Skip to content
All-In PodcastAll-In Podcast

Why Bernie's AI moratorium would hand China the win

A federal AI moratorium proposed by Sanders cannot freeze the math; China's lithography program and Future of Life funding define the real stakes.

Jason CalacanishostDavid FriedberghostChamath PalihapitiyahostBernie Sanders (clip)guestHost (uncertain which bestie)host
Dec 19, 20251h 30mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:00 – 0:19

    Bestie intros!

    1. JC

      All right, everybody. Welcome back to your favorite podcast, the number one podcast, in fact, in the entire universe, the All-In Podcast with your besties. We're all here. It's the original quartet. Everybody loves when the core four are here, and we have a docket for you today.

  2. 0:19 – 17:15

    Bernie Sanders calls for AI datacenter moratorium, how to solve AI's negative perception

    1. JC

      Guys, we gotta start out with Bernie Sanders. I know this is becoming a bit repetitive, but AI is the topic.

    2. DS

      (laughs)

    3. JC

      We reached a new level of retardation this week that we cannot avoid. Bernie Sanders has a major decel pitch, a moratorium on new AI data centers. Here's his argument. Number one, the billionaires are pushing AI because they want more money and power. Number two-

    4. DS

      (laughs)

    5. JC

      ... it's gonna be massive unemployment, and he cites Gates, Staria, Elon saying that AI- AI would replace most jobs. Number three, he, uh, has an interesting point actually. AI is harmful to kids because it decreases social interaction. Actually, we kind of agree with that one, I think, across the board. But here's his pitch, Sax. His pitch is-

    6. DS

      Speak for yourself.

    7. JC

      (laughs) You want your kids using these chat bots? I don't think you do.

    8. DS

      No, I've- I've talked to my kids about it and-

    9. JC

      Oh, okay. Well, we'll get into it, huh?

    10. DS

      I mean, we can get into it, but I've talked to them about it.

    11. JC

      We'll get into it, huh?

    12. DS

      Look, if you actually look at the data, if you look at what kids are doing, it's so much more interactive and engaging for them to talk to their friends on Snap or to watch videos on TikTok. Those things are super engaging-

    13. JC

      They're playing Roblox together.

    14. DS

      ... whereas doing research on an AI chat bot is just, it's way less engaging. And you see this in the data. So I'm not saying there's not an issue there. You wanna pay attention to the way that these technologies are shaping the minds of young kids. But I think p- people get a little bit confused. What they're really talking about is social media, and then they attribute all the ills of social media over to these new AI chat apps, and they are a little different. You know, when I talk to-

    15. JC

      Yeah, I would-

    16. DS

      When I ask my kids, like, "How much do you use these things? Are they addictive?" They said, "No, they're more just really useful." It's like Google. It's like-

    17. JC

      Yeah. There's two-

    18. DS

      ... I couldn't-

    19. JC

      There's two-

    20. DS

      I couldn't do school without it.

    21. JC

      Yeah, there's two pieces here. One, using AI to be smarter and to learn stuff and ask questions. Absolutely fantastic, phenomenal. I think we'll all agree on that. There is, and I don't know if this is actually what he was referencing, but there's Character.AI in a- in a long tail of spicy chats where people... We talked about getting one-shotted and these parasocial relationships. That's, I think, what he's referring to, but maybe I'm reading it wrong.

    22. DS

      Yeah.

    23. JC

      But let's get into it here. His pitch is that we need to slow down so that, quote, "democracy can catch up" and that Congress should put a moratorium on new data centers. That's his solution. He got roasted, obviously, on social media for this. It's the most absurd, I think, solution to, you know, what are valid concerns. Friend of the pod, Ro Khanna, who represents Silicon Valley, he, I guess, supported it a bit, and then he replied to me and says his concerns are not as acute and that he just wants to make sure we use renewable energy, which I guess is, you know, a fine position to have. Polymarket put up a little, uh, market here. "AI data center moratorium passed before 2027." Yeah, we'll see if that comes true or not. But, uh, hey, Sax, you are our SGE. Sax, our SGE, this is your time to shine. (laughs) You have framed the AI debate as we can't lose to China. Bernie framed it as not letting the billionaire class get more power and money by eliminating American jobs. And last week on this very program, Tucker framed it as, what are Americans gonna get out of this, right? That- that was the framing, you know, as of December of 2025. So here's your chance, Sax. Why should Americans who have some concerns about losing their jobs care about beating China? Why should they care about that, Sax?

