The Manosphere Report

0:00How many PRs do you think are going to get pushed to the core structural

0:03internet in 100 days? What's the over -under number?

0:05Because I'll give you a number.

0:06You're going to say zero.

0:07My answer to that is - I'll say like 10 ,000, but it's going to

0:11be a meaningless thing. But if it prevents your browser history from being released to

0:14everybody in the world, Chamath, that may be something that you're willing to, you know,

0:18let 100 days pass on.

0:19I think you got Chamath's attention when you said browser history.

0:21What about the dick pics?

0:25As Chamath is, he's going to release them himself.

0:30We'll let your winners ride.

0:33Rain man, David Sack. And it's sad.

0:38We open source it to the fans and they've just gone crazy with it.

0:41Love you, Wes. I'm the queen of kinwa.

0:43I'm going all in. All right, everybody.

0:46Welcome back to the number one podcast in the world.

0:48David Freeberg is out this week, but in his place, the one, the only, our

0:55fifth bestie, Brian Gerstner. I mean, why don't you ever give me, it puts a

0:58little namaste in your payday anymore.

1:00Absolutely. You used to be the greatest moderator, but now it's just this kind of

1:05label. You know what? These guys beat me up.

1:07They beat me up and they just beat the joy out of me doing this

1:11program. It's because you're a Ro Khanna apologist now.

1:15No. I, we'll get into it.

1:17Okay, save it for the f***ing show.

1:19I'm not a Ro Khanna apologist.

1:21Just because I said like, hey, they've stopped retard maxing and they've started doing like

1:26some logical things. Yeah. Okay.

1:29It was great to be here.

1:30Great to be here. Good to have you.

1:31Good to have you here.

1:33And of course, we have David Sachs is back.

1:37Everybody wants to hear from David Sachs.

1:39We missed you last week, bestie.

1:40We didn't beat the joy out of you.

1:42We just tried to beat some of the hot air.

1:45Any fluff that you can put on the show that just involves you talking and

1:50saying nothing is, that's the stuff we got to cut out.

1:53Yeah. Fair enough. Okay. Yeah, we'll cut it right out.

1:57We'll cut it out and then we'll just put a promo in for thesyndicate .com.

2:00Thank you. Oh, Jesus. Also with us, Jamal Paliapati is here.

2:04How's your retard maxing going since last week?

2:06Did you have a retard maxing full weekend?

2:09Did you have a good full weekend of just smoking cigars in the back deck

2:12and not ruminating about all the chaos you've caused in the last 20 years?

2:17I think I've done generally more good than not.

2:22Oh, you have, but there's been some chaotic moments.

2:25Of course. Don't think about it, Jamal.

2:26You can't, bro, you can't have ups without downs, man.

2:29It's like, what are you there to do?

2:30Just like placate everybody and be a loser?

2:32Are you there to be a winner?

2:34Yes, you're in the arena, but have you stopped going to there after realizing that

2:39ruminating? What's up with this sudden interest in retard maxing?

2:43Are you like that clavicular for retard maxing?

2:47No, the world finally caught up with me.

2:49That's it. What do you, I mean, I've been retard maxing this whole time.

2:52They just didn't have a name for it, guys.

2:54Oh, okay. Eli's videos are really good.

2:56I watched two more this week.

2:58What, take us through what's so appealing about not ruminating, smoking a cigar, and just

3:04living your life. Because what he says actually works at every level of society and

3:09every sort of thing that you may want to achieve, even if you're trying to

3:13like climb the rungs, you very quickly learn that the more you want something, the

3:20less you're going to get it.

3:21And I think that's like his real message is let go, live life, and just

3:27try stuff or don't try stuff.

3:29And I think that that detachment is really healthy for people.

3:32I like it. I like it a lot.

3:34Who's the guy who says this?

3:36I actually didn't know. Elisha Law.

3:38But Eli, I think, is how he goes by.

3:41But he's fantastic. He's got a YouTube channel.

3:44Mark Andreessen found him. And he's like, this guy is the new guy.

3:49Modern day philosopher. He gives you a roadmap for how to live your life, right?

3:53A new age sage. What's the name of the guy, the character's name from Dune?

3:57I was into girls. I didn't read these books.

3:59I was dating girls. He's the Lisan Al -Gaib of the modern internet.

4:04This is why we need Freeberg here, is to explain these deep holes.

4:08All right, listen. We got a lot to get to.

4:10The basic point is build something and don't ruminate, okay?

4:13Ruminating is just not worth it.

4:15Just everybody go forward. Just do stuff.

4:17Stop blathering in your own head.

4:18Just do stuff. Absolutely. All right, listen.

4:21Speaking of doing stuff, Anthropic is withholding its newest model, Mythos.

4:25I'm using the Greek pronunciation.

4:27Its newest model, Mythos, saying it is far too dangerous for any of us to

4:33have access to it, according to the company.

4:34The model autonomously found thousands of vulnerabilities, including bugs, in every major operating system and

4:40web browser. This little study they did included 20 -year -old exploits that had been

4:47missed by security audits for decades.

4:49Some examples, they found a 27 -year -old vulnerability in OpenBSD, used in firewalls and

4:54critical infrastructure. They found a 16 -year -old bug in FFmpeg.

4:58That was missed by automated tools after 5 million scans.

5:02The Linux kernel, all kinds of bugs they found.

5:07They released a hype video hyping up why they were not going to share this

5:13model. Here's Daria. Come on the program anytime, brother.

5:15But as a side effect of being good at code, it's also good at cyber.

5:19The model that we're experimenting with is, by and large, as good as a professional

5:26human at identifying bugs. It's good for us because we can find more vulnerabilities sooner

5:31and we can fix them.

5:33It has the ability to chain together vulnerabilities.

5:35So what this means is you find two vulnerabilities, either of which doesn't really get

5:40you very much independently. But this model is able to create exploits out of three,

5:45four, sometimes five vulnerabilities that in sequence give you some kind of very sophisticated end

5:49outcome. All right, Brad. By the way, that set they're using there, that's the same

5:53room those guys play Dungeons and Dragons in every Sunday.

5:56Brad, you're an investor in this company.

6:00Is this virtue signaling or is it reality?

6:04Is this a good move by them to not release this model and be thoughtful,

6:08give it to a handful of people and just find all the bugs it can

6:12before releasing it to the public?

6:14And we've got a lot more issues to discuss about this.

6:16I actually think they deserve a ton of credit here.

6:19And let me walk you through why.

6:20Right. The company could have just released Mythos, broken a lot of core things.

6:24Oftentimes in Silicon Valley, we say move fast and break things.

6:28In this case, it means just releasing the model to move further ahead of your

6:32competition. But here the company realized it would wreak havoc.

6:35They ran their own vulnerability testing.

6:37They saw that it would allow offensive hacking and people to expose browsers and browser

6:42history, expose credit cards, you know, on the Internet.

6:46So, you know, what I like about this is they didn't need government to hold

6:50their hand on this. We have plenty of government regulations.

6:53They know it's in the best long -term interest of the company and the industry.

6:57You know, so they set up Project Glasswing.

6:59It's an AI -driven, you know, kind of cyber coalition.

7:03Apple, Microsoft, Google, Amazon, J .P.

7:06Morgan, 40 of the most important companies.

7:08And their goal is very simple.

7:10Let's spend 100 days using advanced AI to find and to fix and to harden

7:15these software vulnerabilities. before hackers exploit them.

7:19Now, what I think this represents, Jason, is a threshold that we're crossing.

7:24Mythos and Spud, which is going to be out from OpenAI any day now, which

7:29is the first Blackwell -trained model at OpenAI, they represent the beginning of what I

7:35would call AGI models. These are models with massive step function improvements and intelligence, and

7:42they're just too smart to be released immediately.

7:45And by the way, there was nothing that said that every time you finish a

7:49model, you got to immediately release it GA.

7:53So they set up this idea of sandboxing, building defensive alliances in order to move

7:59away from that regime. I think it shows, and Saxon and I have talked about

8:02this a lot, so I'm interested to hear what he thinks.

8:05It shows you can trust the industry and market forces in coordination with the government.

8:10They were talking to the government about this, but they're not relying on some top

8:15-down regulation in order to do this.

8:18They laid out a blueprint that seems to me very pragmatic that now that we're

8:22at this threshold, we're going to sandbox these things.

8:25I think that OpenAI will end up doing the same thing.

8:28I think Google will end up doing the same thing.

8:30It's an aggressive way to keep the pressure on and win the race at AI

8:35while making the trade -offs to protect safety.

8:39So I think you're always going to have to make these trade -offs.

8:41I think in this case, it was a great move by Dario and team, and

8:45I think they deserve a lot of credit.

