a substitute for AI engineering labor. In which case maybe this rush to major in AI at MIT and UCSD maybe reverse itself, unwind itself, and everyone goes back to majoring in the humanities like they used to.
- [80:31]
A
And we had the conversation in the past about should you learn to code, should you learn. And there comes vibe coding. One of the conversations we had yesterday up in Redmond. We'll, we'll hear about it later this week was the importance of, of studying philosophy. All right, let's talk about the next, the next job boom, which are in data centers where the gold rush is for construction workers. So AI data center construction boom is making welders, electricians and supervisors earn between 100k and $225,000. And these AI companies need that kind of labor. There's a national shortage of 450,000 skilled trade workers. And it's significant. This is a alternate career path where you don't come out with hundreds of thousand dollars in debt. You come out with the ability to earn immediately.
- [81:28]
How long will this, this opportunity last before Optimus 4 or 5 or Figure 6 comes in and does this work for us? I don't know, maybe it's five years, 10 years, something in that realm. Thoughts?
- [81:47]
B
Agreed. I think that is the multi trillion dollar elephant in the room. As with college majors flocking to AI in this case right now, if, if you can pursue a career in the so called skilled trades to facilitate tiling the earth with compute. Drink your whatever. Again, I think that that's potentially very promising local strategy. But of course five to ten years out, and I agree, Peter, with your, your timelines, we're going to see humanoid robot substitution effects.
- [82:17]
A
All right, let's jump in. Yeah, go ahead.
- [82:20]
C
The economics and dynamics of AI data centers are really similar to fracking actually when you think about it, even the financial structures and these booms in these industrial kind of areas. So I think it'll last longer.
- [82:33]
A
But let's see, I find this next article. So Amazon eyes expanding its network after talks with USPS stall. So you may not know this, but the US Post Office is one of Amazon's main delivery carriers. So the US Post Office is delivering Amazon packages last miles in rural areas. It's been a significant about a $6 billion per year contract between the two and that contract is breaking down. My prediction is the US Post Office will be put out of its own misery and Amazon will get a contract from the government. Today the US Post Office has about an 80 billion per year operating budget and it's losing 7 to 10 billion dollars per year annually.
- [83:23]
Thoughts? Comments Gents.
- [83:26]
What happens when we have.
- [83:27]
B
Last mile of robotic delivery services? I think we have to prepare for that imminent future here and that is probably best done by the private sector drones.
- [83:36]
A
You know, we saw an article probably about a month ago that Amazon is giving its drivers now augmented reality glasses, right? And it's saying to the drivers, okay, wear these glasses. We'll warn you about if there's a dog in that, in that apartment building or that house, it will show you where to drop the package and so forth. And I think what's really going on here is that Amazon is collecting all of the last mile or the last hundred meter data and, and being able to train its future robots, right? Autonomous trucks, Autonomous robots doing that last 100. I keep wanting to say 100ft. I hate the fact that we use feet and pounds in the United States. It really drives me up the wall. And it drives me up the wall that science fiction writers are using that as well. Damn it. We switched over to metric back in the 60s. Pain in the ass. Anyway, yeah, this is going to be an interesting battle. What's your under over on how long the post office lasts? Anybody?
- [84:37]
E
Well, this is an interesting bellwether because it should have been privatized probably 30, 40 years ago. Everybody knows that. But it's written in the Constitution. Constitution. And you know, nobody wants to mess with the Constitution. But it's so obvious like space travel is getting or space launches are getting privatized. It's not in the Constitution because space didn't exist when the Constitution was written. So it can just move over to SpaceX and Blue Origin.
- [85:00]
A
By the way, look at the FedEx line, right? I mean FedEx had such an amazing lead. Fred Smith was such an extraordinary entrepreneur and it's been just slowly on a decline. All right.
- [85:15]
My guess is US Post office has at max five years left. I don't know if anybody wants to.
- [85:20]
D
Yeah, that's how it would have about the same.
- [85:23]
B
It'll take an act of Congress and.
- [85:26]
E
2/3 ratification of the states. It's A structure like this is a good case study. It's not that big a deal, but it's a great chance to learn. What are we going to do that's blatantly stupid because of legacy structure? And how is that going to get questions? So, yeah.
