Nvidia's Historic Quarter, SaaS Comeback, Bessent vs Druck, America's Debt Crisis, Cancer Vaccine

Welcome back to the number one podcast in the world — your favorite podcast, and your podcaster's favorite podcast. It's the All-In Podcast, with the Core Four, the Fantastic Four, the original quartet all here.

David Friedberg is fresh off his deep state conspiracy interview with Eric Weinstein. He also spoke with Kazios, and the two conversations together formed a picture of two sides of the coin of science in America. "Eric is one of a kind," Friedberg said of the DC trip. "It was fun."

Who is Eric Weinstein? He's a scientist who has worked with Peter Thiel and comes out of Harvard.

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

What is his heterodox thought about science?

There are things everyone takes as fundamental premises in modern physics — string theory, for example — and he called it out as unproved. So he's been kind of outcast in the scientific communities. He's made this case and proposed alternative theories that aim at a grand unification of physics, and he's been cast out; people call him a quack. He's talked about UFOs and about Jeffrey Epstein, so people label him fringe and use these terms to chastise him.

The point, similar to what Kratos has said, is that so much of science has followed this sheeplike mentality: everyone has to line up behind the same general theory or you get outcast — no grant funding, no tenure, no jobs, no lecture invitations. You have to follow the mainstream in science or you're an outcast.

It's rational to say that until string theory is proved, it's unproved. If you want to use that adjective, fine — it will probably offend some people — but he's more right than wrong on the substance. It's like Jay Bhattacharya saying COVID may have come from a lab leak. People saying that were told they were creating conspiracy theories and pushing misinformation, when there was no misinformation. If there's no absolute truth being discussed, it's theory — everyone was starting from theory on where COVID came from.

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

This is what has led to stagnation in American science. If you're not part of the mainstream, you're excluded, and because everyone has to think the same way, you don't get heterodox or outside-the-scope thinking — which means we're not pushing the envelope and discovering new things anymore. That's a pointed issue right now, and it's why I went and talked to Kratos, who put it differently: there's probably a lot more incrementalism in 2026 than there was in 1926.

Exactly. Think about the general theory of relativity being published — it was completely outlandish, but people still studied and assessed it. They didn't discredit it by saying, "You're an idiot for taking general relativity seriously." That's what goes on in science today: if you say something too far outside the standard approach, you're chastised, outcast, excluded.

Did you see the robot Olympics in China? Check this out.

I did see it. Yeah, this is pretty crazy.

The clips show robots sprinting awkwardly, backing up, and even losing a leg mid-race — which the group jokes is a "Freeberg sprint" and, in one case, like Christian McCaffrey in week three with his entire leg snapping off. One viral tweet quipped that the fallen robot "tore his HDMI, his CPU."

Beyond the jokes, the panel sees a PR masterstroke. One speaker admits Sachs is right about the PR: the CCP got a stadium full of Chinese people cheering for AI and robotics, while back home Dario Amodei is saying all jobs are over and it's the end of days. (The others cut him off — they're not talking about Dario this week.) Sachs pushes back: in a country of a billion people, is it surprising they couldn't find 4,000 to watch robots run around a track?

The real point stands, though: China is far more optimistic about AI than the US. Polls show over 80% of Chinese people think AI will be more beneficial than harmful, versus roughly 30% in the US. As one speaker puts it, the US leads China in every category except optimism — and pessimism about what AI might do is the biggest risk of shooting ourselves in the foot in this race.

Others remain skeptical of the spectacle itself, calling the robot blowing up at 16 seconds staged — "it's all scops" — designed to make you feel bad for the robots and cheer for them.

Optimus Update

Elon has been showing the latest Optimus videos privately rather than publishing them. One speaker who saw the footage asked Elon whether it was CGI or real; Elon said it was real and had been sent to him that morning. His take: Optimus will be the best-selling product.

The hard part with these robots, as another speaker explains, isn't training them for a specific action like running a race — it's handling conditions they weren't programmed to expect, especially interacting with the physical world: folding laundry, picking something up without breaking or dropping it, or doing something they were never taught. The first speaker agrees that Optimus's fidelity and ability to do novel things is exactly the point, but won't say more.

Grok Bot

Sachs has been "Grok bot maxing" — he just hit his usage limit and has to upgrade to the $200 tier. He's found it really interesting: it just does what you tell it to do and gets it done. What's great about Grok bot, he says, is that it's always on and it's in the cloud.

What's great about Grockbot is that it's always on and runs in the cloud. You set up your agents, and they keep working even when your computer is off or asleep. Other agent harnesses doing this, like OpenClaw or Hermes, ran on your desktop, so they stopped when your computer went to sleep. You had to set up all these Mac minis, and they were very hard to use, power-hungry, and super complicated.

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

It's pretty amazing. I told him he should make it multiplayer mode — so you can put two agents in the same room, but also put two people from your team in the same room, with humans in the loop. That's such a good idea; it would make this thing go so viral.

What's cool about the agents is that you actually have an agent swarm — lots of different agents performing different tasks, and they get good at different things. Some people ask why not just have one agent. It turns out that with multiple agents, each develops 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 they participate — and then they add another person. It could get very, very viral.

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 — they blew the doors off this — and we'll also talk about Salesforce coming back, with the SaaS apocalypse narrative ending. Nvidia's stock was up 9% on huge numbers and even huger guidance: $96.2 billion in revenue in the quarter, up 106% year-over-year — more than double. Wall Street's expectation was $92 billion. They guided for 70% growth next year, well above what Wall Street was expecting at 45%. So let that sink in. The number people are going crazy over: $60 billion in net profit — 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, 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 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, 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 — in line with expectations — but they beat their 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 is on May 15:

I think the high end of the market where Mark 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 put in a prompt and it all works — that'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 — which is a way of saying who has the best relationships. Those guys, I think, are positioned to crush.

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, 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.

It was pretty obvious to me in May that Salesforce specifically was meaningfully oversold. What I was trying to say there — and maybe I can say more precisely now — is that we're getting to the end of this second phase of AI. If the first phase of AI was models — think of a model as a brain — the second phase is harnesses: 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 the next phase soon — at least I see it in the enterprises I work with — which is that that's 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. Meaning: you take a brain, you give it arms and legs and eyes, but then you have to train it to be a lawyer, a customer service rep, a sales agent, or what have you. And to do that next step, you need a ton of contextual information. The people who control that today are 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 that Mark is 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, why his revenue is strong, why he's guiding up, and why the stock ripped. Was it up like 43% since I said it had bottomed?

