Daniel Gross’ AGI predictions, SpaceX IPO news, Trump takes control of US chip exports | Diet TBPN

John and I are both very sick.

Sick of not podcasting. Let's go. We are actually very sick.

We are deathly ill.

But that's the beauty of watching The Daily Show live. There's no risk of me getting you sick because you're on the other side of a screen.

And that's the beauty of The Daily Show. There's no— you don't ever have sick days. Yeah, there's no days off because there's no days off. Why is no one talking about Daniel Gross?

No one. Literally no one.

Really? No one? Gross.

Uh, he— no, he should be a household name. He— you should get into a taxi, and they should be, oh, did you see how incredibly dialed Daniel Gross's AGI trades were in January 2024? Yes. Great website. It's what is it? Danielgross.com/AGItrades. The classic Times New Roman serif font, 12-point, just hammering it out in the vanilla HTML. It's interesting because a lot of them are framed as just open-ended questions, but if you think about you know, if you believe in AGI and then you go through the questions like you will see exactly what happened over the last 2 years. And this has been the like the underpinning thesis of Situational Awareness in many ways. Daniel Gross was a, was an anchor of the fund. That of course is—

and it's particularly relevant today because I'll read the intro.

Yes.

This is January 14th, 2024. Over 2 years ago, he says, I think we can all agree that GPT-4 completes many tasks at human-level proficiency. It is imperfect in odd ways. It can write software like a smart MIT undergrad but can't do basic task planning like an entry-level EA. It speaks all languages but can barely do math. Suppose the progress doesn't stop. Just like GPT-4 was better than 3, GPT-5 is capable of basic agentic behavior, i.e., able to accept a task, work on it for a while, and return results.

Nailed it.

And of course, today OpenAI has released GPT-5.4, which does exactly this quite well. Yeah, the reviews are coming in and they are quite, quite good. Daniel continues, some modest fraction of Upwork tasks can now be done with a handful of electrons. Suppose everyone has an agent like this they can hire. Suppose everyone has 1,000 agents like this they can hire. What does one do in a world like this?

So, so let's go through them one by one. There's 17, 18 questions, something like that. So he kicks it off with an easy question. In a post-AGI world, where does the value accrue? People were debating this at the time. Application layer versus foundation model layer versus infrastructure layer. Value has definitely accrued to the infrastructure layer.

I mean, you're even post-AGI, John's new haircut is is paying.

Paul Graham wrote a whole piece about how brands are important. But just looking back in the last 2 years, value has clearly accrued to the infrastructure layer. So that means chips, packaging, power, et cetera. And this is the situational awareness trade by and large. Nvidia basically captures, and did for a while, more than 100% of the profits from the AI boom because so many of the other companies in the AI space were losing money.. And so the foundation labs are losing money. NVIDIA's profit margin went from like 30% to 60% gross margins. It's borne out in market cap. So NVIDIA added $3.2 trillion in market cap from when Daniel wrote this piece. And I remember listening to him on Stratechery talk to Ben Thompson alongside Nat Friedman and say, yeah, based on, based on ChatGPT, NVIDIA seems sort of undervalued. And I was like, oh, well, if they're saying it on Ben Thompson's Stratechery podcast, it's obviously priced in. Everyone knows this. And I was wildly wrong. Never, never doubt yourself. What's interesting is that the rest of the platforms did not see, you know, 3x gains. They did not add $3 trillion in value. Microsoft from January 2024 to today is only up 4%. Amazon's up 30%. And then you do have OpenAI, Anthropic, XAI in the private markets. The gains there, they've been huge and staggering. And it's like the fastest growing companies in private markets history shaking the venture capital world you're either in You're in or you're out. It's a huge deal.

But you add them all up, it's only $1.4 trillion. It's just so crazy when you think it— Satya in many ways went on such a generational run. And you got a 4% pop. If you hadn't looked at the stock price at all, you would think, oh, it's got to be up, what, 40%? No, it's up 4%.

