Nvidia的历史性季度、SaaS复苏、Bessent对阵Druck、美国债务危机、癌症疫苗

1. Science Corner: Moderna's mRNA cancer vaccine

All right, everybody. Welcome back to the number-one podcast in the world, your favorite podcast, your podcaster's favorite podcast. Yes, it's the All-In Podcast. The Core Four, the Fantastic Four, the original quartet is here: David Friedberg, hot off his deep-state conspiracy interview with Eric Weinstein. Amazing. You did a little run to D.C. Great episode, Friedberg.

It was fun. Eric is one of a kind. I saw Kratsios [?], too. Kind of two sides of the coin of science in America was the basis of those 2 interviews. They were fun.

Who is Eric Weinstein? He's a scientist who works with Peter Thiel. He's done some really—

Yeah, and he had a podcast for a hot minute. He is a heterodox thinker in science.

What is his heterodox thought about science?

There are all these things that everyone takes as fundamental premises in modern physics—string theory, for example—and he called that out. So he's been kind of outcast in the scientific communities.

It is, right? He's made this case and proposed alternative theories that kind of create a grand unified theory of physics. Then he's been outcast by the scientific community, and people say that he's a quack. He's talked about UFOs and Jeffrey Epstein, and people say, "Well, this guy's fringe," or they use these different terms to chastise him.

The point is similar to what Kratsios [?] has said: so much of science has followed this sheeplike mentality where everyone has to line up and agree to the same general theory, or you get outcast. You don't get grant funding, you don't get tenure, you don't get jobs, and you don't get invited to give lectures. You have to follow the mainstream in science, or you're outcast.

I know, but it's very rational to say that until string theory is proved, it's unproved. If you want to use the adjective, that's fine. It'll probably offend some people, but he's more right on the substance of what he's saying than he is wrong. It's unproved, right?

It's like Jay Bhattacharya being like, "Hey, COVID may have come from a lab leak." When people were saying this, they were told they were creating conspiracy theories and pushing misinformation, when there was no misinformation. If there's no absolute truth that you're talking about, it's theory. Everyone was starting with a theory on where COVID came from.

You're talking about a level of corruption in a very narrow, specific realm of science, which is the funding and grant-approval apparatus, and getting tenure and professorships. If you don't follow the prevailing methods and theory, if you don't follow the mainstream, you're outcast. By the way, I buy that. I don't know Eric.

This is what has led to stagnation in science in America, because if you're not part of the mainstream, you get excluded. Everyone now has to think in the same way, so you don't have heterodox thinking or thinking outside the scope. That means we're not pushing the envelope and discovering new things anymore, and that's a very pointed issue right now.

That's why I went and talked to Kratsios [?], who said differently. There's probably a lot of incrementalism in 2026 that didn't exist in 1926. Totally. It's exactly right. Think about the general theory of relativity being published. It was completely outlandish, but people still studied and assessed it.

2. China's Robot Olympics, Optimus update, Grok Bot

They didn't discredit it and say, "Oh, you're an idiot." That's what goes on today in science: if you say stuff that's too far outside the realm of the standard approach to things, you're chastised, outcast, and excluded.

Did you guys see the Olympics? The robot Olympics, Sacks?

I did see it. Yeah, this is pretty crazy. They had these Olympics in China.

Oh, my God, they copied how Friedberg actually runs. Look at that. Look at that. Yes, that is definitely a Friedberg sprint.

The Friedberg sprint.

Doing the Friedberg sprint. Look at this one. He backs up and then—boom.

That's my sprint. That's like me skiing right there.

Yeah, look. And there's Jason setting records at 42. This is Jason right here.

Here's Jason with his shirt off at the poker game. Really great. Put your shirt back on.

You know, I was thinking about this, Sacks. You know what I realized? I think you're 100% right about the PR. I realized that the Chinese Communist Party is brilliant at PR. Look what they did: they got a bunch of Chinese people to fill a stadium and cheer for AI and robotics. They're cheering for it and rooting for it.

Then we got Dario saying all jobs are over, we're going to be the last company in the world, and it's the end of days.

I get back on Dario. We're not talking about Dario this week. We're not talking about Dario this week.

Your grand unifying theory is that in a country of 1 billion people, they couldn't find 4,000 to show up to watch a bunch of robots run around a track. Are you really surprised by that unifying theory?

It is true that China is much more optimistic about AI than we are. AI optimism in China is over 80%, meaning they poll on the question, "Do you think AI will be more beneficial than harmful?" Over 80% of Chinese people say yes. In the U.S., that number is like 30%.

I've been saying for a while that the U.S. is ahead of China in every category except 1, which is optimism. That is the biggest risk to us winning this AI race against China: doing something that will shoot ourselves in the foot because we're so pessimistic about what AI might do.

I think it's all psyops. They're making you feel—do you see that? Show this, the robot blowing up at 16 seconds, Nick. This is all staged. It's a psyop. They make you feel bad for the robots. They make you cheer for them. This is going to be a major thing.

No, it's the red robot. The funniest thing was when the red robot lost its legs. There were a couple of tweets that I thought were hilarious. One guy was like, "Oh, my God, he tore his HDMI, his CPU." The second one—which is not funny because I know Christian McCaffrey and he's a great guy—was, "He's Christian McCaffrey in Week 3," and you just see the entire leg snap off. Very, very mean.

Congratulations to the CCP. By the way, have you guys seen the latest Optimus videos from Elon by chance? Has he shown any of you guys?

Haven't seen it.

He's not publishing publicly. I saw the latest Optimus. It would win half of these things already. You guys see it? He's making sick progress, and he's not showing it publicly, but I saw it.

Without the sparks coming out of its midsection.

No sparks, but I saw a video that I said to him, "Is this CGI, or is this real?" And he said, "No, this is real. They just sent it to me this morning." It's like, "Whoa." Optimus will be the best-selling product, I think.

Yeah. The hard part, as I understand it, with these robots is that you can train them to do a specific action, like run in that race or jump or whatever. But the hard part is when they encounter a condition that they weren't programmed to expect, specifically when they have to interact with something in the physical world. That's when it gets tricky, right?

Can they fold laundry or pick something up without breaking it or dropping it? Specifically, can they do something they weren't taught to do? That's where it gets tricky.

Yeah. The fidelity of Optimus and its ability to do novel things—you kind of nailed it. I don't want to say any more.

You've been playing with Grokbot, though, right, Sacks?

Yeah. Oh, yeah.

You've been Grokbot-maxxing.

I just reached my Grokbot usage limit and have to upgrade.

$200.

Yeah, it's been really interesting, actually.

Yeah. What? It just does what you tell it to do, right? It just gets it done.

Yeah. What's great about Grokbot is that it's always on. It's in the cloud. You set up your agents, and then they can keep working even when your computer is off or goes to sleep.

The other agent harnesses that were doing this, like OpenClaw or Hermes, were just on your desktop. If your computer went to sleep, they would stop. You had to do all these Mac Minis, and they were also very hard to use, very powerful, and super complicated.

This is by far the easiest. It's still a lot of work to get it set up and working perfectly, but it's fun, it's the easiest it's ever been, and you can really see where it's all going.

Yeah, it's pretty amazing.

I told him he should make it multiplayer mode, so you can put 2 agents in the same room. You could also put 2 people from your team in the same room and have humans in the loop.

Such a good idea. That will make this thing go so freaking viral. Yes, if you can just add—because what's cool about the agents is that you actually have an agent swarm. You have lots of different agents performing different tasks, and they get good at different things.

Some people ask, "Why don't you just have 1 agent?" It turns out that if you have multiple agents, the agents develop more context and expertise, so they get more specific. You could then use each of those agents as a channel, invite people to it, and have them participate in that. Then they add another person.

Yeah, then it could get very viral.

3. Nvidia and Salesforce rip after big earnings: AI Capex Bubble and SaaSpocalypse narratives get reversed

Yeah, it’s a great product. All right, let’s get to work. We’ve got a lot going on. Nvidia just had a record-setting earnings report. I mean, they blew the doors off this. We’ll also talk a little bit about Salesforce coming back and the SaaS apocalypse ending.

Nvidia’s stock was up 9% on huge numbers and even huger guidance: $96.2 billion in revenue in the quarter. Nvidia’s revenue was up 106% year over year, more than double. Wall Street’s expectation was $92 billion. They guided for 70% growth next year, which is well above what Wall Street was expecting at 45%. So, let that sink in.

The number people are going crazy over is $60 billion in net profit. That is the most profitable core-business quarter of any public company ever. Google and Amazon have had higher one-off quarters, obviously. Here’s your chart: Nvidia net income and revenue quarterly, just blowing the doors off of it. The stock is up; it’s reaching its all-time high again, getting close.

Heading into earnings, Nvidia was only up 12% year to date, on par with the S&P 500, but it was lagging the broader chip index. I guess the headwind, Chamath, is that people believe there’s competition coming. We’ll talk about that. Nvidia is the world’s most valuable company at $5.5 trillion.

Polymarket says there’s a 79% chance Nvidia ends the year as number one. Here’s your Polymarket, our great partner Polymarket: 79% chance, with $6.3 million in volume.

In other news, Salesforce was up over 20% on Thursday after also reporting a big quarter. Revenue was $11.3 billion, up 11% year over year. That was in line with expectations, but they beat adjusted EPS by 80%—$5.90 versus $3.27. They got a big jump because they are major investors in Anthropic. They raised their full-year guidance to $46 billion.