- [85:42]
A
All right, let's move on to space, a fun subject. There are four space stations under development in the US today, Vast which is going to which is being launched by SpaceX, Axiom Space Star Lab and Blue Origin Orbital Leaf. Just throwing this out because it shows finally we're going from government to truly commercial inhabitation. Alex, do you want to add anything here?
- [86:11]
B
Yeah. It's not a coincidence that there are four separate private space stations that are about to launch. These are actually all causally related to a NASA program. When we speak of privatizing the government. NASA in 2021 started the Commercial LEO Low Earth Orbit Destinations Program with ultimately one and a half billion dollars in funding. And the SpaceX surge to space that we saw was in part the result of another analogous NASA program to try to commercialize commercial crew program.
- [86:43]
A
Yeah, that's correct, yeah. In fact, the commercial crew program was what saved SpaceX. Right. SpaceX had three launch failures of their Falcon 1. They got the fourth one finally to orbit after Elon literally borrowed money to be able to put that together. And in Christmas was it, 2008, he won a billion dollar plus contract from NASA to go forward with Falcon 9, which is today the most successful launch vehicle on the planet by like an order of magnitude.
- [87:15]
B
Yeah, that's right. So the commercial LEO destinations program was spun up in part because the International Space Station is going to need to be deorbited sometime soon. And there was a desire for a private US space presence to succeed the iss. So I'm very optimistic about all of these and other private space stations. I think we're going to see an exponential rise of humans in low Earth orbit.
- [87:37]
A
You know what I'm excited about as well, Alex? Jared Isaacman. I cannot wait. So Jared Isaacman is back on the docket to be our NASA administrator. I'm not sure when the Congressional hearings finalized, you know, several days ago. Oh, is he in finally?
- [87:55]
B
Well, there has to be a vote, but the hearing was several days ago.
- [87:59]
A
Okay, so I've been texting with Jared and he's agreed to come on the pod as soon as the confirmation is done. So excited about that. He is brilliant, absolutely brilliant. I've known him for a long time. I took him to Russia to watch the Soyuz flights from some of our commercial launches there. All right, continuing on.
- [88:19]
D
Wait, I have a quick comment. Yeah, this, this space station thing reminds me of a date put out by a member, one of our NASA astronauts telling us the most interesting date in the world for him was October 31, 2000. And it was on that date that the first human being lifted off for the International Space Station. And since that date we've always had some, at least one human being off planet.
- [88:43]
B
It.
- [88:45]
D
First molecules are kind of drifting off this thing.
- [88:48]
A
So this next article is a bit of a surprise that SpaceX is considering a 2026 IPO. I mean, I've had this conversation with Elon. He was always resistant to take SpaceX public for a number of reasons. When you're a public company, you have to disclose all the details. Doesn't want to disclose all of his details and how he operates to his competition. But the other thing in particular was if you're a public company and you're spending a whole bunch of money to build Mars vehicles to go and colonize the Martian surface, is that something which your shareholders are going to support?
- [89:28]
Listen, I'm a SpaceX investor. I've held it from the very beginning. I would love to see it public. I always thought that what Elon was going to do was spin out Starlink and take that public and keep the launch capability.
- [89:40]
D
That was the conventional wisdom.
- [89:42]
A
Yeah, I do think.
- [89:45]
B
Oh, go ahead Iman. Sorry.
- [89:46]
C
No, no. I mean like Elon's a smart guy, he's got a million GPUs and so they have more AI lawyers than anyone to attack the stupid people that come after them. I mean serious, this is like SpaceX, XAI, Tesla, all will base basically have full AI teams top to bottom. Like you can criticize their strategy. The AI will just clap back at you. You can sue them for the silly stuff. AI will just clap back. It's a big difference in the way that you can actually run public companies.
- [90:16]
A
Yeah, you take SpaceX public and you take X public and Elon leaps over the trillion dollar mark in terms of personal net worth. Okay.