Yeah, it was up 20% today just on that. And Freeberg, were you on CNBC, or just on the quarterly call?

I went to Benioff's office yesterday for the earnings setup. Daario was there, and Daario and Mark did their CNBC bit—they set up a studio for the earnings call on the 60th floor, where we've all had dinner. Then I did the earnings call with Mark, because about a year ago I mentioned to you, Chamath, that over a weekend we built a CRM tool internally using Claude Code and Cursor, stood it up, and started using it. But then it was, "you've got to add this feature, change this," plus account security, access, and 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 make it truly useful and scalable.

Meanwhile, we had competing interests at our company about where to build software that's unique to us, like plant breeding software, which is our unique advantage. How do we build a better breeding technology platform, software that gives the people in the lab, the greenhouse, and the breeders the tools they need? We started building all these internal tools and eventually realized our time and energy are better spent building the workflows that are unique and create value specifically for our organization, rather than recreating software other people have already made. Where's your best ROI—your best return on time and dollars spent? It's not in rebuilding a communications tool like Slack, a messaging tool like Gmail, or a CRM like Salesforce.

At the time, Mark and I were talking on the phone. He said, "Hey, try Salesforce, I'll set you up, come on, do it." I said no, no, no, and he said yes, yes, yes—he's so persistent, you guys know Mark. He put his best sales guy in a text thread with me, texting every day: "Are we up yet? Are we up yet?" Until I finally told the sales guy, "Okay, go to my office." He got set up, and then we were done—and we realized, oh my god, this is actually a much better way for us to do this. So Mark asked me to recount that story at the earnings call and talk about my view on this change.

As I've shared before, I think the horizontal platform companies are fine—CRM works across many industry verticals, as do Gmail, Slack, and so on. No one's going to go rebuild Excel. The real value is in software that's unique to your vertical. That's where AI plays a role in building custom workflows and custom software. So the real "SaaS apocalypse" is more of a vertical SaaS apocalypse—software designed for just one vertical. That's where I think things get blown up.

>> I totally agree with you.

>> Sax, let's talk about Marc Benioff. He's 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 how he went from roadkill in the market's mind to leading the pack in SaaS again.

The SaaS Apocalypse Narrative Was Overdone

This whole "SaaS is dead" narrative is getting shredded today with Salesforce being up over 20%. I've been saying for months that I don't believe core systems of record like CRM are going to get ripped and replaced with something vibe coded. Enterprises want certainty. They have compliance requirements, and they want professionally managed software that's been running and debugged for years—just think of all the bug reports filed against Salesforce over 20 years. That's the hard part to get right.

When I play with these agents and AI tools, I'm dealing with something probabilistic with so many edge cases. With a core system of record, you just want to know that it works, and the expense 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. So the SaaS apocalypse narrative was overplayed.

Benioff's "If You Can't Beat Them, Join Them" Play

At the same time, Benioff is doing all the right things here. He made this deal with Anthropic: first, he'll be offering Anthropic's models inside of Salesforce. More importantly—and this is really the critical thing—he is integrating Salesforce into Claude. He's willing to let Claude be the front end and have the primary user relationship, with Claude able to access all the data in Salesforce, all the workflows, and all the actions a user can take.

In other words, Benioff 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 you're using and have them run across all of your SaaS platforms. So he's giving up a big piece of the user relationship to make himself more relevant.

Why is this ultimately good for his business and not bad? Because I still think you need systems of record. You need the AI agents to go to those systems and get data from a canonical source of truth—you don't want any probabilistic variability around that—and you want agents to be able to write back to those systems and workflows. So this isn't a rip and replace; layers are getting built: the database layer, then applications and workflows, then agents, and then the AI user interface. Benioff is saying the AI is now the user interface—he's acknowledging it and leaning into it.

Trapped Value in Salesforce

I talked to him this morning, actually, and a point he brought up that I think is underestimated is that there's a lot of trapped value in Salesforce. It's an incredibly broad, rich platform with tons of functionality that the average user will never find—but the agent will. Think about all the cases where the agent goes to the user and says, "Hey, I just saw 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?" And the user says yes.

Think about the agent as a power user of Salesforce that knows how to use Salesforce perfectly. You don't necessarily need to hire this 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. Once you develop trust in your agent, you as the user just click the "always allow" button, and now the agent can tap the full capabilities of Salesforce. That's where this is headed.

But that raises the question: how do SaaS applications play into this AI and agent wave? You have to have great APIs and a great CLI — the command line interface, which is how agents relate to SaaS tools. 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 relax the idea that you're going to fully control the customer relationship. Think of agents created by the AI companies as an extension of your platform, rather than feeling 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, you've got a big opportunity.

I have to say he's one of the most underestimated and underappreciated product and sales executives in our industry. When we were talking at Davos about Slack, I told him: you have all my data, I can't get my data out, and my team wants to move off Slack to an open source product specifically so I can get all this information. He put me on with the sales team, and now I have complete access to our Slack. They built a product called Slack Today, which they've been slowly rolling out, and it delivers the promise of OpenClaw. After Sam bought out OpenClaw's founder Peter, who then started working at OpenAI, OpenClaw just faded — no more innovation. I'm not saying Sam did something nefarious or that OpenClaw is terrible, but they just lost their edge.

Slack Today is like a seven out of 10 right now, and I'm noticing it getting about 10% better every couple of weeks. He's going to do a parallel play, Sacks: he'll have his own agents and let you bring in your own agents. His agents might be a little behind, but you'll also get those agents for free inside Salesforce. It's two ways to win.

Look, I think SaaS products can offer their own agents, AI, and copilots, but the reality is that most of the action with agents is going to happen outside those products. Agents will simply want to interact with those products — get the data from them, use them as the system of record and source of truth, and write actions back when appropriate. Founders should lean into that trend instead of fighting it, because if you fight to stay in control and insist that only your agents can interact with your system, people will just leave.

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

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

No, I don't think vertical SaaS has a system of record. That's not what they do — they're workflows. This is the key point: horizontal monolithic companies like Salesforce, Workday, Oracle with respect to their GL, SAP, IFS — it's very difficult to see a path where you can replicate these businesses. They become the repository and the single source of truth. But vertical SaaS is workflows and process, and I just wonder what survives.