    24. DS

      Why should they care about beating China? Because AI is a profound technology that's gonna have huge economic and national security implications. And the thing that Bernie gets wrong is that he can't stop the progress. I mean, he can't stop China from making progress. We can stop progress in the US, but it's not gonna stop China from advancing these technologies. A lot of this is just math, and just because we stop doesn't mean China's gonna stop. So this would be the biggest own goal ever if we took our leadership in the AI race and just handed it to China. But look, I appreciate Bernie's honesty in a way, because he is actually telling the truth about what he wants. And I remember just a week ago when President Trump signed an executive order to advance a national AI framework, a lot of critics and a lot of Democrats were saying this was a violation of states' rights and we need to support the states in regulating AI. And here Bernie is acknowledging that this is not about states' rights, because he's saying that there needs to be a moratorium on new data centers, even if the states want them. So he doesn't support states' rights either. He just wants all the progress to stop. And I think this is really the truth of the matter, is that all this talk about states' rights or affordability, it's all a red herring, and what's really going on is there's a large and growing contingent of people who just want all the progress to stop. But again, we can't stop China from making progress, so all they would be doing is ceding leadership of this AI race to China. What people like Bernie really want is they want the US to become like Europe. You know, Europe has half the share of global GDP that they had 30 years ago, and that's because of their hostility towards innovation and technological progress, and that's kind of where Bernie wants us to be, is he wants us to go down the- the path of Europe. And the reason, he says, is because God forbid someone gets rich. Well, look, capitalism sometimes results in the unequal distribution of wealth, but socialism always results in the equal distribution of poverty and misery. And if the US stops developing AI, if we hand this leadership to other countries, it'll make the United States poorer.It'll make the American people poorer, and it will cede our leadership globally. We will not be the preeminent power in the world anymore.

    25. JC

      Okay.

    26. DS

      So I'll stop there.

    27. JC

      No, no, it's great. Thank you, SGE. David Sacks (laughs) over here. There seems to be a couple of conversations, unfortunately we weren't here last week, but we had a really nice, uh, visit from our pal, Tucker, and we talked about this. He seemed to think there was a bit of a communication problem with selling to the public the value of these technologies. Do you agree with that? And what do you think is going on here in terms of the cross-conversations? One group wants to de-sell, has some concerns, environmental, you know, job displacement, et cetera. And then there, is there an issue where the China, beating China doesn't resonate with the American populous? You've been harping on about the rise of socialism for four-plus years on this very podcast. So from where you sit, what's the actual disconnect in these multiple conversations that are going on concurrently, and- and seemingly coming to a head here at the end of 2025?

    28. DF

      Well, if you ask any of the politicians that are making these proclamations about the, quote, "tech barons" and try and actually talk to them about what is the AI technology delivering, what is it not delivering, you kinda typically hit a wall pretty quickly. It's very hard for folks to articulate what is actually happening. There's a tremendous amount of capital being put at risk against infrastructure to build out the capacity for many companies, the entire economy, the entire industry, to deliver AI tools. There isn't an aggregation at this stage of value creation, with the exception of NVIDIA, which has really got this $3 trillion market cap creation that's happened in the last couple years. Peter Thiel said this best. He's like, "Look at all the money that's going into AI. There's really only one company that's making any money, and that's NVIDIA." Like, at this point, the jury's still out. We don't even know what AI is. It's sort of like when the internet was happening, you know, everyone thought these fiber optic switch companies were gonna make all the money. Turns out that was wrong. It was the end applications that made all the money, and they competed in many different markets from Google to Amazon to Uber. You can go down the list of all the beneficiaries of the core infrastructure technology of the internet that was built out. So what is really going on? Well, AI is the new lightning rod for fear and for divisiveness that ultimately breeds compliance and control, which is where these politicians are trying to drive the- the populous and the voting conditions in the United States, and that's what's going on right now. There's a lot of fear about, "Oh, putting this data center in my town is gonna do X or Y or Z," with no real conversation about the truth of that matter. There's a lot of fear about wealth creation being aggregated in the hand of a few, when, as we saw with the internet, it benefited the many. And that fearmongering is a very similar tactic that we've seen in the prior generations, where policies were misstated, fear was used, and then voting control allowed folks to come to power that were looking for power. So I think it's just the lightning rod at the moment.

    29. JC

      Okay, Chamath, what's your take on this big picture of what's going on here?