- [90:27]
B
Also maybe just comment quickly. I'm also not sure the historic story that's Starlink would spin out and do its own ipo. I think with the rise of orbital data centers, I think that muddies the water somewhat in terms of Starlink as pure communication service versus Starlink as a predecessor to orbital data centers and putting compute up there and comms capabilities. So in that sense, I think I could imagine a scenario where orbital data centers are actually pulling all of SpaceX to go public, not just spin off Starlink.
- [90:58]
A
Yeah, and that's a relatively new part of the conversation.
- [91:01]
B
That's right. It's very recent.
- [91:04]
A
I love this competition, you know, it's fun. My mission has always been open up space and I built so many companies on the space theme and it does the nine year old in me so proud and gives me such contentment that two of the wealthiest humans on the planet are battling it out to open the space frontier. So Blue Origin plans to start flying cargo to the moon in early 2026 using its Glenn heavy lift rocket, which by the way recently did a launch and full recovery of its first stage. And it's backed by a multi billion dollar NASA contract for targeted human landings in 2028. The government has always wanted dual suppliers. So for most a human, for most of the America spaceflight industry for the 70s, 80s, 90s it was Boeing and Lockheed Martin competing for this. Here comes SpaceX which becomes the dominant player and now the government wants a number two and it looks like it's going to be Blue Origin, which is super exciting. Alex, thoughts on this?
- [92:16]
B
This was basically the plot of season three of the television show for all mankind.
- [92:20]
A
I love that show.
- [92:22]
B
Yeah, it's a wonderful, Wonderful show. The three way race, in the case of season three, it was a three way race to Mars. In this case it's a three way race between SpaceX, Blue Origin and China to land humans again on the surface of Mars by 2028 or earlier. And I think this resumption of a space race which was dormant for 50 plus years and maybe also had collateral downsides for the rest of the economy in terms of overall innovation, it's coming back to life, we're back in the space race again and one might hope we'll see a lot of growth and.
- [92:54]
A
Innovation come out of it. Yeah, so the nine year old in me is so happy just to put some numbers and size against it. So NASA's 2025 Artemis budget. Artemis is their lunar program, Their human lunar program is $7.8 billion. Let's look at that compared to the Apollo program. So 1966, NASA's budget was about.
- [93:17]
Actually the Apollo budget was about 3 billion of NASA's 5.9 budget. So Apollo was half of NASA's budget. And if you adjusted the Apollo budget to today's dollars, it would be about 35 to 40 billion dollars. So compare that to the 7.8 billion that we're spending in Artemis. We were spending about A half a percent of the US GDP on the Apollo program back in the 60s, pretty impressive. And the reason we don't have to do that anymore of course is commercialization and technology. We brought the price down orders of magnitude. I find this article hilarious. Sam Altman enters rocket business to compete with Elon and SpaceX. You know what's fascinating is Elon goes for BCI and Sam goes for bci. Elon goes for space and Sam goes for space. There's probably a few other areas. Any particular thoughts on this one, gentlemen?
- [94:18]
E
Well, I'm really curious to see how the Code Red interacts with Sam has cut a deal in every single dimension related to AI including Jony. I've with the wearable device and Stargate with the data center and Broadcom with the chips, the tpu. So he's casted out in every direction and more power to him.
- [94:39]
A
That's Sam personally versus OpenAI, right?
- [94:42]
E
It's a mix. It's a mix. The bigger deals are OpenAI and then there's about 3, 400 personal deals that are all the use cases and components. But it's a massive metric in the samoverse, a massive network of connected parts. But now you've got this Code Red where hey wait, the thing that matters at the middle of it is these AI benchmarks and we're now off the chart on polymarket. Code Red. Code Red. So I'd be very curious to see what that means because he has a lot of talent but there's still a limited supply, it's not infinite. And so that all gets drawn back into the middle. That's going to cut some of the things on the edges.
- [95:20]
A
To give some more detail here, the company he's in discussions with is a company called Stoke Space and it's founded by two former Blue Origin propulsion engineers. It's as with every launch company needs to be fully reusable. It's a two stage fully reusable rocket. It's never flown. Right. They're using something called a ring shaped aerospike engine which also has never flown. So it's a little bit of a risky bet. But if Sam wants to enter the orbital data capabilities, I think having space launch capability and of course let's not forget Eric Schmidt is also in the rocket business.
