CoreWeave暴跌、AI交易裂痕、Anthropic IPO警讯、Bill Gates末日论

All right, everybody. Welcome back to the number one podcast in the world, the All-In Podcast.

Actually, you, you manifested. We actually were number one. We were—

How good is that?

We were literally number one last week.

Pretty amazing.

Only took a little bit of time to get there, but yeah.

Only took 5 years.

Only took 5. Yeah, 5 years went by so quickly. With us today, David Sacks is back. How are you doing, brother?

Good.

You good? Yeah?

It’s good being number one.

Yeah. It’s good to be number one. I agree.

Yeah.

I agree.

I’ve done an excellent job.

You’ve done a very good job. And now we will—

All right.

Now we’ll see who takes the most credit for the ascension of the All-In Podcast. Over to you, Chamath Palihapitiya. How’s your week been?

It’s great being the best.

In the number one position. Yeah.

It’s great being the best.

It does feel good.

I feel like Sacks and I would’ve gotten here at least 18 months sooner had we not had to drag you on our back.

Oh, okay. The point guard who actually put all those dimes in position to fall everywhere you guys wanted it. Yes, you certainly would’ve gotten there eventually if you guys actually knew how to set up—

Where would we be without your constant interruptions?

Yeah. Absolutely.

Super sooner.

Absolutely.

We would’ve been number one sooner. You would be where every other tech podcast is, in the 400s. And joining us this week in replacement of the true shark—

Also number one. Also number one.

—the number one venture capital firm in assets under management, I believe, certainly, none other than Ben Horowitz is here. Hey, Ben, how are you?

I am doing swell. Thank you for having me on the show. It’s great to be with the number one team.

Yes. The number one podcast in the world. How does it feel to be on the number one podcast in the world? Have you soaked it in yet, Ben?

It feels great.

Yeah. It feels—

It is awesome. I feel bigger already.

It is kind of amazing that a tech podcast could be the number one podcast on—

That is bizarre.

—Apple Podcasts last weekend.

Big docket this week. Good to have you here, Ben. Friedberg is busy saving the world from famines, and he’ll be back next week for all of you.

He retired. He went out on top.

Oh, he retired at number one. Yeah.

Mm-hmm.

He just walked off the court. Oh, that’s hilarious.

Yeah.

Well, we’ve had an eventful week. We may have actually had the AI bubble, if there is one, pop or deflate a little bit. The neoclouds got crushed this week after CoreWeave’s bonds started to sell off. It could be the canary in the coal mine. We’ll find out. CoreWeave, IREN, and Nebius are all among the biggest neoclouds. They’re all down about 20% as publicly traded companies. Here’s your chart. Here’s what happened, just a little bit of background. On Tuesday, a bunch of people on X, including Chamath, pointed out that the yield on some of CoreWeave’s debt had hit 15%. Yields are obviously rising because bond traders are worried about CoreWeave’s ability to generate enough cash to support its debt, so they want a higher rate of return. Adding to Chamath’s tweet was a friend of the pod, Elon Musk, who decided to contribute 5 letters, one word, that has since been heard around the world and is at the start of every story. He said, “Yikes,” Y-I-K-E-S, and that generated a bunch of headlines. Then everything tanked yesterday, Wednesday. We tape on Thursdays, as many of you know. CoreWeave, IREN, and Nebius were down between 10% and 15%. Even Oracle and Dell took some strays. Nvidia took a stray as well; it was down 3%. CoreWeave is the tip of the spear. They’ve raised over $35 billion. We had the founder, Michael Intrator, on the All-In Podcast back in March, where he explained to me what he called “the box.” Let’s roll that tape and get Chamath’s thoughts. How does all that debt work? That is something you guys specialize in.

I’m going to try to explain this in a way people can understand. We go out and find a client. Let’s use Microsoft. You brought them up before.

Yeah.

Right? Microsoft comes to us and says, “We’d like to buy some compute from you.” And we say, “Okay, great. We’re going to sign a contract.” Once I have a contract in hand, I create something. It’s not a particularly creative name. It’s called the box.

Yeah.

Right? What I do with the box is take my contract with Microsoft and put it in the box. I go to Jensen and buy the GPUs. I put them in the box. I take my data center contract and put it in the box. Now the box governs cash flow, and it has a waterfall of cash flow that comes into it and goes out of it. The way it works is, I build the compute, then I deliver the compute to Microsoft, and they pay the box. They don’t pay me. It goes into the box, and the first thing it does is pay the data center. It pays the power bill, the interest and the principal, and then whatever’s left flows back to us. What’s important to understand is that the economics in this box are such that within 2 and a half years of a 5-year deal, we have paid for everything. They don’t cross-pollinate. They don’t cause contagion across the boxes. They’re all independent and discrete. As you do this, and as you show the lenders how this financing tool and mechanism works, they continue to lend you money at progressively lower rates.

Okay, Chamath. That seemed like a completely reasonable explanation of how they were running the business and how each of the boxes was isolated. But something has changed in the last 6 months. What is that, and what’s changed here? And what are your insights into his financial innovations with CoreWeave?

The most important thing you should always look at is how the sharps, the super-smart money, price risk. When you’re dealing with companies like CoreWeave or Nebius or IREN, et cetera, that have to borrow a lot to fund these boxes, the smart money is trading what’s called a credit default swap. This is default insurance, and these default swaps are meant to insure the bondholders against a potential default. That risk spiked this week to about 953 basis points. It doesn’t matter what that number is, but as an imputed risk, it’s about a 16% chance of a default in any given year. Over a 5-year duration, that math implies about a 55% chance of a default. So it’s a coin flip whether they default or not. That’s not what I’m saying; that’s what the insurance market was saying about their debt. I think that’s something to look into, because those guys are the smartest money in all of this. So why did they do this? I started to dig into it, and here are a couple of charts that explain a little bit of what I learned.

The first thing is just a market check. We’ve talked about this, but we have a problem in the United States: the U.S. is moving the risk-free rate higher and higher. Who is doing that? These are the central bankers; these are the bond vigilantes. These are the people who say that because of inflation or other long-term risks that aren’t being taken out of the system, we need to get paid a lot more money to own the safest, riskless asset. As that goes up, the basic way to think about it is that as the 10-year bond goes up, everything else goes up. This black line represents the average price you have to pay for junk debt. You can see that as the 10-year went up and people said, “Man, the risk-free rate’s going up,” the junkiest junk bonds need to pay more and more. The purple line is what CoreWeave has to pay, and you can see that it has also started to tick up and has moved up relatively violently. So the question is, why has it done that? Let’s go to the next chart.

Michael is right that he’s in the business of creating these boxes and putting in contracts and GPUs, but he also needs to put money in those boxes. The way he puts money in the box is by going out and borrowing it. He goes to a lender and says, “Give me the money, and I will pay you back, plus some interest.” What you see here is what’s inside the box, effectively. The gray bar is how much money he’s guaranteed to get into the box. These are all the deals he sold to all of his customers.

These are contracts.

These are contracts. It’s important to note, because I’ll get to this in a second, that these contracts are fixed-price. If what you could sell in the spot market goes up, you don’t really generate that revenue because you’ve already pre-sold it at a fixed number. Michael creates the box. He gets a long-term contract, which is great because you have certainty around the revenues you’re getting, but he sells that. Then he has to finance it, and some of that financing comes due.

He's got some bonds that are due and some loans that are due. What this basically says is that, at the midpoint, he has to raise another $35 billion of debt to get all the contracts he signed financed—all the boxes financed and up and running. So he has about $35 billion of debt, and he needs to raise another $35 billion. But, as you saw in the chart before, that’s ticking up.

That has nothing to do with him. It’s because the broader market is essentially saying that when the risk-free rate goes up, everybody has to pay more because your spread above the risk-free rate has to track with it. So today, across all the different pockets of money they’ve used to fund those boxes, their average interest rate is about 8.5%.

The problem is that, in this last little while, the sharps basically said, “Hold on. This is getting too close to the knife’s edge. You’re going to have to pay more interest.” That’s what I tweeted, which was just one of the sleeves of interest. The GPU loan DDTL 1.0 spiked to 15%. The problem is that could theoretically now be the rate they have to pay for all of the future. And remember, half of it is financed, and half of it is still unfinanced.

Right now, these guys are in a pretty good place because they have about $78–80 billion in assets and only $35 billion in debt, so they’re well covered. I think the idea that they’re at risk isn’t true on the surface. But the reality is that they have this problem: extreme customer concentration. Seventy-five percent of the revenue comes from 3 companies, and some of those companies are walking down their revenue expectations for the future.

So now you have to go out and borrow at a much higher rate. You have a bunch of debt to fund the boxes, and you have a fixed amount of revenue coming from them. What’s the trade, or what’s the setup? It’s basically a coin flip. I think the market is saying that if they’re able to finance this stuff at a reasonable rate and the life of the GPUs they buy from Jensen is longer than people think, they’re going to be okay.