8:46Sax, when you look at this, we had Emil Michael on the program a couple

8:50of weeks ago. It might have been four or five weeks ago.

8:51And we had a very thoughtful discussion about, hey, if the government is going to

8:56have these tools, you know, Anthropic wants to withhold them.

9:00And, you know, what is the proper relationship there?

9:03You have to think that the government, and I know you don't speak for all

9:07parts of the government. If you were just going to run through the game theory,

9:11they must have gone to the government and said, listen, this thing is so powerful,

9:14it can put together two or three hacks, create a novel attack vector.

9:18And this is incredibly dangerous.

9:20What if China has it?

9:22And if this thing is as powerful as Dario says it is, then this is

9:26an offensive weapon as well for us to take out, let's just pick, you know,

9:31a prescient issue, the North Korea's ballistic missile program.

9:35This is equivalent, the way it's being described as the Manhattan Project, perhaps.

9:40So what are the chances, two -part question for you, Sax, that China already has

9:45this and is using it?

9:46And do you think Dario is doing the right thing by regulating themselves?

9:51I think Anthropic has proven that it's very good at two things.

9:56One is product releases. The second is scaring people.

10:00And we've seen a pattern in their previous releases of, at the same time, they

10:05roll out a new model or new model card, something like that.

10:08They also roll out some study showing really the worst possible implication of where the

10:15technology could lead. We saw this last year, about a year ago, they rolled out

10:19this blackmail study where supposedly the new model could blackmail users.

10:24There's been a whole bunch of these things.

10:26Actually, I went back to Grok and I just asked, hey, give me examples where

10:30Anthropic has basically used scare tactics.

10:33And it's a pattern, okay?

10:35It's a pattern. Okay. These guys, I'm not saying it's not sincere, but they have

10:41a proven pattern of using fear as a way to market their new products.

10:46And if you think back to, again, my favorite example is this blackmail study where

10:51they prompted the model over 200 times to the result they wanted.

10:56And that result was clearly reverse engineered, and it got them the headlines they wanted.

11:02And I would say the proof that it's reverse engineered is we're now a year

11:05later, there's a bunch of open source models out there that have the same level

11:10of capability that that anthropic model had.

11:14And have you seen any examples of blackmail in the wild?

11:16I don't think so. So in other words, if that study were true in the

11:22sense of being a likely outcome of that model, I think you would see examples

11:26in the wild of that behavior.

11:28We haven't seen any of that in the past year.

11:30Now, let's talk about this specific example with cyber hacking.

11:34I actually think that this one is more on the legitimate side.

11:39I mean, look, the reason why I bring this up is anytime anthropic is scaring

11:42people, you have to ask, is this a tactic?

11:44Is this part of their chicken little routine?

11:46Or is it real? Are they crying wolf or not?

11:49I actually would give them credit in this case, and say, this is more on

11:53the real side. It just makes sense, right?

11:56So that as the coding models become more and more capable, they're more capable of

12:00finding bugs, that means they're more capable of finding vulnerabilities.

12:03And like one of their engineers said, that means they're more capable of stringing together

12:06multiple vulnerabilities and creating an exploit.

12:09And so I do think that over, say, the next six months, we're going to

12:12have this, call it one -time period of catching up where AI -driven cyber is

12:20going to be able to detect a whole range of bugs that maybe have been

12:23dormant over the past 20 years across a wide range of systems.

12:28And so I do think that there is real risk here.

12:31And I do think, therefore, that having this pre -release period makes a lot of

12:35sense where they're giving the capability to all these software companies that have existing code

12:40bases to use the tool to detect the vulnerabilities for themselves so they can patch

12:45them before these capabilities are widely available.

12:49And by the way, it won't just be Anthropic that makes these capabilities available.

12:53We know that, like, let's say the Chinese open -source models like Kimi K2, it's

12:57about six months behind. So we have a window here of maybe six months where

13:02we're still in this pre -release period where I think companies that have large code

13:07bases can get advanced access to this model.

13:11And I guess OpenAI is going to release a similar thing in the next few

13:15weeks. I do think that every company or IT department or CISO that is managing

13:22code bases should take this seriously and use the next few months to detect any,

13:29again, like dormant bugs or vulnerabilities and roll out patches.

13:33If everybody does their job and reacts the right way, then I do not think

13:37it will be the doomsday scenario that Anthropic is sort of portraying.

13:41But it's one of these things where the fear might end up being a good

13:45thing in order to drive the correct behavior.

13:50So I ultimately think this is going to work out fine, but you do need

13:54everyone to kind of pay attention, use the capabilities, fix the bugs.

13:59Then we're going to get into a big arms race between AI being used for

14:02cyber offense and AI being used for cyber defense, but it'll be a more normal

14:06sort of a period. Chamath, we have Dario and a number of the participants here

14:13taking this super seriously. They're making a big statement.

14:17Zach's very nuanced, I think, take there.

14:20What's your take on how do these companies have it both ways?

14:23Hey, this shouldn't be regulated.

14:26This should be regulated. If this is, in fact, a cataclysmic, oh my God, they're

14:31going to hack everything. What if the China Chinese have this right now, that would

14:35speak to more government, either coordination, regulation, or some kind of relationship between the CIA,

14:42the FBI for domestic stuff and these companies, because there, it is a non -zero

14:48chance that the Chinese have an equal capability here.

14:51We're assuming they're behind, but who knows what they're doing behind closed doors.

14:55So what's your take on this?

14:56Is it the boy who cried wolf or is this the real deal now?

15:00I think it's mostly theater.

15:02Okay. In February of 2019, when Dario was still at OpenAI, they did the same

15:11thing with GPT -2. That was a 1 .5 billion parameter model, which sounds like

15:18a total fart in the wind in 2026.

15:21But at that time, this 1 .5 billion parameter model was supposed to be the

15:26end of days. And it was supposed to unleash this torrent of spam and misinformation.

15:31And that was the big bugaboo at the time.

15:33And so what happened, they went through this methodical rollout over six or nine months,

15:37they started releasing the smaller parameter models, and then they scaled up to the big

15:411 .5 billion parameter model.

15:43And at the end of it, it was a huge nothing burger.

15:47If you actually think that Mythos is capable of doing what it says it can

15:51do, two things are true.

15:54One is a very sophisticated hacker can probably do those things right now with Opus.

16:00And two, if these exploits are this easy to find, whether you use Opus or

16:08whether you use Mythos, the reality is you'd have to shut down the internet for

16:11about five years to patch them all.

16:14So when you see like a large multi -trillion dollar G -SIP bank, it's a

16:20bit of theater. Why? What do you think they can actually accomplish in two months?

16:24Do you actually think that if there's these vulnerabilities, it's all going to get fixed?

16:29Let's give them six months.

16:31Let's give them nine months.

16:32But the reality is that capitalism moves forward, the funding needs moves forward, and the

16:38need for these guys to build adoption moves forward.

16:42And that's going to supersede what this is.

16:46So I do think that Sachs is right, that they have figured out a very

16:50clever go -to -market muscle here, and a go -to -market motion that activates hyper

16:57-attention and hyper -usage. And so I give them tremendous credit.

17:01And I'll maintain what I've maintained before.

17:04Anthropic is shooting the lights out right now.

17:07This is like Steph Curry going bananas.

17:09From everywhere on the court, these guys are hucking threes.

17:12Clay Thompson. It's all net.

17:14Okay, so huge kudos to Anthropic.

17:18But we've seen it before.

17:20We saw it when these folks were the principal architects at OpenAI.

17:24We're now seeing the same playbook here.

17:27I think we'll look back.

17:28And I think what we'll say are these two things.

17:30One is, if we're really going to patch all these security holes, we need to

17:33shut down the internet for some number of years, honestly, literally years.

17:38And the second is, an advanced hacker can probably do this today with Opus if

17:43they really wanted to. Okay.

17:46Hey, Brad, I'll get you in here for the last word.

17:49I'm going to go with, yeah, maybe they did cry wolf before.

17:53But based on what I see with these models advancing and using them, and I'm

17:57using a lot of the open source ones right now from China, I think that

18:01this is like code red kind of moment.

18:03This is DEFCON. Like, we should be taking this deadly seriously.

18:06And I think these companies got to coordinate with the CIA.

18:09And this is equally a defensive as awful.

18:12expensive opportunity. Do you think this is...

18:15You're asking for the nationalization of AI now.

18:17No, actually, I don't think it should be nationalized, although I did see people sort

18:22of insinuating that. I think these companies need to build a group, Brad, that work

18:28and coordinate with the CIA.

18:29I assume that they're already doing this.

18:31I'm assuming, you know, Emile Michael and, you know, Trump and everybody have these people

18:36in a room and that they've given the DEF CON and said, hey, how can

18:39our government use this to stop bad actors?