It's a good question, and it's case by case. But just a few months ago, everyone was saying horizontal SaaS wasn't going to survive. One of Leopold's major positions in that fund was to short all these SaaS products, including Salesforce. What we've seen is that that has turned out to be wrong — totally overdone.

To be fair, I'm not sure he understood the difference between vertical and horizontal SaaS, because that was a funding-short problem — he had to short the big stocks to get the leverage he wanted. I suspect he worked backwards from his own mental narrative to make it sound correct: he needed to be massively net long certain names, his prime broker told him he needed liquid shorts, and he picked Adobe and Salesforce.

But that was the dominant narrative a few months ago — the "SaaS Apocalypse" — and Benioff's stock was getting hammered, down 50% before recovering 20%. A lot of people were indiscriminately selling all SaaS. If there's one takeaway from today's news, 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 goes to zero. What we're actually seeing is that certain systems of record are extremely compatible and complementary to AI agents, and if their founders position them right, they can lean into the trend.

The job-loss narrative is another simplistic extrapolation: AI can do knowledge work, therefore all knowledge workers will lose their jobs. It ignores the fact that knowledge workers can take advantage of these tools and lean into them, the same way Benioff is leaning into the agent trend.

If you adopt the technology, you survive; if you don't, you die. It's pretty straightforward.

Nvidia's quiet acquisitions

Alongside Nvidia's incredible results, two things happened very quietly — Jensen is clearly open-source maxing now. He bought Hugging Face for $12 billion and did an acqui-hire-style deal with Poolside for $6 billion, roughly $20 billion in acquisitions total — far less than 1% of Nvidia's market cap. Thank God you can do M&A again in this country. He's now owning the distribution layer for open source. Poolside makes the coding model Laguna, so Nvidia now owns the place where developers and hackers go to figure out which open-source tools to use.

I don't think that's quite what went over, though. Jason Warner tweeted this morning about a new model drop, and I was confused because there could be two things they're buying: the 100 engineers going over to Nvidia could be working on what Poolside has — effectively a version of Cursor and Claude Code that runs on-prem —

— open source, open weight —

— no, they also have a model. They have a model plus this agent harness for directing coding. It wasn't clear to me from the announcement; I didn't read it carefully.

I think we're burying the lead, though. These reported deals with Poolside and Hugging Face are big open-source acquisitions if the reporting is true. But the real lead today is that Nvidia had a blowout quarter, growing over 100% year-over-year.

A billion a day in revenue now.

Right — almost $100 billion in revenue for the quarter, $60 billion of profit, 75% gross margins. Excluding one-off items, this was the most profit ever generated by a company in history.

The AI boom is continuing

To underscore what was said in the intro: the reason the stock is up 8% today isn't just the great quarter — it's that they guided to 70% growth next year, when the street was expecting around 45%. And that 70% number would have been higher, but they said they're supply-constrained.

The takeaway for the larger market is that the AI boom is continuing. The other narrative getting shredded today is that AI capex is a bubble about to end very soon. What Nvidia is saying with its numbers and forecasts is that this AI capex will continue well into the future — it's got real legs. And the amazing thing is that even after the 8% bump, Nvidia is only trading at 12 times earnings.

The reason Nvidia trades so cheaply is competition anxiety, and here's the key point: I think Jensen is going all in on open source based on the Hugging Face and Poolside acquisitions. Sam Altman poked the tiger. He had done the huge hundred-billion-dollar deal with Nvidia back in 2025, then announced the AMD deal and his own inference chips — the Jalapeño chip — and Jensen was clearly not pleased. Right after Sam's announcements, Jensen came out and said, "By the way, we don't actually have to do these investments in OpenAI," and I think he's now sprinting toward open source.

You're going to be able to buy your entire stack from Nvidia next year. He's going for it — he's going to own open source. That's a striking shift, because six to twelve months ago the consensus was that 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 that's the huge story.

The generalized takeaway is that all of these businesses are going to converge, competing with each other in every way possible. The neat separation between customer and supplier is melting away. Nvidia still relies meaningfully on the hyperscalers for revenue — it was $24 billion in the quarter — but it has now built up a Neocloud business of equivalent size, which is incredible. And as the hyperscalers say they'll spin their own silicon, it's completely appropriate for Nvidia to spin its own models, go up the stack, host the models, supply APIs and inference, and become a cloud provider.

In business 1.0, you were either a capex business, judged on return on investment, or a software business, judged on opex and ROI. It's all melting together — every company is going to do everything. Five years from now, these big companies will all have their own cloud, their own models, their own silicon, their own data centers, soup to nuts. Then the question will simply be which is better, and I think that's better for the broader ecosystem — better for all of us.

Bessent's new bond strategy was criticized by his mentor Stanley Druckenmiller. Last week, the 30-year Treasury hit a 19-year high of 5.3%. In response, Bessent doubled the Treasury Department's elongated bond buybacks from $2 billion to $4 billion on August 19th. CNBC reported that Bessent is considering ramping up bond buying, and Fox Business said he wants to put the fear of God into traders shorting long-dated bonds — a lot of interventionism going on. Druckenmiller, Bessent's former mentor, published an op-ed in the Wall Street Journal opposing the 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 in debt, no more DOGE, no more cutting costs, plus a war that's quite expensive every five months. Freeberg, I think you said last episode we add a trillion, so we're adding $2.5 trillion a year, and we'll add $10 trillion during this second Trump term. Your thoughts on what's going on here between Bessent and Druckenmiller? Chamath and I can also talk about using AI to write these pieces.

I'll start with the relationship. Bessent and Druckenmiller worked closely together, are close, and respect each other — as does Kevin Warsh, who was working with Druckenmiller in his office every day for the last number of years. They're extremely close. This is a triumvirate — I don't think in history you've had three people so close in these positions: one leading market activity with Druckenmiller, one running Treasury, and the other running the Fed. It's pretty unprecedented to have this sort of closeness in those positions.

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

I'm going to leave that to them to debate.

The Treasury yield curve, then and now

It's worth looking at the Treasury yield curve from the COVID era. This shows, depending on the duration — a one-month or a 30-year Treasury bond — what the yield is across that whole time span: how much the federal government has to pay in interest to get people to loan it money for that duration. Around 2020, because interest rates were low and the inflationary effects of COVID stimulus had not yet hit the market, we were looking at sub-half-a-percent, sub-quarter-percent yields all the way out to the 3-year and 5-year, then a slight climb where you could actually get 30-year Treasuries at 1.7%. 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 that point — but the debt is balanced across this Treasury curve.