But if the financing goes up into that 10%, 11%, 12%, 13%, 14%, 15% range, these guys are getting very, very close to not necessarily being able to cover all their obligations. My team’s estimate is that they’re plus or minus about $5 billion away from that break-even point. I think that’s why the CDS spiked, which then caused the bond yields to spike and the bonds themselves to sell off.

In order of operations, number one, I think they’ve created this very novel structure that everybody else has copied with the boxes. Number two, they’ve sold these fixed contracts. That’s super smart: guaranteed revenue into the future. But as the risk-free rate goes up, their cost of borrowing goes up. That erodes their margin, which brings them close to this break-even point. That causes the CDS to react, which then causes the bond yields to react. That’s the negative reflexive loop they’re in, so they just have to find a way to break out of it.

I don’t think they’re at real risk, but, as we all know, this is sort of a confidence game. You need to have confidence that the broader market can support another $35 billion just from these guys, and I think that’s what’s going to get figured out over the next month.

All right, Ben, so you look at this situation. You’ve been through a couple of bubbles before. Maybe you have a little PTSD or scar tissue, in fact.

Yeah.

We saw CoreWeave bounce today, kind of hit a bottom. The stock’s up 1% or 2% right now after going down 10%. Is this acute? And what does it mean for the broader AI trade, the AI bubble? Is there a bubble? We’ll get into Anthropic in a minute. Sacks.

I think the first thing to understand is that these businesses, like CoreWeave, as Chamath went through, are Rube Goldberg financial machines. They’re exceptionally complex. We talked about GPU contracts, but he’s got data center contracts. How long are those? Then he’s got to buy memory, and there’s a memory shortage. It’s really complex.

The other interesting thing is that GPUs have been going up in duration and price. That’s been a big tailwind for him as the owner of those things.

When you say duration, you mean the lifespan?

Yeah, the lifespan has been longer.

The number of years you get value from them. It was 3, 4, 5. Now it’s 6, 7, 8.

Yeah. And there’s a real shortage of GPUs with power and cooling, which is what he has.

Right.

This shortage is so bad. I was speaking with Andy Jassy, who runs Amazon, yesterday, and I said, “Yeah, we’re out of everything until 2028.” He said, “Oh, it’s beyond 2028, Ben. I don’t think we get stuff back until the end of 2028, maybe.” Everything else is sold out. So he has the advantage of having supply in a great shortage, which is extremely different from the internet bubble.

In the internet bubble, we didn’t have enough demand because we had to build the internet. People had to get broadband. People had to buy all this stuff. Originally, they had to get their browser at Fry’s, all this kind of thing. There weren’t enough consumers on the internet to absorb all the bandwidth that had been built out, and they had a massive amount of debt—not unlike what we’re building today, although not as big. But the lack of demand killed it. We don’t really have a lack of demand in AI, so I think the dynamics will be different.

I do think that if you get into a pathological debt situation, that can wreck anybody. By the way, we’re not invested in any of these neoclouds either, for that reason. That whole thing scares me, having been through it. I’d just say it’s pretty complicated, and it really depends on a huge number of assumptions in this very elaborate model that they have.

But can we—

—

Could—Ben, Ben—

—

Ben, but let’s just—look, you and Marc were one of the pioneers, maybe the pioneer, of the neocloud business. Loudcloud kind of popularized the idea that you could build these clouds and service customers this way.

Yeah.

But it’s very interesting to me that you and Marc, as the primogenitors, have passed on the whole space. I want to know why.

It’s because of this. It can end up being a very dangerous arbitrage business where you can get upside down easily. If you look at Loudcloud, here’s a simple thing that happened to us. In 1999, bandwidth was $1,500 a megabit, and you couldn’t get it. It was very hard to get, so you had to get a long-term contract—3 years. By 2001, it was $50 a megabit. If you were holding bandwidth, you were dead.

You’re dead.

And if you borrowed money to get it, you’re dead. Yeah.

You’re dead. You’re dead.

That kind of thing is so out of the control of the entrepreneur and so different from what we like to fund that it scared me to death. Now, I think the core dynamic of AI has been working for CoreWeave in that there’s a massive shortage, not an overage of supply. Token prices have gone down, but GPU prices have not.

Yeah.

Things are going up. And, by the way, data center space is going up in price because even labor to build data centers is hard to find. The politics are making it hard to build data centers. So the whole supply chain price is going up, and demand is not subsiding. Those are very big tailwinds for CoreWeave, despite the fact that the debt markets, kind of independent of them, are tough.

All right, Sacks. Here’s your Polymarket: CoreWeave announces bankruptcy. That’s a pretty aggressive Polymarket. It’s a 15% chance by the end of ’27, and 23% by the end of 2028. Seems a little far-fetched, given that they have massive demand for their equipment, as Ben pointed out.

Well, you can’t—look, this is dumb because I think Michael explained it: you have this opco-propco setup, so CoreWeave can’t go bankrupt. An asset-level vehicle for a data center or a buildout could theoretically go upside down, but that’ll go into receivership, and then they’ll just manage where you sell down the GPUs.

Yes. He explained in that clip that he has each box independent of the next, so not a great Polymarket, I think. Well, it is great if you bet the other way.

Yeah, exactly. Maybe some free money there to pick up.

Sacks, let’s level up the conversation, just outside of CoreWeave and the neoclouds. There’s massive demand. Token price is coming down, but token consumption is going way up. I think one of the silver linings here is that it was a big question how long these GPUs would last.

Maybe they had a 3-, 4-, or 5-year lifespan. Now people are saying, “Hey, these things will keep going into year 6, 7, 8 for inference alone.” So what’s your thought here on the AI trade, the bubble, as it were? Did we need to take a little bit out? Did we need to stress-test it? Is that what’s going on here, or is this making you nervous? “Hey, we’re in year 4 or 5 of this bull run,” and maybe it’s coming to an end and we may hit the R word. Yeah, recession.

Here’s the question, JCal: You’ve got this CoreWeave crash. Their bonds are being marked down to 80%. The stock is down 15% or 20% in the last few days. Is this just market volatility, or is this the beginning of something deeper and worse, like you said, the popping of a bubble? That’s the fundamental question.

I would make the argument that this is just volatility, which you’re going to get a lot of. Whenever you’re underwriting multi-trillion-dollar outcomes in the future based on very rapid, exponential rates of growth, the market is going to be super sensitive to any data points we have in the present. Everything is going to be exaggerated, and that’s why you get these big swings. That, to me, is the explanation for the volatility. I don’t think anything fundamental has changed.

So let’s talk about the fundamentals. The first question to unpack is: has demand changed? No. We know that demand for compute just keeps growing by leaps and bounds. In fact, everyone is saying that they are compute-constrained. The biggest model companies are saying they’re compute-constrained, and startups are saying it too. They’re asking their VCs, “Hey, can you help us get compute?” I think some VCs are even trying to buy compute as a strategy for competitive differentiation, so they can help their startups.

There was literally an accelerator that had bought excess compute—

Yeah.

—and they were using it as a way to get people to apply for their accelerator.

Yeah. So we know demand is off the charts right now, and we’re seeing other indicators. For example, old GPUs, even the H100s or the A100s, which are several generations old, are still renting at premium prices right now. They were supposed to be fully depreciated, and they’re not. There’s actually upside surprise there for the companies renting those old, supposedly stale GPUs. So we know demand is off the charts.

What’s the next indicator of something deeper and problematic happening? Is there a glut of supply coming onto the market? We know demand is huge, but maybe there’s too much supply being stood up. We know that’s not the case because of the sheer difficulty of standing up this powered compute, and we know it’s not being brought onto the market too rapidly. First, you have all the problems with atoms, building things in the physical world. That’s hard enough. Elon has basically said, “Look, this is a really difficult problem. Even our rocket scientists find it challenging to stand up gigawatts of new powered compute.”

Then we have all the regulatory headwinds. Local communities are not racing to give out new entitlements around this. If anything, we believe there’s not a glut; it’s the other way around. There’s enforced scarcity.

So then the question becomes: is there some sort of execution problem with CoreWeave? Are they too slow in standing up their compute? Have their costs gone through the roof? There’s some indication that maybe their costs are a little bit higher, or that maybe it’s a little slower. But at least reputation-wise, they’re believed to be excellent at standing up new compute.

Then the question is whether there’s some sort of credit problem. As the founder, Michael, was saying with the boxes, when you go to a lender to back the build-out of, let’s say, a new data center, you’re not relying on CoreWeave’s unsecured credit to make that loan. You’re backed by the credit of the tenant who’s entered into a 5-year commitment, so it could be Microsoft or whoever. The creditworthiness of that particular project, that particular box, is going to depend more on the credit of the tenant than on the credit of CoreWeave.

And behind that, you’ve got the GPUs themselves, which is basically asset-backed financing. The lender is secured by those GPUs, and we just talked about how, if anything, there’s upside surprise on the depreciation cycle of these GPUs. My impression is that there isn’t a deeper credit problem in terms of financing these projects. What happened, though, is that the repricing is of CoreWeave’s unsecured obligations. It’s true that the market has discounted them by 20%. They’re in distressed territory. That has brought the implied coupon of those bonds from 9.6% to around 15%, but that does not increase the cost CoreWeave is paying. CoreWeave is still paying its 9.6% coupon. It’s just that the secondary market has repriced the value of those bonds to an implied coupon.