18:42And this is already being coordinated with the CIA and the FBI.

18:46I am 100 % certain of that, that Dario went to them and said, look

18:49what we found. This is the real deal.

18:52I'll give you the last word on this, Brad, since you're an investor in both

18:54companies. You know them quite well.

18:56The Frontier Model Forum, which was put together in 23, is cooperating on anti and

19:02adversarial distillation stuff as we speak, right?

19:05They don't want to make it easy on, you know, so Google and OpenAI and

19:10Anthropic, they're coordinating on this stuff.

19:13You know, there are times where I've pushed back on Anthropic because I thought it

19:16was, you know, perhaps regulatory capture or something else.

19:19This is very different in my mind, right?

19:21He could have easily, Dario could have easily come out and said, oh my God,

19:24we passed a threshold. We need to have a government moratorium.

19:27Remember, even our friend Elon called for a six -month moratorium in 2023 because of

19:32civilization risk. This guy didn't do that.

19:35Instead, he said, okay, what should we do?

19:37I'm going to get 40 of the leading companies together.

19:39We're going to spend 100 days sandboxing, hardening the systems, and then we're going to

19:44keep pushing forward. What do you honestly think is going to get accomplished in 100

19:47days? How many PRs do you think are going to get pushed to the core

19:51structural internet in 100 days?

19:52What's the over -under number?

19:54Because I'll give you a number.

19:55You're going to say zero.

19:56No, no, no. I'll say like 10 ,000, but it's going to be a meaningless

20:00thing. But if it prevents your browser history from being released to everybody in the

20:03world, Chamath, that may be something that you're willing to, you know, let 100 days

20:07pass on. I think you got Chamath's attention when you said browser history.

20:10What about the dick pics?

20:13As Chamath is, he's going to release them himself.

20:16Right now, Chamath's like, hey, Chinese hackers, here are my dick pics.

20:20Please put them out. We have to be out there complimenting when they're doing the

20:25right things or relying on the market rather than running to the nanny state and

20:28saying, do more of this.

20:29So this to me was just an example of a good balance.

20:33I'm sure we're going to have plenty of debates about this in the future.

20:36But, you know, this is one I would like to see more of.

20:38This is why, to use your word, Jake, I tried to have a more nuanced

20:41take, is because we have no choice but to take this seriously.

20:45Whether it's total theater, whether it's fear mongering, and they do have a pattern around

20:50this, we can't take the risk, right?

20:52And it does logically make sense that as these models become more and more capable

20:58at coding, they're going to get better at cyber, and there's going to be that

21:01one time period where you're moving from pre -AI to post -AI, and you need

21:06a patch for that. So my guess is we're going to see a lot of

21:08patches over the next few months.

21:10I think that that will resolve the problem.

21:13I think this is a case where I'm going to give them the benefit of

21:17the doubt. I think that, you know, I've criticized them in the past.

21:21I think that blackmail study was embarrassing to the level of being a hoax.

21:25But I think in this case, I'm going to give them credit and say that

21:29I think that it's legit.

21:31So it's not the anthropic hoax.

21:33This could be legit. I, you know, looking at this - Or we have no

21:36choice but to treat it that way.

21:38Of course. Yeah. I mean, even if two things could be true at the same

21:42time, Sax, they could have used this tactic before.

21:44It could be performative, like the video with the dramatic music in the background.

21:48It does have a little bit of - drama to it.

21:51And the way they presented it is very dramatic.

21:54But it does make logical sense that the one company that made the bet on

22:00code, bigger than anybody else, would be the one who would discover this quickest.

22:04And, you know, in 100 days, that's a pretty good, that's a pretty big advantage

22:08versus the hackers. But let me think one more point there, Chamath.

22:12The most important thing that people haven't talked about here is the amount of code

22:17being pushed right now because of these tools is 10x, 100x in most organizations.

22:22So we need to have this type of security embedded in these new coding tools

22:27to do it in real time.

22:29That's the opportunity. There should be real time correcting of this.

22:32If this is real, they pick the wrong companies.

22:35Meaning, there are energy companies, folks that control nuclear reactors.

22:42There are airplane companies that are flying hundreds of 1000s of people in essentially manufactured

22:48missiles of like, streaming gas going at 500 miles an hour.

22:53None of those companies were the ones that were included in this.

22:56And so I think if you really thought that this was end of days, at

23:02a minimum, we can agree, maybe we should have expanded the circle a touch.

23:07Well, maybe those are customers of the ones they're including here.

23:10Anyway, this is a really important story.

23:12We'll obviously track it in the coming weeks to see what turns out to be

23:16reality. And Dario, do come on the program at some point.

23:19Hey, Brad, will you get Dario to come on the program?

23:21I've invited him like three times.

23:22I got his phone number.

23:23He's ghosted me. I don't know why.

23:24Wait, he's ignored you? I literally got an introduction from the number, like one of

23:29the number one venture capitalists in the world is on the cap table very early.

23:32He just won't respond. I don't know why.

23:35I would tell you Dario's podcast with Dworkish, who I think is an excellent podcaster.

23:40I've listened to that three or four times, taken notes every time.

23:43It is a really exceptional piece, really exceptional piece of work by, by, by them.

23:48All right, let's keep moving.

23:49We've got a lot on the job.

23:50You may once again be tarred with your affiliation with us.

23:54Poor you. I mean, I don't care.

23:56Literally. I, I've got friends on both sides of the aisle.

23:59I have friends. Of course you do.

24:00Even J -Cow. Even J -Cow has friends everywhere.

24:04Let me ask Brad a question here, which is while we're on the topic of

24:07Anthropic, there was a really interesting story or tweet, I guess you could say by

24:11the founder of OpenClaw that Peter, Peter.

24:15Yeah. What's his name? Peter Steinberger.

24:17Steinberger. Yeah. Renowned coder created OpenClaw, which is kind of the thing that launched the

24:23soul agent era. Now you, I guess you could say any event, he said that

24:28Anthropic was cutting off his access to, was it to, was to Claw?

24:33Is that the next topic?

24:34This is on the docket.

24:35It's a little bit nuanced.

24:37Everybody using OpenClaw would take their $200 a month subscription to Anthropic, which was essentially

24:42like a, people were using more tokens and it's an average.

24:46The people from OpenClaw, it is very verbose and those people are a hundred X

24:52the usage of the average subscriber.

24:54So he said, you can't use your $200, you have to use the API.

24:57You move from the $200 plan to the API, add a zero to your token

25:01use. So, or more. And so they essentially ankled OpenClaw and then 10 days later

25:09or less, they released or announced their new agent technology, which is a, according to

25:14them, a safer, better version of OpenClaw.

25:17So, Hey, all's fair in love and war.

25:19And they have basically shot a huge cannon across the bow of OpenClaw.

25:25Can you just explain that exactly?

25:27So, so I think you're right that they...

25:29systematically copied feature by feature of OpenClaw, incorporated that into Clawed.

25:34And then the coup de grace was basically cutting off OpenClaw.

25:38Can you just explain exactly what they did?

25:41Okay, very simply. When you buy a subscription to these services, they have blended your

25:47usage across many users. So there's, you know, nine out of 10 users use less

25:52than the tokens they're paying for and the top 10 % use much more.

25:56When OpenClawed became a phenomenon, the number one open source project in history on GitHub,

26:01with all of this usage, people went crazy.

26:04And you heard me talking about how crazy I went for it.

26:07Those people with the $200 subscriptions were using $2 ,000, $20 ,000 worth of tokens.

26:12So they said, you can no longer use your subscription to, you know, either your

26:17professional or enterprise subscription at $200 and plug that into your OpenClawed.

26:22You now have to go to the API and pay per usage.

26:25So no more like unlimited essentially.

26:28But if you use Anthropics' own agent harness, are you part of the bundled flat

26:32rate? You can assume that that's what they'll do, which if you were thinking on

26:36an antitrust level, might be token dumping or price dumping.

26:40I'm not saying like I'm ratting them off.

26:42No, it's like bundling, isn't it?

26:43Well, price dumping or bundling.

26:45When you price something under the market price in antitrust, that would be price dumping,

26:50right? And if you were to bundle, it would be like the bundling issue.

26:54Critically important, you can use OpenClawed via Claude API.

26:57And every company has a right to set the price for its products.

27:00It's just saying that you were under their current regime, they were selling dollars for

27:05$0 .10 via OpenClawed because these were such power users.

27:08And now they're just saying we have to price this rationally, but we're happy to

27:12have you guys use the API.

27:13Okay, okay. But Brad, when you use the OpenClawed competitor that Anthropic now offers, are

27:20they subsidizing that? Are you paying?

27:22We don't know yet because it's in closed bin.