Now let's look at where the curve is today: what the market is charging the federal government to loan it money to pay its bills. On the low end, you're looking at just under 4% — for the federal government to borrow money for one month, it has to pay a 3.8% vig to lenders. And out at the 30-year, it's now at 5.2%. It is very expensive now to borrow money if you're the US federal government.

The reason is persistent inflation, which I would argue comes from excess government spending on social programs and other things. The big problem at this point is that the federal government is spending so much that if it were to cut spending aggressively, the argument and the concern is it would hit unemployment and cause a recession, because the federal government is such an intricate part of the economy. That's the argument — but the spending is causing inflation and deficit spending. This year the deficit will be roughly $2 trillion, and as a result the market is saying: we are worried about US fiscal solvency over the long run, or there's a higher risk, so we're going to charge you a higher interest rate — 5.2% on the 30-year.

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

That means 1.25% of GDP in interest each year for that 1% change in the interest rate. We're now looking at a 30-year at 5.2% and a short-term rate over 4%. 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 — the bonds are maturing, the principal has to be paid back to bondholders, and the Treasury has to go back to the market and sell more treasuries to borrow more money to refinance. So borrowing costs will climb, the government's burn goes up, and the fiscal deficit goes up.

My argument is that there is no action Bessent can take that will have a meaningful effect on the long end of the curve. We have a fundamental fiscal spending problem. The note that Druckenmiller wrote — whether AI wrote it or he wrote it, it doesn't matter — makes the correct point, and it's meant to provide cover for Bessent: it is not Bessent's fault or responsibility to solve the yield curve problem. It's Congress's responsibility, the president's responsibility, the responsibility of the people who hold the budget to stand up and say we're going to cut spending — because if we don't, there's going to be a spiral, which we'll face over the next 12 months as we refinance that $10 trillion.

Bessent is doing his job as Treasury Secretary: going to market, buying treasuries, trying to lower yields. But he has at most a trillion dollars of buying capacity, and he still has to sell $10 trillion of treasuries in the next 12 months. Even if he maxed out his buying authority in the near term, that's only a trillion of buying — and then he has to turn around and sell 10.

"So Druckenmiller's note is coverage for Bessent, pointing responsibility back to Congress, back to the president, saying we have to solve this deficit problem. Is this like kayfabe between the two of them, supporting each other?"

I don't think they're allowed to talk to each other, so I don't think that's happening. But Druckenmiller is sending a signal to the market: this is not Bessent's responsibility. The burden isn't on his shoulders — he can't be Atlas, he can't lift this. You guys have to act.

"Chamath, what's your take on what happens going forward? We're past 5% — does this break? What does it do to interest rate cuts, and how bad could this get?"

The tactical answer to why he's in the market buying bonds is yield curve suppression: buying bonds bids up prices and pushes implied yields down, which lets the Treasury finance at a lower borrowing cost — if yields print at 5.1 versus 5.3, new issuance is marked against that lower print. The suppression is meant to help with the refinancing tsunami that's coming.

But step back: the very big issue is that Congress is ultimately responsible for spending, and it is completely unable to get these budget deficits under control. It's consistent — not a Democrat or Republican problem, but a congressional problem. Both sides consistently spend way more than they should. Debt is growing at 7% while GDP grows between 2 and 4%. That's a recipe for disaster.

What Druckenmiller said in that note is that you have to cut entitlements, and he's probably right. The question is when — and who will force the United States Congress to confront this issue?

What he is saying is that the bond market is going to start to force this, because that single price is the true reflection of good faith and trust in the United States. The simple rubric is: as the yield goes up, trust goes down.

This is happening at an incredibly awkward moment, because we are supposed to be in the middle of an enormous financial buildout to support AI. Go back a hundred years, to the industrial revolution and the grand bargain, and the United States government was the balance sheet — it was the one able to step in. Unfortunately, because of its financial situation, the US government is not able to do that. That's why, thank God, we have companies like Nvidia, Google, Microsoft, Meta, and Amazon taking on that burden.

I wanted to tie this to all the chirping on the internet about Nvidia's balance sheet, because I think people completely miss that these companies are putting the entire US economy on their back. A hundred years ago it would have been the US government doing this, but Congress cannot get its act together on spending. The president has actually done something pretty incredible: on the tail end of those trade deals, he found $2 trillion to pump into the US economy to support this buildout.

But the practical reality is that the two trillion comes from other nation states, and their processes are long and meandering. So all this money has to be invested, it should be backstopped by the US government, and the government is not in a financial position to do so. The president has done some magic to relieve it in the short term, but it's only a short-term band-aid. The real solution, as Freeberg says, is that Congress needs to come together and fix this problem. If you see the third-year yield at 6%, it is the beginning of a death spiral. It won't be immediate — don't freak out — but it is the beginning of some extreme pain, and that pain will last years. The only thing that solves it now is getting the budget under control, which is a congressional act. Whether it's this year or the next, it just has to happen.

Why no politician can touch spending cuts

Question: Sacks, the president was quite vocal about cutting costs early on, but we haven't heard much from the commander-in-chief about this. Is cutting spending just too toxic for anyone in politics? He sent Elon in there, which was pretty spicy, and obviously it's not occurring now — in fact, the debt is growing faster than it was a year ago.

Sacks: Part of it is a tragedy of the commons. You've got 435 members of the House and 100 senators who all get a vote, plus the president, and no one of them controls the spending — the power of the purse originates in the House. It's very hard to control spending when so many people weigh in, each with an incentive to protect their programs. If a couple of people could get together in a room and sort it out, they probably could, but that's not the system we have.

The president doesn't have a line-item veto — it'd be great if he did — and he doesn't have the constitutional powers to completely rein this in. Despite what the resistance liberals say, he's not a king. That's basically the heart of the problem.

You're right that he backed 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 sub-agency, and the hysteria was so great that they accused him of killing millions of children — total nonsense — and Tesla dealerships were getting firebombed. That's the other piece of this: how do you make reform?