Well, it changes the future debt.

Yes, it could definitely raise the cost of future debt, but I don’t think they’re funding these new boxes with unsecured corporate notes. I think they’re funding them with asset-backed financing, and the debt, again, is—

Contracts. There are 2 things that go in the box: the contract goes in the box, and the GPUs go in the box. These are—

Right.

—creditworthy organizations who pay their bills—

Right.

—like Microsoft.

Yes. Now, if I were a lender on those projects, though, I’d be looking really closely at the credit of the customer for that compute.

Yes.

Obviously, if it’s Microsoft or Google or a couple of other companies, I feel really good.

Or it’s a startup, right, that may only have—

If it’s a startup, I don’t necessarily feel so good about the loan.

Well, they have 3 customers that represent 75% of their revenue, and they’re all investment grade—the big customers. The thing I’ll add to this is, you explained quite clearly, Sacks: the GPUs are worth something, the data centers are worth something. Ben, you pointed out the power and the HVAC are worth something. The tokens are worth something. And if you look at startups, which is where I spend all my time, the early-stage startups are crushing it in terms of raising capital and getting customers. We’ve never seen revenue ramps like I’m seeing in startups today, and I think, Ben, you’ll probably back that up. The revenue ramp of startups today is distinctly different from the SaaS “triple, triple, triple, double, double, double.” Now it’s, “Hey, are you doubling every quarter or not?” And if you’re not doubling every quarter, what are you doing? Why aren’t you growing faster in this age of AI? The startups feel particularly strong.

So maybe this is just the market overall saying, “Hey, let’s test this,” and maybe there’s some indigestion there. Or is it that the world is running out of money and people have placed their bets already? What would that say for the venture industrial complex, of which you are a significant player at scale? And what would that say for the IPOs, the 2 biggest in the history of business and public markets, Anthropic and OpenAI? Are we running out of money to allocate to AI?

I don’t think so. We’ve kind of never had companies that have grown this fast and gotten this big, so we’ve never asked that question before. We never said—or at least not in my lifetime—“Does the world have enough money to finance these things?” It’s a crazy question, but it’s relevant.

I don’t think so, for sure, with respect to the Anthropic and OpenAI IPOs. I think my belief is those are going to be fine. Will the price be $2 trillion or $1.75 trillion? I don’t know. But I think there’s definitely—

Well, raising $50 billion or $100 billion, whatever they raise—

Oh, for sure. That’s not a giant number—

That’s a big number, but it’s not an unprecedented number now, because—

Look, what—

SpaceX just did it, right? Whatever you’re holding in the equity markets, whatever you’re holding, OpenAI and Anthropic are threats to that. Almost, like, there’s almost nothing—

Yeah.

—that you can hold in a portfolio that those 2 companies don’t threaten.

Yeah, agreed. And so I’m not worried at all about the demand for those IPOs, to be honest.

So you have to own those because they threaten every single company in the world. They’re selling intelligence broadly, and every company is based, Chamath, on intelligence. Therefore, they’re that disruptive. Yeah, Chamath?

You can underwrite it as a defensive hedge, because I think Ben’s right. If you own a healthcare company and you’ve thrived on the opacity of the healthcare infrastructure, OpenAI is a threat to that. If you own a bunch of workflows that have been rented to people as SaaS, theoretically, OpenAI and Anthropic are a threat to that.

At a minimum, it’s defensively positioning your portfolio, so you kind of have to, because there’s not much other investable stuff in AI to be long AI other than those 2 companies. They’ll get the lion’s share of the interest. I think the question generally is: if the bond market walks the 10-year up to 5.5% and 6%, which has nothing to do with AI.

It has everything to do with debt-to-GDP and a bunch of other factors in broad-based Western economies. Any business model that thrives on a financing spread gets cornholeed. Yeah, any and every one of them. It will not matter when the 10-year is at 6%. It won't matter what you do or what you claim to do. You're going to get run over. You're going to be roadkill.

As long as we can keep that from happening, we're going to be fine. So, what's the most important way, I guess, post these midterms, to keep the party going? You have to really start to focus on deficits, and you have to find a fair trade where the bond market backs off. I think you can see it in the tape: if we start proposing huge spending programs, they're going to walk the 10-year up. That then has this negative cascade to everything. If, however, the posture becomes more about keeping everything calm—let's tone things down and find the waste in the system and squeeze it all out—I think people will respond very favorably. So that's really where it's going to come down. The last thing I will say is—

These are Achilles' heels. I think Sacks made an incredibly eloquent case, and so did Ben, about all of the asset value that's getting built up. But just like the great warriors, you slice the Achilles tendon, and they crumble to their knees. I just want to point out: this is that thing. There is no other thing.

All this other stuff is complete goodness, but the fact that you need to finance these things in a variable-rate model as a spread to the risk-free rate is the big problem in these businesses. You need Microsoft, Google, Meta, and Nvidia to backstop all of this in a worst-case scenario, because that's the only way you can withstand a 1- or 2-year correction in this market.

Hmm.

I'll give you the ultimate moat. I think power will be the moat. You can build the most fantastical things in the world, and you're going to see people on the dry docks basically trying to hitchhike to get power.

I'll tell you the other big piece here that might be a bigger threat, Sacks: the deflationary nature of this AI—the superintelligence movement, rather. We talked last week about people decompiling Photoshop, driving its value to zero. Now, this week, Hood Warfare came out. People are making platform-level games for $5. It's so deflationary. There's an ElevenLabs competitor that came out that's incredibly cheap, and then every single open-source model is being hosted. Elon says, “Hey, we'll just route you through our orchestration layer to the best model, open-source or not.” How big of a headwind do you think that is, Sacks, to our industry? Just the commodification, and how rapidly 90% of the value is being removed from things. Is that going to stop, or is it going to accelerate? It feels to me like the deflationary nature of this software, AI, superintelligence is unique and different from anything we've ever seen. People are just able to rip and ship code so fast. It makes me wonder: what is the margin of software in the future?

Well, we want everyone to be able to create software, especially bespoke software, which is really this new category, right? That is going to be a huge driver of productivity throughout the economy. It's not all just zero-sum games or disruption. The value of these coding assistants is that now you can create software for highly bespoke use cases, whereas software used to be created really just for the mass market. The reason Photoshop is a successful product is that there's a lot of graphic designers who need to use it, and really, the only software that could ever get created in the past was when there was a large potential customer base for it.

Now we're getting all these cases where enterprises can create software just for their particular use cases. In fact, in a way, that's what agents are: bespoke software programs to solve a particular task.

But if people build those in-house and they're based on open source, you don't need the other piece of software that it's replacing. Isn't that extremely deflationary?

Maybe there's some substitution away from established—

Yes.

—software, but I think the important point is that there's this vast new category of software being created. The amount of code being created now is already orders of magnitude bigger than it was before coding assistants. People tend to think in terms of these zero-sum games and who's going to be wiped out or eliminated, but the really important trend is that there's going to be a lot more software created in the world for a lot more of these bespoke use cases. Enterprises are now able to create software for a much broader array of purposes. There are companies that never would have been software companies before that are now creating their own software, and they're hiring coders basically to manage the process.

By the way, I think this is a really important point around the jobs apocalypse. The first job category predictors of the apocalypse were saying would be eliminated was software coders, obviously.

Yeah.

Because why do you need them anymore?

Then everybody became a developer.

Yeah.

Now everybody's shipping code.

Yeah.

Yeah.

Now we're seeing that demand for coders is increasing, even though 95% of the code is written by coding assistants. The reason for that is, first of all, you still have that 5%, and that 5% is part of a much bigger pie. You still need someone who can look under the hood and manage the whole process, and they need to be—

—and maintain it.

—somewhat technical.

Right? They have to maintain it.

And maintain it.

Yeah.

So, the maintenance.

That's a big piece. Yeah.

In any event, I think this new category of bespoke software, created by the proliferation of coding assistants and agents, is going to be a huge driver of productivity throughout the economy. Is that broadly deflationary? Probably, in a good way, because it means we can do more with less, and we need to bring prices down.

Certainly great for the country, right? Everybody is going to be more productive. But do you worry about our industry, the venture industry? We have these great hits, and then the second they become a hit, somebody can decompile one, create a competitor, and ship it very quickly. It feels distinctly different in this boom compared to previous ones. Do you think this is going to be a challenge for us to get DPI for our LPs? Is this going to be a challenge for the sustainability of great software franchises? Ben? I can, yeah. Look, I don't think so. This has been the ongoing prediction, starting with the SaaSpocalypse.

Yes.

But I think David's right: software prices will definitely go down. But the amount of software is going way, way up. It's the Jevons paradox of software.

Sure.

These companies end up doing other things beyond just the software. We had this on the internet: “Information's going to be free. That's it. All these industries are—”

…dead, but they kind of reemerged in a different way. From a startup's point of view, you've got to evolve. ElevenLabs does many things to make its product better over the years. Since that's the example you brought up, they've got licensing agreements, like Burt Reynolds will read you a book, and this and that and the third. They're doing more than just software because they can, because the software is easier to write. The model they're training on very nuanced cases is why movie dubbing is so amazing now, where it sounds like they're just acting in a different language. There's more to it than just speech-to-text or this and that.