27:24So in other words, what I'm saying is if they charge for API usage, their

27:29own first -party agent harness or system, then that would be apples to apples.

27:34But if they end up charging the bundled flat rate, let's say, for their stuff,

27:40but then charge the metered rate for third -party stuff, you could make a bundling

27:45argument. Sure, sure. And you could say it's anti -competitive, assuming that Anthropic has dominant

27:50market share in coding, which I think most people would say they do at this

27:54point. And assuming that it's the same product.

27:56I mean, the reason most enterprises will probably use the Anthropic version of this agentic

28:03product is because it meets all of your security parameters, right?

28:07So Altimeter runs a lot of stuff on Anthropic.

28:11They're already integrated within our data warehouse, our data lake, things of that nature.

28:16So just letting OpenClaw loose on the Altimeter data set would not be wise.

28:21And so it's a different fundamental product.

28:23No, I get that. And I think that Anthropic has a huge advantage, let's say,

28:28cloning OpenClaw and just building it into Claude.

28:30I'm not denying that. To me, that would be the reason why they don't need

28:34to do price discrimination is because there's already a very good reason to use the,

28:39let's call it the bundled offering on a featured basis.

28:42But the question I'm specifically asking is whether they're giving themselves a price advantage.

28:48I think Brad is giving the most generous interpretation.

28:51You're taking a more cynical one.

28:52I'm with you, Sachs. I'm 100 % on the cynical side.

28:55OpenClaw is so powerful. It's got so much momentum that not only is Anthropic trying

29:02to ankle it. I believe when Sam Altman bought it, it was, and he didn't

29:06buy OpenClaw itself. The point that this is in theихцию is around 1 -800 -1

29:07-800 -880 -00. It says he's a giant 나뉘 on theloc normally you can buy

29:07iPhone, 1333, or without any So if not clonًt that you can pull whether or

29:07not haverot guy. he hired, Aqua hired Peter.

29:09I believe it was to subvert the open source project to get Peter's next set

29:13of genius ideas inside of OpenAI, as opposed to letting them go there.

29:18People are going to say I'm a conspiracy theorist.

29:19But this is the number one focus.

29:22And let me just give you a list of who is trying to kill OpenClaw

29:26slash compete with them. Obviously, you have Anthropic, but also Perplexity Computer Launch.

29:32It's awesome. I've been using it.

29:34Anthropic has this Claude Managed Agents.

29:37They dropped that on Wednesday, April 8th, yesterday, today's Thursday when we tape.

29:41You guys listen on Fridays.

29:43And then you have Hermes Agent.

29:46That was released on February 25th.

29:48That's also open source and very good.

29:49So that's in the open source camp.

29:51Alibaba is coming out with one that's going to be based on their Quinn model.

29:55Then you have Elon, who said he's got something called Grok Computer coming out of

30:00MacroHard, which is a play on words for Microsoft.

30:02In addition to that, Amazon and Apple are preparing new releases of their retard maxing

30:09assistants, Alexa and Siri, that will be less retarded in this new version.

30:13And then nothing out of Satya and Microsoft yet.

30:17So the number one goal, I believe, in the large language model, frontier model space,

30:23is to kill this open source product.

30:26No, I mean, come on.

30:27Like, why? They're building multifunctioning agents that can move from answering questions to actually doing

30:33something for you. Like, you got to do that because that's what consumers and enterprises

30:38wants. It doesn't mean that it's about killing open claw.

30:41It's just this is an obvious thing that the world wants.

30:42They have the right to do it.

30:44But this is a giant movement to stop it because this is the equivalent of

30:48having an open source Android -like player in the market.

30:51And that could be incredibly disruptive.

30:53These, I believe open source is going to win the day on the large language

30:56models and take 90 % of the token usage.

30:58And I think the entire frontier model space could be undercut by open source.

31:03And I think they realize that SLMs, the smaller language models that are verticalized now,

31:08that will run on, you know, desktops and laptops and is even starting to run

31:13on the top ones, that is their biggest competitive threat.

31:15And I hope it happens.

31:17All due respect to your investments, Brad, I think this technology and the interface is,

31:22you know, he plays bets.

31:23But I think it's imperative that the agent level, which is essentially your entire life,

31:28you don't give that to Anthropic.

31:30You don't give that to OpenAI.

31:32That's your entire business, your entire life.

31:34It is foolish for you, Brad, to give your entire business and all the knowledge

31:38you have to Anthropic through that, unless you're just doing it to boost your investment

31:42in those companies. But I would be very concerned if I was you with putting

31:46all of your knowledge that you've earned over a lifetime into any of these large

31:50language models. All right, J.

31:52Cal, let me ask you, can I ask a question?

31:54Thank you for that impassioned monologue.

31:58Actually, I want to ask all three...

31:58Thanks for coming to my TED Talk.

31:59Yes, thank you for that TED Talk.

32:02I have a yes -no question for each of you.

32:05Do you believe that Anthropic has dominant market share in coding right now?

32:10Yes, no. No. In coding?

32:15Yes, just coding. They had the lead, but not dominant.

32:17I think it's a trillion -dollar market, and these guys have less than 10 %

32:21of it today, so it's hard to make a case that...

32:23What percent of coding tokens do you think that Anthropic is providing the market right

32:29now? Greater than 50%. Yeah, that's true.

32:31Okay, that's called dominant market share.

32:33I don't know about that.

32:35More than 50 % of the market?

32:36You've got to look at what the TAM is.

32:38You've got to look at what the TAM is, David.

32:41Right? There are a lot of people who provide, you know, that are in the

32:44business of helping people right now.

32:45Chamath, you have to take on him.

32:45software. Do you want to be the tiebreaker before we move on to the next?

32:47I'm not saying it's a permanent condition, but if you're telling me that today Anthropic

32:52is delivering over half of the coding tokens, that's clearly a dominant position in the

32:58market for coding. It's an early market.

33:00It could change. If I were representing them, David, I would say nine months ago,

33:04everybody called us out of the game.

33:07We were being destroyed by OpenAI.

33:10In three months now, people are saying we have dominant market position.

33:13This is the fastest changing, most competitive market in the world.

33:17I think it'd be very hard pressed to walk into some district court and make

33:21the case that these guys have somehow already formed a monopoly against Amazon, Google, Microsoft,

33:26OpenAI, et cetera. Well, I'm not saying it's already a permanent monopoly, but I am

33:32just asking about market share.

33:34And I do think you guys all agree they have dominant market share.

33:37Let's get Chamath in here.

33:37Chamath, go ahead. They probably have 50 to 60 % market share because I think

33:40Codex is actually quite broadly used as well.

33:45But that belies the more important point, which is AI enabled coding, I think is

33:51still 5 % of the broad market.

33:52So it's kind of a nothing burger.

33:54Yes, they're leading, but they're leading in something that isn't that big yet.

33:58Now you would say, how could it not be big?

34:00And what I would say is because most of the stuff that's being written is

34:04still white sheet de novo code.

34:07And I think the ugly truth is, I don't care what model you have, but

34:12the long horizon ability for any of these models to actually build enterprise -grade software

34:17is still shit. S -H -I -T shit.

34:21And that's the actual lived experience.

34:24Not for me, but when I call on our customers, half a trillion dollar banks,

34:29hundred billion dollar insurance companies, none of these guys are like, wow, it just works

34:33out of the box. It doesn't work.

34:36So most of it is still hand -tuned.

34:38So until I can honestly tell you that we can point a model at this

34:44with the right guardrails, which I can't today, what I would say is it's a

34:48small market that will become large as these models become better.

34:54But we are in the world where we have 50 years of accumulated tech debt

34:59as a world. And I suspect when you enumerate the number of lines that that

35:04represents, it's hundreds of trillions of lines of just pretty marginal, mediocre code to bad

35:10code. On top of that, we have all these legacy languages.

35:13I'll tell you one of our customers, they have to go and get 60 year

35:17old pensioners to come into the office to interpret cope.

35:21No, I'm not joking. This is a hundred billion dollar a year revenue company.

35:27And that's how they solve these problems.

35:29It's not Opus just solves it.

35:32So I would just keep in mind that most of the tech debt in the

35:36world that exists, 99 % of it is still poorly addressed by these models.

35:40We are untying this Gordian knot.

35:44It's going to take decades to do it right.

35:46So all the breathlessness about all this other stuff, I really think it's not where

35:49the money is. It's not the big time stuff.

35:51And you can tell me, oh yeah, it's going to be the future.

35:54And I would say, tell this business that's a hundred billion dollars a year of

35:58revenue and 50 million billing relationships that all of a sudden you're going to open

36:01claw your way to a solution.

36:03It's bullshit. Not to say that you can't have a great chief of staff and

36:08not to say you can't do some useful stuff and trick trickery and, you know,

36:11have a good knowledge base.