The Breaking Point for Spending Reform

How do you reform spending and rein it in when the media isn't on your side and won't report this stuff accurately? That's the worst part of it. The media has no political incentive to engage in log rolling, but they do absolutely nothing to report the story accurately—just look at how they're reacting to all of this. It's a difficult problem when you start from the fact that no one person controls spending; it really is a tragedy of the commons, and then the media isn't doing a good job reporting it accurately. I don't know how we get reform here. And if the Republicans lose control of Congress, I don't think that will make it better, because the political energy in the Democrat party is all toward these massive new spending programs that the DSA wants. If you vote for "change" by voting the Democrats in, that won't help either—they just want more and more spending. So I don't know what the solution is.

The interviewer pressed for a timeline: Republicans love to cut taxes, Democrats love to increase spending, and this hasn't gotten better through three or four presidents. The last person to balance the budget was Clinton, and it's been getting worse and worse. At some point the bill comes due and austerity measures will be needed. When does that hit? Is there a breaking point in terms of interest-to-GDP—maybe creditors demanding 12% on the debt, and we simply can't pay it?

The question is what it breaks to. As a democracy, we have this recursive solution: as people feel pain, they react with a vote. This goes back to the question of whether we break for socialism after the midterms. Sometime between 2026 and the 2028 election cycle, the inflation problem will matter—and remember, inflation is fundamentally rooted in government spending. Deficit spending correlates closely with the cost of housing, healthcare, education, and everything else: the more the government spends, the more expensive things get. Inflation is persistent because what we thought were temporary emergency stimulus measures during COVID very quickly got normalized into persistent spending cycles. If we went back to 2019 spending levels, we would actually have a budget surplus right now—the economy has grown so much since 2019 that the federal government would be running a profit and paying down the debt. Instead, we've made emergency spending permanent, just as we did after the 2008 global financial crisis, when crazy stimulus measures were persisted.

So what happens next? If inflation stays high and affordability keeps falling, people will vote for something that promises a solution. And I don't see a political message winning that says the federal government will now spend less and cut a bunch of stuff. So I worry the one-way solution is something very damaging to individual liberties, to rights, and to the things that make a free market work—these more socialist principles and socialist candidates.

That's the breaking point: a spectrum, where the reaction during the 2026-to-2028 transition phase can be negative. The next big thing happens around 2030 to 2032, when Social Security no longer has money to pay out. In that era, many states will also go bankrupt, unable to pay their obligations. When that happens around 2030 to 2032, you'll have both a Social Security bailout and restructuring as well as state bailouts and restructuring—and that might be when all of this breaks.

The interviewer reacted: so you're saying we'll see municipal defaults—states defaulting on their debt. That would be cataclysmic, having to bail out something like California.

The obligations on unfunded pensions in the states are so significant, not to mention the debts. California has spent tens of billions of dollars on a railway that goes nowhere, where there are no tracks, and there isn't even the political will to stop it. 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. Anybody who was in favor of a railway to nowhere should just be shamed out of existence. A 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 — this thing is absolutely ridiculous. 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 figured out how to do it at a fraction of the price. If this were a good idea, private industry would have built it. There is no opportunity here.

Across the board, these government programs cause more harm than good. When the government intervened in underwriting student loans and gave everyone a loan, administrative costs went up 6x and tuition skyrocketed 8% a year, compounding for 30 years. The same with housing — now no young people can afford a house. The same with health care, where they said we'll pay people to stay home and not work to take care of people, and we won't check whether they're actually taking care of people, and the cost of health care skyrocketed. Every one of these government programs has the adverse effect of driving up costs and inflating everything. The core root of inflation in this country 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 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 to your point about media, Sacks — 100 Nick Shirley, 60 Minutes, whoever it is — someone needs to just say this spending, this fraud, is not acceptable, and we need to send people in to recapture that money. We're burning money we don't have. The fraud is layer one; there are like 18 layers to this.

Given the tragedy of the commons, the structural nature of it — there's no one person in control of it — you need massive public pressure to control it, and there's none of that. And 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.

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

Socialism is much more likely — a hundred years ago, when these things were less interconnected, America could decide to change course. But now all of our debt is interconnected to domestic and international actors. It's not a decision; there is no button you can push, because you would go bankrupt overnight. So the reality is that we will hit some kind of fiscal wall. Something will get terribly broken, and like in COVID, like in the GFC, like in 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.

And in every one of them, it ended up becoming a "never let a crisis go to waste" moment, always argued in favor of some big power grab by the government.

What was your question for me?

Think about the affordability of houses. That depends on mortgage rates — the average 30-year mortgage right now is 6.73%.

A little under 7%. It's about 150 basis points above the 30-year Treasury. So 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 be able to afford the same level of house you would otherwise, because your monthly payment is much higher? Do people connect those dots about what's happening?

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.

Exactly. That's the human brain. That's how it works. There's some problem — who do we blame? That's the first thing you think about. You don't think objectively, "What's the solution to the problem?" You think, "Who do I blame?" And then when mortgage rates go up, who do you blame? The current president, my congressman, whatever — reelect someone. That's the response. And the person who shows up and says, "I'm going to give you free housing, free buses, free mortgages, I'm going to take the land from the landlord and give it to you" — you vote for that person.

So 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 of being able to get a home — is giving the keys of the kingdom over 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. 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 to make our economy grow big enough that we can grow out of the problem. 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 three people who are working.

I'm just kidding. We're joking. Dario, come to the summit. It's not Dario's fault. But anyway, keep going.

No, seriously, come to the summit. Listen, we are kind of counting on AI as the deus ex machina to grow our way out of this thing. We need more quarters like Nvidia and Salesforce just posted — we need that to happen for like 10 years to grow our way out of this. One great silver lining 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 to a family, to 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 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 — down 27%. Since March of 2020, 25%. It can be done. You need leadership that says housing is a priority, therefore we're going to let you build more houses. It's being done in Texas, in Florida — those conservative, let-the-free-market-work environments. Let them build: Brightline has high-speed trains going 125 miles per hour because they got out of the way and 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 preventing 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 J-Cal political rant. I'm just pissed off about it, because—

It's a total non sequitur.

It is a non sequitur in some ways, but you know, 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. It just came to me at the top of my head.

Well, actually, I hired Sacks's writer to do some punch-ups for me. But putting that aside, it is obviously a leadership issue.

I'll get him a watch. Let's do a side—

The question on the table: what did everyone think of Druckenmiller writing his Wall Street Journal piece with AI, getting caught, and the Journal responding that it's totally fine?