- [96:04]
B
Yeah, I think this vertical integration by hyperscalers into space is probably an inevitability at this point. We're certainly not going to get our Dyson swarms drink without that. But I also think imagine near term future, are we going to get a meta space station? Are we going to get an anthropic space station? Maybe. Maybe clusters in space for fascinating.
- [96:29]
A
Just like we're getting hyperscaler fusion plants.
- [96:33]
B
That's right.
- [96:34]
A
Yeah.
- [96:35]
B
Facebook space station.
- [96:38]
A
No, I'm not going there.
- [96:40]
C
You've got to be full stack to control the likelihood of humanity.
- [96:43]
B
Exactly.
- [96:45]
D
It's a better place to put a name than a sports stadium.
- [96:48]
B
Maybe Instagram Space station. I'm not sure.
- [96:51]
A
Oh God, you guys can be rotors here. All right. Orbital compute energy will be cheaper than on Earth by 2030. So again, I still find this.
- [97:05]
Kind of challenging just because we have so much solar flux on the earth and don't have to worry about launch. But if we can really get the cost of launch down to $100 per kilogram, which is the projection with Starship versus 500 to $1,000 per kilogram, perhaps. Who wants to jump in on this one?
- [97:25]
B
I'll just comment. It could also be even cheaper than this once we get compact fusion online. A lot of these orbital compute projections are assuming that solar is the primary power source for orbital components. Compute doesn't have to be. Once compact fusion is cheap enough and there's no reason to expect it won't be. We can tile low Earth orbit at minimum with compute as well. And it won't require all of these expensive solar panels.
- [97:48]
A
So we're going to throw fusion reactors into orbit?
- [97:51]
B
Yes.
- [97:53]
E
Wait, so the theory there is launching a fusion reactor is cheaper than just a solar panel. The solar panel in space is about 10 times more effective than it is here on Earth. But it's still cheaper to launch.
- [98:04]
A
It's maintenance reactor. It's maintenance.
- [98:06]
D
I think that this I just won the bingo game. We said launching a fusion space.
- [98:13]
A
Saleem wins. No.
- [98:15]
B
So I mean solar and space is still fusion based. It's just using the fusion reactor at the center of our solar system. So I think the question is where do we want fusion power to be located for space based computer and as fusion reactors. And there have been a number of deep space probes that NASA and other organizations have launched that are using fission, for example, ion based propulsion. It's not like nuclear energy is that foreign for space fission thermal has been used by deep space probes for decades. It's not like we don't know how to do it. What's going to be new is compact fusion. In particular, we've put fission based energy in space for decades.
- [98:57]
A
You know Alex, I'm looking at the numbers here and says current terrestrial average is $12 per watt. And we're talking about 6 to $9 per watt in space. That's not enough of a difference for the level of complexity.
- [99:13]
D
Yeah, you're going to need at least a 10x drop in that.
- [99:15]
A
Yeah. So maybe it is compact fusion. I think if we're actually mining the moon, lest I say disassembling the moon to build the beginnings of salim already.
- [99:28]
B
Won the drinking game.
- [99:29]
A
Yeah, well, hey.
- [99:32]
Maybe that occurs. But I love the fact that it's now.
- [99:39]
Orbital data centers that are driving humanity's expansion into space. That's amazing. Would have never gone.
- [99:48]
B
It was going to have to be something. I mean if you look at all the sci fi plots, it was either going to be the discovery again for all mankind. It was, without spoiling too much, it was either going to be ice on the moon or the discovery of microbial life on Mars or something like that. That had to motivate space exploration and development. Who knew that it was going to be data centers? Well, it had to be something.
- [100:08]
A
It had to be. I remember when I was in college, I was at MIT and I was running seds and I put together this brochure on why open the space frontier. And I used to have to rationalize like better materials there. And I mean there's like always this like very soft rationalization. But this is real industry, real basic. People were trying to like what can we manufacture in space that has value here on Earth? Guess what, Peter, Yield.