36:12I'd like that too. But the core things that your lived experience sits on today

36:18is a mess of tech debt that will get very slowly replaced.

36:22And that's just the reality of life.

36:24And there are competitors that are extremely disruptive.

36:28I'll tell you about one.

36:29We talked about BitTensor Tau on this program a couple of weeks ago when we

36:32had the Jensen interview. You brought it up, actually, Chamath.

36:35There's a project that's subnet 62.

36:38It's called Ridges AI. And what they're doing is a competitor that is not only

36:44open source, but anybody can contribute to it.

36:47They spent about a million dollars in Tau, like rewards.

36:49And in 45 days, they hit 80 % of what Claude 4 is.

36:54And they did that in under 45 days.

36:56The way that works is they give rewards for people who, and they can do

37:00this anonymously, make that coding product, which is like Codex or Claude Code, better.

37:06That flywheel is racing right now with participation in the same way Bitcoin is.

37:11So you're going to see a lot of open source and these crypto open source

37:16combinations. And anybody who's not investigated this, I highly recommend you investigate this.

37:23I do think you're right about one specific thing.

37:25I would put zero, literally the probability zero, of any important company worth anything more

37:31than a dollar having and outsourcing their production code to an open source project.

37:36That'll never happen. However, what will happen, though, is when you look at the cost

37:41of training this 10 trillion parameter model on Blackwall, and when you look in the

37:47future, let's just say in six or nine months, that a 15 or 20 trillion

37:52param model is going to get trained on Vera Rubin.

37:55I think, Jason, where you are right, I have zero, and just to be clear,

37:59I have no investments in this at all.

38:02I do, to be super clear.

38:04I'm just observing, because another project other than BitTensor that someone brought up to me

38:08is Venice. The concept of open source training and orchestration is a hugely disruptive idea,

38:16which is the complete orthogonal attack vector to this idea that you have to raise

38:21tens and tens of billions of dollars to train your models.

38:24Because if the capital markets run out of 10 and 20 billion dollar checks to

38:30give people, the only solution is to be totally distributed.

38:33So I tend to agree with you, Jason, that there is going to be, at

38:37some point, a very successful open source project for pre -training.

38:42Absolutely will there never, ever be an open source way where a real company that

38:47has any skin in the game says, here, guys, re -engineer my code base as

38:51an open source project. Never going to happen.

38:53Yeah, I think the coding tools will.

38:55And if you look at the history of open source, Brad, you actually, I think,

38:58had a lot of bets in this space.

38:59Linux, Kubernetes, Apache, Postgres, like Terraform, like these open source projects are deep inside of

39:07enterprises, deep. And we were sitting here 15, 20 years ago, the same argument was

39:11made. Nobody will ever adopt these inside the enterprise.

39:14You got to go with Oracle, whatever.

39:16And fair enough, many people do.

39:18But I think this is this $29 Ridges subscription to do this versus 200.

39:24It's starting to take hold inside of startups.

39:27And that's where I always look at the tip of the spear.

39:29Startups love to, you know, use open source products.

39:33I think this could be the next big thing.

39:35But listen, I invest in things that have a 90 % chance of going to

39:39zero. So do your own research.

39:41No crying in the casino.

39:43Can I just make a final few points?

39:46So just quickly. So number one is, with respect to this market for code, for

39:51code tokens, whatever you want to call it, it might be 5 % today, meaning

39:555 % of the codes AI generated versus human generated.

39:58I think it's going to 95%.

40:00I mean, I bet any amount of money on that.

40:02The only question. is when?

40:03Probably over the next few years.

40:05So that's point number one.

40:06Point number two is, it's possible that if you're the early leader in coding as

40:13an AI model company, let's say you have 50 % to 60 % market share,

40:17you have the most developers using it, therefore you have the most access to code

40:21bases, you might get the most training tokens.

40:25There is a potential flywheel there where you can see the early market leader consolidating

40:30its lead because it's generating the most code tokens and it's getting access to the

40:35most existing code. Now, I'm not saying for sure that's going to happen.

40:38It's possible that the other guys catch up, but I think there is a possibility

40:42of a flywheel there and strong, I guess you'd call it data scale effects, things

40:47like that. So I do believe that the market for coding tokens could be monopolized.

40:52Third, Anthropix revenue run rate is based on what I can tell and what's been

40:57publicly released is the fastest growing revenue run rate at scale that I think we've

41:02ever seen. Perfect segue. It's the next story.

41:05Okay. Maybe pull up the tweets, but this thing is ramping at a rate we've

41:09never seen before. Yeah. We can get into that in a second, but just one

41:14last final point is I think it's pretty clear that where we go from here

41:18is agents and coding gives you a huge step up on agents because, you know,

41:25one of the main things that agents need to do is write code to be

41:28able to enable them to complete tasks.

41:30Correct. And so if it is the case that coding is this huge market that's

41:36going to be dominated by one or two companies, and then that leads to another

41:41huge market, which is agents.

41:43My point is just, I think all these companies need to behave in a very

41:47clean way and not engage in tactics that later the government might say, you know

41:53what? That was anti -competitive.

41:54Everyone should just, I think, play fair, do not engage in discrimination against other people's

41:59products, engage in fair pricing.

42:01I'm not accusing anyone of breaking any of the rules, but what I'm saying is

42:05that eventually the government's going to look at this market with the benefit of 2020

42:09hindsight. And I think everyone should just basically, you know, keep your nose clean.

42:15Keep it tight. Keep it tight and right.

42:17Tight is right. I think it's an excellent point.

42:20Let's talk about the revenue ramp of Anthropic.

42:24This is just unprecedented. Anthropic's revenue run rate has topped 30 billion with a B.

42:31Early 2023, they turned on revenue.

42:33They started charging for API access.

42:35End of 2024, they're at a billion dollar run rate.

42:38February 25, they launched clogged code.

42:40That was the starter's pistol.

42:42Mid 2025, $4 billion run rate.

42:44End of 2025, $9 billion run rate.

42:48Just a couple of months later in April, $30 billion run rate.

42:52Yes, that's right. Triple. And the way they did this is enterprise customers are a

42:58major part of the spend.

43:00Dario announced a couple of months ago that there's over a thousand enterprises paying over

43:051 million annually. This is truly mind boggling when you think about it because those

43:10are the most coveted customers in the world.

43:13These are the big fish that you just, when people are running enterprise software, they

43:18dream, Slack dreamed of getting these million dollar customers.

43:21Salesforce dreams of getting these million dollar customers.

43:24Brad, you're an investor. I guess Sam famously on BG2 asked you to sell your

43:29OpenAI stock back to him.

43:31You didn't, you demurred, but you're an investor in both.

43:35How shocking is it to you to place both of those bets and then see...

43:41One of them come from so far behind, you know, ChatGPT has 900 million users.

43:46I don't think they've passed a billion officially yet, but they are the verb, right?

43:50They're the Uber. They're the Xerox.

43:52They're the Polaroid of AI, but they didn't go after the enterprise.

43:58Dario made that and Dario worked.

44:00He was the co -founder of OpenAI.

44:02He left. And according to the New Yorker story that came out from Ronan Farrow

44:05this week, he was basically left because of his disgust in working with Sam Altman.

44:12Your thoughts, Brad? Before we go down the OpenAI rabbit hole, let's just really contextualize

44:18what's going on here. I have this additional chart.

44:22You showed one. They added $4 billion of revenue in January, $7 billion in February,

44:28$11 billion of annualized run rates, or $10 or $11 billion in March.

44:32Just to put in perspective, that's Databricks plus Palantir combined that they added in a

44:38single month, right? So we started with everybody at the start of the year wringing

44:43their hands, including, you know, Gurley and others saying we're in a big bubble, asking

44:47whether the AI revenues would show up to justify all of this investment.

44:51And bam, you have the largest revenue explosion in the history of technology.

44:56So the company's plans were to end the year at about a $30 billion exit

45:01run rate. They got there by the end of March, right?

45:05And I suspect that it's continuing in April.

45:08So you have to ask what's going on and what's the big so what?

45:11The first thing for me is that model and product capability just hit this threshold

45:16we talked about earlier, near AGI, whatever the hell you want to call it.

45:19And everybody like Altimeter said, damn, this is so good.

45:23I have to have it.

45:24This is no longer about my IT budget.

45:26This is about labor augmentation and labor replacement.

45:29And by the way, co -work is growing even faster than Claude Goad at the

45:35same stage of development. So what it showed is we have a near infinite TAM.

45:41It turns out that the TAM for intelligence is radically different than anything that we've

45:45seen before. And I think the best example of this, right, this is millions of

45:51self -interested parties, consumers, enterprises, a thousand now over a million dollars, right?