That was Daniel Gross's next question. What happens to NVIDIA and Microsoft? These are the two interesting players at the time, some of the biggest companies, the most AI-aligned. Absolutely crushed. Revenue tripled from $60 billion in fiscal year 2024 to $215.9 billion in fiscal year 2026. Microsoft has been far less dominant. So Azure growth actually is accelerating. It's at 40% year over year, but the stock only returned 4%. As we said, the market punished the $80 billion in AI CapEx that Satya Nadella has been telling to investors because everyone's asking, well, I could be in NVIDIA and they're not, they don't need to really invest that much 'cause they're a fabless semiconductor design company. Their gross margins are increasing, they're printing cash. And you are saying, okay, you gotta spend $80 billion and we don't know when we're gonna get the profits from that. So there's an open question there. So when it comes to the picks and shovels trade, you don't wanna tie yourself up to an individual startup or a foundation model lab, you just want to own the simplest thing that value will accrue to, Nvidia all the way. They were the clear winner of the picks and shovels trade. Microsoft's infrastructure play, I think it's a good decision. It's just the bets have yet to pay off for shareholders. Much more pointed question Daniel Gross asked, is copper mispriced? Was copper mispriced in January of 2024? The answer was, oh yeah, majorly. Now, raw materials don't move like meme stocks. So the actual move, copper was $3.75 a pound in January of 2024. 2 years later, it went to an all-time high of $6.61 a pound. So it's not like it even doubled. But that type of move in a basic material that we've been mining for hundreds of years is remarkable. And that's because AI uses a lot of copper. Nvidia's GB200 NVL72 server rack uses over 5,000 copper cables. There's 72 GPUs wired together, but you need 5,000 copper wires to get it all to work together. If you stretched out all all of the copper wire in an NVL72 from one side to another, it would go 2 miles long. And this is one server rack. Like this is like, it's not like going across the data center. This is not taking the data from AWS East to California. This is within that one server, you stretch out the copper wire, you're going 2 miles. It's an incredible amount. A single 100-megawatt data center, which is a data center for ants by modern standards, needs around 3,000 tons of copper I think that's like $500,000 or something like that when you multiply it all out. Oh, no, it's data centers broadly will be using 500,000 tons of copper annually in a few years. And people are actually saying that copper is the new oil, but there are a bunch of things that are also the new oil. So in the AI buildout, so it's so complex, there's bottlenecks everywhere. So you got to take that with a grain of salt. But copper was certainly—

Today it's looking like oil is the new oil.

Oil is the new oil.

What is crude at?

80. 80. So it was at 70 before the Iran war broke out. There were predictions that it would go to above 100. Oil is another commodity that does not move in the same fashion as a meme stock or a—

Except for today.

Except for today. Big move today.

Up 80%.