- [100:34]
E
This is the moment. This is the moment you've been waiting for since in undergrad. I mean you should be like a kid the candy store. But it totally makes sense. And it always drove me nuts when, you know, like Ramon Cepeta, when he graduated, our buddy, he went to Ford and he was working on wire harnesses and rearview mirror motors for Ford. And like, why do we need like another electric component in a Ford? Why don't you work on space or something foundational that changes humanity? It's like, well, because you know, it's kind of like your iPhone now. A new feature for this massive installed base is economically incredibly valuable even though it's marginal for society. So it sucks up way too much great talent and something really important like space data centers doesn't get worked on. But you need an economic crack that starts the whole process. And this is, this is it. We finally have it, you know, after. And it is so much better of a storyline than, than for all mankind. You know, actually it's, it's a lot later.
- [101:33]
A
I still, I'm still betting, I'm still betting on Asteroid mining. I mean everything we hold of value on Earth. Metals, minerals, energy, real estates and infinite quantities in space. So you know those, those nickel iron asteroids that are worth trillions of dollars in platinum group metal. Or those carbonaceous chondrites we're going to mine for oxygen and hydrogen for fuel.
- [101:54]
D
Can I burst your bubble there, Peter?
- [101:56]
A
Oh, don't do it.
- [101:57]
D
I think AI transformation of material science will ride around all the scarcities around that.
- [102:04]
A
I don't know.
- [102:05]
B
Or maybe it'll accelerate it. I don't know. When Peter, when you were founding SEDs, did you, I'm guessing not foresee the plot twist that the killer app would for space would actually be like getting enough compute available to do generative cat videos doing funny things.
- [102:21]
A
That I was not able to project that far ahead, to be honest.
- [102:25]
B
Yeah. So who knows what the asteroid belt will actually end up being used for.
- [102:29]
A
So you know, lest we'll assemble it.
- [102:31]
D
For compute, obviously.
- [102:34]
A
Lest we leave the Chinese out of this China. A company called CosmoSpace is planning to build and add AI data centers in space. They're putting up a supercomputing cluster with three modules. One's got 100 megawatt level energy, the other is 10 terabits per second comms and the third is 10x operations per second. 10 exaflops, 10 to the 18th level compute module. Alex, what do you think about this?
- [103:06]
B
I think we're seeing a race to build Dyson swarms. It's as simple as that. It's not just a race to the moon. It's not just a race to tile the Earth. It is a race to put as much AI accelerated compute into low Earth orbit as possible. And coma space emerged from nowhere, I admit. Never heard from them several months or heard of them several months ago. And I don't think an obscure or otherwise obscure Chinese provider is going to be the last story we hear for Chinese orbital AI compute. We're going to see probably a dozen different vendors from China. We'll see as just discussed, dozen hyperscalers in the west. And the concern maybe becomes like overpopulation on Mars, making sure that all of these Dyson swarms remain interoperable.
- [103:51]
A
Can we just point out to all of our listeners if you've been following us on moonshots? This conversation of orbital data centers did not exist four months ago in any way, shape or form. It was there. I'm sure people were speaking about it, but it's now become a weekly conversation over the last three months. It literally came onto the scene with A vengeance. It's extraordinary.
- [104:15]
E
Well, and the technology clearly works. It's proven technology now, so it's just a question of launch costs. That's the only missing link and it looks promising.
- [104:26]
C
They figured out dissipation side of things. I think that was still outstanding.
- [104:30]
E
Yeah, no, no, they got.
- [104:33]
A
There is still an issue of how you can get efficient energy dissipation on the back end. In fact, that was a subject. It was proposed as an X prize this year at Visioneering.
- [104:45]
C
I think they'll get an efficient. It'll make more sense. But the thing that always gets me with this is it's horribly insecure. Like you had the proposal by Eric Schmidt and others. Like pounds of things go on, race, go against data centers, space data centers, think they'll go up and they'll just start disappearing. Honestly, yeah.
- [105:03]
E
I'll give you the counterargument and I totally agree, by the way, so I don't want to. But just to give you the counter argument, the hardware depreciates in three years anyway, so you only need it to be secure for three years. So I think the counter argument is that the US Space Force will basically guarantee enough safety that you can get three years of hard work out of it before something bad happens. And that's all you need to pay off.