    30. CP

      I think it's important to understand that politicians have an incredible sense of self-preservation. So the question is, why would Bernie Sanders think he would not look totally foolish in putting that video out? And the reason it went viral is not because it sounded so crazy. It was that, to some faction and percentage of people, it sounded rational and reasonable, and the problem that highlights is that we have a huge perception issue in AI. We have a handful of companies. All the PR that you see from those handful of companies is a bunch of circular deal-making, a bunch of capital that flows from one to the other. It causes these stocks to go up, of which a small percentage of people benefit, and at the tail end of it, it's accompanied by a completely different set of articles that everybody also reads about the Sword of Damocles that's about to fall on their head, whether it's electricity prices, or whether it's their jobs, or whether it's the jobs of their children, or the quality of their education. So we have a big perception problem. So the question at hand is, how do we fix it? How do we get back to the place where a video talking about stopping all progress would seem as laughable as it should be? And I think you can go back to the Gilded Age, and you can ask the question, how did the industrialist leaders of that era respond through that 1880 to 1920 age of industrialization, when you had all of this technological upheaval accompanied by a handful of people with incredible success, right, Andrew Carnegie, John D. Rockefeller, Henry Ford? What did they do? And I think the lesson we can borrow from them is we now need to be on the forward foot as an industry. Enough of the stupid haircuts, dumb watches, ugly clothes, ostentatious displays of wealth. We've all done it. I've been a- guilty of it. It has to stop. Absolutely stop. And instead, we need to start to use a percentage of the balance sheets of these companies in order to benefit as many Americans as possible. That is the absolute minimum. Andrew Carnegie built 2,500 libraries. The idea was, as he built the railroads, you're going to scale GDP, you're going to scale education and knowledge. Those libraries are artifacts that allowed people to feel a dividend from that industrial revolution. John D. Rockefeller took all of his wealth and invested it in institutions and universities. Henry Ford specifically focused on wages. We need to selforganize better, and we need to be more on the forward foot. We need to start doing things that are practically measurable by tens of millions of American citizens, and it starts to beat back the perception problem that we have because, as Sacks pointed out last week, the data does not support...... the misperception. But the problem is, and Sacks and I were talking about this last week, the misperception is gaining steam. And so, I think we need to use these tools that we have to provide some more practical benefits that every man, woman, and child can feel. Otherwise, this thing is gonna build and you'll see crazier and crazier versions of this Bernie Sanders video.

  3. 17:15 – 32:39

    Anti-AI astroturfing: Is the AI sentiment manufactured by special interests?

    1. DS

      Now, why do people wanna sabotage it? There's a really interesting article on Semaphore recently that described how AI critics were funding journalism fellowships at major publications like NBC News, Bloomberg, Time, The Verge, LA Times which, by the way, I mean, I track these things, they're relentlessly negative about AI. These fellowships were funded by Future of Life Institute, which is a doomer group that thinks that AI is gonna become sentient and replace humans, and they were funded by a donation by Vitalik Buterin from Ethereum. It's an interesting story actually. He donated his Dogecoins to Future of Life, they ended up being worth $600 million.

    2. JC

      (laughs)

    3. CP

      Dogecoins?

    4. DS

      Doge and then remember how there was, like, those other dog coins?

    5. CP

      Oh, Shiba Inu.

    6. JC

      Oh, Shiba Inu.

    7. CP

      Shiba Inu. Yeah, yeah.

    8. JC

      Oh, Shiba Inu. Oh, he had a collection of dog coins. Got it. (laughs)

    9. DS

      He had a collection of dog coins. Apparently, people, like, airdropped them to him as, like, a promotional thing.

    10. JC

      ... got it.

    11. DS

      I don't think he wanted them, so he's like, "How do I get rid of them?" So he donates them to Future of Life, and they end up being worth $600 million. So by this-

    12. JC

      Oh, my God.

    13. DS

      ... almost, like, this accident, this Doomer think tank ends up with a-

    14. JC

      (laughs)

    15. DS

      ... $600 million war chest. And they've been funding these journalism fellowships. They've been funding grants for academics to study AI. Obviously, that's gonna end up being very negative. And they are funding a lot of these NIMBY organizations that are opposing data centers, because their goal is just to get it to stop. That's their goal. They just want the development to stop. And you can't, I don't think, underestimate how much of an impact this has had on the public discourse. But if you look at their actual claims, like for example, the water use claims, it's a total hoax. I mean, these-

    16. JC

      Total hoax.

    17. DS

      ... AI data centers do not use a lot of-

    18. JC

      Yeah, that was one-

    19. DS

      ... a lot of water.

    20. JC

      ... of the things I wanted to get into, because yeah, compared to a golf course, they're not using that much or-

    21. DS

      I think, Jason-

    22. JC

      ... or maybe walnuts and, you know, uh-

    23. DS

      No, because-

    24. JC

      ... almonds in Central California. So I think it is worth maybe just unpacking-

    25. DS

      Jason, they're throwing a lot of spaghetti at the wall. Look, you have to remember, who was- who was Ida Tarbell? 'Cause it's quite interesting in the context of this. Ida Tarbell was this American writer, again, in the Gilded Age. She was part of the Muckrakers, which was a group of journalists that were doing investigative research around the abuses of the Industrial Revolution, labor abuses and the like, and she took on Standard Oil. Exactly. If you look at that example, I'm sure that the people in this current generation who have a Tarbell Fellowship, what they're living out is this desire to tell a story about exploitation and wrong and misdoing. The problem is that these things don't factually hold together, but unfortunately, it adds to the perception problem that we have, and that has been growing. As Sacks said, like so many of these things have been debunked, but it's playing whack-a-mole, Jason, because they'll throw the water thing out. It'll get debunked. Next week, we'll probably something about liquid cooling and all the PFAS and forever chemicals. We'll have to debunk that. Then we'll go to, like, air cooling and how that's a problem. Then we'll go to something else. Eventually, it'll be the upper land grouse-

    26. JC

      (laughs)

    27. DS

      ... but they're not gonna stop.