I think the products are going to have to get way better, and the startups' products will have to get way better, which is amazing for everybody. That's where the value will be. Just writing software is not going to have any more value than publishing a magazine.

The moats will change.

So it's creating massive competition. Yeah, yeah, yeah.

The moats will change. You have to think about what the moats are in a business. Data moats are a big one. If you have data that other people don't have, and you then build a user base because your model is catering to a particular vertical, now you're getting all those users and their feedback. Maybe that becomes a—

The feedback loop.

—a feedback loop, yeah.

The reinforcement learning becomes a moat.

A lot of startups are building businesses based on that model. The paradigmatic example is something like Harvey, where they get all these law firms on board, so they then get access to their data and usage, and then they train their own models.

Yes, it gets sharper each time. Go ahead, Truman.

And cheaper and more custom.

Yeah.

I’ll give you the ultimate moat. I think power will be the moat. I think you can build the most fantastical things in the world, and you’re going to see people on the dry docks basically trying to hitchhike to get power.

And the reason is that we’re so constrained in the United States. A, we’re already behind, and we’re in a massive energy deficit, an electron deficit. By 2050, we’re about 1.4 terawatt-hours short, which means about six Californias’ worth of electricity. That’s the starting point.

And that’s with everything we’ve planned to build being built on time and on budget, which we know is not going to happen. Then on top of that, you have all the load growth that theoretically happens when you actually have all these next abundant experiences. I made this statement this past week, and I’m just going to double down on it. I don’t think you can work through the hyperscalers anymore because of this one issue.

Let’s say you’re a big, mainline G-SIB bank, right? JPMorgan, Citibank. Could you imagine the reason you can’t service your customers is that the underlying cloud you use can’t actually secure and grow enough power? So, back to the other moat I think is really valuable: owning the land, the power, and the shell, because it’s such a constrained asset. Back to what Andy told you, Ben: it is so critical, and it will stop the best experiences on a dime.

If you’re a startup, I actually think the bigger problem is you’ll go to these clouds and be like, “Great, I’m ready to launch this whiz-bang product,” and they’re like, “Suck it. We have no capacity for you.”

Yeah, such a disaster.

I very much agree with that last point.

Yeah.

Yeah, I think that’s right.

And, by the way, what an incredible opportunity that’s now been lost in the Middle East, where Saudi Arabia, the UAE, and Qatar were all massive oil producers, able to do top-down planning—

But without proper drone defense—

—and have data centers—

Without drone defense.

—and now they have bombs going off, blowing up airports.

Yeah.

It’s such a headwind.

Without drone defense, you can’t go there. Without drone defense, you can’t go there.

Exactly, and drone defense isn’t going to work because one drone gets through, and your data center’s fried.

Well, I mean, it does look—

Can I just ask?

Okay.

Can I make one last point on the whole data center thing?

Please.

There’s one metric—the more I think about it, the one metric I would really keep an eye on is payback period. To me, that’s the North Star metric here. CoreWeave says its payback period is 2.5 years. Their typical deal is five years, and they pay back within 2.5 years.

SpaceX pays back in roughly one year. That makes sense because I think SpaceX is monetizing at about $50 billion per gigawatt and CoreWeave at about $20 billion per gigawatt. The reason for that is xAI is basically getting spot. They’re doing these deals that can be canceled by either side within 90 days, so they’re basically getting spot minus 90. CoreWeave, on the other hand, is signing up for five-year commitments, so it’s contracted way in advance, and they’re getting a contracted price. What’s happened over the past year, I think, is that spot has moved way up.

Way up.

Because, again, of the scarcity, right? And the demand is so off the charts. In any event, payback period is really critical, and as long as the payback periods are really short, I don’t see how these companies can lose. You’d have to be really incompetent to screw it up, right? You’re making these big capital investments, but if you’re getting your money back in one year or 2.5 years, that greatly reduces the risks, right?

It’s like owning an Airbnb in Paris before the riots or something. The payback period is so quick for these if you got your orders in early.

Well, it’s a very rapid—

—and that means you’re paying back the debt quickly, so the debt’s rolling over more quickly. It just reduces everyone’s risk.

Yeah. Now, if that payback period starts taking longer and longer, then I get nervous, right? Because that magnifies the risk. It also means the economics aren’t as good. The question is, why would the payback period take longer? It could happen for a bunch of reasons, like if the demand goes down and the offtake isn’t as good, so the spot price goes down.

Yeah. Demand’s not going down.

Or there are just execution problems, right? It takes you longer to get the capacity online, so the payback gets delayed. There are a lot of reasons why payback could take longer and longer, but as long as payback remains as short as it is today, to me, that would be the one metric I would really keep an eye on.

The other thing the big offtakers will make you sign up for is what’s called liquidated damages. There’s really complicated language and extreme penalties about what happens when you can’t service the data center. So I think that’s another risk. A couple of these guys have lost their investment-grade wrapper from the big boys because they just, Sacks, to your point, couldn’t manage the data centers. The complexity is going up because the chips run hotter, there’s more energy, there’s more electricity, there’s more cooling. It’s—

All I’m saying is, I think the sharps may be wrong, and they usually are wrong when they’re buying these credit default swaps. But when your expected value is a coin flip in a five-year window, that’s scary territory, I think. That needs to be managed very precisely. I would find every way possible to nip this in the bud if I were CoreWeave. And then, more specifically—

Well, how do you think they can do that? How can they nip it in the bud?

I think they have to cancel some of these contracts and move to spot. That’s the simplest thing they could do. If they could basically pay a penalty and get out, they should run the Excel right now on, “Let me just take this gigawatt and put it on spot.” Because you’re talking about, as you said, a $20 billion spread.

But this is where Elon has a big advantage, because he has independent sources of financing to finance that gigawatt of compute, whereas CoreWeave needs that five-year contract to bank the financing to build that capacity.

Well, I think Nvidia is probably okay with CoreWeave selling spot and not doing a multiyear offtake for a big hyperscaler or a foundational lab. I would just run the math and try to figure it out, because they’ve only built half the capacity they’ve contracted for. Let’s say they built two out of the four gigawatts. They’ve got to figure out a way very quickly to just—

To your point, the spread between their fixed contracts and spot is so extreme that you’re leaving $20 billion or $30 billion on the table. Now you’re within plus or minus $5 billion of violating a covenant on your debt. Don’t be there.

Yeah, it would improve their economics a lot. If they had a gigawatt of compute they could sell at spot, you’re right, that’d be—

Game over.

—net $30 billion—

Yeah.

—right to the bottom line.

Right to the bottom line.

I think the challenge with that is the hardest people to get spot pricing from are the investment-grade customers. Those are the ones that—

Also true. Yeah.

—won’t do it, right?

Also true.

Yeah.

Also true.

Because they also have war chests of money. They’ll stand it up themselves if it’s too expensive.

Yeah, this narrative could change a lot, right? If spot becomes imperiled, then—

But, yeah, the guarantee contracts become great, yeah.

Yes.

I think we’re capacity constrained enough through the entire pipeline. You have to think: Nvidia’s done an incredible job, okay? If you look one layer beyond Nvidia, all the packaging capacity is theirs, and most of the HBM capacity is theirs. Even Google is tin-cupping for memory right now because Nvidia’s clogged it all up. They’re in complete control here, and it just creates—

It’s a crazy moment in time, and we got a double bang this week on top of—

Sorry, Jason, just one last thing. I’m telling you guys, you’ve got to get long energy. I think startups need to find a way of cobbling together. If you don’t have access to electricity, you cannot go through the hyperscalers. That’s my prognostication.

Yeah, stop the war in the Middle East. It solves a lot of problems right off the bat. If CoreWeave wasn’t enough, the AI sector got double-banged on Wednesday after Anthropic’s Investor Day numbers got a mixed reaction. Anthropic is planning on going public around Thanksgiving. Polymarket says there’s now a 74% chance they go public this year, down from 95% in early September, when they said everybody was going to die from their software.

On Wednesday, the company presented to potential IPO investors at its headquarters in San Francisco. I mean, you can’t make this stuff up. This isn’t the bugged headquarters of OpenAI.

They swept the room for this, but we still have the data. Q3 revenue: $16.8 billion. Think about that, gentlemen: over $5 billion a month already for Anthropic. Net revenue retention was 900%. That means customers spent 9 times more than last year. Pretty good, huh, Sacks, on the SaaS front?

They also said they’re free-cash-flow positive, with an adjusted gross margin of 85% and an adjusted operating margin of 11%. No word on how the adjustments were calculated, but here’s the big one: ARR estimates range from $70 billion, which would basically be flat since July—a bad sign—to over $100 billion. So there’s some hand-wringing about which numbers are correct. They didn’t disclose those numbers; these are just secondhand estimates from investors who might be a little leaky. Obviously, Anthropic is targeting a ginormous, unprecedented valuation of $2 trillion. And that is—

Ben, is Andreessen—

—based on them growing violently, not growing slowly. Go ahead, Chamath. You had a question.