45:58It's not that there was some great go to market and anthropic that all of

46:02a sudden, you know, they snuck up and blew everybody away.

46:04No, it was companies demanding the product.

46:07They're getting throttled on the product.

46:08Why? Because it's so good.

46:10It makes them better at their business.

46:12We are all self -interested actors.

46:14And when millions of those people are all making the same decision, there's a huge

46:18tell. And the tell here is that the TAM is as big as Dario and

46:23Sam and others have been saying.

46:25We knew intelligence was going to scale on the exponential.

46:28The question was whether revenue will scale on the exponential.

46:31And that's what we're seeing.

46:32And remember, they're doing this with only one and a half to two gigawatts of

46:37compute, right? These guys are massively compute constrained.

46:41They're each going to be adding three gigawatts of compute this year.

46:44And so that will unlock, they would be growing even faster, but for that.

46:49And then Jason, to your point about the open source models that we all want

46:53to be a part of this solution.

46:55I've talked to a lot of big companies, 65 to 70 % of their token

46:58consumption is open source model, right?

47:01Are these cheap Chinese and other tokens?

47:04So these revenue ramps are happening while the world is already using open source.

47:09This is not frontier only.

47:11This is frontier plus open source.

47:13We're going to see massive token optimization over the course of the year.

47:17But what happens on this Jevons pair?

47:19The unit cost of intelligence is plummeting, not the cost of tokens.

47:26The unit cost of intelligence is plummeting because the capabilities of these models is so

47:30much better. I look at what it does for Altimeter day in and day out.

47:34I talked to a major company yesterday.

47:37They're on a run rate to do 100 million of token consumption this year on

47:41about $5 billion in OpEx.

47:43They think that we're now nearing peak employment in their company, but that their intelligence

47:48consumption. Let's not call it token consumption because tokens may go up a lot, but

47:54their intelligence consumption is going to go up a lot.

47:58I would leave you with this.

48:00We're early to Chema's point.

48:03We have low penetration of the Global 2000.

48:05We have low penetration of the use cases.

48:08We have low penetration of within the use cases that they're already using, and the

48:14models are only getting better.

48:15I think when you look out toward the end of the year, I would not

48:19be shocked if you see Anthropic exiting this year at $80 to $100 billion in

48:25revenue. By the way, doing it at the same time, the OpenAI, who is also

48:29on the wave, they'll be releasing an incredible model imminently.

48:34They're going to be on that wave, and you're going to see an inflection in

48:37their revenues as well. Okay, Chamath, question one has been answered.

48:41The question of, hey, does this stuff actually have utility?

48:45That went from a question mark to an explanation point.

48:46Of course, it's got utility.

48:48People are getting value from it, and it might be variable.

48:50Some people get more value than others.

48:51Number two, the revenue ramp was a big question.

48:54Now that's turned into an explanation point.

48:56The final piece of the puzzle that you've brought up many times is, can this

48:59be profitable? These companies are burning through a large amount of cash.

49:05What is your take on when these companies can get out of the J -curve?

49:09We talked about this, I think, three episodes ago.

49:11I estimated like we're going to be looking at $400 or $500 billion in investment

49:15into these data centers at a minimum, and then they have to climb out of

49:20that to get to profitability.

49:22What are your thoughts on these becoming profitable companies?

49:26Do you remember that investor that published this list, Jason, where he put all of

49:33the terms you talk about when one of the terms you can't talk about is

49:37profit? It's a list where it's like, if you can't talk about free cash flow,

49:41you talk about EBITDA. When you can't talk about EBITDA, you talk about margin.

49:47When you can't talk about that, you talk about revenue.

49:49And then when you can't talk about revenue, you talk about gross revenue.

49:53Bookings. So you can kind of figure out, I think, where we are in any

50:00part of any cycle by just indexing into what does everybody talk about.

50:05I think where we are is we are between gross revenue and net revenue.

50:11That's where the discussion is.

50:13There was another article, I think, today, and I think maybe it was the information

50:18that tried to categorize and distinguish that Anthropic presents gross.

50:24OpenAI presents net. They're different.

50:27We don't know what the various take rates are.

50:30So they're saying that there's a difference.

50:32If it's not true, there's been no clarity provided by these companies.

50:35So at a minimum, you have this confusion where there's the breathless talk.

50:39Then there's people that don't even know the difference between actual recognized revenue and run

50:44rate revenue. Totally. And how to multiply.

50:46I mean, so we're definitely there, okay?

50:48We can quibble about the details, but we are not at the place where people

50:51are like, oh, here's your steady state, you know, free cash flow margin.

50:54And here's what your EBIT is.

50:56We're never, we're, we're years from that.

50:58They're going to have token maxing EBITDA, like a cumulative EBITDA at the WeWork.

51:02The thing that we need to understand is how gross margin negative is this revenue

51:06growth. We don't know that.

51:08And at least we don't as outsiders.

51:11Brad might know. Brad may know.

51:13I would tell you, think about this.

51:15What are their big cost inputs?

51:17The number one cost input is the cost to compute.

51:20Cost to compute. Right? I just told you they only have a gigawatt and a

51:24half of compute. And they had that gigawatt and a half of compute, whether they

51:27have a billion in revenue or whether they have 80 billion in revenue.

51:31So you might actually expect to see these companies, their gross margins are exploding higher.

51:36Like the fastest increase in gross margins I've probably seen out of any technology company.

51:41So this is not gross margin negative, you're saying?

51:43No, definitely not gross margin negative.

51:45And what I would tell you is - So then they must be hugely profitable

51:48then. Well, you may see accidental, what I call it accidental profitability.

51:52They may not be able to spend this revenue fast enough, Chamath, on compute.

51:57And remember, it's only 2 ,500 people.

52:00Google crossed this revenue threshold when they had 120 ,000 people.

52:05These guys have 2 ,500 people.

52:07So the only thing you can really spend money on, right, is compute.

52:10And they can't stand up to compute fast enough.

52:12But none of this foots to me then, to be honest.

52:15Because if you were on a threshold of 90 % plus gross margin - I'm

52:20not saying it's there. I'm not saying it's 90 % plus.

52:23I'm just saying it's gone from meaningfully negative 18 months ago to, you know, very,

52:29very positive. I've seen rumored out there 50 to 60%.

52:32Going well is what you're saying.

52:33Right. The trend is there.

52:34Let me just say this.

52:37I think if you're an incumbent, you want the cost of compute to go down.

52:42I think if you're not an incumbent, so specifically, who do I mean?

52:45Meta, Google, and SpaceX. I think those three people who have all three of them,

52:54well, sorry, Meta and Google have a Fortress balance sheet.

52:57I think by the end of June, SpaceX will also have a Fortress balance sheet.

53:01What they will want to do is they will want to make this a compute

53:04problem because they will control the conditions on the field.

53:08You already see this today.

53:10Meta's models today, what people's general reviews are, it's okay.

53:14But the one thing that people say is it's incredibly performant.

53:18The model quality is okay, but the performance is great, which speaks to Meta's huge

53:22advantage. They have a massive compute infrastructure.

53:24So if you're not OpenAI and Anthropic, they'll want to make this a capital problem

53:29because then they can win it.

53:31If you're Anthropic and OpenAI, you want this thing to be as efficient as possible.

53:36I think where we are is very much in the early innings.

53:38And we're bumbling around talking about gross margins and revenues.

53:42We are not at profitability.

53:44And what is true for Facebook and what was true for Google was irrespective of

53:50where they got to a billion, who the f*** cares?

53:52They were profitable by year three and they never looked back.

53:57I was there. I remember.

53:58It was glorious. The cost of building AI totally stipulate is radically higher than the

54:07cost of building retrieval at Google, right?

54:09Like it's just a fundamentally more expensive problem.

54:12But I will tell you that there's a lot of thought out there about negative

54:15gross margins. I mean, Jason, you started this segment by saying they're burning through large

54:19amounts of cash. I think people are going to be shocked at how low the

54:23burn levels are at these companies.

54:25At Anthropic or OpenAI? Yes, and I would say at OpenAI as well, like if

54:29they're on, you know, if they do $50 billion this year, again, just look at

54:33the number of people they have.

54:35Revenue per people is pretty low.

54:36And the inference cost is plummeting.

54:38Inference cost is down by 90 % year over year.

54:41And so just finally, I want to respond to this point about gross versus net,

54:47this tweet that Chamath was referencing.

54:49Okay, so there's a certain percentage, a smallish percentage of Anthropics revenue, right, that they

54:55distribute through the hyperscalers. And like a lot of arrangements, whether it's Snowflake or Databricks

54:59or others, you pay a commission, right, on that.

55:03I will just tell you that you're talking single digit percentage of total revenue of

55:07these companies. So the gross versus net thing isn't what's being reported.