You called it. Let’s give Chamath his victory lap. Here he goes—the victory lap clip: Chamath on May 15. What’s your take on this? You’ve been pretty critical of the SaaS space in previous episodes.

I think the high end of the market, where Marc operates, where the large monoliths operate, is quite safe. What people are finding is, “Hey, hold on a second. This is a lot harder than we thought.” It’s not like, “Boop, boop, boop, put in a prompt,” and beep-boop, it all works. It’s not how it works.

I think we’re a little oversold. Now, I think this consolidation and the rerating can happen in the opposite direction. So, what is the opposite trade? The opposite trade is: Who has constructive net dollar retention? Who has negative churn that’s been really predictable? That’s a way of saying, who has the best relationships? Those guys, I think, are positioned to crush.

All right, there you have it, Chamath. Your thoughts on Nvidia and Salesforce blowing the doors off their quarterly earnings reports.

There’s a lot there. Maybe we’ll start with Salesforce first, and then I’ll go to Nvidia, because I think the Nvidia one is actually a good segue into talking about the national state of finance.

I don’t know. It was pretty obvious to me in May that Salesforce specifically was meaningfully oversold. I think what I was trying to say there, and maybe I can say it more precisely now, is that we’re getting to the end of this second phase of AI.

If the first phase of AI were models—and think of a model as a brain—the second phase of AI is harnesses, right? Agents, which is the analogous equivalent of giving a brain a pair of eyes and hands, a notebook for memory, and a keyboard to type things on. You’re taking this primordial object that was really good at Q&A and making it more useful, almost like an autonomous being.

But we’re starting this next phase soon. At least, I see it in the enterprises that I work with, which is that this is still insufficient. What you now need to do is make that autonomous agent much more informed, give it context, and allow it to do a job much more intelligently.

You take a brain, give it arms and legs and limbs and eyes, et cetera, but then you have to train it to be a lawyer, a customer-service rep, a sales agent, or what have you. To do that next step, you need a ton of contextual information.

The people who control that today—the large systems of record—what I said before and what I’ll double down on now is that those guys hold an incredibly special place in the ecosystem if they do it right. I think what you’re seeing is Marc finding a way to play well with the models, support harnesses, but also build his own, and now start to push into this next phase.

I think that’s why his net dollar retention is strong. That’s why his revenue is strong. That’s why he’s guiding up. That’s why the stock ripped. I think it was up 43% since I said it had bottomed.

Yeah, it was up 20% today just on that. And Friedberg, I think you were in—were you on CNBC, or were you just on the quarterly call?

I went to Benioff’s office for the earnings setup yesterday. Dario was there. Dario and Marc did their CNBC bit. They set up a studio for the earnings call on the 60th floor, where we’ve all had dinner up there. Then I did the earnings call with Marc.

About a year ago, I think I mentioned this to you, Chamath, we kind of built a CRM tool internally over a weekend using Claude Code and Cursor and stuff. We spun it together, stood it up, and started using it. Then it was, “You’ve got to add this feature,” and, “You’ve got to change this,” and then account and security and access. Where’s the data repository that makes it protected? There was a lot of back-and-forth on all the features we ultimately needed to build to make it truly useful and scalable.

Meanwhile, we’ve got competing interests at our company in terms of where we can build software that’s unique to our company, like plant-breeding software, which is what we do. Our unique advantage as a company is plant breeding. So, how do we build a better breeding-technology platform? How do we build software that gives the people in the lab and in the greenhouse and the breeders tools?

We started building all these internal tools and eventually realized that our time, efforts, and energy are better spent building the workflows that are unique and create value for our organization, rather than recreating a bunch of software that other people have already made. Where’s your best ROI—your best return on time spent and your best return on dollars spent? It’s not on rebuilding a communications tool like Slack, a messaging tool like Gmail, or a CRM tool like Salesforce.

At the time, Marc and I were talking on the phone, and he was like, “Hey, try Salesforce. I’ll set you up. Come on, do it.” I was like, “No, no, no.” Then he was like, “Yes, yes, yes.” He’s so persistent. You guys know Marc. He’s so persistent. He put his best sales guy on a text with me and kept texting every day: “Are we up yet? Are we up yet? Are we up yet?”

Finally, I said to the sales guy, “Okay, go to my office.” He goes to my office, gets set up, and we’re done. Then we realized, “Oh my God, this is actually a much better way for us to do this.”

Marc asked me to come and recount that story at the earnings call and talk a little bit about my view on this change. So, I do think—and I’ve shared this in the past—that the horizontal-platform companies, these tools that are horizontal—CRM works across many different industry verticals, and Gmail works across many industry verticals, and Slack, and so on. Excel—no one’s going to go rebuild Excel.

The real value is in the software that’s unique for your vertical. That’s where AI plays a role in building custom workflows and custom software. That’s kind of how I view this: the real SaaS apocalypse is more of a vertical SaaS apocalypse, which is software designed for just one vertical.

Um, and that’s where I think things get blown up.

I totally agree with you.

A platform like Salesforce—

I totally agree with you. I totally agree with you, Jason.

Sacks, let’s talk about Marc Benioff. He is just an incredible salesperson who fought through the SaaS apocalypse, figured out a way to avoid being put in the bucket of SaaS roadkill, and then massively integrated AI into the platform. You’ve known him for a long time. Maybe you could talk about him as an executive and what we can learn from what he just did in terms of going from roadkill in the market’s mind to leading the pack in SaaS again, let’s say.

This narrative of the SaaS apocalypse was totally overdone. This whole “SaaS is dead” narrative is just getting shredded today, with Salesforce being up over 20%. You could go back and look at some of the clips I’ve been saying for months. I do not believe that core systems of record like CRM are going to get ripped out and replaced with something vibe-coded. I’ve been saying that for a long time.

Enterprises want certainty. They have compliance requirements. They want professionally managed software that’s been running and debugged for years. Just think about all the bug reports that have been filed against Salesforce for over 20 years, right? That’s the hard part to get right.

When I play with all these agents and AI, you’re dealing with something that’s probabilistic, and there are so many edge cases. When you’re dealing with a core system of record, you just want to know that it works. The expense of it is low enough that you don’t want to jeopardize one of your core systems by ripping it out and replacing it with something built in-house in a DIY way.

I just think this whole SaaS apocalypse narrative was overplayed. At the same time, to your point, Jason, I think Benioff is doing all the right things here. You could say it’s a “if you can’t beat them, join them” approach. He made this deal with Anthropic.

What is that deal?

I mean, first of all, he's going to be offering Anthropic's models inside Salesforce. More importantly, I think this is really the critical thing: he is integrating Salesforce into Claude. He's willing to allow Claude to be the front end and have the primary user relationship, and Claude is now going to be able to access all of the data in Salesforce, all of the workflows, and all the actions that a user can take.

In other words, Benioff here is willing to risk disintermediation in order to ride this AI wave because he knows it's what the customer ultimately wants. You don't want to go into Salesforce or any SaaS application to write your agents. You want to create agents in the AI program that you're using, and you want those agents to be able to run across all of your SaaS platforms.

So he's willing to give up a big piece of the user relationship in order to make himself more relevant. Now, why is this ultimately good for his business and not bad? Because, like I said, I still think that you need systems of record. You need the AI agents to basically go to those systems and get the data from a canonical source of truth.

You just don't want there to be any probabilistic variability around that, and you want AI agents to be able to write back to those systems. You want to be able to write back to those workflows. So what's happening here is that it's not a rip-and-replace. Layers are getting built, right?

You've got the database layer, then you've got applications and workflows, then you've got agents, and then you've kind of got the AI user interface. Benioff is saying that AI is now the user interface. He's basically acknowledging that he's leaning into it.

Now, here's the thing: I talked to him this morning, actually, and a point he brought up with me that I think is underestimated is that he says there's a lot of trapped value in Salesforce. We know it's got this incredibly broad, rich platform with tons of functionality. The average user is never going to find all those things, but the agent will.

Think about all the cases where now the agent is going to go to the user and say, “Hey, I just saw that there are these actions you could take in Salesforce that seem to line up with what you're telling me to do. Do you want me to do this?”

Yes, modify Salesforce as the system of record.

Yes, sir.

Yes. Exactly.

Because now, think about the agent as a power user of Salesforce that knows how to use Salesforce perfectly. You don't necessarily need to hire a team of highly priced consultants to figure out Salesforce. The agent knows how to do it, so it can unlock the full value of Salesforce and all these actions.

After you develop trust in your agent, you, the user, are just going to click the “Always allow” button, and now the agent can basically tap the full capabilities of Salesforce. So, I mean, that's where this is headed.

But now it raises the question: How do SaaS applications play into this AI and agent wave? You have to have great APIs, and you have to have a great CLI—the command-line interface—which is how the agents are going to relate to the SaaS tools. I think in the past, most SaaS products spent a lot of time on their user interface.

Now they really have to think about the agent interface, not just the user interface. You have to build the best agent interface. And you've got to, I think, relax the idea that you're going to fully control the customer relationship.

You have to think of agents created by the AI companies as being an extension of your platform and not feel threatened that you're losing the customer relationship. If you can do all those things and become a source of truth and a system of record, then you've got a big opportunity.