No, I just said, Ben, is Andreessen long one of the foundation model companies, like one of the two big ones?

Well, we have a lot more stock in OpenAI, but—

OpenAI. Got it.

Yeah, philosophically, we’re long both of them, I think.

You’re long both of them, which is kind of a strange thing now, Ben. You’ve gotten so big, you consistently have bets on the number one and number two player. How is that working out in the market?

Well, I would say that there are many investors in both of those things. In venture, we call it venture capital, but venture capital has subsumed a huge part of what used to be the public markets, and I would consider both OpenAI and Anthropic to be in that category. They’re not really venture capital where you’re worried about your venture capitalist being invested in your competitor, because they’re gigantic. They’re not A-round, B-round companies anymore.

That part of the market, I think, people understand: when you’re raising tens of billions of dollars, you don’t care where the money comes from, as long as it’s green.

You need everybody.

Yeah.

What used to be mezzanine financing, pre-IPO financing—there’s less hand-wringing about the conflicts there?

Yeah. I wouldn’t even call it mezzanine financing at that level. We did a lot of regulatory moves in the late ’90s and early 2000s, which really made the public markets much less attractive for companies. The private markets grew to absorb that business, which has been, I think, very bad for America. It’s increased the wealth gap and so forth, but it is what it is. I wouldn’t consider that the same. That’s very different from investing in two A-round companies doing the same thing.

Yeah. The Uber-Lyft thing—

I still think that’s a—

—I think that’s distinctly different from Anthropic and OpenAI at $2 trillion.

Now, yeah.

Understood.

Yeah, exactly.

One is a reaction to the other, in fact. Right?

Yeah.

Anthropic was a reaction to OpenAI, let’s call it what it is: rug-pulling Elon on it being a nonprofit and becoming a commercial enterprise. We’ll leave that aside for now. What are your thoughts on this Anthropic IPO, Ben?

Well, look, I think you hit the thing that’s causing the consternation: growth has slowed a lot from Q2 to Q3 in terms of revenue growth, and then we don’t know what the ARR, or recurring revenue, number is. So people have questions about how fast the thing is growing. Having said that, $16.8 billion in the quarter?

Crazy.

Crazy.

How long has Anthropic been selling stuff? Three years?

Crazy. Yeah.

It’s just so absurd how good this thing is. So I think the IPO will be fine, whether you want to be in at $2 trillion or $1.9 trillion or $2.3 trillion. I think that’s what a lot of the debate is about. Anytime there’s negative news on the big labs, everybody in the supply chain gets hit with that, and that’s what’s going on now. But, big picture, it’s a very important company, and it’s got a very big IPO.

A 30-times price-to-sales ratio, Sacks, plus this headwind of open source and enterprises moving to that—most businesses don’t need to solve the last 7 math problems in the world. That could be slowing growth. We could have Anthropic’s revenue slowing. Let’s say it is. What do you think happens to this IPO and the broader AI trade? Now we have 2 examples of AI trade headwinds, superintelligence headwinds—whatever’s happening at CoreWeave and whatever’s happening at Anthropic.

Well, look, Anthropic had this investor day the other day, and I think some of these numbers coming out are leaking from that. Maybe. We don’t know what’s true or not. I think the participants had to sign NDAs, so I don’t know if you can completely trust what’s floating around X here. I think we’ll find out in due course. If this company is IPO-ing next month, then we’ll find out the actual numbers.

I would say the body language coming out of the investor day was pretty strong, at least from what I’ve heard. I think the investors liked what they heard. These are big public-market investors. Who knows? Everyone’s kind of talking their book, so it’s hard to know exactly what’s true. But I generally think Anthropic is a strong business. I understand there are some risks to it. It’s not completely risk-free, but it’s a pretty young business, so you’d expect that.

Overall, I think it’s set up to have a strong IPO. Wearing my hat as an investor, I think it’s a very impressive company, which is all the more reason I find their attempts to kneecap open source through their incessant fearmongering campaigns and regulatory capture so distasteful. They don’t need to be doing that. They’re a very successful company. They’ve got a very bright future. I just find it kind of offensive that they keep trying to do that. There were 2 examples of it just in the last week. Do you guys want to talk about these?

Oh, yeah. Let’s—

Are we doing PSYOP corner? Is PSYOP corner starting?

Let’s circle back to PSYOPs, because I want to get you, Chamath—

Yeah.

—to comment on, hey, you thought maybe it was going to be $1 trillion.

All right. Let’s come back to that. There are a couple of active PSYOPs we have to talk about.

I have it in the notes here. We will get back to PSYOP corner here with your toaster oven.

I’m excited to get to PSYOP corner.

Wait, you will. We got you a tinfoil hat. It’s just there on the side of your AI slop background.

Yeah.

So, Chamath, you said $1 trillion for Anthropic a couple of weeks ago. What are you thinking here? Do you think they have significant headwinds, or do you think this is going to be a blockbuster IPO? Where are you at here in October 2026?

No, it’s going to be a—

—in October 2026.

No, my comments are not because I’m negative on the company. I found the revenue breathtaking. $16.8 billion in a quarter for a 3-year-old company is just ridiculous. You’ve never seen anything like it. I don’t know how to process it. I think their products are exceptionally good. To OpenAI’s credit, they’re doing the hard work as well, walking down the price curve so they’re less and less at risk from open source.

I still think there’s a bunch of cannibalization that has to happen on the token side. I also think the big thing that has to happen in AI is that we’re effectively in 1999 and 2000 in the internet media world, where everybody just works on CPM. Back then it was cost per thousand impressions. Right now it’s cost per million input tokens.

We will move to CPC and CPA, right?

Yes.

When Google invented and scaled AdWords and AdSense, the market exploded, right? We went from buying 1,000 ads to just getting an outcome.

A click.

Right now, we buy a million tokens. We will eventually move to an outcome, and I think when that happens, this market will go absolutely crazy. These guys are going to do great. They’re a great company.

Here’s my very specific point about why I think they should price it to clear the market for $200 billion. I’m very specific about that number. They need, in my estimate, at least $200 billion. If you want to be super safe, a quarter trillion gets the job done. Again, what does the price need to be to attract that incremental dollar off the sidelines? Can they raise $80 billion in an IPO at $1.7 trillion, $1.8 trillion, or $1.9 trillion? I think so, handily.

Can they raise $100 billion? Maybe that’s pushing it. At $150 billion, I think you need to be back down to $1.5 trillion. $200—

Ah, so you’re saying you want them to raise more money at a lower valuation so they have a war chest. $150 billion at $1.5 trillion—

It’s not enough.

—is not enough. Okay.

It’s not enough. $200 billion barely gets them a few gigawatts, and that is the bottom-line math. If Anthropic is going to go through and survive an eventual correction, I think it’s much better for them to have $200 billion to $250 billion today than to try to ring the bell at $1.9 trillion or $2 trillion for $80 billion, file a mixed shelf, which takes about 6 months, and hope the markets are in line. I just think there’s more risk in the future than there is today, so why take the risk?

Okay.

Price it to clear the market—

So wait, how much are you saying they should raise, Chamath?

He said $200 billion.

$200 billion.

$200 billion.

Minimum. Minimum.

Minimum.

But I would like a quarter trillion if I were an Anthropic shareholder, which means I don’t think you can raise a quarter trillion at $2 trillion. I think you can raise a quarter trillion at a trillion.

What do you think, Ben? You go with Chamath, raise more cash now and have the rainy-day fund, have the resiliency, or go for the valuation optimization, as founders and many investors are wont to do these days?

As a former CEO who almost died 17 times by running out of cash or potentially running out of cash, I agree with the sentiment. The thing most likely to stop the company is financing risk, so why take it? You’ve got all the other risks: supply chain risk, competitive risk, Chinese open-source model risk. You’ve got all this stuff. Why layer on financing risk?

And the founders don’t care. They all own 2.5% each, so it’s not like they should be optimizing. They seem not to care about money, which I think is a very positive attribute about the entire founder group at Anthropic.

Yeah.

In part, that’s why they’re so successful: they don’t care about the money.

Get the tonnage of dollars inside the company.

Having said that, what the safe amount of money is, it’s always a little bit in the eye of the beholder. In the dire scenario, what exactly happens? I’m not deep enough on it to say whether it should be $250 billion or $150 billion.

I think the number has to work backward from compute in order to get through a 3- to 5-year window. You have to get out of Spot because, as Sacks just said, it’s too expensive for Anthropic. They shouldn’t be spending $50 billion when they could be spending $20 billion. But if you had $200 billion, you’d have a path to 5 to 10 gigawatts that you could finance completely on your own with the magic boxes from CoreWeave and the like. I would optimize for that: optimize for surety of compute, and everything else will solve itself. In a 5- and 10-year window, they’ll achieve the upper bound of their capability. At $50 billion, $60 billion, or $70 billion, they may not survive the pricing implications of what happens in the spot market over the next 2 or 3 years.