55:11Like the apples for apples is pretty easy.

55:13And if you want to be conservative on it, take down Anthropics revenue by, you

55:17know, five to 10%, which, you know, again, I don't, I think it's better to

55:21gross up OpenAI's revenue, but any way you do it, I just think it's a

55:24distraction from what's really, what's really going on here.

55:27Happy to. Zach, do you have any thoughts on this massive revenue ramp?

55:31Yeah, I mean, I want to go back to a point that Brad made, because

55:34I think it was just really important.

55:36And I want to just underline it.

55:38Consider where we were at the beginning of the year.

55:40And what everybody was saying is that AI was a big bubble.

55:44And the evidence they would point to was the fact that hundreds of billions of

55:49dollars was going into CapEx that needed to be spent on these data centers.

55:53And there was no evidence of significant revenue to justify that spend.

55:58Where was the ROI? By the way, as an aside, the same doomers who were

56:02saying that AI was in a bubble, were also the ones who were saying that

56:05AI was so powerful, it's going to put us all out of work.

56:08And it's going to, you know, take over from humanity.

56:11I mean, in other words, they couldn't decide if AI was too powerful or not

56:15powerful enough. But putting aside that contradiction, they clearly were making this case that AI

56:21was this big bubble and that there'd be no payoff or justification for this massive

56:27CapEx that's being spent. And I think we're starting to see here, there is justification

56:31for it. We're seeing it just in this one vertical of AI, which is coding.

56:36We're, again, seeing the fastest revenue growth in history.

56:40It's utterly unprecedented. And this is just one category or vertical of AI.

56:46We know that agents are coming next.

56:48And the enterprise adoption of that is going to be absolutely massive.

56:52So I guess what I'm saying is that this is early proof for, I think,

56:58the thing that makes Silicon Valley special, which is we're willing to basically bet on

57:03things that just intuitively, on a gut level, we know are the next big thing.

57:08We're not that spreadsheet driven, actually.

57:11Silicon Valley believes that if you build it, they will come and is willing to

57:15finance that build out. And that's basically what's been happening.

57:18Again, just the top four hyperscalers, $350 billion of expected CapEx this year.

57:24On its way, I think Jensen said $1 trillion by 2030.

57:27So Silicon Valley, whether it's big companies, whether it's founders, are always willing to bet

57:31on this next big thing.

57:33They're not like Wall Street.

57:34They don't need, you know, to tell them where to go.

57:38They know where the technology is going and they make their bets based on that.

57:42And I think that there is going to be a big payoff for this.

57:45And I think it's the thing that's going to make our economy and the United

57:49States in general remain extremely dynamic and in the lead on this thing is that

57:55we are willing to make those kinds of bets.

57:56And I think it's going to pay off big time.

57:59Yeah, clearly. Hey, Brad, you didn't answer my question about the vibes over at OpenAI

58:05versus Quad. OpenAI is, I wouldn't say reeling, but there's a lot of hand -wringing

58:12going on. A lot of employees leaving, a lot of...

58:15people who are wondering like is our strategy the winning strategy of like consumer first

58:20they shut down Sora you know unwinding the Disney deal and really trying to get

58:26the company focused and it's kind of like I mean listen the New Yorker story

58:29was a bit of a rehash I don't think we have to go into the

58:31blow by blow because we covered here three years ago but the truth is a

58:36lot of the great founders co -founders of OpenAI and a lot of the great

58:40contributors are now at Anthropic and other large language models and in the secondary market

58:48OpenAI is trading lower than the last valuation and Anthropic is trading significantly above the

58:54380 billion dollar so maybe talk a little bit about this competition this Microsoft versus

59:00Apple this Google versus Facebook well let's let's start with immense credit where credit is

59:06due Anthropic was literally counted out of the game last year yeah right and here

59:11they come over the last 12 months and and they've kicked OpenAI's ass over the

59:15last 90 days right and what did Anthropic do Anthropic made choices no multimodal no

59:22video no hardware no chips no building data centers they said we're just going to

59:26focus on coding and co -work we think that is the path to AGI and

59:30and and and ASI they executed their butts off they took the lead 2500 people

59:35tight pulling on the ore in the same direction but I think you would be

59:41seriously foolish to count out OpenAI right and I think we're why we're at peak

59:46OpenAI FUD and I'll tell you it starts with great researchers and great models and

59:50I think when you see the spud model they're about ready to release I think

59:54it's going to be an excellent model shows that they're firmly on the wave if

59:59you look at what's going on with codex incredible ramp on codex fastest ramping model

1:00:05with 5 .4 I think 5 .5 or spud whatever we're going to call it's

1:00:08going to be an even faster ramp have you seen spud have you used it

1:00:12have you gotten a preview people are using spud right so it is being previewed

1:00:17and so so you're talking to people who've used it and what are they telling

1:00:20you they're telling us that it's an incredible model on par with mythos right and

1:00:25that it's a very usable model in terms of um how it's packaged I will

1:00:30say that back to David's point now this is the most important point I think

1:00:35anybody can take away here this is not zero sum the TAM of intelligence is

1:00:43dramatically larger than any TAM we've ever seen in our investing careers over the last

1:00:47two decades right and if you're on the wave which OpenAI is you are going

1:00:52to be selling into the world's biggest TAM they are going to build a very

1:00:56big company I'm a buyer of the shares today notwithstanding all of the vibes that

1:01:01you describe I think these companies are firmly on the wave they are jarred they

1:01:07are sitting there saying what did we do wrong and how do we get our

1:01:10mojo back they want to compete it is embarrassing to people on the research team

1:01:14and the product team over there so I'm not saying there's not a real awakening

1:01:18occurring there but I think that's what the case is and by the way to

1:01:22Chema's point do not count out meta right I think meta is absolutely in this

1:01:27game Google is absolutely in this game Elon is absolutely in this game and if

1:01:30you're on got some stuff dropping shortly that's going to be very impressive if you're

1:01:34on team America the fact that we have five frontier models competing against each other

1:01:40and David made sure they weren't throttled by excessive government regulation we have mythos come

1:01:45out it's a self -imposed safe harbor you know to harden our system is it

1:01:50wasn't a call for moratoriums or getting the government involved We have the type of

1:01:54competition that's causing us to accelerate our lead against the rest of the world.

1:01:59We can't take our eye off the prize.

1:02:00We got to stop adversarial distillation, and we need to make sure that we're distributing

1:02:04our products around the world.

1:02:06But I view this as really good for Team America.

1:02:09Well said. And here is your polymarket IPOs before 2027.

1:02:14Obviously, SpaceX at 95 percent, Cerebrus at 94 percent.

1:02:19And hey, number five on this list, 51 percent chance that Anthropic goes out before

1:02:25the end of the year, 44 percent chance that OpenAI comes out before then.

1:02:30All right. Here is the closing market cap for Anthropic on Polymarket, only $158 ,000

1:02:38in volume. So Chamath, when you put in 400K, you're going to really tilt this

1:02:42market. 78 percent chance that it's above 600 billion.

1:02:4719 percent chance that it doesn't go out.

1:02:50So it's looking like this will be a decent investment for you.

1:02:54Brad, what valuation did you get into Anthropica?

1:02:57We first invested, and I believe it was the $130 or $150 billion round.

1:03:03So this will be a 7X, 5X for Altimeter.

1:03:06Al, please, congratulations. I mean, listen, again, there are lots of people who were there

1:03:12before us and who are on the board and who are going to do better

1:03:14than us. What did you put in, 50, 100?

1:03:17What did you put in?

1:03:18No, we've got billions in both companies.

1:03:20Billions in both companies. Oh, my Lord.

1:03:23Yum, yum. I think there's this existential thing going on in venture today.

1:03:28David could talk about it as well.

1:03:29I mean, people can't. They're extraordinarily nervous about.

1:03:33You look at the IGV stock index, down 30 % year to date, down 5

1:03:39% today, all software stocks plummeting, right?

1:03:42Venture capitalists are terrified to invest money in anything other than these frontier models and

1:03:49things like SpaceX or military modernization.

1:03:52Finding something that's out of harm's way of AI, right, where you can count on

1:03:57the terminal value to Chamas Insights over the last few weeks is very difficult to

1:04:02do. That's why you see this crowding.

1:04:03So we've taken a barbell approach, right?

1:04:06We've got a lot in what we think are the most important companies that are

1:04:09on the frontier. And then we're betting on really small teams that we think have

1:04:13very defensible businesses in a world of, you know, AGI.

1:04:17But it's tricky. What happens to all these enterprise software companies?

1:04:20Do they become PE takeouts?

1:04:21Do they get consolidated or do they just have to adopt these AI technologies and

1:04:28solve this problem of, hey, the frontier model is just going to solve for whatever

1:04:34these niche software companies do?