But typically, as prices go up, the firms drill more and the prices reach equilibrium. US drillers aren't rushing to increase oil production is on the front of of the Wall Street Journal business section today. The Middle East is on the cusp of a prolonged conflict that could push oil prices to heights not seen in 4 years. For now, American oil drillers are sitting this one out. The US oil benchmark settled at $74.66 a barrel Wednesday, the highest front-month settlement since the 12-day exchange of Israeli, Iranian, and American strikes last June. But West Texas oilmen, it makes But to West Texas oilmen, it makes little sense to add expensive rigs and boost production when the war could be short-lived and crude prices drop. So that's what's happened in the oil markets. Anyway, moving over to real estate, is San Francisco the new Detroit? And I'm not exactly sure what he means by new Detroit, because new Detroit meaning like the old Detroit when it was Motor City and they were building amazing cars there, or the new Detroit in the sense of like Detroit today is a hollowed-out shell of what it used to be, a former boomtown. The one thing is clear is that SF is back. SF is completely booming. Office vacancy fell from 36.9% to 33.5%. OpenAI has 1 million square feet of offices. Anthropic has a 25-story tower. Sierra, an application layer company, signed 300,000 square feet of office space in San Francisco. The Bay Area received 78% of AI venture capital in the first half of 2025. And there is a flip side to this. So overall, employment in San Francisco is still down relative to the pre-pandemic. Some people left. Some legacy companies aren't hiring as much. All the hiring is happening at the AI startup lab area, but housing prices remain strong. It's certainly not a hollow shell by any means. And if you've ever visited San Francisco, you can tell that it's cleaner and safer than it has been in previous years. So AI overall was not a total rethinking of San Francisco and It did not become open source or the, or the hub of activity in tech didn't move to Miami or New York or Austin or Los Angeles. Like San Francisco is still where it's at. The next question he asked, how does AI change wealth inequality? It's sort of too soon to tell. The data is not entirely clear. The data hasn't moved that much, but there are some interesting studies. So the International Monetary Fund released a working paper in 2025 that said that AI could reduce wage inequality, reduce wage inequality. So the amount of money that people make on an annual basis could be reduced. What they say in this paper is that high-income tasks, the job of a lawyer, the job of an executive, the job of a management consultant, those will be automated before the job of the machinist, the gardener, the street sweeper. And so you will see wages at the high end fall while wages at the low end remain stable. And also at the bottom end, you have the minimum wage. You will see higher-end wages go down, the bottom wages will stay the same, and that will reduce wage inequality. On the other side, AI could worsen wealth inequality because it's concentrating capital returns into tech owners. So the OECD found that wage growth was actually the strongest in low-skilled occupation. Assemblers. I didn't even know that was a job. But if you're an assembler, you've seen your wages increase by 11.6%. You know who's had it the hardest? CEOs. The high-skilled workers. Those chief executives saw their wages increase by just 2.7%. It's rough out there for a CEO. Brutal. This is mostly because of minimum wage increases. But in general, we're seeing wage inequality decrease, but wealth inequality increase because there is incredible stock market concentration right now. So the Mag Seven, the 7 biggest tech companies in America. They now comprise 32% of the S&P 500 market cap, and they drove 42% of total returns in 2025. So if you're invested in Tesla, Meta, Microsoft, Apple, Google, Amazon, etc., Nvidia, you did very well. Your wealth increased. But if you weren't in those and you had a broad index or you didn't have a lot of capital to begin with, you were left behind. So that's increasing wealth.

Paul Graham shared 2 2 days ago. Companies grow fast now. That's the reason economic inequality is increasing, not some sinister policy shift. He said, when companies grow fast, it makes founders doubly rich. The company not only hits a given revenue number sooner, but it is more valuable when it hits it because the value of the company will be a multiple of the growth rate. People who don't understand the math of valuations can't imagine that founders would get so rich naturally, whereas to founders and investors, it's the most obvious thing in the world. This is one of the reasons there's such a disconnect between the tech world and politicians.

So wealth inequality, AI will probably change it. At the very least, it will change it as much as previous technology, technological booms. If you were the founder of Instagram, you did very well because there was a lot of wealth creation all of a sudden in one major tech boom, mobile. AI is broader and bigger already than mobile. So moving over to energy and data centers, if it If it does become an energy game, what's the trade? It did become an energy game, and the trade was buy everything, basically, because every energy thing basically did very well. Anyone who got the trade correct did very well. Vistra returned 321%. It was the second-best performing S&P 500 stock of 2024. Wow. You know who beat them in 2024? Palantir. Palantir mooned. They were already in the S&P 500. They did very well, but Vistra was sort of the secular energy energy winner. Constellation Energy tripled in size after ChatGPT's launch. NRG Energy gained 95% in 2025 alone.

Nikola surged 700% in 12 months, and that's without, I think, even having any producing any energy. No, just the idea that they will produce energy someday. Yeah. Across the next step, across the entire data center supply chain, which components are hardest to scale up 10x? What is the chip on wafer on substrate of data center?