    28. JC

      Sacks, how do you- do you have a plan to reframe this to the American public? You're explaining how these bad things are happening and all the evil forces at work behind the scene and debunking all that, but-

    29. DS

      Well, look, it's not evil forces, but I don't think people understand the extent to which the discourse has been impacted by a few anti-AI tech billionaires.

    30. JC

      But you've brought it up-

  4. 32:39 – 51:44

    Economy: Unemployment, jobs, inflation, are we primed for a "Golden Age" or not?

    1. JC

      Topic number two. Economic numbers are out. It's a mixed bag. Unemployment rate is up and government payrolls are down, inflation somewhere in between. Let's get into the numbers here. Fun with numbers. Unemployment rate rose to 4.6% from 4.4% in September, and 4% in January of 2025, when President Trump took over. US economy added 64,000 jobs in, uh, new jobs in November. October saw about 104,000 job losses, but 162,000 of those job losses came from the federal government. You remember the Doge buybacks that friend of the pod, Elon Musk, was involved in. Uh, those all took effect at the end of September, so there was a little bit of a balloon payment that occurred. An analyst for Moody's says, quote, "It's a frozen job market. There's not much hiring, there's not much firing happening." That tracks with what I'm seeing on the ground. Meanwhile, inflation came in better than expected at 2.7%, beating the 3.1% expectation, but still far away from the 2% target for the Fed. Trump gave an 18-minute, very loud, emphatic address to the nation last night, and it was not that we're invading Venezuela. (laughs) It was, uh, mostly him talking about his accomplishments in terms of, uh, prices and affordability, and, uh, a little bit of admonishing Biden. Here's a 20-second clip.

    2. B(

      The last administration and their allies in Congress looted our treasury for trillions of dollars, driving up prices and everything at levels never seen before. I am bringing those high prices down and bringing them down very fast.

    3. JC

      Sax, your thoughts on the economic data here in the fourth quarter?

    4. DS

      Well, you're- you're painting it as- you're painting the economic data as mixed. I don't know how you have a economic report that's better than what we just had today. So first of all, we saw that CPI came in at 2.7%, like you said, but the expectations were 3.1%. That's why the market is rallying today in a big way. Core inflation is down to 2.6%. These are significant beats, and this puts core CPI inflation in the US at its lowest level since March of 2021. So since the whole COVID thing. And if you listen to Kevin Hassett, he said that over the past three months, core inflation has been running at 1.6%. So the trend line is going down even further. So it looks to me like inflation is rolling over, and that's just about a solved problem, which is really good news for what it implies for interest rates, because it implies that interest rates are coming down. And that's gonna bring down things like mortgage costs and the costs of financing a car payment and things like that. Now, you talk about unemployment. I think the unemployment news is very, very good. So if you look at the data from September to November, the overall employment is down 41,000. But why was that? Private employment was up 121,000, but government employment in the last two months has declined by 162,000. So this is what the media was focused on, is they're trying to claim that unemployment was up, but actually, it's just that the government jobs decreased. Why did the government jobs decrease? Well, you remember that when Doge went in and made their cuts at the beginning of the year, they offered people a buyout as of October 1st. Now, some people took the buyout right away, but a lot more waited until the last possible day, which was October 1st. And that's why you saw this big spike in October unemployment. But those are government jobs that are being cut. And moreover, there are people who voluntarily wanted to take the buyout and they took that deal. So again, I think the employment picture is looking actually quite good.And if you believe what we believe, you don't want an excessive number of government jobs. And President Trump has presided over, I think, the first decrease in the federal workforce in decades. We've seen a 10.7% decrease in federal workers in 2025. Basically, the number's gone from 2.4 million to 2.15 million. I know that'll make Freyberg very happy. But look, what we had during the Biden years is we had 9% inflation. We had a very weak economy. We did have a recession. It was that two-quarter shallow recession, and the media even tried to start redefining what a recession was to avoid those headlines. But what the Biden administration did in response to that is they went hog wild with government hiring, and what we're seeing now is inflation is now coming back down, and we're seeing the number of government workers come down to a more reasonable level, and the private economy is making up for it. We still have a relatively historically low unemployment rate. So, those numbers look good. Then you look at the deficit. We've reduced the deficit year over year by 600 billion. That will help bring interest rates down. And then on prices, you got the lowest gas prices in five years, below $3 nationally. And then finally, on wages, real wages are up by over $1,000 on average, and, um, it's $1,300 for factory workers, $1,800 for construction workers. And again, that's a big change from the Biden years, where you saw that in real terms, wages went down by about $3,000 on average per worker. So look, I mean, it seems to me like we're on the cusp of a golden age here. I don't see how the numbers could really be better. And then on top of it-

    5. JC

      Okay.