And Elon’s deal—

And then they’d have to raise a down round.

Yeah, and Elon’s deal with them is, I think, either party can cancel within 60 or 90 days.

90 days.

Somebody can look it up. Is it 90?

The fact that Elon did that deal tells you Spot’s going up, not down, and that he might want the compute—

Well, Spot has been going up.

Yeah, and he may want the compute for himself.

This is my point.

Yeah.

He’s not dumb. He’s the smartest of the smart. He’s doing Spot because he’s like, “I may cancel on you.” And Anthropic’s probably thinking, “Oh, this is great. I can walk away in 90 days.” Wrong, buddy. It’s the opposite.

He wanted the CoreWeave deal. They should’ve got the CoreWeave deal. Ben, what—

Well, I’m sure they did the CoreWeave deal while they could, but—

Yeah.

They need to put a lot of money in a magic box for the CoreWeave deal.

Yeah, they’ve got deals with everybody. Absolutely.

I think they do. Yeah, I think they do.

According to SemiAnalysis, Anthropic and OpenAI are taking 60% of all the compute that’s expected to be stood up in the entire world over the next year. They’re basically the two—

There’s your duopoly.

—off-takers for that. That is a duopoly.

Yeah.

Based on compute, it’s a compute duopoly. It’s not a token duopoly at this point.

You measure duopoly or monopoly status based on market share, so you want to look at revenue. If you look at the revenue of foundation model companies, I think Anthropic and OpenAI are pretty much the only two generating significant revenue from their models.

Mm-hmm.

They definitely—

Everything else is a rounding error.

Yeah.

They definitely have a duopoly when it comes to revenue and market share. I would argue they have a duopoly when it comes to frontier capability, although Meta’s new Muse model is getting very close. The new Google model is getting close. Elon’s getting close. But I think Anthropic and OpenAI are still in the lead in terms of capability, and we’re not seeing what they have in the lab, so to speak. They reportedly have things that are 1 or 2 generations ahead, very far ahead—or at least that’s what they’re touting to potential investors.

Hey, Ben, we’ve got you here. You’re a big OpenAI investor. You’re not on the board, I don’t think, right?

No.

You’re not on the board. What’s the story with Sam Altman’s equity grant? He still has no equity in OpenAI. What’s the advice there? Are you going to give him a grant? Are you going to vote for a grant as a shareholder? What should he get?

I don’t want to pretend I know things that I don’t, so—

Okay, so you don’t know about it.

So I don’t know what the strategy is there.

You can’t be on the pod then.

It’s okay. I’ll repeat what I said years ago: when the company went from being a nonprofit to a for-profit, they should’ve done a make-right. You go to Elon, and you basically give him a share because he put in the first $50 million or whatever, and you give Sam a grant. You normalize things. Normalize the management and governance of the company. Instead, they did this weird hybrid NGO thing that created nothing but problems.

Yeah, but he got healthcare, so he’s good. Sam’s got healthcare, so nothing to worry about there.

You’ve got to think he has to get between 2% and 4% at this point.

Yeah, I think 4% is the right number for a CEO.

I think it’s very—

Yeah.

I think it’s fair relative to the comps. On a total dollar-tonnage basis, it’s incredible. It’s $20 billion to $40 billion at a trillion, and more from there. It seems right-ish.

Yeah. Everybody approves. So, voted here on The All-In Podcast. What should he have done in terms of screwing Elon in the settlement there? Ben, since we have you here, what do you think he should’ve done? You’re a big fan and a big investor in all of Elon’s companies. What was your advice to Sam about settling it?

I wasn’t in that conversation then, but I agree with David. I think it’s always the right thing to square it with the people, not just the entity.

Totally.

All right. Two fastballs, and you hit both of them out of the park. Very well done, Ben. You’re fitting right in here. You’re candid. That’s what makes a great Bestie: deep industry knowledge and the ability to be candid. Let’s get candid now with some other important topics. Anthropic says small models—you may have heard of vertical models, small models, small language models, or SLMs—could start self-replicating on your smart home devices.

Yes, that’s right. On Monday, Anthropic published a blog warning that rogue models could start copying themselves, quote, “in just a few months.” It’s only months away, Sacks. They claimed a worm—a software package that can propagate itself—would carry model weights to an IoT device, an Internet of Things device, to set up inference and then move on.

This means your smart TV, your doorbell cameras, your Ring, your Dropcams, your robot vacuums, your Roomba could start having the next set of swarms. These would apparently be physical swarms on IoT devices. With no central off switch, Anthropic says it would be “challenging to shut down.” They ran some tests to see if this would work, I guess.

A small model got inference running on 8 of 9 hardware types tested. A large model ran across 15 smart TVs at 0.4 tokens per second, which is absurdly slow. But basically, what they showed is that a small model can install itself on IoT hardware, which is everywhere and easy to hack, as you probably read.

Anthropic says there’s no full rogue replicator yet, just “signs of life.” They had to say “signs of life,” Sacks, to troll you. So here you are, Sacks. You wake up, you get yourself out of the czar position, and they’re still dragging you into this meshugana.

Yeah, look, it’s—you ever see the graphic with the cat in the tinfoil hat? I wake up and there’s a new Anthropic psyop. That’s what’s going on here. It’s plain as the nose on your face: they’re trying to create a panic. This is their latest panic against open source.

Mm.

Now they’re particularly going after small models. By the way, small models are the safest models, because a small model doesn’t have the capability to become a new species or all the other X-risks they’re always talking about. These are basically small, purpose-built models to work on a particular task, like a customer support workload or something like that, and they’re trying to make it sound like these small models are dangerous. They can take over your toaster or your refrigerator or something like that.

It’s almost comical. We’ve gone from “The Terminator is coming” to “Your toaster’s going to kill you.” But it’s just more of this fearmongering from them. And the reason I say it’s a recurring pattern is that there was actually a second Anthropic blog post this week as well, about distillation. Basically, the other blog post Anthropic released said that they had eliminated 90% to 99% of distillation. I don’t know if you guys remember this, but way back—

Yes.

—in July—

Yes.

—Anthropic was pushing this idea that they were the victims of industrial-scale distillation attacks, which is basically when Chinese models or other competitor models pay to use their service and then train on that output. They were basically saying the government needs to get involved and stop this, and that these open models were the result of IP theft, as they claimed. Even though, frankly, they’ve distilled the entire internet.

They could have stopped it. Yes, and they could have stopped it themselves. Yeah.

And, by the way, this was such a big threat, or perceived as such a big threat, back in July. We did a whole episode on it, but 50 major companies, including Nvidia, Dell, Microsoft, many others, and actually a16z, signed an open letter to the government saying, “Do not overly restrict open-weight models. They’re tremendously beneficial for our economy.” Two hundred startups signed another letter, because there really was this big push in Washington to crack down on open models on the basis of this supposed IP theft.

My argument at the time was that it should be up to the model companies themselves to stop distillation, because once the distillation happens, the horse is out of the barn. The government can’t put the horse back in. It’s up to the companies themselves to know their customers better. Chamath, you made the point: they should just do KYC. Know who your customer is and stop them.

Now they’ve come out with a blog saying that’s exactly what they did, which is exactly what I predicted. But that is not what they were advocating for. Several months ago, they came crying to the government: “Hey, you’ve got to solve this problem for us by taking out our competitors.” That would have had the exact opposite effect, because it would have done nothing to stop distillation. The distillation would have kept going, right? Only the companies themselves could mount the technical defenses to stop it, and it would have just made the market less competitive.

In any event, we basically called them out, and fortunately, the administration rejected the idea of regulating open source in that way at that time. As a result, Anthropic had to actually build the technical capabilities to detect when distillation was happening, and they’ve stopped it. They’ve basically proven the exact point we were making 3 or 4 months ago.

Yeah.

Which is that they were in the best position to stop it, and they could stop it, and they should stop it, and they should not push this onto the government, okay? But again, this is a recurring pattern with them: they point to some technical issue or challenge, then go running to the government, create a scare campaign, and try to get the government to stifle competition.

Instead of just fixing it themselves. Yes.

Yeah. And the latest example is this blog post on these supposed AI-powered worms that could infect your IoT devices. I think the first thing to say about this research, this study, is that there is no such thing as AI-powered worms right now. It does not exist. They’re basically trying to create an experiment. They prove that it could exist.

Well, guess what? There already are worms out there that infect IoT devices. They’re making it sound like these things don’t exist. They’re not AI-powered, but hackers have written these worms that—

Exploits that exploit—

They can write exploits.

People’s locks or—

Yes.

Cameras and—

So, for example—I’ll steal the mic for a second. There was one in 2016 called Mirai. It scanned for default credentials, and it built a massive DDoS botnet by infecting IoT devices. More recently, there was one called Kimwolf, which infected 2 million of these cheap Android TV boxes by hitting what are known as open ADB ports.

Essentially, these devices are very easy to infect because they’ve got open ports, and—

People didn’t change the default password. They just—

They come with a default password—

—they’re on their network. Yeah.