- [105:26]
A
So what about solar flares? One of our subscribers asked that question, what happens when you have solar flares hitting these and what happens when there's an EMP that hits them as well? All of this gets knocked out.
- [105:41]
C
Actually, I've got a very interesting thing. So we were training on thousands of a 1/ hundreds a few years back and we kept getting errors exactly at the time of solar activity because it was basically messing with the EMCC memory.
- [105:55]
D
Wow.
- [105:55]
A
Yeah.
- [105:56]
C
We'll find out how these things sustain in the vacuum.
- [106:00]
E
You know this better than anyone, Imad, but you know, a little inside scoop on the million compute clusters that are being built now. They have errors, you know, here on Earth too. And you have to solve that problem in order to have coherent training anyway, so the error rate goes up a lot in space, but you have to have a process for backing off and you can't, you know, right now everybody does checkpoints and rollbacks, but you can't, you know, invest, you know, an hour of a million GPUs at millions and millions of dollars and say, oh wait, we have an error. We're going to roll back all of those GPUs for an hour. So you have to do it, you know, unit by unit. And so that work is well underway. So presumably that'll work fine in space too, and you can tolerate the error rate.
- [106:40]
A
All right.
- [106:41]
B
The other comment, of course, is that those models just want to learn. So for AI compute workloads, to the extent we're doing training or inference, they can be structured to be fault tolerant.
- [106:50]
C
Yeah, Microsoft was actually the leader in that and now Google's the leader in that. It's just seamless the way that it flips. It'll be interesting once we get out of space.
- [106:57]
A
Our last topic here. We're going to dive into robotics just for a few stories.
- [107:03]
Here we are after an AI push. The Trump administration is now looking to robots. So robotics are the focus for a 2026 executive order to accelerate US robotic development.
- [107:17]
We're seeing this in China, right? China is crowning its winners. It's got huge investments into the robotic industry. In fact, in our last pod, we talked about the fact that there is a, at least what the Chinese are calling a robot.
- [107:34]
What do you call it, a robotic bubble. Going on today with over 150 Chinese robot companies.
- [107:43]
I find this one fascinating. After the acceleration, major national robot strategies are coming online. Robotic firms are likely to have tax credits, subsidies, protection against trade measures, something along the CHIPS Act. Once again.
- [108:02]
Thoughts, gentlemen?
- [108:04]
B
I'll comment that the zeitgeist at Neurips this year was that humanoid robotics is the next big thing for AI after agents. I think this is how re industrialization of the US and the west happens. I think this is probably the best path for radically increasing economic growth and automating the 2/3 of the services sector that relies on physical intervention. This is instrumentally convergent for the future that we want.
- [108:31]
A
Yeah, just to remind folks from our last pod, we talked about the fact that China installed 54% of the world's total robots last year. So again, massive, massive push. I want to show a couple of quick videos here just for fun to close us out.
- [108:54]
We saw in the last week a little bit of Optimus versus figure competition. Elon posted this image of Optimus walking. Let's take a look. Here it comes. Running along, jogging. And then we had Brett posting this one of figure running across.
- [109:16]
I have to say they look pretty natural compared to where they were six months ago. Which one did you like better? Let me play this again. Here comes Optimus. Optimus coming along. I don't know, it kind of looks like figure is doing a better job running to me. What do you guys think?
- [109:34]
B
I think they're both incredible. And I'D also throw out maybe a request to the audience for the show. If you're interested in supporting robot athletics in the United States, either as a host or as a vendor or in some other capacity, please reach out to me. I'd like to do what I can to ensure U.S. dominance and Western dominance in general with humanoid robots via robot athletics.
- [109:58]
A
Yeah. Well, let's take a look at Chinese dominance with this video, then we can talk about it. So we saw last week the T800 humanoid robot from Engine AI in China. This is 5 foot 8 inches tall.
- [110:13]
Incredible capabilities. They put out a new video that I wanted us to take a look at here because it's a little bit shocking. All right, this is.