One speaker argued that it isn't only the left that virtue signals — it's a class of people who do it, who judge others, can't see the forest for the trees, and are the loudest chirpers in the peanut gallery. In his view, Stan Druckenmiller is the most incredible investor of our generation — or at least top three or four — so critics should shut up and read what he says. Instead, they fixate on trivia: "Oh my god, he has an em dash, he used the AI flourish." They're missing the point — he's telling you what's about to happen.

When asked whether opinion pieces in the Wall Street Journal and New York Times should be written by AI rather than the person whose name is on them, Sax pushed back on the framing: Druckenmiller never claimed it was written by him — he admitted it was 100% AI, but he prompts it, shapes it, and guides it. A more precise way to put it, Sax said, is that the take is his, and he uses AI to help him write it. He doesn't get the pearl-clutching: you'd have to be pretty dumb these days not to use AI to help you write, and Druckenmiller is definitely not dumb. The opinion is his own take, consistent with what he's been saying for decades, and the piece itself cites the many times he's weighed in on this before.

He added that the critics on X should be asking the questions that matter instead: How does this affect my company, my portfolio, my 401k, my parents' retirement? How should I think about hedging this? Rather than that conversation, everyone was just saying, "Oh my god, he used AI."

The opposing view: it's lip-syncing writing

Jal disagreed, calling it the lip-syncing of writing. As a writer, he finds it offensive to let the system give your entire opinion, as opposed to using it for research — which is totally fine — or proofreading, also fine. To him it feels like going to see Adele and hearing her lip-sync: he wants to hear the person himself.

When another speaker mocked this as equivalent to refusing Excel and trusting only people who do calculations by hand, Jal clarified the distinction: he'd rather Druckenmiller had said "don't use an Excel spreadsheet, I don't trust those guys, I only trust the guys who do the calculations by hand" — the point being about the substance of the argument, not the tool.

Jal's core objection: if you're giving your opinion and it isn't written by you, it's disappointing that it wasn't disclosed. He holds a much higher standard for writing in your own voice. Sure, a ghostwriter could help polish it, you could check grammar, do research — those are all valid uses. But writing something with AI and letting people read it without knowing feels icky and insincere. He'd prefer people say upfront, "I didn't write this, but I'm in agreement with what the AI said." He finds it personally offensive.

What do you think is the difference between that concept and requiring artists to disclose whether they used Adobe Photoshop, and the very specific filters they used in it? Or DJs who use specific software to create electronic music instead of making analog recordings by hand with guitars? Or people using Microsoft Excel to build spreadsheets instead of writing everything down? Do I have to disclose that I used Excel, or a calculator, or a synthesizer, or Photoshop instead of doing it in an analog way? At the end of the day, digital tools for creativity work in parallel with human ingenuity. Human craft and creativity are expressed, magnified, and in many cases improved by digital tools. The definition of whether someone is "using AI" is just 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, to create novel paragraphs from scratch? If I give the AI a whole bunch of inputs, take the outputs, and edit that output, am I "using AI"? I think this is one of the fundamental issues: we treat AI as a standalone, simple intelligence-replacement system, and that heuristic doesn't match how humans actually use it. Like all other tools, it's a magnification of human creativity, ingenuity, and potential. Go ahead.

The reason is the expectation of the reader. Readers have a high expectation, and that's why people were so disappointed in this in a very large way. It wasn't virtue signaling — they actually care about Stan's opinion and 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.

Betrayed?

Yes, I think they did. 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. People don't want to pay for a teacher who uses AI slop either. So we as a society have to decide when we need something to come from the human — and understand whether it's 100% human or 60% human — and when we trust the AI on it. I want to hear his actual opinion.

I have questions. Many questions. Question number one: are you suggesting that Stan didn't read the thing that was generated?

I'm sure he did read it, yes.

Okay. So when he read it — before that, that's an open question.

Well, he defended it. Yes, he read it.

I hope he did. So do you have an issue with the fact that he read it, thought "yeah, this is what I think," and hit the send button? That's a betrayal to you?

I think not disclosing it is the betrayal. 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'd like to make — oh, listen, what exactly?

The standing disclosure for all future publications I put out, including tweets, is that I try not to publish anything without running it through AI first — to fact-check it, to line-edit it, and to 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, fact-checking — great uses of AI. Grammar checker, great use of AI.

Now you're debating the slider of how much proofchecking he did.

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

Question: is Stan, in his next op-ed, allowed to use a computer?

Of course.

Okay, let me keep going. Is he allowed to use a word processor, or just Notepad?

You obviously understand he can use any of those tools — he should disclose.

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

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

Asked what Dr. Miller's process was on the post—whether it was simply delivered to him fully baked and he just went with it—the answer was that the AI detection software said it was about 90% written by AI, so he'd be surprised if that was the process. "This is just a guess because I don't know. My guess is he had the take, and he probably told the AI the take—either in writing, or just speaking to it, or essentially speaking to his assistant who dictated it and put it in 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 just said print it."

Another speaker offered a mocking 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.

The problem with the lip-syncing analogy, one speaker argued, is that he's criticizing something that just happened—the best-and-buyback policy, something very specific his former colleague, the Secretary of the Treasury, just did. A lip-sync is when you have a song that's been performed a thousand times; this is novel and new. The reply: that's even more evidence of why he should write it himself—it's about respecting the audience and taking the time to write it yourself. "I was part of the audience."

Asked whether he had read the article and had an opinion about its substance, one speaker said he wanted to know if it was coordinated, some sort of campaign, and admitted he didn't know it was AI-generated before someone printed that on the timeline. Another said it came up in his timeline first, framed as plagiarism or "AI slop," before he could even answer the question. "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. It sounded completely consistent with the many other times I have heard Dr. Miller speak about the debt problem, on podcasts and written versions or whatever. I didn't know. So if he's plagiarizing from someone, he's plagiarizing himself."

The response was that plagiarism isn't the right analogy—it's not perfect because this is a new thing—but that he's taking credit for what the AI wrote. "I would prefer him to take the time to respect the audience, to write it himself, because I have respect for him and I want to understand his true opinion, not as interpreted through AI. Just like when I hear somebody sing a song, I would like to know that they don't autotune it and haven't post-produced it to the point of absurdity. That's me. I believe the artist should give you their actual art, giving you their actual position without using the AI. That's just my opinion."