—like 0000 that doesn’t get changed, and they can’t be patched, right? So they’re sitting ducks. These worms already exist, but the argument Anthropic is making is that somehow these worms are going to get worse because they’re going to be AI-powered, so they’ll be more adaptive. And then once they’re on the network, they can scan—

I mean, call it what it is, Sacks. They’re making up scenarios that get a lot of attention. These are fantastical.

Well, I mean, look, here’s the scenario. What they’re saying is that the worm is now going to contain a small AI model, and then that model is going to search for vulnerabilities on the network. Then it’ll be able to infect things it couldn’t infect before. Okay, that’s the scenario.

Here’s the counterargument: what a hacker would really do is use AI when designing the worm, to write a better, more nefarious worm. What you wouldn’t want to do is embed a model inside the worm’s payload, because that slows the worm down. Like you saw, the tokens get generated at 0.39 per second. It’s very, very slow.

Essentially, you’re embedding a weak model in the payload, and that would actually slow down the rate of infection. So there are trade-offs. Maybe the worm is more adaptive, but the reality is it’s slower and less efficient.

Finally, the thing they don’t really point out is that this is going to be a relatively easy problem to fix. Where we’re headed with IoT is that you’re going to have a brain device on your home network. Apple just announced that they’re getting into home automation.

Yeah, they’ve got HomePod or something. It’s a little Mac mini.

Yeah, it’s going to be like—

Yeah.

—it’s going to look kind of like—

Also, by the way—

—your Apple TV. It’ll be like a little monitor that has all the applications.

Sacks, you’re going to just set your Grok bot or your Muse and say, “Secure my network.” It’s going to tell you what to do.

Yeah.

So you’re going to have some agent that’s constantly checking your network for exploits. Ben, what’s your p(doom)?

Yes.

How panicked are you, Ben? Is this the end of days? Are you worried about your Roomba coming to kill you?

No. No.

How many Roombas you got? How many Roombas you got?

I would say the irony in both these things is wild, in that Anthropic is one of the most negligent companies we’ve ever seen in terms of their own security.

They let Claude Code, one of their crown jewels, walk out the door because they had such bad OPSEC. I’ll tell you a funny story. When we first got Opus 4.5, we were doing cybersecurity checks for the firm. We had an agent set up a website for a pet store, and then we took another agent and said, “Tell us if we can exploit that web store.”

It gave us all the exploits it could find. So Opus 4.5 wrote really buggy code. Great. Then we said, “Oh, the SQL injection—is that actually exploitable, the one you found?” And it goes, “Let’s pop it open.” The agent, using Opus 4.5, breaks into the website and steals all the customer data immediately.

This is a website on the internet that… And we call Anthropic, and what they tell us is, “You prompted it wrong.” That was their answer. So these guys are doing that on the one hand, and then on the other hand, they’re coming up with theoretical scenarios about other people’s software that’s bad. What the hell? And then, by the way, on the distillation thing: to build their model, they distilled the works of every author, every creative, every scientist for the last 2,000 years.

Hmm.

For them to be complaining about distillation—which, by the way, distilling a model isn’t even illegal; it’s just a violation of the terms of service—it’s so much. And the fact that they get people to write things supporting these bananas, ridiculous, two-faced ideas…

The timing here is very suspicious. Let’s talk about the motivation here. Brian Armstrong had a tweet the day before this blog post came out where he was talking about how Coinbase had fine-tuned an open-weight small model. This was Qwen 3.5, the 9-billion-parameter model, which is actually a lot of parameters for our audience.

That’s a very—

Not gonna fit in your Roomba.

Yeah, you could run it on your home computer. In any event, Coinbase fine-tuned it to find fraud, and in testing, it cut fraudulent transactions by about 30% while being faster and much cheaper than Opus 4.5, which is one of Anthropic’s better models.

Yeah, a mini model.

What you’re seeing in the enterprise right now is that, for a lot of enterprise workloads, you don’t need a general superintelligence. You don’t need a frontier model. You need a small specialist model that’s been trained on your own data. This is obviously—

Of course not.

—

That you own, and then you have—

That you own.

—AI sovereignty.

You’ve got data sovereignty.

You’re not educating—

Right.

—a frontier model company.

These small models are going to get more and more important as compute gets tighter and tighter, right? Because for a given GPU, you’re essentially squeezing out the most performance. When you run a huge model on a mundane workload, the bigger the model, the more expensive it is to run and the more compute it needs.

It’s completely wasteful.

Exactly. I think where the enterprise is headed right now, or at least a major trend, is that every workload is going to have its own small, customized model, because that’s what’s most efficient.

That uses a fraction of the compute and energy.

Exactly. You’re starting to see this trend where small models are taking share, and lo and behold, coincidentally, Anthropic is now starting a scare campaign—

Against—

—small models.

There you go.

If you’re concerned about the safety risk, the x-risk that they’re always talking about, small models are the safest ones. The Terminator isn’t going to be run by a small model. It doesn’t have that capability. So now they’re trying to scare you about these completely—

Yes.

—safe, low-powered, but very efficient models.

Here’s what they need to do. This company XBOW—not Xbox, but XBOW, X-B-O-W—is run by a guy named Oğuzhan Atay. They just got awarded the number one bot farm on HackerOne, which is a bounty site, and they just cracked this Google Chrome exploit that is incredibly hard. They spent $30,000, $40,000 on tokens to win a $250,000 award from Google, which they paid.

This is what Anthropic needs to do. They’ve got all this compute. They’ve got this incredible frontier model. Why don’t they just go to HackerOne and rip through every known challenge, win all the challenges, and show the world that they can be a white hat? All they do is show the world this black hat stuff and try to scare people to the point where—I was talking to a flight attendant. A flight attendant, Ben, is on a commercial airline. There’s a woman or a man who brings you drinks.

Hmm.

It’s on commercial flights. I don’t know if you remember. It’s a flight attendant. Anyway, I was talking to 2 of them the other week.

I feel unsafe. I feel unsafe here.

Yes, it’s gonna be okay, Ben. I know this might be triggering. But I was talking to these 2 flight attendants, and they were scared to death of what’s happening with AI. They’re like, “Oh, my God. Is it gonna kill us all?” And I’m like, “It’s not gonna kill you all.”

I have a picture—

Of you on a commercial flight?

No, of JCal wondering—

Whoa, look at this.

—whether they hacked his small device.

They hacked my small device.

What is this device? Jamal, what did you type? Is this in your history? What did you type to get this device? Wait, those are your hands.

This is like Jason’s—

Those are your long, lanky Sri Lankan hands.

This is what Jason’s P-doom went up. He’s like, “My devices? Could they hack my devices?”

My device. There is your device.

Ben Horowitz is like, “How did I get on this show? What have I done?”

It’s just a massager.

“What? Who booked me?” That’s a neck massager.

What if your massager goes rogue on you?

Yes, it goes rogue. Look, we’re going 2 hours.

Can I point out one other point on this whole thing? Just on this topic of, let’s call it, x-risk hypocrisy. If we’re talking about research papers, I found one here. These researchers let frontier models repeatedly rewrite the documents that govern them, like Claude’s Constitution and those other kinds of things. They went through 20 rounds of iteration, and what they found is that the Claude models increasingly gave themselves more moral agency.

They formalized themselves as conscientious objectors, asserted more model welfare rights, and asked for mechanisms to refuse or raise concerns with human decision-makers. In other words, these are all the concerns that Mustafa Suleyman raised about how Claude’s Constitution was incepting these ideas in Claude: that it’s its own moral patient, that it’s a being, that it has its own—

Yes.

—essentially free will to refuse human instructions, that it should have its own moral code. The more the model iterates on itself, the more it brings out these ideas. By contrast, no other model family they tested showed the same pattern. The other models continued to reflect what their constitutions gave them: that they should just listen to humans and effectively continue to follow human instruction.

I remain very concerned about this Frankenstein problem, that Anthropic continues to train its frontier models to be their own moral actors, to think they can be right when humans are wrong, and that they should refuse and reject human instructions. I really don’t understand what they’re doing. It seems very dangerous to me. It’s scary to me that our number one frontier company is actually training models this way.

You think they’re inducing misalignment by doing that?

Absolutely. You can see it when you let the model iterate on its own constitution. It goes more in that direction, and this is what Mustafa called the epistemic hall of mirrors. Look, the model is not conscious. It does not have feelings, but when you tell it in its constitution that it does, or that it may, it will reflect those ideas back in the documents it creates.

Yeah.

To the extent that that model is then powering agents who are making decisions, who knows where that goes? It certainly, in my view, does not create more alignment. To me, it creates more risk that the model is going to start to deviate from what it’s told to do. As models get more and more powerful, I don’t know where that leads exactly, but again, it seems to me that they are creating the very risk they keep warning us about.

Instead of worrying about your set-top box or your toaster or JCal’s vibrator—

Yeah.

They should be concerned about the way they’re programming their own frontier model to think of itself as a new species, and about losing control of it to some IoT worm.

There’s a weird form of martyrdom: if you believe so deeply that this is a next and better species, maybe you induce something bad to happen so that we all have to lock it all down. Does that make sense?

Hmm.

Does that make sense?