- [110:23]
E
We are 100% sure this is real, by the way.
- [110:25]
A
Yeah, they claim it's real. And this is a follow on.
- [110:32]
But. So here we are with this T800 robot, basically.
- [110:40]
Kickboxing. But check this out. When he goes up against a human opponent.
- [110:48]
Like, wow.
- [110:53]
Kind of scary.
- [110:55]
D
I'll go back to my standard comment that having a robot doing kickboxing is not a great marketing message.
- [111:01]
C
Yeah, I think calling it a T800 is also not a great marketing message.
- [111:06]
E
What about all the skulls that they put in there on their. Can I do a little headline?
- [111:11]
A
Yeah, we love your rant, Celine. Go for it.
- [111:12]
D
Look, a human being has evolved for 4 billion years, which is an optimized strategy.
- [111:19]
A
Well, 200. 200,000 years as a human being.
- [111:22]
D
Whatever. For survival. We have the human structure to survive and being able to quickly pick fruits off trees. We have opposable thumbs and whatever. I mean, a wheel is so much more energy efficient than walking. It's ridiculous. I think, for God's sakes, at least put those little wheels in the bottom of the robot like the kids with the wheels in their sneakers so they can be more efficient as battery power seems to be huge limiting factor in this. So why don't we have a wheel along with the leg so they can do both when it's needed? This just having robots copy human beings seems to be the stupidest thing in the world. It feels to me like when we first had tv, we're doing radio announcers and we're doing television reading the same scripts.
- [112:03]
A
We're going to be doing a podcast. We're going to be doing a podcast from Figures Headquarters in Palo Alto in January. You're not invited.
- [112:14]
C
Yeah, I don't know. I think this is really interesting.
- [112:17]
D
You guys kickbox the robot.
- [112:19]
C
This is really interesting. So the entire chest cavity of the Robot's actually a battery here. But this is really interesting because the 450 max joint torque, that's the really interesting part. Basically this thing can punch harder than the gorilla, like four times. Mike Tyson. Do you really want those to be walking around in the streets? Like, are they going to have to do regulations on the max joint torque? Also, if you actually look at the full video, which is real, they even show a behind the scenes one and you look at the previous video. So there's something called Sim2Real, which can basically model human actions in a robot. So we have full, almost real steel that move with Hugh Jackman teleoperation capabilities now in robots. And soon it'll be policy learning. Like this robot can do freaking Kung fu with great.
- [113:07]
A
UFC is coming.
- [113:08]
C
Punch through.
- [113:10]
A
UFC is coming. We're going to see Tesla bot. You know, basically Optimus versus T800. I mean, it's going to be Olympic level sports. It's going to be amazing.
- [113:19]
C
Olympics versus ufc. I think that'll be exciting. But we have to actually ask, do we actually want to have regulations around the max joint? Talk of humanoids in the street. Because I'm fine with these being in the fighting arena. I think it's fantastic. Or in industry, but I don't really feel comfortable them walking around.
- [113:35]
A
Yeah, the challenge comes.
- [113:37]
D
No, no, the kitchen triggered me in the kitchen.
- [113:39]
A
The challenge comes when they enter warfare. Right. I mean, this is Terminator in sort of the pure sense of robots on the battlefield. And it's a scary direction for us to take humanity.
- [113:53]
B
Yeah, I think it's all of the above. It'll be warfare, it'll be in the kitchen, it'll be on the street. And you'll see, I think governance and governments at different levels, whether it's municipal or national or international regulations for the parameters of what the rules of engagement are. Making an omelet versus fighting a war.
- [114:14]
A
All right.
- [114:16]
I want to just do a.
- [114:19]
Thanks to CJ Trueheart who gave us our first song on the Moonshot Mates. This is an outro piece called the Exponential. But before I play our outro piece, gentlemen, it's been a blast to spend time with you guys again. I love this. Imad, it was wonderful to have you as a fifth here today. Grateful for you. What's your week ahead look like, Iman?
- [114:48]
C
Lots of policy work and more agent stuff. We've got lots of releases coming. Exciting times.
- [114:53]
A
And how was Japan? You were there for FII Japan.