I’ve got to say, he's one of the most underestimated and underappreciated product and sales executives in our industry. I was talking to him when we were at Davos, Sacks, about Slack, and he was like, “Tell me about what you're doing.” And I was like, “Listen, you have all my data. I can't get my data. My team wants to move off Slack. They want to use this open-source product specifically so I can get all this information.”

He, again to your point, Friedberg, put me in touch with the sales team, and now I have complete god access to our Slack. And they built a product called Slack AI, which they've been slowly rolling out. It does what OpenClaw promised to do, and what Sam did by buying out the founder of OpenClaw right after he bought OpenClaw's founder, Peter, and had him start working at OpenAI—OpenClaw just faded. There was no more innovation.

I'm not trying to say this to suggest Sam did something nefarious, or that OpenClaw is terrible or anything, but they just lost their edge. And now the version of Slack AI today is like a 7 out of 10, and it's getting 10% better every couple of weeks, I'm noticing.

So he's going to do a parallel stack. He's going to have his own agents, and he's going to let you come in with your agents. His agents might be a little bit behind, but you're also going to get those agents for free inside of Salesforce. It's like a two-way win.

I mean, look, I think that SaaS products can offer their own agents, and they can offer their own AI and copilots and all that. But I think the reality is that most of the action with agents is going to happen outside those products. Users will simply want to interact with those products, get the data from them, use them as the system of record and the source of truth, and then write actions back to them when appropriate.

I just think that founders should lean into that trend instead of fighting it, because if you fight to stay in control and say that only our agents are the way to interact with our system, then people will leave.

So are you long vertical SaaS, or are you short it like Friedberg and I are?

I think it's specific to the company, what its moats are, how good a system of record it is, and all those things.

Well, no, I don't think vertical SaaS has a system of record. I don't think that's what they do. I think their workflows are processes, and this is the key point: These horizontal, monolithic companies like Salesforce are just unique. Workday and Oracle with respect to their general ledger, SAP, IFS—it's very difficult to see a path where you can replicate these businesses and have them become the repository and the single source of truth, as you said.

But vertical SaaS is workflows and processes, and I just wonder what survives.

Well, I think it's a good question, and I think it's case by case, but look, just a few months ago, everyone was saying that horizontal SaaS wasn't going to survive. And one of Liupold's major positions in that fund was to short all these SaaS products, including Salesforce. And what we've seen is that has turned out to be wrong or totally over.

To be fair, I don't know that he understood the difference between vertical and horizontal SaaS, because that was a funding short problem. He had to go to the big stocks in order to be able to short enough to get the leverage that he wanted.

So I think he may have just worked backward from his own mental narrative to make it all sound correct. But I suspect it was a leverage trap where he was saying, “I need to be massively net long these other names,” and his prime broker was saying, “Well, you’ve got to find liquid shorts.” And he was like, “Okay, Adobe and Salesforce.”

Well, but look, I mean, to be fair, I think that was the dominant narrative as of just a few months ago with the SaaS apocalypse, and Benioff's stock was getting hammered by the market. It was down 50%, and now it's back up 20%.

So I think a lot of people were just indiscriminately selling all SaaS. And if I had to say there's one takeaway from the news today, it's that simple extrapolations of the future based on these trends don't work out.

The simple extrapolation here was: AI agents can code, therefore all software is going to zero. That was the simplistic extrapolation. Now what we're seeing is, no, actually, there are certain systems of record that are extremely compatible and complementary to AI agents, and if they're positioned right by their founders, they can lean into this trend.

By the way, I think the job-loss narrative is another one of those simplistic extrapolations. The job-loss narrative is: AI can do knowledge work, therefore knowledge workers are all going to lose their jobs. It completely ignores the fact that there are ways for knowledge workers to take advantage of the tools and lean into it.

So, in the same way that Benioff is taking advantage of the agent trend, knowledge workers are taking advantage of it, and they're going to be able to lean into it.

If you adopt the technology, you survive. If you don't, you die. I mean, it's pretty straightforward.

I just wanted to also point out, with Nvidia's incredible results, 2 things that happened very quietly. Jensen is clearly open-source-maxxing now. He bought Hugging Face for $12 billion, and he did one of these acqui-hire deals with Poolside for $6 billion—about $20 billion in acquisitions—which is far less than 1% of Nvidia's market cap.

And thank God you can do M&A again in this country. But he's now owning the distribution layer for open source. If you don't know Poolside, they make the coding model Laguna. So now Nvidia has the place where developers and hackers go to figure out which open-source models to use.

Sorry. I don't think that's what went on. My understanding is that, because Jason Warner tweeted this morning about a new model drop, there could be 2 things that they're buying. The 100 engineers who are going over to Nvidia could be working on what Poolside has, which is effectively a version of Cursor and Claude Code that runs on-premise, that runs in—

Open source, open weight.

No, then they also have a model. What I'm saying is they have a model, and then they also have this agent harness that you can direct for coding. I don't—it wasn't clear to me in the announcement. I didn't read it carefully.

What is it that they bought?

I think we're burying the lead here, by the way. They're reportedly doing these deals with Poolside and Hugging Face, and those are big open-source acquisitions if the reporting is true. But look, here is the lead today, I think: first of all, Nvidia had a blowout quarter, doing over 100% year-over-year.

A billion a day in revenue now.

Yeah. Almost $100 billion in revenue just for the quarter, $60 billion of profit, 75% gross margins. I mean, excluding one-off items, this was the most profit ever generated by a company in history. Jason, let me just underscore what you said in the intro.

Which is really, I think, the reason why the stock is up 8% today: not just the great quarter, but the fact that they guided to 70% growth next year, whereas the Street was only expecting around 45%. So that's pretty amazing. And that 70% number would have been higher, but they said they were supply-constrained.

Yes.

So, what is the takeaway? I think for the larger market here, it is that the AI boom is continuing. The other narrative that's getting shredded today is that this AI capex is a bubble. It's basically going to end very soon. And what Nvidia is basically saying with its numbers and forecasts is that, actually, this AI capex is going to continue well into the future. It's got real legs. And the amazing thing is that after these numbers, even despite the 8% bump, Nvidia is only trading at 12 times earnings.

Yes. And the reason is that people think about competition—obviously, Cerebras—and the point I was going to make earlier. I think Jensen is going all in on open source based on these 2 acquisitions, Hugging Face and Poolside. And I think Sam Altman poked the tiger. If you look at what he did, he had done that huge $100 billion deal with Nvidia back in 2025, and then he announced the AMD deal. Jensen was clearly not pleased about that, and he talked about, right after Sam made this announcement, that he's going to make his own inference chips—

Jalapeño.

Jalapeño—and that he was going to do the deal with AMD. Then Jensen came out and was like, "Well, by the way, we don't actually have to do these investments in OpenAI." And I think he's speed-running open source, and you're going to be able to buy from Nvidia next year just your entire stack. He's going for it. He is going to own open source. And that is a really interesting moment in time for us to look at, because we were sitting here 6 months ago, 12 months ago: "Oh, open source would never catch up." He's doing an open-source self-driving project, and now he owns Hugging Face, the leading open-source indexer where everybody goes to try the new models. I think this is actually the huge story. I think the generalized takeaway should be that you're going to see all of these businesses converge—

—and they're all going to be competing with each other in every which way possible. So this idea that you have this very delineated separation of customer and supplier, I think, is melting away. Nvidia still relies meaningfully on the hyperscalers for its revenue. I think it was $24 billion in the quarter, but they now have built up a neocloud business that was equivalent, which is incredible, right?

Yes.

And I think what you're going to see is, as more people try to compete with them—so now your customers, the hyperscalers, are saying, "We're going to spin our own silicon." It's completely appropriate for Nvidia to say, "Well, we're going to spin our own models." They do that. And then you're going to see them go up the stack and host the model. Then you're going to see them supply APIs and supply inference, and they're going to be a cloud provider.

And so, to your point, Jason, I think the right way to think about it is, in Business 1.0, you used to be a capex business, right? And people would say, "Okay, what's your return on investment?" Or you'd be a software business, and you're like, "Okay, what's your opex, right? And what's your return on investment?" It's all melting together. Every company is going to do everything. You're going to look at these big companies in 5 years. They're all going to have their own cloud. They're all going to have their own models. They're all going to have their own silicon. They're all going to have their own data centers. Soup to nuts. They're going to go top to bottom. And then it's going to be about which is better. And that, I think, is actually better for the broader ecosystem. It's better for all of us.

4. Bessent gets called out by Druckenmiller for bond market interference

Absolutely. All right, let's keep moving through the docket. Bessent's new bond strategy got criticized by his mentor, Stanley Druckenmiller. Last week, the 30-year Treasury hit a 19-year high of 5.3%. You've been talking about this a whole bunch, Friedberg, so I'm interested to get your take on it.

In response, Bessent doubled the Treasury Department's long-dated bond buybacks from $2 billion to $4 billion on August 19. CNBC reported that Bessent is considering ramping up bond buying, and Fox Business said that he wants to put the fear of God into traders who are shorting long-dated bonds. So, a lot of interventionism going on here. Stanley Druckenmiller, Bessent's former mentor, published an op-ed in The Wall Street Journal opposing his plan. His main point was that Bessent is wrong for trying to manipulate prices rather than fixing the core reason yields are rising.