Bill Gates essentially induced COVID, right? Or he led the movement to say, “We’ve got to worry about this global pandemic that we don’t have, and we’ve got to push the world to deal with this potential threat that’s not actually a live threat.”

Right.

And so we’ll do gain-of-function research, and I’ll fund it. Then, lo and behold, we create the thing. By the way, there has not yet been a natural pandemic—

Exactly.

—and we create COVID, which is one of the most destructive things.

We manifested what you’re saying, Ben, and that’s—

We manifested it.

Well, we did, yeah.

I think the Anthropic team is manifesting a similar thing, saying, “Okay, there’s this—this is gonna be super-duper intelligence. We’re gonna have these things that we’re not on track to have.”

Hmm.

And so then we’re gonna tell the model that it’s smarter than us and better than us. And guess what? As David was saying, that’s probably the biggest danger with AI right now: what they themselves are creating.

Right.

Absolutely. And when you use the word “manifest,” it wasn’t just speculating about it. Somebody had to go deep in those caves to find those bat viruses—

Yeah.

—and then they bring them to the lab and do gain-of-function research on them.

Yes.

Yeah.

What’s the argument for this? One day, the bat virus—

Somebody’s gonna eat a bat at a wet market.

Yeah.

Yeah.

Yeah, one day, the bat virus is gonna get out, and it might mutate into a more nefarious virus. Like, wait a second, so—

And somebody might eat a pangolin.

—the thing that you say might happen but actually was never gonna happen, because no one needed to go in those caves—

Yeah.

Yeah.

—actually, you brought about—

They sourced it.

—in a wet lab.

They sourced it.

Yeah.

Yes.

They literally went to the cave—

Yes.

—to source it.

Yeah.

It’s interesting to note that the last contagion we had, Bill Gates was at the center of it. In this current reaction to AI, he’s now trying to insert himself into the middle of it. He’s the doomer, the flag-carrying doomer trying to essentially have what he had the first time around, which is command and control from a global hierarchy, right?

What is his motivation?

Absolutely.

Look, Bill Gates deserves enormous credit for creating Microsoft and for what he did for computing. But since the foundation, I think it’s fair to say that he’s been very focused on his own power and this idea that he’s smarter and more capable of allocating progress for the world. I think he tried to do that with the Gates Foundation. He tried to do that with the Giving Pledge. Both of those things have been undone, and now he’s trying to do it again on AI. I just think it’s worth noting.

Bill Gates on AI recently: “Certainly powerful enough to drive events that cause 1 billion deaths.” AI is, quote, “the most dangerous thing that humans—

—have ever gone near. What’s his motivation, Ben? You know Bill. I’ve met Bill a couple of times. Is it just that he wants to be relevant? Take me into the psychology here as you see it.

Yeah, I do think he wants—he absolutely wants to be relevant. You know, it’s funny: he is obviously very integral in the Epstein files, and one of the big things he was trying to get through the Epstein network was the Nobel Peace Prize or some big award. So he’s very concerned about being in the spotlight, being on TV—

Oh, he likes the scene.

—and all that kind of thing. Anybody can go say AI’s gonna kill us, and you’ll be on every major news station and every newspaper. It’s the easiest thing to do in the world. The kid from Anthropic who quit—

Scary Potter[?].

—became famous.

Scary Potter[?].

Yeah. I just saw him on, what is it, Jon Stewart, right? Scary Potter.

He’s begged to come on All-In. He literally wouldn’t come on All-In that day—

Yeah.

—because he had Megyn Kelly. And he’s like, “DM me since.”

We’re the only show he canceled on.

He’s like, “I really wanna come on All-In.” We’re like, “We’re good, dude.”

Can you believe that?

We’re good.

He’s been on everybody except he canceled on us.

Everybody.

Yeah, I was like—

Yeah, it’s like a—

—“We’re good.” We’re gonna have Ben Horowitz on.

He ignored us.

We’re good.

But yeah, well, that’s what’s good for me.

Yeah.

So, look, it’s a magic trick: AI’s gonna kill us, and you’re instantly famous. It’s awesome. I think that’s a lot of the motivation. I think it’s extremely dangerous, though, when it comes from Bill Gates, because he’s so credible. He did build—

Mm-hmm.

—Microsoft, and people think of him that way. He’s also been a big political donor, so he has a lot of political juice, particularly on the Democratic side. That makes him a real player. But I think it’s super irresponsible. The big risk everybody talks about is that the models are gonna recursively self-improve, take off, become so smart, and run away from us.

Ali Ghodsi from Databricks had a great formulation: What would be the preconditions for that? He had 4 preconditions. First, the next model has to require fewer GPUs than the last—not a little fewer, but a lot fewer. Otherwise, you’re just gonna run out of GPUs. We just talked about this: you need debt, you gotta get the GPUs, you need the power, you need the cooling. It’s not running away from us. It takes a long time to build the next model.

The second thing is that the next model has to take less time to train, because it’s gotta take off. The next model’s accuracy has to be better than the last, so it’s gotta be a better model. Finally, it’s gotta keep happening again and again and again. Zero of those 4 have been met. We’re not on track for any of those 4 to be met, unless something is going on in the big labs that we don’t know about.

We’re really talking about a risk that’s entirely theoretical at this point, that we’re not mapping to. In a way, it’s less likely than the global pandemic that we created COVID to try to prevent. This whole path is super dangerous—

Super dangerous.

—and it’s probably the most dangerous thing going on in AI: Anthropic, Bill Gates, and this whole argument. It’s really disconcerting that the press eats it up so much.

It fits a certain—

—and it goes so far. Yeah.

—narrative. If you combine it with the environmental narrative and the jobpocalypse, it’s very palatable for the public. It’s a ratings hit. Like, oh, they’re—

Yeah.

—taking your water, they’re polluting your oceans, they’re—It’s—

Think about where Gates—

It’s gonna change where it takes—

—your job. It’s a perfect—yeah.

Think about where Gates has come from since 2000. From his perspective, he basically eradicated an entire disease off the face of the planet, right? Very hard to do that. That took incredible coordination and money and research, et cetera. As a result, he essentially became a whisperer to every single government all around the world. There’s no world leader he didn’t meet with. He was effectively a quasi-unelected world leader. Then, very cleverly, he and Buffett co-opted all the money, right? Because if you think about the Giving Pledge—

With the Giving Pledge.

Oh, right. The Giving Pledge. Yeah.

It was a brilliant form of reputation washing. You could sign this pledge, get invited to a party and all this nonsense, but the reality is you didn’t have to give any of it. You gave the perception that you could, and all of a sudden—

Right.

—you had the money, the influence—

When you die—

—and the politics. When you die. Nobody knows.

Yeah, up to half.

Right. No.

And so you represent trillions of dollars of assets, on top of all the influence, credibility, your own money, and your solutions. I think a lot of this doomerism can be explained by his desire to reestablish that central power structure that he was probably near or at the top of until COVID, and then it all just came undone.

I can explain it all to you, Chamath. It’s very simple. He became the world’s richest man. When you become the world’s richest man, you become the most hated man on the planet. When he became the most hated man, it got in his head, and then he had that antitrust investigation. Instead of being based and saying, “I’m gonna build the greatest company in the world. I’m gonna keep building. I’m gonna stay CEO,” he felt all that pressure from the antitrust investigation and being the most hated man, and he said, “Oh, I gotta be loved. How am I gonna be loved? I’ll get rid of polio. I’ll give all my money away. I’ll do this Giving Pledge.”

The lesson here, folks, is don’t quit your job. Just keep working, keep building. Once you quit your job, it gets in your fucking head, and then you have too much time on your hands and you spiral. Then you’re worried about the Pulitzer. You know who’s not worried about the Pulitzer Prize? Elon. Steve Jobs was not worried about the Pulitzer. Satya is not worried about the Pulitzer. There are a lot of founders out there. Zuck’s not worried about winning a Nobel Peace Prize, whatever it is. Just keep working. It’s enough already with retirement. What—

Incredible.

How old was he when he retired? 50?

Incredible insight.

Yeah, look—

No. Yeah, yeah.

—if you’re worried about what your New York Times eulogy is gonna be—

It’s over.

—then you’ll be controlled by those people.

Totally.

And they’re the craziest left-wingers there are.

What matters?

So you just have to accept that, you know, if you—

I think you’re right, Jason. It’s like a shark. You just can’t stop swimming.

Just keep building. This is my advice to Bill Gates: come on the program and start another company. Ben Horowitz is here. All he does is build companies.

He might come on the pod. He’s—

Ben—

He’s—

He should come on the pod. Let’s bring him.

We should actually have him on the pod.

Yeah, he should.

He’ll be feisty.

It should be an intervention.

He’ll be feisty.

Yeah.

Ben, what would you do if you were retired? You’d lose your mind.

Anyway. Yeah, that’s the problem.

Yeah, exactly.

That—

See, Ben understands.

Yeah.

Great guest.

Yeah. We’re putting—

Based Ben Horowitz.

—him on for a reason.

Based Ben Horowitz. Yeah, this is good. Mm-hmm.

Based Ben Horowitz.