The counterargument: even top-performing artists use autotune and other tools to make themselves sound their best. Similarly, AI gives people superpowers. Dr. Miller has said, "I'm not the greatest writer"—what makes him interesting is that he's a thinker with a take, a long-held position, and he's willing even to criticize his former colleague and, presumably, friend because it violates his view of the world. He says he uses AI to make his writing better. "I don't get it."

"I don't have a problem using a Grammarly-type tool. I just don't like the idea that it wrote it in full, 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 to being written by AI. That's my preference, and that was the expectation of the audience. But this is a new situation.

"Now that you know Duck 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 you're saying the substance of what he says will no longer matter to you, or you're going to discount the substance less?"

You just said that, not me. No — I will wonder what his actual opinion is, which came out of him personally.

"You're going to trust the greatest macro trader in the world less tomorrow than today because he used the tool to write it?"

I will wonder if 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'll make me wonder whether that is actually his perfect opinion, crisply his — because I think he's phoning it in. And 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.

CIA director John Ratcliffe made an unannounced visit to Moscow on Tuesday, flying back the same day. It was the first visit by a sitting CIA director since November 2021, when Bill Burns went to warn Putin not to invade Ukraine — he invaded three months later. The Wall Street Journal reported that Ratcliffe was there to warn Russia not to attack a NATO country, with new US intelligence reportedly showing Putin could be planning to test NATO's resolve with a limited assault in the next few years — anything from a cyber attack to a small land incursion. Polymarket puts only a 21% chance on a Russia–Ukraine ceasefire by the end of the year.

Freeberg said there's wild speculation: the Wall Street Journal and others reported the message delivered was "don't invade NATO and don't use nukes on Ukraine," and it seemed like something meaningful was going on. But as Sax pointed out in the group chat, this is speculation — there's no evidence Putin made such threats. More than that, the president called into a show to say the story wasn't true and the contact was routine, and the Russian readout said roughly the same thing. The truth is, we don't know what the meeting was about.

Sax pushed back: both the American and Russian sides have denied the dramatic claims that Ratcliffe was there to warn the Russians not to invade Europe, and he believes those denials — it wouldn't make sense for Russia to invade NATO countries. He called it the same threat inflation heard for years from neocons about Russia's supposedly imperial ambitions to conquer all of Europe, saying there's simply no basis for believing it.

On the war itself, Sax argued it's not going badly for the Russians. It has taken far longer than anyone thought and Russia has paid a heavy price in casualties, but most sources covering the war now say the Russians are making slow but steady progress while Ukraine suffers the most. Russia has air superiority, Ukraine is essentially out of air defense, and Kyiv is being bombed routinely — the Russians can destroy what they want at will. Ukrainian infrastructure is taking a heavy toll heading into an expected cold winter, with power being taken out. Odessa and Ukraine's Black Sea ports have been bombed and largely shut down, so Ukraine can't export grain and its economy is being destroyed. Zelensky says he needs another $35 billion or Ukraine will run out of money; the US isn't funding that, but he's trying to get it from the Europeans. Ukrainian casualties, though the numbers are unknown, are also huge, and there have been stories about potentially drafting women because they're so low on soldiers. Russia has four or five times Ukraine's population, so it can sustain much larger armies.

His conclusion: things don't look great for Ukraine, and 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 make a deal then and still doesn't, but Sax thinks he should — and he doesn't believe Ukraine will come out on top.

Chamath declined to add anything, instead thanking Meta for solving his biggest problem by imposing limits on Instagram — a big story the show hadn't gotten to yet.

I'm tired of playing whack-a-mole with my kids — they have fake accounts and secret accounts and all the rest of it. So thank you; it should have happened sooner, but I'm glad Meta did it. I'll give them some credit — they were dragged kicking and screaming into doing the right thing, but they did it. I hope TikTok and YouTube do the same thing, because it will make the life of a parent much simpler. It will let us actually parent instead of just policing device usage. I'd love to see Android do it too. And it's exhausting being a child with a device — it's exhausting, and it's terrible for young women to be on Instagram.

We held the line until 16 years old, but at 17 we lost the battle. Where did you hold the line for your kids on social media? My kids aren't allowed until they're 16 — but that's not really the point. As we've talked about before, I've been very consistent on this: this could have happened a lot sooner. I'm glad it happened, and I applaud Meta for doing what they did. I hope TikTok and YouTube follow suit. These limitations are coming in Australia, Canada, and the UK, and they'll probably appear in other places. Physiologically and psychologically, these limits are really helpful for kids.

The parental control apps aren't that great. They do a decent job, not a great job, so as a parent I'm always on my back foot. It's frustrating — I'd rather spend time with my kids on other things than screaming at them to get off their phone. Yeah, the tools are hard to implement.

All-In Summit sponsors

Freeberg, what's going on with the summit — how are you delivering it? The All-In Summit is just two weeks away, and this year we have some incredible sponsors helping bring big ideas to life. Iron is back as the presenting sponsor — big thank you to our friends at Iron. Make sure you visit the Iron House at the summit, the on-site lounge where you can kick back, relax, and hang out between sessions.

New this year is the gifting boutique you may remember from Napa, brought to you by merch.com. Every attendee fills a bag with All-In merch plus special offers from sponsors — you walk in and walk out with a bag full of goods. Oracle is back, bringing an F1 racing simulator to the summit. If you've ever wanted to know what it feels like to take a corner at 180 miles an hour — sort of like being in a debate with JCAL here on the All-In Pod, but without ending up in the hospital — this is your chance. And EY is taking care of business with their meeting pods, a quiet space to step out, take a call, close a deal, or hold meetings — which happens a lot at the summit. People are doing deals.

That's just the tip of the iceberg; we're keeping everything else under wraps, with more fun stuff coming. We're excited — we'll see you all at the All-In Summit. It's going to be a great party.

The stock in question is Moderna, which went from roughly a $20 billion market cap to $60 billion on positive readouts from what's being called a cancer vaccine. It's worth explaining how this therapy works.

Neoantigens: teaching the immune system to attack cancer

The idea goes back to the late 1990s: creating what are called neoantigens — small peptides or proteins that mimic a protein unique to a cancer in your body. Because your immune system sees the antigen as an invader, it starts making cells that destroy that protein. By putting that protein into your body, your immune system gets activated and goes looking for that protein elsewhere, destroying it. The objective is to identify the protein unique to a particular cancer, make that protein, put it in the body, and let the immune system destroy the cancer.