America has a spending problem. We've talked about it over and over here: $40 trillion. No more DOGE. No more cutting costs. And, obviously, we've got a war that's quite expensive every 5 months. Friedberg, I think you said last episode, we add $1 trillion. So we're adding $2.5 trillion a year. We'll add $10 trillion during this 2nd Trump term. Your thoughts on what's going on here, Friedberg, in terms of the bond market and Bessent and Druckenmiller going back and forth? Chamath, you and I can also talk about using AI to write pieces—write your pieces. Yeah.

I'll start with the relationship, as you highlighted. Bessent and Druckenmiller worked closely together, and they're close and respect each other. Kevin Warsh is in a similar position. He was working with Druckenmiller for the last number of years in his office every day. They're extremely close.

This is a triumvirate that I don't think, in history, you've had 3 people who are so close in these positions: one leading in market activity with Druckenmiller, one running Treasury, and the other one running the Fed. It's pretty unprecedented to have this sort of closeness in those positions. And, you know, I think—

What's that? Yeah. Right.

Caesar. There's—

You called it a triumvirate.

The triumvirate.

Who's Caesar? Who's Pompey? And who's Crassus?

I'm going to leave that to them to debate. But if you pull up the yield curve from the COVID era, I think it's worth just taking a look at the Treasury yield curve. This shows, depending on the duration—is it a 1-month or a 30-year Treasury bond?—across that whole time duration, what's the yield? How much does the federal government have to pay in interest to get people to loan it money for that duration?

This is right around the COVID era, in 2020. You can see that because interest rates were low and the inflationary effects of stimulus that came about during COVID had not yet hit the market, we were looking at sub-half a percent, sub-a-quarter percent, all the way out to the 3-year and 5-year, and then a slight climb up where you could actually get 30-year Treasuries at 1.7%.

Again, going back in history, it's always easy to rewrite history. The federal government should have rolled all of its debt into 30 years at this point. But the debt is balanced across this Treasury curve. So let's look at where the Treasury yield curve is today. This is what the market is charging the federal government to loan it money to pay its bills.

And on the low end of the Treasury curve, you're looking at just under 4%. So for the federal government to borrow money for 1 month, it has to pay a 3.8% vig to the lenders for that money. And if you go out to the 30-year, it's now at 5.2%. So it is very expensive now to borrow money if you're the U.S. federal government.

The reason is persistent inflation. I would argue that it's because of excess government spending on social programs and other things. And the big problem at this point is that the federal government is spending so much that, if the federal government were to cut spending aggressively, the argument and the concern is that it would hit unemployment and cause a recession because the federal government is such an intricate part of the economy.

Now, that's the argument. But it's causing inflation, and it is causing deficit spending. So this year, the deficit will be roughly, call it, $2 trillion, and as a result, the market is saying, "We are worried about U.S. fiscal solvency over the long run," or that there's a higher risk. As a result, we're going to charge you a higher interest rate: 5.2% on the 30-year.

Now, what does this mean for the federal government? Well, today, the federal government's average cost of debt is 3.4%. That's what we're paying in interest, on average, on the $40 trillion of debt that the federal government has outstanding. For every 1% change in the interest rate, the U.S. government has to pay 1.25% of GDP in excess interest each year—1.25% of GDP in interest each year for that 1% change in the interest rate.

And we're now looking at a 30-year at 5.2% and a short-term rate over 4%. So the federal government has a problem because over the next 12 months, it has to refinance $10 trillion of debt. That debt is coming due. Those bonds are now due. It has to pay the principal back to the bondholders, and it has to go back to the Treasury market and sell more Treasuries to borrow more money to refinance.

So the borrowing cost is now going to climb. When that borrowing cost climbs, the federal government’s burn goes up and the fiscal deficit goes up. My theory and my argument is that there is no action Bessent can take that’s actually going to have a meaningful effect on the long end of the curve. We have a fundamental fiscal spending problem with the federal government right now.

I think that the note that Druckenmiller wrote—whether AI wrote it or he wrote it—doesn’t matter. The point of the note is correct, and the note is meant to provide cover to Bessent. He is saying it is not Bessent’s fault or Bessent’s responsibility to solve the yield curve problem. It is Congress’s responsibility. It is the president’s responsibility. It is the responsibility of the holders of the budget and the accounts to stand up and say, “We are going to cut spending,” because if we don’t cut spending, there is going to be a spiral, which we are now going to face over the next 12 months as we have to refinance $10 trillion of debt.

Bessent’s doing his job as Treasury secretary. He’s saying, “I’m going to the market. I’m going to buy Treasuries. I’m going to try and lower the yield.” But he has, at maximum, a trillion dollars he could use to buy debt. He’s still got to go sell $10 trillion of Treasuries in the next 12 months. Ten trillion—he’s got to sell. So even if he maxed out his buying authority in the near term, that’s only a trillion of buying, and then he’s got to turn around and sell 10.

So the market is the market, and Druckenmiller’s note is coverage for Bessent. It’s pointing responsibility back to Congress and back to the president to say, “Guys, we have to solve this deficit problem.” So, is this kind of performative between the two of them, and they’re kind of supporting each other?

I don’t think they’re allowed to talk. I don’t think they do talk. I don’t think Druckenmiller’s allowed to talk to Kevin, or that any of them are allowed to talk to each other. So I don’t think that’s happening. But I think that Druckenmiller is sending his signal to the market and saying, “Guys, this is not Bessent’s responsibility.” Got it.

The burden’s not on his shoulders. He can’t be Atlas. He can’t lift this earth. He can’t solve this problem. You guys have to act.

I think it’s coverage for—

Bessent. Got it. Okay, that’s what I was getting at. Chamath, what’s your take here on what’s going to happen going forward? We’re past 5%. Does this break? What does it do to interest-rate cuts, and how bad could this get if we can’t get it under control?

The tactical answer to the question that you asked before is: Why is he in the market buying bonds? It’s called yield-curve suppression. When you buy bonds, you bid up the prices and the implied yields go down, which allows Scott Bessent to then finance at a lower borrowing cost. If yields are 5.1% versus 5.3% and you’re issuing bonds, they’re marked against that 5.1% print versus 5.3%. Yield suppression is trying to help with what Friedberg talked about, which is this refinancing tsunami that is coming.

But look, just take a step back for a second. I think the very big issue is that Congress is ultimately responsible for spending. Congress is completely unable to do anything to get these budget deficits under control. It is consistent—it’s not a Democratic or Republican problem. It is a congressional problem, meaning both sides consistently spend way more than they should. So you have the debt growing at 7%, and you have GDP between 2% and 4%. That’s a recipe for disaster.

What Druckenmiller said in that note is that you have to cut entitlements. He’s probably right. The question is, when? And who will force the United States Congress to confront the issue? What he is saying is that the bond market is going to start to do this, because that single price is the true reflection of the good faith and trust in the United States. As that yield goes up, the very simple rubric is: trust is going down. Yield goes up, trust goes down.

The other thing I would say is that this is happening at an incredibly awkward moment. We’re supposed to be in the middle of an enormous financial buildout to support AI. If you go all the way back, like 100 years, to the Industrial Revolution and the grand bargain, all of this stuff, the United States government was the balance sheet, right? They were the ones that were able to step in. Unfortunately, because of its financial situation, the U.S. government is not able to do that.

That’s why, thank God, we have companies like Nvidia, Google, Microsoft, Meta, and Amazon that take on that burden. I wanted to tie it to this other thing because there’s so much chirping on the internet about Nvidia’s balance sheet and so on. I think people completely miss that these guys are putting the entire U.S. economy on their backs. A hundred years ago, it would have been the U.S. government that did this, but the U.S. government is not in a position to do it because Congress cannot get its act together on spending.

The president has actually done something pretty incredible, which is, on the tail end of those trade deals, he was able to find $2 trillion to pump into the U.S. economy to support this buildup. But the practical reality is that $2 trillion comes from other nation-states, and their processes are long and meandering. So you have a situation where all this money has to be invested. It should be backstopped by the U.S. government, but it’s not in a financial situation to do so.

The president has done some magic to try to relieve it in the short term, but it’s only a short-term Band-Aid. The real solution, as Friedberg says, is that Congress needs to come together and realize that it must now fix this problem. Because if you see the 30-year at 6%, it is the beginning of a death spiral. It’s not going to be immediate, so don’t freak out, but it is the beginning of some extreme pain. That pain will last years.

The only thing that solves it now is getting the budget under control, which is a congressional act. It must happen. Whether it’s this year or next year, it just has to happen.

Sacks, the president was quite vocal about cutting costs early on in the presidency, but we haven’t heard much from the commander in chief about this. Is the issue that no politician can touch cutting spending? It’s just too toxic because Trump’s in his final term. He’s a lame duck by definition, right? It’s the second and only term, hopefully, unless he decides to take a third. But he hasn’t been vocal about this. Is it just too radioactive for anybody in politics?

You got to spend a little time—you dipped your toe there in D.C. Is it just too toxic for anybody to take on spending cuts? He sent Elon in there. That was pretty spicy, and it’s obviously not occurring now. We’re not cutting things massively. In fact, the debt’s growing faster than it had a year ago. So what’s your thought here, Sacks?

Yeah. Look, part of it is that it’s a tragedy of the commons. You’ve got 435 members of the House, 100 senators, and then you’ve got the president, and no one of them controls the spending. In fact, the power of the purse originates in the House and then goes from there. So it is very, very hard to control spending when you have so many people weighing in.