    6. DS

      ... again, you've got this AI tailwind, which I think is a huge positive, not a negative, which is adding roughly 2% GDP growth every year because of huge CapEx investment, with so far no job loss associated with that. So, I would just say sit back and enjoy this. I think we're headed for a gangbusters 2026. Rates are coming down, inflation's coming down, and you're also getting tax cuts going into effect next year because of the Big Beautiful Bill. No tax on tips, no tax on overtime, no tax on Social Security, plus the standard deductions being beefed up. So, people haven't even felt the benefit of those tax cuts that's coming in in April. I don't see how things could be much better.

    7. JC

      Okay. Freyberg, any thoughts here on the data as it's come in? Or Chamath, I'll, I'll go to either one of you.

    8. CP

      Directionally, I think it's really positive. You wanna see private sector jobs growing, and the shrinkage of government sector jobs has profound impacts, not just on the budget, but it starts to create obvious air pockets where you get a little bit more rational regulation. You can deregulate in the appropriate places. You can shift the burden and the responsibility to private industry. So, it's more than just what's in the numbers that I think is positive. I've told you this story before, Jason. To do some of the things that I've wanted to do... I'll give you an example. When we started a battery business five years ago, when we filed with the DOE, the Department of Energy, these are 700-page reports. You spend millions of dollars and you hire teams of 50 and 100 people, and what are we trying to do? We're trying to build a battery business in the United States so that we can de-lever from the Chinese. To the extent that we could do that a little bit simpler and a little bit easier because there's fewer folks, and so the burden goes to state and local regulators, which we all- already have to deal with. These are just generally good things for productivity. They're generally good things for GDP. It allows us to invest more aggressively in the things that help A- America, so it's trending in the right direction. By the way, Sax didn't mention this explicitly, but I'll say it. The other thing in '26 is you get a bunch of tax cuts that kick in. No tax on tips, no-

    9. DS

      Yeah.

    10. CP

      ... tax on overtime, the deductibility of the cost of interest for things like car loans. These are big stimulative actions for the-

    11. DS

      Accelerated depreciation.

    12. CP

      Accelerated depreciation. You have big stimulative actions for the United States economy.

    13. JC

      Yeah, I think what you guys are missing, respectfully-

    14. DS

      (laughs)

    15. JC

      ... is that, uh, you know, uh, Trump-

    16. CP

      Of course.

    17. JC

      ... promised something completely different. He said, "Starting on day one, we will end inflation and make America affordable again to bring down the prices of all goods." And when he was running for election, he promised the Americans that he would reduce prices. Now, what's happened is prices have not come down. They've gone up 2.5 to 3.1%. But what really matters, and I think you guys are doing a great job, you know, pro-administration, part of the administration, what you're missing is the American people don't believe you, and the American people are experiencing something different. And when you look at the, uh, approval rating, Trump's approval rating is at its historic lows, and specifically on inflation. Pull up the silver bullet and meta-analysis there. This is the disconnect, and this is why it seemed tone-deaf last night when Trump was telling everybody it's great, and when you say it's the golden age, Americans aren't experiencing that. That's just not what they experience. They are experiencing grocery prices that have continued to go up, and they are experiencing unemployment that's gone up 15%. Whether it's the federal employees, whether it's private sector and you debate the numbers or there's a golden age coming because of no tax on tips, these are all talking points. What the American people remember is the Trump administration said prices would go down. Prices have gone up. That jobs and manufacturing, there'd be all these incredible manufacturing things, obviously that's gonna take years. That has not happened yet. So I hope for the best. I hope inflation goes down to 2, and I hope they turn this around, but the American people don't believe the Trump administration, and it's a big difference between promising stuff in 2024 and delivering it. And here we are in the first year, and the delivery that's come in from the Trump administration on inflation and on the economy is not good if you're in the bottom half of society that doesn't own equities. Period, full stop.

    18. DS

      So first of all, prices have come down. You look at gas prices are the lowest in five years, they're below $3 nationally. Now, I don't know how you want a new administration to come in and literally affect every single change on day one. How do you do that? What the pre-

    19. JC

      Oh, no, that's just what he said.

    20. DS

      No, what the president said is they would start working on it.

    21. JC

      That's just what he said. Now, I'm not saying it's possible.