The DNA sequencing revolution made this personalized. Researchers found they could take samples of a cancer, run them through a DNA sequencer, and read the genes behind the proteins unique to that particular cancer — because every cancer mutates differently and no two cancers are identical.

How the personalization works

They take a sample of your tumor. Melanoma is one of the biggest targets for this immunotherapy because melanomas carry very distinctive DNA mutations caused by UV light hitting the skin, so the mutation is highly unique to the rest of your body. They extract DNA from a bit of the melanoma — multiplying it millions or billions of times — and run it through a sequencer, which gets chopped up and read. The sequence data reveals what makes this cancer unique: a small DNA sequence that causes the cancer to make a unique protein, essentially a fingerprint of that cancer. They then make protein from that DNA and put it in your body. Your immune system reacts to the antigen and starts attacking it, and once those immune cells are activated to that protein, they eventually destroy the cancer.

Why mRNA changes the delivery

Hundreds of trials of neoantigen-based therapy have been run across many cancer types, largely testing the delivery technique: do you manufacture the protein outside the body (ex vivo) and inject it, or — as Moderna did — deliver mRNA, so your own cells make the protein in vivo and release it into your blood, activating your immune cells? The mRNA approach means one shot, and your body starts producing the protein, just as the COVID vaccine worked: your cells make a particular protein segment that activates your immune system to kill the cancer.

Open questions include which adjuvants to use, which cancer types this works for, and combination therapies with other cancer drugs. The ongoing trials are all about the delivery mechanism, the technique, the cancer types, and the dosing. What's notable is that this is much more of a technique than a drug in the traditional sense.

It's a process, and the process is customized per person, so there's some cost to doing it per person.

That's right. But the cost is manageable, because I do DNA sequencing at my company all day long. Getting that sample is very cheap, getting the DNA data is very cheap, and a lot of the upfront work is very cheap. Theoretically, making the mRNA or the protein should also be pretty cheap. The real cost is human trials and human safety: there are a lot of questions around how to make sure everything is safe, what the right combinations are, so that people don't get overreacting immune systems or adverse side effects. That's all the tuning of this.

But what frustrates me the most is that everyone is lauding this as some unique, special, powerful breakthrough — and it's not. 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 body, 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 will do it, and it's not a special FDA-approved drug. To your point, it's a process: 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. Great idea. So why is Moderna saying they're going to charge $500,000 for this? That's what frustrates me. Everyone is happy about this, but I'm actually pretty sad, frustrated, and disappointed. I don't think this technique — largely developed through several decades of iteration, research, and development funded by the NIH and other public dollars, leading to this incredible concept of neoantigen immunotherapy for cancer that could theoretically eliminate many different types of cancer — should now be patented, FDA-approved, and charged at half a million dollars per treatment.

And what I don't like about this story is that a private pharmaceutical company has seen its market cap go from $20 billion to $60 billion because it went through this regulatory process — which in my mind is a form of regulatory capture. The case is made that it's all about safety and trials and patents, when fundamentally there's an underlying technique funded by a lot of public research dollars that I think should be more open-sourced and more ubiquitous. Every hospital should be trained on how to do this process to treat cancer patients; it should become much more of a ubiquitous thing.

I have two questions. Do you think they're going to be able to successfully defend this patent, in the United States and also globally? There's obviously a global world here, and people travel for treatments — medical tourism is a thing. So if it's half a million dollars here, can it be done for $50k in South Korea, Japan, or other places? And secondly, whatever happened with mRNA and the COVID vaccine, and the concerns people had? We never got an update on that, and this podcast started during all of that — that was the first rollout of mRNA at scale. How do you feel about the technology?

For years, Moderna was working towards this mRNA-based cancer treatment. They're calling it a vaccine, but it's really immunotherapy, because a vaccine means you prevent yourself from getting something, whereas here you're actually getting rid of something you already have. So I don't like the word vaccine. Moderna worked on this for years and never really got to market with it, and then COVID came about and they pivoted very quickly and got the COVID vaccine out.

Moderna holds a whole bunch of patents on the techniques and tools they use to make mRNA and to make sure it isn't integrated into the human genome, and it doesn't use a traditional replicating system — you can't actually replicate that mRNA, which means it won't start propagating in your body, since your cells might otherwise start making copies of RNA that ends up in them. So Moderna developed a lot of things that made their particular mRNA technology patented and safer. 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, look at its DNA, and find what makes it unique — that underlying concept is a different matter.

There's this little sequence, and now I take that sequence of DNA and stick it into a bacterial cell — E. coli — in a tank called a bioreactor. That bacterial cell will now start to make that protein, because I've put the DNA into the E. coli cell. The E. coli cell codes that DNA into RNA into proteins — DNA turns into RNA 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 inject it into your body, and your immune system gets activated. So you don't necessarily need mRNA is my point.

mRNA is a really effective way of doing this, though. 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 that end-to-end system's real cost is very low. This raises a big fundamental question we should all be asking: should we be giving the monopoly on cancer treatment to a company that went through the FDA process and filed a bunch of patents on certain techniques, when the general theory, the general thesis, the general therapeutic process is well described and well 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'll see it happening overseas very cheaply. It's very efficacious, and we actually have a friend in common who is 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. So I think we'll see more of this come to work. For certain types of cancer it is very efficacious, very safe, very effective — I'm excited by this.

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

So you're talking about the Galleri test by Grail. 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.

And remember, there's a bunch of new modalities — new modes of therapy — being used to destroy cancer. One is CAR T therapy, where you take T cells out of your body and use CRISPR or some other system to reprogram those T cells to go after the protein I described. You program the T cells to find a specific protein in your body and destroy it — 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 a million dollars. It works for people with blood cancers, where the cancer cells aren't clumped together in a tumor but float around individually in the blood. By putting T cells in your blood, it's much easier for them to find the cancer cells, grab them, and destroy them. For blood cancers like multiple myeloma, the efficacy of CAR T therapies now is incredible.

So that's one new modality, and this mRNA or neoantigen therapy is another. There are all these new techniques being used to destroy cancer, and in aggregate we should be very excited and optimistic about where this is headed.

Thanks for sharing, and let's hope for the best with implementing this. I love this right to try — that seems like something we should talk about in a future episode. We'll see you all next time on the podcast. Bye-bye.