1:04:36I think the market's probably being a little too pessimistic with respect to at least

1:04:40some of these software companies.

1:04:42I mean, obviously, there's going to be big differences in the quality of the moats

1:04:46of these companies. And so, look, software is going to be a lot cheaper and

1:04:52easier to generate, but I'm not sure that was the competitive advantage of a lot

1:04:56of these companies. So there's probably a little bit of the baby being thrown out

1:04:59with the bathwater right now, and there probably are some value buys in enterprise software.

1:05:04I think the interesting question here, and we've been talking about this for a couple

1:05:08of years in the pod, is just where you see the AI value capture being

1:05:13in terms of layer of the stack.

1:05:15Remember where we started. It was really just the chip layer of the stack was

1:05:18where all the value capture was.

1:05:20It was basically NVIDIA was the first company to be worth multiple trillions of dollars

1:05:24because of AI. And for a while, it looked like that's where all the value

1:05:29capture was going to be because OpenAI, for example, was losing.

1:05:32so much money and Anthropic wasn't on the radar as much.

1:05:35Now we're seeing, well, wait a second, you know, it's not just the chip companies.

1:05:40It's also the hyperscalers are now benefiting.

1:05:42And now we're seeing at the model layer, it looks like Anthropic and OpenAI, they're

1:05:47all going to be huge beneficiaries.

1:05:48I think the next question is at the application layer of the stack.

1:05:52Okay, well now does all that value capture just get eaten by the model companies

1:05:55or are there applications that get turbocharged?

1:05:58I guess you could say that Palantir is already one of them.

1:06:02Right. It's an application company that's been turbocharged by these model capabilities.

1:06:06Who else will be a big beneficiary?

1:06:09Again, is it all going to be at the model layer or will you see

1:06:13an explosion of value at the application layer?

1:06:15I'm hoping obviously that it'll be at all layers the stack you see beneficiaries.

1:06:20But to me, that's a really interesting question right now.

1:06:23Yeah. What happens to Salesforce, HubSpot, Oracle, right down the line, David Chamath, your thoughts

1:06:28here on the layers here and where the value is captured?

1:06:34It's too early to tell.

1:06:35Too early to tell, right.

1:06:36And energy you kind of put into sort of data center as well, but that's

1:06:40obviously been a clear winner.

1:06:42A little housekeeping here. Liquidity, put a little Tiffany in here, producer Nick, is sold

1:06:48out. There's a wait list of hundreds of people, but it is what it is,

1:06:52folks. If you snooze, you'll lose.

1:06:54And top tier speakers are coming.

1:06:57It's going to be great.

1:06:58We'll get an update from Chamath, but I think, Brad, you're going to be joining

1:07:00us again. Yes, for Liquidity?

1:07:02I have an update. That's probably not your headliner, though.

1:07:05I'm probably not your headliner.

1:07:06No, but you always score so high.

1:07:08Every event you've spoken at, you've been either number one, two.

1:07:11I don't think you've ever dropped to three.

1:07:13Go ahead, Chamath. Make your announcement here.

1:07:16Nat sent me an article from Wikipedia about penile links when you guys are talking

1:07:21about Dick Bits. Okay, this is breaking news.

1:07:22Showing me that I'm in the large category, top 5%.

1:07:26She highlighted it. Top 5%.

1:07:28Okay. And that's with, is that with Nano Banana or without?

1:07:32Is that with the generator banana?

1:07:33She just texted, dummy, it's Claude.

1:07:35My apologies, Claude. Oh. All right.

1:07:38This is why Chamath isn't afraid of the cybers, because nothing's going to come out

1:07:42that's more embarrassing than what he says himself on the pod.

1:07:44It's like Bezos. When Bezos got hacked, he's like, guys, I got hacked.

1:07:48So I saw the agenda for this thing.

1:07:50It's incredible. Congrats to you guys.

1:07:52I mean, like the, like just the fun of being in Napa, all the poker,

1:07:56all the dining experience. This is five star all the way.

1:08:00It looks really cool. Six star.

1:08:01It's a mom level, because Chamath was, dare I say, belligerent in his demands.

1:08:08He said, this has to be six star or I will not show up, Jake.

1:08:12Al. I said, okay, boss, boss, get to work.

1:08:15And Chamath, what do you got?

1:08:17And no mids. This is all elite.

1:08:19And for the hundreds of people who are on the wait list, I am sorry,

1:08:21but we have a capacity issue.

1:08:23We'll try to get you in for next year.

1:08:24But Chamath, give us some updates here.

1:08:26You have any updates that you want to share?

1:08:27Because you are running programming for Liquidity 2026 up in Yon.

1:08:31Look, it's going really well.

1:08:34Really excited to hear all of these great folks speak.

1:08:37I think the next two will release today, Brad Gerstner and Thomas Lafont of KOTU.

1:08:43Ooh, KOTU. That's a great get.

1:08:45We also have, I think, three people confirmed for their best ideas pitch.

1:08:49Really interesting folks. They each run between one and six or seven billion.

1:08:55Awesome. Superstar compounders early in their career.

1:08:58This is the new zone.

1:08:59This is the new zone, Chamath.

1:09:00It's great. So right now, we have Bill Ackman.

1:09:03We have Andre Karpathy. We have Dan Loeb.

1:09:06We have Thomas Lafont. We have Brad Gerstner.

1:09:08We have Sarah Fryer. And more.

1:09:10to come we will announce more there might be one or two surprises jay cal

1:09:13and a couple and a couple of surprises yeah we we don't announce all the

1:09:18speakers jay cal's got a couple of surprises coming and if you didn't get in

1:09:23to liquidity apologies you're on the wait list we are going to be hosting the

1:09:28fifth annual all -in summit in los angeles september 13th to the 15th sax you're

1:09:36going to come to that all in .com sessions sax you should come to that

1:09:41i've been advised that i can attend business i can be in the state for

1:09:44business reasons okay there you go then we'll see you at liquidity and the summit

1:09:48correct that's that's big news now we just got a bunch of sax stands who

1:09:52are racing uh and now we're gonna get sax at this is what happens every

1:09:56year behind the scenes sax at the last minute says oh i have four speakers

1:10:00and i have 72 people who need tickets and then the whole team has to

1:10:04like do a fire drill 48 hours before the event okay here we go guys

1:10:08we're going to go to the third rail here we got to catch up on

1:10:11the iran war here's the latest two weeks into a ceasefire i've started just two

1:10:17days ago at the taping of this vp jd vance friend of the pod is

1:10:22a and some special consultants uh wikoff and friend of the pod jared kushner are

1:10:27headed to islamabad the capital of pakistan for talks this very weekend so while you're

1:10:33listening to this event they are going to be working on the peace deal easter

1:10:36sunday trump posted a truth stating open the straight and crazy bastards or you're going

1:10:42to be living in hell just watch praise be to allah on tuesday morning trump

1:10:47posted uh another threat on social media a whole civilization will die tonight never to

1:10:53be brought back again i don't want that to happen but it probably will tweets

1:10:57were obviously discussed uh a lot over the last week he gave him an 8

1:11:01p .m deadline at 6 30 p .m potis announced on truth social that he

1:11:06had agreed the president trump had agreed to a two -week ceasefire if iran opens

1:11:12the straight he also said hey listen we got the straight maybe there'll be a

1:11:16toll booth but we'll take the majority of the toll and we'll split it with

1:11:19iran here's the quote we received a 10 point proposal from iran and we believe

1:11:24it's a workable it is a workable basis on which to negotiate and apparently netanyahu

1:11:30took the ceasefire to mean level lebanon dropping 160 bombs in 10 minutes yesterday sacks

1:11:37uh you were out last week everybody wants to know your position on the war

1:11:40i'll hand it off to you what are your thoughts on how on the two

1:11:44-week ceasefire and everything that's occurred up until this point well look i have to

1:11:48preface what i'm about to say which is i'm not part of the foreign policy

1:11:53team at the white house and the last time i commented on the war on

1:11:57the show it somehow made international headlines that trump advisor says xyz and i'm not

1:12:06a trump advisor on this issue i think that'd be a fair headline to write

1:12:09if it was a technology issue but this is not so whatever i say is

1:12:14just my personal opinion but then the media is going to somehow portray it or

1:12:17attribute it to the white house or try and create an issue out of it

1:12:21so i feel like i'm limited in what i can say except that to say

1:12:25that i think it's terrific that we have the ceasefire i think it's great that

1:12:31there's going to be this meeting in islamabad to hammer it out and i think

1:12:36what the president's accomplished so far with the ceasefire is it's a great thing because

1:12:41what happens with these wars is they take on a life of their own meaning

1:12:45they tend to go up the escalation ladder right there's a lot of pockets and