    22. DS

      They would start working on it.

    23. JC

      Uh, hold on a sec. I'm not saying it's possible. That's what he said.

    24. DS

      No, he said he'd start working on it on day one, and so he did, and look at the results. Inflation's now down to 2.7%, way below expectations. I don't know how you want a better report than this.

    25. JC

      Okay.

    26. DS

      Unemployment is very low. The only reason-

    27. DF

      I do remember the, uh, the economic target was three, three, and three, 3% GDP growth, 3% deficit to GDP, and 3% inflation. They're below 3% inflation, they're above 3% GDP growth, and the deficit to GDP is not where it needs to be. So cutting is still ahead. But I looked at the budget for 2026 for a number of the departments in the Cabinet over the last couple days, and they do have big cuts that they're trying to implement across the federal government. We're gonna be interviewing Scott Bessent tomorrow morning. That'll come out shortly after this episode airs. But this is the catch-up conversation with Scott on how's the three, three, three plan going, but on two of the three metrics, J Cale, it does appear like things are good. I'm not trying to be a spokesperson for the Trump administration, but I'd say if you look at this just from the raw economic data and the, the goals of the administration, following that, there should be a flow through in terms of job growth, a flow through in terms of affordability, in terms of wage growth, all those other sort of economic indicators that actually affect people on Main Street every day. Uh, that's all still TBD, and the narrative is still to be written. But those high-level economic goals are still, you know, are starting to kind of fall within the, the framework of what they set out to do, which is generally good, right?

    28. JC

      Yeah, I'm just-

    29. DS

      Yeah.

    30. JC

      ... pointing out the disconnect. Again, this is the disconnect-

  5. 51:44 – 59:56

    Dog Corner!

    1. JC

      to.

    2. DF

      Do you guys wanna hear a crazy personal story for two minutes as a palate cleanser?

    3. JC

      Oh.

    4. DS

      Three, two, go.

    5. DF

      Oh, yes, as a palate cleanser.

    6. JC

      Love it.

    7. HB

      This is an amuse-bouche?

    8. DF

      Okay, this is an amuse-bouche.

    9. HB

      What is it, like a little sorbet?

    10. JC

      A little, a little pamplemousse sorbet?

    11. DF

      All right.

    12. JC

      Quince?

    13. DF

      This is the craziest story-

    14. HB

      Grapefruit?

    15. DF

      ... in my life right now.

    16. JC

      Okay.

    17. DF

      Pull up the photo, Nick, the first photo. You guys remember-

    18. JC

      Crazy story coming up.

    19. DF

      ... my dog, Monty?

    20. JC

      Yeah, this is the virtue signal dog.

    21. DF

      No.

    22. HB

      He's not the virtue signal dog.

    23. DF

      No, this is the best dog. This is... Oh, that's...

    24. JC

      Oh, Monty's the good one, yeah.

    25. HB

      This is the only one I love.

    26. DF

      So, Monty, Monty, Monty was my dog for 10 years. I got him 10 years ago in January. And he died suddenly in May. He, turns out he had a tumor on his heart, and his heart burst open in front of me and he died. It was the worst thing. He was with me every second of every day-

    27. JC

      Mm.

    28. DF

      ... that I wasn't at work. He slept on my bed, he was in my office all day. You guys have seen him a million times in the background-

    29. JC

      Yeah.

    30. DF

      ... during the All-In Pod. My best friend in the world, the greatest s- thing I've ever come across in my life. So, he died suddenly. It was very brutal, very hard. So I'm at my house the other day. My next-door neighbor walks in and he's got this dog on his leash. Pull up this photo. This is this dog. So my next-door neighbor finds this dog at Rocket Dog Rescue. It was pulled out of a kill shelter. The vet estimates this dog is like a decade old. Today, I live probably 40 miles away from where we got Monty. He was found wandering the streets of the Mission District in San Francisco. This random dog here was found at some random kill shelter somewhere in California, pulled out of a kill shelter a few days ago, put into Rocket Dog, and rescued by my next-door neighbor. Brings him over. He's walking him in the street. I'm like, "Oh my God, that dog looks just like my old dog, Monty." We ran a DNA test.

  6. 59:56 – 1:19:43

    China's major AI breakthrough: lithography, impact on the AI race?