- [114:57]
C
Fantastic. Huge amounts of corporate and government interest in using AI to help accelerate the Way forward. And so again, hopefully some announcements about that soon.
- [115:07]
A
Yeah. And Dave and awg, you're about to hop your flights back to Boston, I gather.
- [115:13]
E
I think Collectively we covered 12 countries in the last week and a half in this group. So it'll be nice to be home for at least a week.
- [115:21]
A
Nice and same for you, Salim. Chance to stay home.
- [115:25]
D
Yes, I'm here for a bit. I just got back. So I'm preparing for the big online meaning of life session where if people are interested, come armed with any question you have about life and any question.
- [115:37]
A
When is it?
- [115:38]
D
Salim is December 17, 11am Eastern. Will go for several hours on metaphysics, philosophy.
- [115:43]
A
Yeah. And when Salim says go for several hours, like six to eight hours. Well, it's a big topic.
- [115:49]
C
There's lots to cover.
- [115:51]
A
It is for sure.
- [115:53]
D
This is an example. We start off on a conversation of what is truth and have it broken down to a two by two framework that just dissense making it allows us to have a decent conversation. What do we mean? Do we mean by that?
- [116:05]
A
Well, on Monday I'm heading up to the buck in the Bay Area to talk about longevity and AI. My favorite one two punch. And with that, let's listen to the music of CJ Trueheart as we wrap this episode. Gentlemen, see you on the next episode of Moonshots. Thank you to all our subscribers. If you haven't subscribed yet, please do. We're now putting out more than one episode a week just because the speed is moving so rapidly. So if you want to know when the episodes drop a quick hit, subscribe and let's listen to cj.
- [116:58]
E
Linear thinking held us down crawling centuries slow.
- [117:06]
But something shifted in the code.
- [117:08]
A
Now watch the numbers grow.
- [117:13]
1 becomes.
- [117:14]
E
2 becomes 4 the double and never.
- [117:16]
A
Ends Deceptively flat it first then vertically ascends. Can you feel it building the pressure in your chest? This is the moment where the future manifest.
- [117:32]
E
We're rising exponential.
- [117:37]
A
Straight up to the.
- [117:38]
C
Sky.
- [117:41]
A
Every second accelerating. This the top time to be alive.
- [117:48]
Digitize, disrupt, democratize the dream. Nothing's ever been this fast. We're breaking through the seam on the curve of infinite rocket boost. And now the inflection point is here.
- [118:02]
C
And we're never.
- [118:04]
Coming.
- [118:06]
Down.
- [118:10]
A
All right, if you're a music producer using AI and you want to give us an outro, just go ahead and let us know. And please, next, when you're watching this and you have questions, please post them in the comments. We are going to do more AMA in the next couple of sessions. Gentlemen, moonshot mates Dave Awg Mr. Exo Imod thank you guys having a fantastic week. Every week my team and I study the top 10 technology metatrends that will transform industries over the decade ahead. I cover trends ranging from humanoid robotics, AGI and quantum computing to transport energy, longevity and more. There's no fluff, only the most important stuff that matters that impacts our lives, our companies and our careers. If you want me to share these metatrends with you, I write a newsletter twice a week, sending it out as a short 2 minute read via email. And if you want to discover the most important meta trends ten years before anyone else, this report's for you. Readers include founders and CEOs from the world's most disruptive companies and entrepreneurs building the world's most disruptive tech. It's not for you if you don't want to be informed about what's coming, why it matters, and how you can benefit from it. To subscribe for free, go to dashmandis.com metatrends to gain access to the trends 10 years before anyone else. Alright, now back to this episode.
- [119:44]
E
So you're about to make a trade.
- [119:45]
A
Based on a friend's text, but which.
- [119:48]
E
You do you listen to is it we could buy a house in Tulum.
- [119:55]
A
Get optioning those options.
- [119:59]
We could lose everything.
- [120:02]
E
Or let's do a little research, get.
- [120:05]
B
Your head in the trade and make.
- [120:06]
C
The investment decision that's right for you.
- [120:08]
E
Learn more@finra.org TradeSmart.