If it were just a couple of people getting together in a room and sorting this out, they probably could do it. But when so many people have an incentive to basically protect their programs, it gets very, very hard. So that’s the tragedy-of-the-commons part of it.

The reality is that the president doesn’t have a line-item veto. It’d be great if he did, but he doesn’t have the constitutional powers to completely rein this in. Despite what the resistance liberals say, he’s not a king. He doesn’t have those kinds of powers. That’s basically the heart of the problem.

You’re right, he did back DOGE to the hilt. He’s the only president to ever back something like that. And what happened? Elon got to the point of cutting one agency or one subagency, and the hysteria was so great that they accused him of killing millions of children. It was total nonsense. But Tesla dealerships were getting firebombed.

The other piece of this is: How do you reform spending? How do you rein it in when the media isn’t on your side and is not willing to report this stuff accurately? That’s the worst part of it. The media doesn’t have a political incentive to engage in logrolling, but it does absolutely nothing to report the story accurately. Just think about how the media is reacting to all of that.

So it’s really a difficult problem when you start off with the fact that no one person controls it. It really is this tragedy of the commons, and then you don’t have the media doing a good job of reporting this accurately. I don’t know how we get reform on this.

Let me just say that if the Republicans lose control of Congress, I don’t think that’s going to make it better, because you see now that the political energy in the Democratic Party is all toward these massive new spending programs that the DSA wants. So if you vote for “change” by voting the Democrats in, that’s not going to help either, because again, they just want more and more spending. I don’t know what the solution to this is.

Friedberg, give us a timeline here. Obviously, the Republicans love to cut taxes, and the Democrats love to increase spending. And here we are. This has not gotten better through 3 or 4 presidents.

The last person who did this was, I guess, Clinton, who balanced the budget, but this has been getting worse and worse. At some point, we're all in agreement: this is going to come due, and we're going to have to have some austerity measures. When does that hit, in your mind? Is there a breaking point in terms of interest to GDP? Maybe the people on the other side just say, “Hey, I want 12%. You guys have bad debt. I want 12% on it,” and we just say we can't pay 12% and have to go into full-on austerity measures. Is there a breaking point? If so, when?

Well, the question is, what does it break to? We, as a democracy, have this recursive solution: as people feel some sort of pain, they react with a vote. This goes back to this point about whether we break for socialism, because what ends up happening post-midterms?

I think sometime between 2026 and the election cycle in 2028, the inflation problem will become acute. Remember, the inflation problem is fundamentally rooted in government spending. Deficit spending is correlated very nicely with the cost of housing, the cost of healthcare, the cost of education, and the cost of everything else. The more the government spends on various things, the more expensive things get.

Inflation is persistent because what we thought were temporary emergency stimulus measures during COVID very quickly became normalized and turned into persistent spending cycles. If we went back to 2019, we would actually have a budget surplus right now as a country. Our economy has grown so much since 2019 that, if we had the same spending levels as we did in 2019, our federal government would be making a profit and paying down the debt.

The reality is that we've taken this emergency spending and made it permanent. We did the same thing, by the way, after the global financial crisis in 2008. There were some crazy stimulus things that happened, and we persisted with that spending. So the question, Jason, is really: What happens next?

If inflation remains high and people's affordability continues to decline, they're going to vote for something that gives them a solution. I don't see a political message winning that says, “The federal government will now spend less. We're going to cut a bunch of stuff.” So I worry that the one-way solution here is something very damaging to individual liberties, to a lot of rights, and to a lot of the things that make a free market work—and that is some of these more socialist principles and socialist candidates.

I think that's kind of the breaking point. I think it's more of a spectrum, and the reaction during that transition phase, from 2026 to 2028, can be negative. Then the next big thing is going to happen around 2030 to 2032. That's really where Social Security doesn't have any money to pay anymore.

I think during that era is also when a lot of the states are going to become bankrupt. They're not going to be able to pay their obligations. When the states can't pay their obligations around 2030 to 2032, you're going to have both the Social Security bailout and restructuring, as well as the state bailouts and restructuring. That might be the time that all of this stuff breaks.

Wow. I mean, so you're saying we'll see munis? Yeah, and people will default on their debt in the states. That would be cataclysmic—to have to bail out something like California.

Well, the obligations on the unfunded pensions in the states are so significant. And not to mention just the debts. I mean, look, California has tens of billions of dollars spent on this stupid railway that goes nowhere, where there are no tracks. And they're talking about the political will to even stop that. It would be the easiest cut in the world.

Just think about the idiocy. No one is saying we shouldn't do this. Why is no one saying we shouldn't do this?

That's what I'm saying about the media.

A trillion-dollar train. Anybody who was in favor of a railway to nowhere should just be shamed out of existence.

Trillion dollars for a train from San Francisco to Fresno that costs more than an airline ticket is so idiotic. It's like the movie Don’t Look Up. Look up. This thing is absolutely ridiculous. Turn this thing off. Why is no one even just saying, “Turn it off”? It tells you everything you need to know.

A private company built Brightline in Florida, and they seem to have figured out how to do it at a fraction of the price. What are we even talking about here? Private industry, if this was a good idea, would have built it. Yes, Friedberg. Private industry would see an opportunity, and they would have built it. There is no opportunity here.

Across the board, these government programs cause more harm than good. When the government intervenes in underwriting student loans and gives everyone a loan, administrative costs go up by 6 times, and tuition skyrockets by 8% a year, compounding for 30 years, because the government said, “We'll underwrite any student loan.”

The same thing happened with housing, and now no young people can afford a house. The same thing happened with healthcare, where they said, “We'll pay people to stay home and not work to take care of people, and we won't check on whether or not they're actually taking care of people.” The cost of healthcare skyrocketed.

You go across every one of these government programs, and every one of them has the adverse effect of driving up costs and inflating everything. The core root of inflation in this country—the core root of it—is government spending.

The reason we are going to end up becoming a socialist country is that we aren't reining in government spending and looking at it in the mirror and saying, “Guys, this is idiotic. What are we doing?” Now the markets, to Chamath's point, are telling us this is idiotic. We are not going to underwrite this anymore.

And we need, to your point about the media, Sacks, 100 Nick Shirleys, 60 Minutes, whoever it is, to just say, “Hey, this spending, this fraud is not acceptable, and we need to send people in to go recapture that money.” This is like we're burning money that we don't have. The fraud is layer 1. Yes, of course.

And given the tragedy of the commons and the structural nature of it—there's no one person who's in control of it—you need massive public pressure to control it, and there's none of that. There's no effort to create that public pressure by the mainstream media. So how do you even get out of this?

I don't think it happens that way. It has to be an acute moment.

Yeah. You're saying there has to be a crisis and then people react. But then, to Friedberg's point, we may get socialism because people may not connect what the real causes of the crisis are. And Friedberg, that's my question for you.

No, socialism is much less likely now than it was 100 years ago, when these things were less interconnected. America can't decide to do that. Do you understand why? All of our debt is interconnected to domestic actors and international actors. It's not a decision. There is no button you can push to do that, because you would go bankrupt overnight.

I think the reality is that we will hit some kind of fiscal wall. Something will get terribly broken, and, like in COVID, the GFC, or 9/11, those are the only moments where you can have structural changes in expectations. The structural changes that have had the most impact on our lives happened in these acute moments: 9/11, the GFC, and COVID.

Yeah, and every one of them ended up becoming a “never let a crisis go to waste” moment, where it was always argued in favor of some big expansion of power by the government.

What was your question for me?

Okay, so think about the affordability of houses. That depends on mortgage rates. The 30-year mortgage right now, I'm looking at the average, is 6.73%—a little under 7%. It's about 150 basis points above the 30-year Treasury, right?

Do you think the average person can connect the dots here—that higher spending means higher Treasury rates, which means higher mortgage rates, and that leads you to not being able to afford the same level of house because your monthly payment is much higher? Do people connect those dots about what's happening?

No. No. But people want to blame someone. Every time there's a crisis—this group, even us, when we get together on a podcast, something bad happens—we always have this question: Who is to blame?

It's the first thing a human does.

Yeah, exactly.

And literally, that's human. That's the human brain.

No, I mean, that's the human brain. That's how it works. There's some problem, and the first thing you think about is, “Who do we blame?”

You don't think objectively, “What's the solution to the problem?” You think, “Who do I blame?” That's the first thing you want to do. And then the question is, when mortgage rates go up, who do you blame? The current president, my congressman, blah blah blah—reelect someone. That's the response.

The person who shows up and says, “I'm going to give you free housing. I'm going to give you a free bus. I'm going to give you free mortgages. I'm going to take away the land from the landlord and give it to you”—you vote for that person.

I do worry that the mechanistic response to what you're describing, Sacks, which is a very problematic situation for people—the lack of affordability and being able to get a home—is to give the keys to the kingdom to someone who's going to put more fuel on the fire.

The fire is government spending, and the fuel is more government spending to try to solve a problem that's caused by the fire. I think that's really where I worry about this.

So I think I know how we can blame Dario.

Okay, circle it around.

What did he do?

Here's the thing. Our only hope is AI, right? Because only AI can create the exponential growth necessary—

Yes.

—to make our economy grow big enough that we can grow out of the problem.

Yes.

And if we hold back AI by creating some new regulatory apparatus that takes years to approve new model releases, then we're cooked.