    1. JC

      All right, according to a Reuters report, Freeburg, China has built a prototype of ASML's EUV machine. Hey, if this is true, this report, it's, uh, it could have, uh, profound impact on the AI race that we talked about. For those of you who don't know, ASML is the Dutch company worth $400 billion, 22nd largest market cap in the world. And, uh, they are the only company that makes EUV machines, uh, also called lithography. That stands for, uh, extreme ultraviolet. That's how you make H100s-Here's a photo of one of these giant machines. They cost $250 million to make and they take six months to make. Filled with all kinds of important parts, Carl Zeiss lenses, et cetera. Since 2018, these machines have been limited for sale by the Dutch. Why? Why can't China buy these machines? Well, in 2018, Trump's Secretary of State, Mike Pompeo, put pressure on the Dutch to impose export controls on China and, uh, later that same year, they did. So according to Reuters in this news story, Friedberg, China used former ASML engineers, allegedly, to reverse engineer the machines. Reuters has pointed out that the machine in China has not yet produced any working chips. It's a prototype. The CCP is targeting 2028, uh, for working chips. Some sources say maybe 2030 is more likely. You've been talking about China lithography on the show. We can play the Chariots of Fire music that Friedberg insists we play when he does his victory lap come up. Here's his victory lap, Episode 224.

    2. DF

      Last year, China announced and began a $37 billion investment in developing their own three nanometer, uh, chip technology. China, uh, made a claim that this investment they had made was starting to pay off and they had developed their own EUV system. And their big semiconductor company is called the Semiconductor Manufacturing International Corporation or SMIC in China. They launched a seven nanometer chip with Huawei in their Mate 60 Pro, which is sort of like their iPhone competitor in China. And so they're proclaiming that they've already got this EUV technology from what I understand, and Sacks would know better than I, it sounds like there was a lot of reverse engineering and workaround of existing technology in order to deliver that system.

    3. JC

      Got it. Who do you think has the best chance of challenging NVIDIA?

    4. DF

      My early prediction for 2026 is Huawei, where I think that there's lithography technology that exists in China that is not publicly discussed that is going to be deployed in Huawei and all these fabs that they're building in mainland China.

    5. JC

      So announcements 2026, impact '27?

    6. DF

      Probably fair.

    7. JC

      There you have it, first. First on All-In, Friedberg, congratulations on your weekly victory lap here on the show.

    8. DF

      Well, I don't, I don't, I don't... I honestly don't think that the Reuters story is necessarily news 'cause I think it's a little bit of a narrow scope in trying to describe what happened, which is that they, quote, "Stole ASML technology 'cause scientists are working on it." That's not really the full scope of what's been going on in China for a number of years and it really understates the technological progress that China's independently made in using other systems to try and achieve lithography parity. So if you go back a number of years, the current kind of investment vehicle is actually Chinese banks put capital into a fund that's roughly $48 billion, uh, US dollars, which is actually what's called the National Integrated Circuit Industry Investment Fund Phase 3. So there was Phase 1, which started in 2014, which focused on manufacturing, so developing fabs with groups like SMIC. Phase 2 was stood up in 2019, which focused primarily on design and materials. And then this Phase 3 fund was established in 2024 explicitly to pivot to what they called choke points or bottlenecks in the manufacturing process. And state media has confirmed the existence of this Phase 3 fund and the intentionality of the Phase 3 fund to try and replicate lithography technology. A couple of months ago, if you pull up the link, Nick, here's a publication on this paper. So this is the leading research group out of China that's been backing into lithography technology using AI-driven systems. They published this paper, which was a very good summary of where they are in July. Was this in Nature? No, it was in Light Science and Applications. And this paper is called Advancements and Challenges in Inverse Lithography Technology: A Review of Artificial Intelligence-Based Approaches. If you read into the paper and then you read the other publications from this particular research team at Tsinghua University, they have made a number of breakthroughs in using AI to try and reestablish alternative systems in the traditional Zeiss optics and other control systems that are used in the ASML platform. And the way that they've done it, I, I won't spend too much time on it, but there's a bunch of neural networks that they've used, trained, and have made discoveries from, including doing things like working with suboptimal optics, meaning you can have optics that have a degree of diffusion, and then how do you actually recreate three nanometer printing techniques or three nanometer masks by having shadowy effects accounted for in the AI. So you almost, like, assume that the shadowy effects are gonna come out and you print differently than you otherwise would. You don't need the degree of precision that you need and you don't necessarily need to go get the Zeiss optics. There was another paper that was published early this year by another Chinese research team that actually demonstrated how they did use AI to discover a novel method for doing optics. So there is a very broad and well-funded effort that's been underway for over a decade, but just in the last couple of years with artificial intelligence-based approaches, they've made a series of breakthroughs. It is very likely the case that Huawei already has a lithography system that they'll be putting into production and this is why I kind of talked about it-

    9. JC

      Mm-hmm.

    10. DF

      ... when we were in Vegas and pointed it out earlier this year. But I, again, I think Reuters just sort of maybe caught on to some narrow segment, but this is a bigger story and a bigger set of progress that's been made over quite some time.

    11. JC

      Ramification, Sacks? Here and, and what's your thinking? Obviously you've talked a bunch about, "Hey, let's sell our stack in there." And they're clearly working on their own stack. What, what, what's the, what's the ramification of this if it should go?