Yeah. So chip on wafer on substrate, that is TSMC's secret sauce. That is the most gating factor in scaling TSMC's 2-nanometer chip production. It's what allows them to package HBM and the GPU all on one chip. And TSMC reported that they were like sold out like years in advance. It was a huge bottleneck. Bottleneck. In the data center world, the biggest bottleneck was probably power transformers. We heard a lot about this. Lead times for new power transformers reached over 3 years in some cases with a 30% supply shortfall. High voltage circuit breaker—

Yeah, remember my mom was in escrow on a condo. Really? And closing was contingent on the development getting a transformer. She waited like 6 weeks or something like that. They were like, we don't know when we're going to get it. And she ended up just backing out.

Transformers, the cost surged 150% since 2020. It's a 100-year-old technology. It's not, it's not new. It's not crazy innovation. Overnight success. Overnight success. But it became the binding constraint on how fast data centers could connect to the grid. Nations. Who wins and loses? Take a wild guess. Who won? America, baby. America won. Next question. But it's really true. Like, the stats are crazy. The United States is truly the dominant winner of the last 2 years in the AI era. $109 billion in private AI investment in 2024 alone. Way more now. If you look at $200 billion CapEx guide from AWS, $110 billion going to OpenAI, $30 billion going to Anthropic, tons of money going into XAI. Like, the money is truly flowing in America. In 2024, China had just $9.3 billion invested in private AI companies. There's $470 billion cumulative since 2013, more than all other countries combined. The US produced 40 notable AI models in 2024 versus China's 15. Now, the game's not over. Lots of countries are making big investments now.

And you can't forget France's €30 million.

They put in €30 million. Yeah. The question is, if we are displacing software engineers, will software engineers become machinists? This is what Daniel Gross asks. He says, what is the Euclidean distance of reskilling in prior revolutions and how does AGI compare? The typist became an executive assistant. Can the software engineer become a machinist? We're not seeing this yet. Software engineers aren't grabbing blue-collar jobs just yet, but there is a divergence that's starting to show up in the data between software developers and programmers. So programming jobs are in decline, but software engineering jobs are actually growing. And so I think what's happening is that as coding models and agent systems allow for building systems at higher levels of abstraction, demand for AI engineers has grown 143%. And so if you're going to hire someone to help you build software, you want them to be like AI native. You want them to go beyond full stack, which typically meant you could do front end in JavaScript and back end in Python. And now full stack or AI engineer means everything from prompt to design to product development to deployment to operations and DevOps to database migrations to front end and back end, because all of that is going to be handled by agentic systems.

I am pretty bullish on like the like the light blue color, like thesis, which is that like you'll have robotics, but you'll just have like, yeah, so you'll be in the machine shop, but you're just using an iPad. You're not actually working on stuff.

Yeah, I think we drop you in a Nike factory, you're running that place in like 2 weeks, guaranteed. I'm not kidding. I'm so bullish on young people who can use all the tools and actually go and understand the leverage that they're getting from the modern systems natively and not need to fall back to old habits.

Yeah, there was a good journal article yesterday about the Colgate, like, head of AI. Yeah. And it was very interesting because I was like, oh, this is like, I feel like I could do a pretty good job at this.

You're like, ah, yeah.

But, you know, he's just basically telling everyone like, yeah, guys, you actually got to use AI, like, despite what these random surveys might say.

Guy has had one job in his life working on a podcast and thinks he could go dominate in the big toothpaste space. I love the corporate athlete.

You are a corporate athlete. We got to get him. We got to get the Colgate guy on the show. Is lifelong learning worth investing in? Something worth doing beyond the economic value of mastering the task? It's a very abstract question. It's a very personal question. I tend to think yes. I tend to think that on Maslow's hierarchy of needs, you know, you need food, shelter, family, friends, clout or whatever. At a certain point, learning a skill for the sake of learning that skill is edifying in even in a world. You know, it's like going to the gym. Like I was listening to Sam Altman talk about how kids will never, kids that are born today will never be smarter than the smartest AI systems. And that's weird, that's different. But I was thinking about like, But like, I was— there was a time hundreds of years ago when you could actually be the best thing in the world, the best option for like carrying lumber before we created machinery, like before we created the car. Like you could be the fastest person. And then as soon as the ZR1 was released, you didn't stand a chance. And I had to explain this to my son at some point. He's very interested in cheetahs. And I was telling him like, cheetahs are the fastest animal. And he was like, but is it faster than a car? And I was like, not even close. Like, like me in a car, I'm smoking that cheetah. I am slower than that cheetah, but I'm way faster.

Yeah. So in my Cadillac, DG says something worth doing beyond the economic value of mastering the task. Yes. And the way I would kind of flip this, I feel like most of the gains in my life have come from mastering myself or understanding myself, learning about myself, and then understanding the world, and then combining those two things in a way. It hasn't been about mastering any one specific task. I still think there's tremendous, tremendous gains to, again, understanding yourself, understanding the world, how do these two things fit together.

Even in an ASI world where there is no economic value to any task mastery, I still think it is worth investing in lifelong learning. Because that will be edifying, that will be satisfying.

Yeah, just enjoyable.

It, you know, it's becoming a painter, becoming someone who gets joy purely out of the process of painting the landscape. When cameras exist, when generative AI exists, it can still be edifying and valuable to sit there and look at a landscape and paint it, even if the result is not economically valuable. The bigger question and the interesting, like, hot take question he has in here is, does node process matter if the country has more energy? So China has not been able to get their semiconductor manufacturing, their fab champion SMIC. They have SMIC, which is the TSMC equivalent. They have Huawei, which is the NVIDIA equivalent, and they have SME, which is their ASML equivalent. They have not been able to clone TSMC to the level of Taiwan Semiconductor, but they can produce 14-nanometer chips. And so the question is, usually they take those chips and they put them in just random, you know, consumer electronics goods that they ship over here. So if you go to China and you want a co-packer or manufacturer to build you something and you're like, yeah, I want it to connect to the internet or I want it to have a speaker inside, they're like, great, we will fab a 14-nanometer chip or something even less frontier than that. But can they just marshal a ton of lagging edge capacity and then just say, hey, you know what, we have cheap energy, we're burning coal, we have nuclear, we have the Three Gorges Dam, we have hydropower, we have so much free energy, let's just spend 10 times as much energy on the lagging edge. Can we achieve AGI that way? And it seems like no. It seems like leading edge nodes are incredibly important. You can't just throw a ton of lagging edge 14-nanometer chips at the problem, at least with current architectures. Now, this might change. No front tier model to date has been trained on hardware older than 5 nanometer. So China's best effort, Huawei's Ascend 910C, is on SMIC's 7 nanometer class DUV process. They're not at, it's extreme ultraviolet lithography. They're a deep, deep ultraviolet lithography process that's competitive for inference, but requires dramatically more chips and energy for training at scale. So brute forcing AI progress through disruption disregard for energy consumption, it feels like it still hits economic walls at some point.

And last but certainly not least, DG asks, what is the likely Taiwan event and what would be a leading indicator for it? Taiwan blockade would be the biggest trigger, but Taiwan Strait tensions are already escalating. China conducted joint Sword 2024B exercises in October of 2024 surrounding Taiwan with coordinated military operations. In December 2025, Justice Mission 2025 deployed over 100 aircraft, 90 crossing the median line. 13 warships and 27 rockets fired from Fujian. 10 rockets landed in Taiwan's contiguous zone. Contiguous zone. Contiguous, sorry.

Um, 12 to 24 nautical miles offshore. So they're just like firing rockets and being like, they didn't hit land, they're just— you can see them from the beach, probably 10 miles offshore, 12 miles offshore.

TSMC is planning ahead, working on a fab complex in Arizona, which Tyler visited. Should be able to handle 30% of total advanced chip production at scale, but it's on a knife's edge.

If you want to know roughly how far that is, you've seen Catalina Island off the coast of LA. Catalina Island's, I think, 26 miles. And so imagine—

So very visible.

Yeah, you can see Catalina Island from Long Beach and imagine a missile coming and landing like halfway between where you are and Catalina Island. Crazy.

Very. Yeah. The question is, how does Iran in the conflict there update China's thinking on this.

There's—

yeah, some of the intel accounts have been sharing, who knows if it's true, China actually has operatives in Iran kind of learning in the same way that the US has learned from Ukraine, specifically in regards to missile capabilities. So yeah, it just feels like a day hasn't passed so far since this was published that Taiwan, a Taiwan event feels less likely than it did the day before. President Trump gave a phone interview in which he said, I fired Anthropic. This Anthropic is in trouble because I fired them like dogs, which you don't fire.

You don't fire dogs. I would have fired my dog years ago. He's the least effective dog.

Yeah. Isn't the phrase he, he died like a dog?

Yes. Yes. He's just adapting that.

But he said Anthropic to the farm.

No, if he's comparing them to dogs, it's man's best friend. This is bullish.

This also coincided with Anthropic, I guess, officially being designated as a supply chain risk, which is, again, not something that I've seen a single person in the industry actually push for. Maze says, imagine autonomous weapons powered by this. You got a screenshot from Claude Opus 4.6. Mays says, tell me a color and I'll try to guess it. Claude says blue. Mays guesses blue. Claude says nailed it. Nailed it. The Trump administration is drafting rules that would require US approval for nearly all AI chip exports, giving Washington sweeping power over companies like Nvidia and AMD. The draft framework sets licensing rules based on shipment size, from simplified reviews for small orders to government-level approval for massive deployments, potentially tying exports to security guarantees or US investments. Officials say the goal is to make American AI the global standard while controlling critical infrastructure through delays or strict conditions.

Even in this context, like the Anthropic thing is so interesting because like if you take this as saying the US is going to be more restrictive, like that's what's always been said.

Yeah, exactly.

Exactly. It's like, and that's what this reads like on paper. He almost like agrees with.

Yeah.

Totally. And in action, it's like completely different.

Is this a piece on the chessboard? Is this a chip that's being traded metaphorically to Dario in some ways where, where, you know, it's like, how do I win over the labs, get them to work with me more effectively? Well, I'll give on the thing we agree on, which is export, export controls. So I will be more aggressive about export controls.

Robin shared SpaceX probably has revenues less than $20 billion and loses money after the merger with XAI. But is targeting a $1.75 trillion IPO. Alphaville explored how Musk might try to pull off what could be the biggest bag holder operation in history.

I love it. That's such a good one.

Star Trek's warp drive allows a starship to bend spacetime and exceed the speed of light without breaking Einstein's general theory of relativity. As SpaceX prepares to go public at a reported $1.75 trillion valuation, Elon Musk is attempting a comparable feat, defying the laws of financial gravity in pursuit of something the equity capital markets have never witnessed. Elon has been extremely consistent in being able to defy the financial laws of gravity. For early investors, bankers, employees, and advisors, the fees, returns, prestige, and bragging rights would be stratospheric at the mooted valuation. Even a tiny 3% float would comfortably eclipse the US IPO record set by Alibaba's $25 billion float in 2014. A rumored $50 billion raise sounds formidable until you consider how much larger public markets have become since Alibaba went public. Back then, Apple, then the world's most valuable publicly traded company, was worth around $600 billion. By the end of 2025, 10 companies had surpassed $1 trillion, and Nvidia today around $4.5 trillion. Musk is arguably the most impactful entrepreneur of his generation. And has built a remarkable company. The question isn't whether there will be demand for SpaceX stock, but rather what IPO investors would— should pay. Based on what has been reported, this price being contemplated is truly out of this world. The super jumbo deal promises not to be a conventional exercise in price discovery. Musk has reportedly timed it to coincide with the alignment of the planets Jupiter, Venus, and Mercury. But But that's one of the more humdrum features.

Head over to Culshi to track when will SpaceX officially announce an IPO. Currently, before August 1st, 2026 is at 79%. So maybe September, maybe October, but it's coming. Anyway, thank you for watching. We'll see you tomorrow.

Have a wonderful evening.

Goodbye. Goodbye.