Completely. And Canada and Putin—those are the 3 people who are working—

I'm just kidding. We're joking. Dario Amodei, come to the summit.

That is not Dario's fault. But anyway—

Come on, keep going.

No, seriously, come to the summit. Listen, I want to just tell people: We are kind of counting on AI as the deus ex machina to grow our way out of this thing. We really need more quarters like the ones Nvidia and Salesforce just posted. We need that to happen for 10 years to grow our way out of this thing.

One of the great silver linings is that we're seeing people migrate around the country or to other countries as a way to vote against out-of-control spending. I lived in New York, Los Angeles, and San Francisco, and I did not appreciate how evil these socialists and limousine liberals were until I got to Texas.

After a couple of years in Texas, I have seen what having affordable homes does for a family and an individual. You have people in those cities who are so miserable because they can't afford a home. They're paying exorbitant rents, and these people who already own their Victorian homes, et cetera, in San Francisco refuse to build.

In the great state of Texas, if you look at this chart, housing prices since the 2022 peak in Austin are down—down, not up, Sacks—27%. Since March 2020, they're up 25%. It can be done. You need to have leadership that says housing is a priority. Therefore, we're going to let you build more houses.

It's being done in Texas and in Florida. Those conservative, let-the-free-market-work environments let them build Brightline. They have high-speed trains going 125 miles per hour because they let private industry figure out how to build high-speed trains.

This is a distinct issue in certain groups where they want to stop progress. They are stifling the creation of homes. And when you get homes, people feel great about their lives. People in Texas are happy that they have a home for $250,000. They're everywhere. These things can be solved. It just takes leadership.

And that's what we don't have: leaders saying, "Enough already. We're just going to let people build more houses." All right, let's keep moving through the docket.

I love a good JCal polemic. I'm just pissed off about it because, you know, it's a total non sequitur to—

It is a non sequitur in some ways, but we're dancing around here: Who's going to be the leader who will say—

Did you practice that before?

No, I didn't practice it. I'm just—

This just came to you off the top of your head?

Yes. It's just another—yeah, well, actually, I hired Sacks's writer to do some punch-ups for me. But putting that aside, it is obviously a leadership issue.

That's too expensive.

True.

5. AI writing controversy: Druck used AI to write his WSJ Op-ed

I'll get him a watch. Let's do a side quest here. What did you think of Stanley Druckenmiller writing his piece with AI and getting busted, and then the Wall Street Journal saying, "Hey, it's totally fine"?

I thought it was the left that virtue-signaled, but I actually think it's a class of people that virtue-signal. They judge, and they're judgmental, and they can't see the forest from the trees. They're probably the loudest chirpers in the peanut gallery.

Stanley Druckenmiller is the most incredible investor of our generation. Maybe—sorry—maybe top 3 or 4.

He's up there. Yeah.

And so just shut the fuck up and read what he says. You have an opinion, and you're like, "Oh my God, he has an em dash. Oh my God, he used the AI flourish." It's like you're missing the script. He's telling you what's about to happen.

Sacks, you agree? You think the opinion pieces in the Wall Street Journal and New York Times should be written by AI, not by the person who puts their name on them?

He didn't say it was written by him. He admitted it's 100% AI.

He prompts it. He shapes it. He guides it.

Yeah. A more precise way of putting it is that the take is his, and then he uses AI to help him write it.

Look, I don't get all the pearl-clutching about this. You have to be pretty dumb these days not to use AI to help you write, and he's definitely not dumb. This is his opinion. It's his take. It's consistent with everything he's been saying in the past for decades. In fact, the piece cites the many times—

And he's right.

Yeah. The piece mentions the many times in the past he's weighed in on this. I think that this clearly came from his mind, and he uses AI to help him write it.

The morons on X should be like, "Wait a minute. How does this affect my company? How does it affect my portfolio? How will it affect my 401(k)? What will it do to my parents' retirements? How should I think about hedging this?"

Instead of that conversation, all of you people were like, "Oh my God, he had AI."

I think, JCal, do you have a problem with this? Are you one of the people who has a problem?

I do think it's kind of the lip-syncing of writing. As a writer, I find it offensive to let the system give your entire opinion, as opposed to using it for research, which is totally fine. Use it for proofreading, totally fine. But to me, it feels like I went to see Adele and she's lip-syncing. I really want to hear him.

Don't use Excel. Only go to the guys who do hand calculations.

No, no, no. Here's the thing. I would rather he said, "Don't use an Excel spreadsheet. I don't trust those guys. I only trust the guys who do the calculations by hand."

But if you were going to give your opinion and it's not written by you, to me, that is disappointing if it's not disclosed. I think it's a much higher standard to say, "I'm going to write this in my own voice." Sure, I could have somebody who's a ghostwriter help me polish it. I could check my grammar with it. I could do research with it. Those are all valid uses.

But I don't like this idea that you're going to write something with AI and then I read it and I don't know. That, to me, feels icky, and it feels insincere. I would prefer people just say upfront, "I didn't write this, but I agree with what the AI magic guessing-word box said." I find it personally offensive.

Yeah. Interesting opinion. May I respond?

Sure. I was trying to get your opinion. Go ahead.

What do you think is the difference between that concept and having artists be required to disclose whether or not they used Adobe Photoshop and the very specific filters that they used in Photoshop? Or DJs that use specific software to create electronic music, as opposed to doing analog recordings of the music by hand with guitars? Or people that, for example, are using Microsoft Excel to create spreadsheets, as opposed to writing it all down?

Do I have to disclose that I used Excel, or that I used a calculator, or that I used a synthesizer, or that I used Photoshop, as opposed to doing it in an analog way?

At the end of the day, digital tools for creativity and creative output are used in parallel with human ingenuity. Human craft and human creativity are expressed, magnified, and in many cases improved with digital tools.

And the definition of whether someone is "using AI" is as nebulous as the rest of those questions. Am I using AI to improve my spelling, to improve my grammar, to create novel sentences from scratch, or to create novel paragraphs from scratch? At the end of the day, if I give the AI a whole bunch of inputs, take the outputs, and edit that output, am I "using AI"?

This is, I think, one of the fundamental issues. We treat AI as a standalone, simple kind of intelligence-replacement system, and that heuristic is not accurate with respect to how humans actually use it. It is, like all other tools, a magnification of human creativity, ingenuity, and potential. Go ahead.

The reason is the expectation of the reader. The reader has a high expectation, and this is an important issue. The reason why people were so disappointed in this, in a very large way, was not because they're virtue-signaling. It's because they actually care about Stanley's opinion, and they want to consume his opinion and make sure it's not the AI-slop opinion. That's why people felt betrayed by what he did.

Now, this is a new—

Betrayed?

I think they did, yes. And they also felt the Wall Street Journal betrayed them, because the Wall Street Journal should be checking that anything submitted is not AI slop—the same way professors should be looking at students' work and making sure their papers are not AI slop, because we want them to have learned something.

And they want the same thing when a teacher is teaching. People don't want to pay for a teacher who uses AI slop, either. And so we, as a society, have to decide: When do we need it to come from the human and understand that it's 100% human or 60% human, and when do we think we trust the AI on it? I want to hear his actual opinion.

I have questions. Many questions. Okay, question number 1: Are you suggesting that Stanley didn't read this thing that was generated, or do you think—

I'm sure he did read it. Yes.

Okay, so whether he read it before—that's an open question. Did he—

Okay, I think he—well, he defended it. Yes, he read it.

I hope he did.

Okay, so do you have an issue with the fact that he read it, thought, "Yeah, this is what I think," and then published it?

And hit the send button. That's a betrayal to you.

I think not disclosing it is the betrayal. If you want to write—if you want to give me your opinion as interpreted through the AI magic guessing-word box—you need to tell me.

I understand. I think it's difficult to tell everybody.

I understand. Now, I just want to understand. I'd like to make—

Oh, listen. What exactly?

The standing disclosure for all future publications that I put out, including tweets, is that I try not to publish anything without running it through AI first to fact-check it, line-edit it, and help me make the best argument I can. So this is a standing disclosure: If I've written something and published it, it's probably been through AI.

I'm totally fine with proofing, to be clear. Proofing and fact-checking are great uses of AI. A grammar checker is a great use of AI.

Now you're debating the slider of how much he proofread—

Those are 2 different things. I understand you don't understand this concept because you're not a trained journalist like I am, but I can tell you there's a distinct difference between being proofread and having somebody plagiarize or write something for you. And that's what this is, and that's why people object.

Question: For Stanley Druckenmiller's next op-ed, is he allowed to use a computer?

Of course. Of course.

Okay, let me keep going. Is he allowed to use a word processor, or is it just the notepad?

I mean, you obviously understand he can use any of those tools. He should disclose it.

Wait, what about Grammarly? Is he allowed to use Grammarly?

I encourage him to use Grammarly. I'm a shareholder in Grammarly.

JCal, what do you think his process was on this post? Do you think it was simply delivered to him fully baked and then he just said, “Go with it”?

That's what the AI detection software said: that this was about 90% written by AI. I'd be surprised if that was his process. My guess—and this is just a guess, because I don't know—is that he had the take, and he probably told the AI the take, either in writing or just by speaking to it. He may have essentially spoken to his assistant, who dictated it and put it into the AI. Then he probably got an output, made corrections and edits to it, and that was probably something like his process. I don't really believe this thing was just spit out as AI slop and then he said, “Print it.”

This is JCal's image of what happened: He's sitting in his office in New York, gets in front of a browser, opens ChatGPT, writes, “Yields up, bad,” gets the whole thing printed out, and—

—emails the op-ed. If you can't appreciate the lip-syncing analogy, it would—

How do you lip-sync?

Okay, here's the problem with the lip-sync analogy: He's criticizing something that just happened. It was this Scott Bessent buyback policy, right? He's criticizing his former colleague about something very specific—

Okay—

—that the Secretary of the Treasury just did. What's the lip-sync? A lip-sync is when you have a song that's been performed 1,000 times.

There. This is novel and new.

Even more evidence of why he should write it himself. It's really about respecting the audience and taking the time to write it yourself, not letting AI be part of the process. You can have an editor. That's allowed.

I was like, “Oh, this is a useful data point for me—a reminder.”

Totally fine. Totally—

Wait, can I ask you: Did you read the article and have an opinion about the substance of it?

Yes. I wanted to know if they were coordinated and this was some sort of—did you know it was AI-generated before someone posted that on the timeline?

No, it came up in my timeline first. No, I didn't.

Exactly. It came up first. Hold on. You asked me a question. Let me at least answer it.

It came up first. I saw the plagiarism, the AI slop, whatever people frame it as in journalism. That's how people—

Let me tell you: I read it when it first came out. I didn't see the stuff in the timeline about the software supposedly detecting it.

I got hit by the first—

It sounded completely consistent with the many other times I have heard Stan Druckenmiller speak about the debt problem—on podcasts, in written versions, or whatever. I didn't know. So if he's plagiarizing from someone, he's plagiarizing himself.

Yeah, plagiarism isn't the right analogy here. It's not perfect because this is a new thing, but—

It is taking credit for what the AI wrote. Hold on, let me finish. I would prefer him to take the time to respect the audience and write it himself, because I have respect for him and I want to understand his true opinion, not his opinion as interpreted through AI.

Just like when I hear somebody sing a song, I would like to know that they haven't autotuned it and post-produced it to the point of absurdity. That's me. I believe the artist should give you their actual art and their actual position without using AI. That's just my opinion.

Even top-performing artists use autotune and all these other tools to basically make themselves sound the best. In a similar way, we know that AI gives people superpowers. Stan Druckenmiller said, “I'm not the greatest writer. What makes me interesting is that I'm a thinker. I have a take. It's a long-held position that I've had.” And he's willing to criticize his former colleague and, I assume, friend because it violates his view of the world. He's saying he uses AI to basically make his writing better. I don't get it.

I don't have a problem with using a Grammarly-type tool. I just don't like the idea that it wrote the whole thing, which is what the claim is. Now, we don't know. We'll talk to Stan when he's at the event.

I will tell you, I prefer listening to a band like Mark Knopfler, where he actually can write the song and play the guitar riffs. I prefer that in my art, in an opinion, or in a novel, rather than having it written by AI. That's my preference, and that was the expectation of the audience. But this is new.

Now that you know Druckenmiller will continue to use AI because he said it, will you just ignore what he has to say?

No, but I will not think as highly of him, and I don't think it's respectful.

So are you saying the substance of what he says will no longer matter to you, or are you going to discount the substance?

You just said that. No, you just said that you're going to discount the substance. That's hard for me. I will wonder what his actual opinion is and how much of it was constructed by him personally.

You're going to trust the greatest macro trader in the world less tomorrow than today because he used a tool to write his—

I will wonder whether it's actually his opinion and how much work he did, and I feel it's disrespectful that he didn't do it himself. That's how I feel. It will make me wonder whether that is actually his opinion and whether it is crisply his, because I think he's phoning it in. I just don't like people phoning it in.

It's only 800 words. Write it yourself. Work with a proofreader. It's totally fine to do your research that way. Listen, we're going to agree to disagree on this one.

6. CIA Chief visits Moscow, Meta's settlement and new teen rules

The CIA director made an unannounced visit to Moscow. John Ratcliffe made a surprise visit on Tuesday and flew back the same day—the first visit by a sitting CIA director since November 2021. That was when Bill Burns went to warn Putin not to invade Ukraine. He invaded 3 months later.

The Wall Street Journal is reporting that Ratcliffe was there to warn Russia not to attack a NATO country. New U.S. intelligence reportedly shows Putin could be planning to test NATO's resolve with a limited assault in the next few years—anything from a cyberattack to a small land incursion. Polymarket says there's only a 21% chance of a Russia-Ukraine ceasefire by the end of the year.

Friedberg, you were on fire in the group chat about this. What are your thoughts on what's going on here?

There's wild speculation. The Wall Street Journal and others have reported that the message delivered was, “Don't invade NATO, and don't use nukes on Ukraine.” So there's been this kind of wild speculation about what this means. It just seemed like there was something meaningful going on. But to Sacks's point in our group chat, this is speculation. There's no evidence.

And more than that, I think the president called into a show to basically say that the story was not true and that the contact was routine. The Russian readout said roughly the same thing. So the truth is, we don't know what the meeting was about.

Quite dramatic.

I think it's all speculation.

I know what it was about. Oh, here we go.

Druckenmiller's AI-infused memo to the Wall Street Journal.

Yes, he was warning Putin not to use AI to write his next warning—his next op-ed.

We flew over there to deliver a very strict message: We expect non-AI-tool usage, because we are journalists at our core.

Yeah, I think it's possible, JCal, that the trained journalist got this one wrong and we're engaging in wild speculation and hyperbole without actually knowing what the meeting was about.

Well, speculation is what we do here. It's a podcast. That's the whole point of the show: to speculate about what the hell's going on here.

First of all, both the American and the Russian sides have denied these dramatic, speculative claims that somehow Ratcliffe was there to warn the Russians not to invade Europe. I actually believe those denials because it wouldn't make sense for the Russians to invade NATO countries. This is the same kind of threat inflation that we've heard for years from neocons about Russia's supposedly imperial ambitions to conquer all of Europe. There's simply no basis for believing it.

And frankly, the war is not going badly for the Russians. It's certainly taken a lot longer than anyone thought, and Russia has paid a heavy price in casualties. I don't know that the numbers are what you said, Friedberg, but I think if you look at most of the sources now covering the war, what they're saying is that the Russians are making slow but steady progress. And it's Ukraine that is suffering the most.

First of all, the Russians have air superiority. The Ukrainians are basically out of air defense, so Kyiv is getting bombed on a routine basis. The Russians are able to pretty much destroy whatever they want at will. So Ukrainian infrastructure is paying a heavy toll, and this is coming on the eve of what is expected to be a pretty cold winter. Ukrainian power is getting taken out.

Moreover, Odesa and the Black Sea ports of Ukraine have been bombed and basically shut down. So their economy is paying this incredibly heavy price right now. They're not able to export grain, so their economy is basically getting destroyed, and they're running out of money. Zelensky is running around saying that he needs another $35 billion or they're going to be out of money. The U.S. is not funding that, but he's trying to get that from the Europeans.

And then their casualties, although we don't know the numbers, are also huge. Now there have been stories about them potentially drafting women because they're so low on soldiers. Keep in mind, Russia has 4 or 5 times the population of Ukraine, so they're able to sustain much greater armies. Things do not look great for Ukraine.

And frankly, Zelensky should have listened when President Trump told him, “You don't have the cards in this war. You should make a deal.” Zelensky didn't want to do that. He didn't want to make a deal. He still doesn't want to make a deal. He should make a deal. And frankly, I don't think Ukraine's going to come out on top in this thing.

Chamath, do you have anything you want to add there, or should we just wrap?

Yeah. Thanks to Meta for solving my biggest problem, which is imposing limits on Instagram.

Yeah, okay. We didn't even get to that story. It's a big story.

I'm tired of playing whack-a-mole with my kids, and I can't do it. They have fake accounts and secret accounts and this and that.

Yeah, I will give them some amount of credit. They got dragged, kicking and screaming, to do the right thing.

TikTok and YouTube—I hope TikTok and YouTube do the same thing.

Much simpler. It'll make the life of a parent much simpler. It'll allow us to actually parent versus just policing device usage.

I'd love to see Android.

And it's exhausting being a child with a device.

It's exhausting.

And it's terrible for young women to be on these Instagrams. We held the line at 16 years old, but at 17 we lost the battle.

Where did you hold the line with the kids for social media?

No, my kids aren't allowed until they're 16. But that's not the point. Look, we've talked about this. I've been very consistent about this. This could have happened a lot sooner. I'm glad that it happened. I applaud Meta for doing what they did. I hope that TikTok and YouTube do the same thing. These limitations are coming around in Australia, Canada, and the U.K. They'll probably appear in other places. It's physiologically and psychologically really helpful for kids to have these limits.

Yeah.

These parental-control apps aren't that great. They do a decent job; they don't do a great job. So I'm always on my back foot as a parent, and it's frustrating. I would rather just spend time with my kids on other things than screaming at them to get off their phones.

Yeah, the tools are hard to implement.

Hey, Fred, what's going on with this cancer vaccine and the stock that tripled in the last 2 weeks? We didn't get to it last week, but I know I want to hear about it, and I'm sure the audience does.

Yeah. Moderna stock, I think, is what you're talking about, right?

Yes. I think they went from around a $20 billion market cap to a $60 billion market cap on the positive readouts from this thing that's being called a cancer vaccine.

And I think it's worth highlighting how this cancer-therapy system works. This goes back to the 1990s, when there was this idea that you could create what are called neoantigens. These are small peptides or proteins that mimic a peptide or protein that's unique to a cancer in your body. That antigen, which means that your immune system sees it as an invader, causes your immune system to start making cells that destroy that protein.

By putting that protein in your body, your immune system gets activated, and it gets turned on to go and find that protein in other parts of your body and destroy it. So the objective is to figure out what the protein is that's unique to a particular cancer, make that protein, stick it in your body, and your immune system will go and destroy the cancer that's in your body.

This immunotherapy—this system of basically using an antigen that's designed for a cancer in your body to go and attack it—has been around since, call it, the late 1990s. There was this big revolution in DNA sequencing, where we identified that we could take samples of cancer, put them in a DNA sequencer, and look at the sequence of the genes—the proteins that are made because they're made from genes that are unique to that particular cancer—and therefore make a personalized neoantigen that's personalized to the unique cancer that's in your body, because all these cancers mutate and they're different. No 2 cancers are identical.

How do they get that key for me personally versus you? How do they get that information?

They would take a sample of your tumor, or cancer cells. So let's say you have a melanoma. The reason melanoma is one of the biggest targets for this immunotherapy is that melanomas have very unique DNA mutations because of UV light hitting the skin. The particular cancer that you'll get in a melanoma is going to be very unique with respect to the rest of your body. It's going to be a very unique mutation.

So they take a little bit of the melanoma, they get the DNA out of it, and when you do DNA sequencing, you multiply the DNA by millions or billions of times. You put all that DNA in a DNA sequencer; it gets chopped up and read. Then the sequence data comes back and says, “This is what makes this cancer unique. It's this little sequence of the DNA.”

That sequence of the DNA causes that cancer to make a unique protein that's basically like a key, an ID, or a fingerprint of that cancer. They then take that DNA and make a protein from it. They put that protein in your body, and now your immune system goes crazy and starts attacking that protein because it's like, “Oh, antigen, antigen, I've got to go get it.”

Once those cells are activated—those immune cells are activated to that protein—they'll eventually go destroy that cancer. So this has been successful. There have been hundreds of trials with this neoantigen-based therapy for many different types of cancer, and there are all these trials that have tried to figure out what's the right technique for delivering the neoantigen.

Do we make the protein and then stick it in your body, or, in this case with what Moderna did, make mRNA and stick it in your body? The mRNA causes your own cells to now make that protein and spit it out. It ends up in your blood, and now your immune cells get activated.

So rather than making the protein ex vivo, which means outside of your body, you're making the protein in your body, in vivo, and now your body's making the protein.

So the idea with the mRNA concept is that you can get one shot. That shot goes into your body, and now your body is making that protein, just like we made the COVID vaccine.

Yes.

Your cells would make a particular protein segment that would activate your immune system, and it would go kill that cancer. There are other questions around what the other adjuvants are, what types of cancer you can use this system for, and combination therapies. Are there other cancer drugs that you can use in combination with this?

All the trials that have been going on are about the delivery mechanism, the technique that's used, the types of cancer, the dosing, and getting all of that stuff right. What's interesting about this is that it is much more of a technique than a drug in the traditional sense.

It's a process, right? The process is customized per person, so there's some cost to doing it per person. That's right. But the cost—because I do DNA sequencing at my company all day long—of getting that sample is very cheap. Getting the DNA data is very cheap. Actually doing a lot of the upfront work is very cheap.

Theoretically, making the mRNA or making the protein should also be pretty cheap, but there are human trials and human safety. There are a lot of questions around how you make sure everything's safe and what the right combinations are, so that people don't get overreactive immune systems or have adverse side effects. That's all of the tuning involved in this.

But what frustrates me the most about all of this is that everyone's lauding it as some unique, special, and powerful breakthrough. It's not, because a lot of people are going to places like Montana, getting peptides printed for their particular cancer sequence, making their own neoantigens, putting them in their bodies, and getting cured of their cancer.

You can pay someone $50,000 to do this for you today. There are a lot of clinics that'll do it, and it's not a special FDA-approved drug. To your point, JCal, it's a process. The process is: take the DNA sequence from my cancer, make a protein, put it back in my body, and my immune system goes and destroys the cancer in my body. Great idea.

So why is Moderna saying that they're going to charge $500,000 for this? That's what frustrates me about all this. Everyone's lauding this, and everyone's very happy about it, but I'm actually pretty sad, frustrated, and disappointed.

I don't think that this technique, which has been largely developed through several decades of iteration, research, and development—largely funded by the NIH and other public funding dollars—should now be patented, FDA-approved, and charged at half a million dollars for people to get treated for this.

I have 2 questions. We can end it on that, but do you think they're going to be able to successfully defend this patent in the United States and then also globally? There's obviously a global world here, and people do travel for treatments now. Medical tourism is a thing. If it's half a million dollars here, can it be done for $50,000 in South Korea, Japan, or other places?

And secondly, whatever happened with mRNA and the COVID vaccine and a lot of the concerns people had? We never got an update on that, but obviously this podcast started during all of that. And that was the first rollout of it at scale, correct? For mRNA technology, how do you feel?

For years, Moderna was working toward this mRNA-based cancer vaccine. They're calling it a vaccine, but it's really immunotherapy, because a vaccine means you prevent yourself from getting something. In this case, you're actually getting rid of something you already have, so I don't like the word vaccine.

Moderna worked on this for years, but they never really got to market with it. Then COVID came about, and they pivoted very quickly and got this COVID vaccine out. Moderna has a whole bunch of patents on the techniques and tools they use to make mRNA, to make sure that it's not integrated into the human genome, and to make sure it doesn't actually use a traditional kind of replicating system.

You can't actually replicate that mRNA, which means it won't start propagating in your body, because your cells might typically start making copies of RNA if it ends up in your cells. There was a lot of stuff that Moderna developed that made its particular mRNA technology patented and safer, and so on. I think they do have strong patent coverage on their particular set of tools for mRNA.

But when I described earlier—if I take the sequence of your cancer cell and look at its DNA, and I'm like, “I found what makes it unique”—there's this little sequence. Now I take that sequence of DNA and stick it in a bacterial cell, E. coli, in a little tank called a bioreactor. That bacterial cell will start to make that protein because I've taken the DNA and put it in the E. coli cell.

Now the E. coli cell is coding that DNA into RNA and then into proteins, because that's the sequence: DNA turns into RNA, which turns into proteins. There are billions of copies of those cells, and they make billions of copies of the protein. You isolate the protein out of the bioreactor, and you can inject it into your body.

When you inject that protein into your body, your immune system gets activated. You don't necessarily need mRNA, is my point.

And mRNA is a really effective way of doing this. It's super cheap because you don't have to make all the protein outside the body. You can literally just print the mRNA, get one shot, and you're done.

But the end-to-end system—the real cost of doing that—is very low. This is one of those things where we have a very big, fundamental question that we should all be asking: Should we be giving the monopoly on cancer treatment to this company that went through the FDA process and filed a bunch of patents on certain techniques, when the general theory, the general thesis, and the general therapeutic process are well described, well understood, generally understood, and should be available everywhere?

I think over the next couple of years we're going to see this cancer immunotherapy-based system start to proliferate. You will see this happening overseas very cheaply. It is very efficacious, and we actually have a friend in common who's kind of standing up a business to do this in a very cheap way.

Up in Montana, where you have a right to try, and they can make this stuff.

We're going to see more of this come to market, I think. For certain types of cancer, it is very efficacious, very safe, and very effective. I'm excited by this.

For people who are listening who are normies, all of this cancer treatment requires early detection. That's something people have been working on with a blood test, Galleri, I believe it's called. So getting to the cancer and getting regularly tested is a critical part of this.

Yes. You're talking about a Galleri test by Grail.

The blood test. Yeah.

The blood test. Yeah.

Just early testing is the key piece for all of us to keep in mind: get tested early and often for cancer.

Yeah. And remember, there are a bunch of these new—what are called modalities, new modes of therapy, or therapeutic modalities—that are being used to destroy cancer. One was CAR-T therapy, where you take T cells out of your body.

And you use CRISPR or some other system to reprogram those T cells to go after—remember that protein I described? You actually program the T cells to find a specific protein in your body and destroy it. They go find the protein that's unique to a specific type of cancer cell, and the T cells get put back in your body and destroy it.

CAR-T therapy costs $1 million. For people with blood cancers, where the cancer cells aren't clumped together in a tumor but float around individually in the blood, putting T cells in your blood makes it very easy—or much easier—for the T cells to find the cancer cells, grab them, and destroy them.

For blood cancers like multiple myeloma, the efficacy is incredible with CAR-T therapies. That's a new type of modality. This is a new type of modality, this mRNA or neoantigen therapy. There are all these new techniques being used to destroy cancer, and I think in aggregate we should be very excited and optimistic about where this is headed.

Wow. Thanks for sharing, and let's hope for the best with implementing this. I love this right-to-try concept. That seems to be something we should talk about in a future episode.

We open sourced it to the fans and they've just gone crazy with it.

Oh man, my appetite will meet up.

We should all just get a room and have one big, huge orgy, because they’re all useless. It’s like this sexual tension that we just need to release somehow.

Wet your feet. Her feet.

We need to get Mercury’s back. I’m going all in.