    12. DS

      Well, I've never supported selling EUV lithography to China. It is probably-

    13. JC

      No, no, selling the stack of H100... Whatever the previous, uh, NVIDIA stack was. Yeah.

    14. DS

      EUV lithography, these machines that are made by ASML, is probably our single biggest advantage in the AI race because they're the only machines that are capable of creating, say, two to three nanometer-... chips, semiconductors, and it's sort of the perfect export control because there's only one company that makes these machines. Like you showed, they're gigantic, they're expensive. So if you think about, like, trying to put an export control on a commodity, there'd be a lot of ways to circumvent it because there's a lot of sources for a commodity. But in this case, there's literally only one company that makes these machines. So it's been a tremendous advantage, I'd say, for the West that we have EUV lithography. I'm not surprised that China is trying to reverse engineer it. It's kind of the obvious thing to do. No advantage is forever. It does seem likely that at some point, they'd be able to figure out how to do it. So I think that this article is not altogether a surprise in that sense. I mean, I'm sure that China is trying really, really hard to reverse engineer this. In the meantime however, they've also made substantial progress with DUV lithography. So the previous generation of lithography before EUV was DUV, E is extreme, D is deep. In any event, DUV was supposed to sort of top out at, say, 14 nanometer chips and China was able to push the technology to get to 7 nanometer and now to 5. So they're able to get a lot further using DUV lithography than anyone thought possible, and that is why the Huawei chips, the Ascend chips are... They're not as good as our chips but they're serviceable and they've been able to use their prowess and networking to string more of them together. You have the CloudMatrix 384 technology where Huawei will string together 384 of their Huawei Ascend chips into a rack, and that will perform comparable to an NVIDIA rack but using a lot more power. So the bottom line is China's been able to create a lot of these workarounds to our restrictions but if they figured out how to reverse EUV, that would be, I'd say, a blow because it is a real advantage for us. But I'm not convinced that this Reuters story is the f- the full picture. I think it's just a data point.

    15. JC

      Mm-hmm. Chamath, any thoughts here?

    16. CP

      If you want a very short summary of how the last generation of chips work, the best way to think about this is that the most profitable and valuable chips to date have taken a very compute-centric approach, that's why we've needed these ultra-advanced process nodes. But the reality is that in a world of infinite inference, let's say, the next generation of silicon will not take the same compute-heavy approach and will probably rely on a much more memory-centric architecture that uses a lot of SRAM. The value of that is that you can produce these things, as Sax said, at 14 nanometer, at 7, 5, all of these process nodes that are much less advanced, that I actually think is the future. So in a weird way, if they steal this, which I think we should assume they've already stole it, quite honestly, and they figured it out, the question is how bad is it? That's the really important question. And the thing about a more memory-centric approach for AI inference or these next generation forms of silicon, it demands a lot more of compilers, it demands a lot more of the software. I still think that's where we are lightyears ahead. So I don't think it's the end of the world and we're still in a very good place, but it is bad.

    17. JC

      Friedberg, weren't we supposed to, uh, start onshoring with the CHIPS Act and maybe get trending towards having this production ability in America? And obviously, that's been really hard to do in China and their relationship with Taiwan seems to be... and, and the possibility that they actually take over Taiwan in a military fashion has been this existential concern, or even short-term concern, maybe is a better way of saying it.

    18. DF

      Look, the- the- the- I think the biggest security implication to consider at this moment is that both the US and mainland China are onshoring, uh, manufacturing, and that makes Taiwan less of a strategic point of interest for the United States and for United States industry. Maybe one of the ways to think about the positive aspect of what is going on, which is maybe call it the Balkanization of generally manufacturing, but specifically semiconductor manufacturing in this case, is reduced security concerns. So there's no longer this flashpoint perhaps in Taiwan. If China did invade Taiwan and the US had fab manufacturing, we're less likely to wanna defend Taiwan. And frankly, if China has fabs on mainland, do they really need to invade Taiwan? What's the economic interest in doing so?

    19. JC

      Yeah, that's-

    20. DF

      So-

    21. JC

      ... kind of where I was going with it.

    22. DF

      So our-

    23. JC

      But if they also wanted to jumpstart, Friedberg, they could just go to Taiwan and take the machines. Like, is that actually the plan?

    24. DF

      That's a- a-

    25. JC

      Like-

    26. DF

      I- I don't know what the CCP's plan is. I think that's a-

    27. JC

      Yeah.

    28. DF

      ... pretty, um, simple statement to make. I think just going to Taiwan is probably a lot more complicated than-

    29. JC

      Of course it is.

    30. DF

      ... than it's stated, so-

Episode duration: 1:30:26

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode GIW1yU9zHW8

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome