YouTubeFeed

The Fight Over Open Source AI, Anthropic’s $1.5B Payout, NYC Socialists: Evictions = Violence?

Summary

The besties dig into the week’s biggest AI story: China’s Moonshot AI released the open-source Kimi K3, scoring on par with Opus 4.8 and GPT-5.6 at roughly half the cost, reigniting DeepSeek-moment panic and pushing the debate over banning Chinese open-source models all the way to the White House. Sacks lays out his thesis that Anthropic is running a regulatory-capture play — coining the term “industrial scale distillation attacks” to scare policymakers into a government-protected duopoly — and argues the tell is that Anthropic wants to ban American developers from using Chinese public-domain outputs rather than fixing its own KYC failures. Chamath explains distillation mechanically (asking a model questions and training on its outputs), calls the moment a “valuation preservation game” by closed frontier labs whose models are getting commoditized within weeks of release, and warns that any government intervention forcing US companies to pay 50-100x more for AI than the rest of the world would “tank the stock market.” Friedberg adds a free-speech angle (you can’t easily un-ring the bell on downloaded open-weight files) and a sweeping historical argument that open source AI will do to intelligence what Apache and Firefox did to the web — diffusing value to millions of businesses instead of a handful of gatekeepers.

The second segment turns to Anthropic’s $1.5 billion piracy settlement over pirated Libgen books. Chamath argues the judgment exposes Anthropic’s own hypocrisy: the company insists it’s fair use to train on the world’s copyrighted output without permission, yet claims Chinese distillation of its own model outputs is uniquely off-limits — and notably, its blog post coining “industrial scale distillation attacks” never actually uses the words “IP theft,” because doing so would torch its own fair-use defense in the ongoing New York Times and other copyright suits. Sacks piles on with the Thinking Machines and Cursor Composer 2 examples — both built on the “tainted” Chinese Qwen 2.5 model — to argue American open-source companies would be collateral damage in any crackdown. Friedberg warns this over-reach could invite content creators to demand ownership of Anthropic’s entire revenue stream.

The show closes with Google and Tesla earnings (Google’s CapEx guidance jumping to $195-205B and negative free cash flow for the first time ever, against a 32% 25-year average return on invested capital that makes Chamath more bullish, not less), a SpaceX IPO down 30% from its debut close, and a long “Socialism Corner” segment on NYC’s new rental rip-off law — Friedberg delivers a John Quincy Adams-quoting monologue on private property rights as the foundation of American liberty, while Sacks argues that banning evictions and credit checks will turn rent-controlled buildings into “the equivalent of housing projects” and produce more of New York’s 50,000 “ghost apartments.”

Highlights

”Anthropic would push to ban Chinese access to American models, not American access to Chinese models”

Sacks on the tell that distillation panic is fake

“The tell on this, the way that you know that this whole distillation thing is fake, is because if stopping distillation was their primary objective, Anthropic would push to ban Chinese access to American models, not American access to Chinese models.” — David Sacks, 3:55

Clip command
yt-dlp --download-sections "*3:55-5:41" "https://www.youtube.com/watch?v=wcV0SRPFK9s" --force-keyframes-at-cuts --merge-output-format mp4 -o "wcV0SRPFK9s-3m55s.mp4"

”Why would you buy $50 Coke when you can buy 50 cent Coke?”

Chamath's Coca-Cola analogy for banning open source

“It’s not just beverage choice. You have two beverage choices, but they cost 50 times more than Coke outside of America… So you can buy 50 cent Coke or $50 Coke. Why would you buy $50 Coke when you can buy 50 cent Coke?” — Chamath Palihapitiya, 14:07

Clip command
yt-dlp --download-sections "*13:58-14:24" "https://www.youtube.com/watch?v=wcV0SRPFK9s" --force-keyframes-at-cuts --merge-output-format mp4 -o "wcV0SRPFK9s-14m07s.mp4"

Thinking Machines and Cursor were both built on the “tainted” Chinese model

Sacks on the hypocrisy of the distillation-taint argument

“Thinking Machines. They currently have the best American open source model. You know how it was trained? It was bootstrapped. It was distilled off a Chinese model, Qwen 2.5… if you say that American companies can’t do that or that somehow it’s tainted with IP theft, you are going to basically put a dagger through the heart of the entire American open source ecosystem, and that is exactly what Anthropic wants because they do not want to have the competition.” — David Sacks, 39:46

Clip command
yt-dlp --download-sections "*39:44-41:31" "https://www.youtube.com/watch?v=wcV0SRPFK9s" --force-keyframes-at-cuts --merge-output-format mp4 -o "wcV0SRPFK9s-39m46s.mp4"

”They did not have the chutzpah to claim that it was IP theft”

Chamath on Anthropic's hypocrisy after the $1.5B settlement

“It is still Anthropic’s position, and it’s OpenAI’s position, that they should be able to train on all of these books under fair use if they buy one copy… You should be able to see the total hypocrisy of their point of view… if you go to that blog post, search for the words IP theft. It’s not in there… they did not have the chutzpah to claim that it was IP theft.” — Chamath Palihapitiya, 51:19

Clip command
yt-dlp --download-sections "*51:19-53:49" "https://www.youtube.com/watch?v=wcV0SRPFK9s" --force-keyframes-at-cuts --merge-output-format mp4 -o "wcV0SRPFK9s-51m19s.mp4"

The molecule economy vs. the knowledge economy: China’s long game

Friedberg on China's molecule-conversion strategy

“If the knowledge economy and the services economy gets compressed… the value of the US and the Western economy has been largely degraded. And what’s left is the value of the molecule economy… they have 20x the manufacturing capacity, 8x the electricity production plus all of their other sources of energy. And I think that’s the long game for China.” — David Friedberg, 41:36

Clip command
yt-dlp --download-sections "*41:36-44:36" "https://www.youtube.com/watch?v=wcV0SRPFK9s" --force-keyframes-at-cuts --merge-output-format mp4 -o "wcV0SRPFK9s-41m36s.mp4"

”Property must be secured or liberty cannot exist”

Friedberg's Socialism Corner rant on private property rights

“In 1787, John Quincy Adams published a work called A Defense of the Constitutions of Government of the United States of America… ‘The moment the idea is admitted into society that property is not as sacred as the laws of God… anarchy and tyranny commence.’ And then in 1791: ‘Property must be secured or liberty cannot exist.’” — David Friedberg, 79:04

Clip command
yt-dlp --download-sections "*79:04-82:56" "https://www.youtube.com/watch?v=wcV0SRPFK9s" --force-keyframes-at-cuts --merge-output-format mp4 -o "wcV0SRPFK9s-79m04s.mp4"

Key Points

  • Kimi K3 sparks a new DeepSeek moment (0:00) - China’s Moonshot AI ships an open model on par with Opus 4.8/GPT-5.6 at ~50% lower cost; Polymarket odds of a US open-source ban jumped from 22% to 45% in days.
  • White House palace intrigue (0:59) - Axios reports a possible ban under consideration; Commerce Secretary Lutnick reportedly opposes it, wanting to incentivize better US open models instead.
  • Sacks: regulatory capture, not national security (2:01) - Anthropic went from $10B to $70B+ ARR this year; Sacks argues a company growing that fast doesn’t need government protection from competitors.
  • KYC as the actual fix (8:13) - Chamath: distillation could be curbed with real identity verification on accounts; Anthropic hasn’t done it because it would slow growth.
  • Distillation is industry-standard, not new (15:12) - Friedberg draws the parallel to early Google benchmarking against Yahoo/Microsoft search results — using output, not stealing code.
  • Weights vs. outputs (17:13) - Sacks draws the crucial technical distinction policymakers miss: stealing model weights is theft; learning from outputs (what OpenAI/Anthropic themselves do to the entire internet) is not.
  • The Coca-Cola cost-tax thought experiment (11:47) - Chamath models what happens to a company like Coca-Cola forced to pay 50-100x more for AI than global competitors — a “token tax” that would force a market re-rating.
  • Open source as the internet’s Apache/Firefox moment (20:15) - Friedberg’s historical case: Netscape’s proprietary browser got crushed by open-source Firefox/Apache, and value flowed to millions of businesses instead of gatekeepers.
  • 95% of tasks, many models (35:44) - Chamath’s framing: open/commodity models can already do 95% of AI tasks; frontier labs should capture value in the application layer and vertical products, not try to defend foundation-model moats.
  • Frontier labs are still accelerating (30:41) - Sacks: Anthropic is tracking to $1B+ ARR mid-year vs. $100M in January; OpenAI’s ARR estimate rose from $33B (May) to $41.3B (July).
  • Anthropic and OpenAI “flopping” (32:48) - Sacks accuses the labs of drawing a foul from the government while mid-roadshow, to preempt the obvious investor question about commoditization.
  • Thinking Machines and Cursor built on Qwen 2.5 (39:44) - Both leading American open-source/coding products derive from a Chinese open-weight model, undercutting the “tainted” distillation argument.
  • The $1.5B piracy settlement, explained (51:19) - Anthropic lost not for training on copyrighted books but for pirating them via Libgen without ever buying a single copy.
  • “IP theft” never appears in Anthropic’s own blog post (51:19) - Chamath: the company coined “industrial scale distillation attacks” but deliberately avoided the IP-theft framing because it would undermine its own fair-use defenses.
  • Garry Tan and 200 startups push back (64:33) - Friedberg: if the IP-theft framing sticks, every derivative model built on distilled data — including American startups’ — is tainted, activating the whole startup ecosystem against Anthropic’s position.
  • Google’s CapEx jumps to $195-205B (67:15) - First-ever negative free cash flow quarter; stock down 7-10% on the news despite Google Cloud growing 82% YoY.
  • 32% 25-year average ROIC (69:55) - Chamath uses Google’s historical capital efficiency to argue the CapEx surge is a buy signal, not a red flag.
  • SpaceX down 30% from IPO close (67:38) - Now trading around $1.5T after debuting near $2T, with staggered lock-up expirations adding pressure.
  • NYC’s rental rip-off law (77:38) - Bars application fees for credit checks, requires landlords choose either a credit check or 40x-income standard (not both), legally recognizes tenant unions, and effectively froze rent for a year.
  • “Evictions are violence” (78:10) - A masked activist’s quote at the City Council hearing that triggers Friedberg’s private-property-rights monologue.
  • Austin’s permitting-reform proof point (87:02) - Chamath: relaxed permitting directly drives down rent per unit — the same political capital Mamdani spent on tenant protections could have gone to zoning reform instead.
  • New York’s 50,000 ghost apartments (91:05) - Friedberg: expensive renovation code requirements plus rent control make it rational for landlords to leave units empty rather than re-rent them (and Airbnb is banned in NYC too).

Mentions

Companies

  • Moonshot AI (0:00) - Chinese lab behind Kimi K3, accused by the White House of distilling Anthropic’s Fable model.
  • Anthropic (2:01) - Central subject of the distillation debate and the $1.5B Libgen piracy settlement.
  • OpenAI (9:31) - Also accused of distilling the New York Times; ARR estimate rose to $41.3B in July.
  • Thinking Machines (39:44) - Mira Murati’s startup; its leading American open-source model was bootstrapped off Chinese Qwen 2.5.
  • Cursor (39:46) - Composer 2 was post-trained on Qwen 2.5 plus proprietary coding data.
  • Google (67:06) - Blew past CapEx expectations ($195-205B), Cloud growing 82% YoY, first negative free cash flow quarter ever.
  • Tesla (67:21) - CapEx surged 140% YoY to an expected $25B; stock down ~14% around the taping.
  • SpaceX (67:38) - Trading ~$1.5T, down 30% from its $2T IPO-day close, with staged lock-up expirations.
  • Lovable / ElevenLabs (34:00) - Cited as high-spend Anthropic/OpenAI customers now moving to open-source models.
  • New York Times (9:31) - Suing OpenAI over scraping its site in violation of terms of service.
  • Thomson Reuters / Ross (Westlaw case) (58:40) - Cited as the key precedent where training on a direct competitor’s output crossed the legal line.

Products & Technologies

  • Kimi K3 (0:00) - Chinese open-source model on par with Opus 4.8/GPT-5.6 at ~50% lower cost; Ben Thompson’s cost analysis disputes the “much cheaper” claim.
  • Qwen 2.5 (39:46) - Alibaba’s open-weight model, the base for both Thinking Machines’ and Cursor’s American open-source products.
  • Distillation (8:13) - Chamath’s definition: querying a model at scale and training on its outputs.
  • GPT-6.0 (30:00) - Referenced as Sam Altman’s unreleased next model, said to be heading to Washington for discussion.
  • Codex (32:40) - Cited as reaccelerating OpenAI’s growth and taking market share.

People

  • Michael Kratsios (0:59) - White House official who publicly linked Moonshot AI to Anthropic’s Fable distillation.
  • Howard Lutnick (0:59) - Commerce Secretary reportedly opposed to banning Chinese open models.
  • Mira Murati (39:29) - Thinking Machines founder, whose model is built on Qwen 2.5.
  • Sam Altman (30:38) - Reportedly heading to Washington to discuss GPT-6.0.
  • Garry Tan (64:33) - Y Combinator president; organized a 200-startup letter opposing the IP-theft framing.
  • Ben Thompson (28:54) - Stratechery writer whose cost analysis undercut the “Kimi K3 is much cheaper” narrative.
  • Zohran Mamdani (77:27) - NYC mayor behind the rental rip-off law discussed in Socialism Corner.
  • John Quincy Adams (79:04) - Quoted at length by Friedberg on private property rights as the foundation of liberty.

Surprising Quotes

“They should hire Lina Khan.” — Chamath Palihapitiya, 39:28

“IP for me, but not for thee.” — Jason Calacanis, 53:46

“If you go to Washington and you lay down with the dogs, don’t be surprised if you wake up with the fleas.” — Jason Calacanis, 65:40

“All anarchies end up in tyranny… anarchy and tyranny are one and the same.” — David Friedberg, 79:46

“The interesting question is, if you were forced to live under these rules as a landlord, what would you do?… rents will not go down. Rents will go up.” — Chamath Palihapitiya, 90:04

Transcript

Jason Calacanis: 0:00 All right everybody, welcome back. Episode 282 of the world’s greatest podcast. It’s your podcasters’ favorite podcast, it’s your mom’s favorite podcast, it’s the All-In podcast. With me again, David Sacks from my blog to you, David Friedberg. It was a big week, it was a big week. Uh, the continuing number one story in the world is Kimmy K3. It sparked a debate about banning Chinese open source models here in the United States. And it’s gone all the way to the White House. Last Friday we talked about it here. China’s Moonshot AI released Kimmy K3, open source model obviously. Performance on par, on par. Not six months behind, not 12 months behind, but now on par with models like Opus 4.8 and GPT 5.6. Which in and of itself is extraordinary, but about 50% cheaper. And this has created a bit of a panic similar to the DeepSeek moment that we had here back in early 2025. The White House, David Sacks, has gotten involved. Michael Kratsios, friend of the show, said quote, “We have information that Moonshot AI distilled Anthropic’s fable for the development of its K3 model.” Here’s how the Trump administration has reacted so far. Monday, Axios reported the White House was considering banning Chinese open source models. A couple of weeks ago, David, I think it was three weeks ago, I gave that to you as a hypothetical and here we are. On Wednesday, Wired reported that Howard Lutnick from our Commerce Department, friend of the show, does not want to ban Chinese models. So apparently, palace intrigue, there might be different opinions inside Trump’s White House. And instead they want to incentivize more US frontier labs to develop better open source models. Polymarket says 45% chance US government bans an open source model in 2026. Uh, that was a brand new market, started just… it was at 22% a couple days ago. Sacks, you called it.

David Sacks: 2:01 Two months ago, our first victory fap of the ep. I think where it’s all leading to is an effort to ban open source models. There’s a lot of breadcrumbs leading here. If you look at a lot of the rhetoric around how models need to have guardrails and that with open source models, the guardrails can be removed and therefore they’re dangerous. You see this rhetoric already in Anthropic’s blog posts. Any threat that they describe, they kind of go out of their way to take that shot at open source models. I think again, they’re trying to create ideas or put predicate facts in the public record to justify an action later on. I think it’s just a matter of time before they feel like they’re at a position where maybe they can push for that type of ban directly.

Jason Calacanis: 2:47 All right, there it is Sacks. What’s going on at the White House? What is the administration’s position here? Why are we getting multiple… is the White House testing and and probing to figure out what their position is here? …or is it just this is a super dynamic situation? What’s going on?

Chamath Palihapitiya: 3:03 Well look, I mean, I have it on good authority that there is no decision by the White House to ban open source models and I think they want that known. I think there’s an ongoing conversation happening around what to do about Chinese distillation and we should talk about that. But no decision has been made and the President listens to a chorus of voices, he wants to get advice from as many people as possible, and I’m confident that if everybody weighs in, that the President will make the right decision as he always has with these tech issues. I think his instincts have been absolutely impeccable on this and he’s always supported a let’s say lighter regulation, more open approach, and that’s why I think the US is winning the AI race. So I think that’s kind of where where things stand.

Jason Calacanis: 3:51 Where do you stand? Where do you stand, Sacks? That’s what everybody wants to know.

David Sacks: 3:55 Yeah, I think it’s important for me to make my opinion known in the spirit of again contributing my voice so the President hears all perspectives and then can make the best decision. Look, I think it would be a tragic mistake if the government were to take action against the open source ecosystem. That would do nothing but hurt America’s position in this AI race. It would backfire badly and I think that the key point here is that regardless of what you think about distillation, you cannot punish American developers for it. So you know, you can’t say that American companies and American developers can’t use Chinese contributions to the public domain. That’s just cutting off our nose to spite our face. I mean obviously American companies have to be able to use everything that’s in the public domain because the rest of the world will be using those things. And let me just say, I’ve said for a while that Anthropic is guilty of regulatory capture attempts to seek government protection. This is a company that is the fastest growing tech company at scale in history. They started the year at 10 billion of ARR, they’re now over 70 billion of ARR. This is not a company that needs government protection. This is not a company that is under threat from competitors, whether they’re Chinese or otherwise. And yet they have been very successful at trying to panic everybody into thinking that they need some sort of government protection. And the tell on this, the way that you know that this whole distillation thing is fake, is because if stopping distillation was their primary objective, Anthropic would push to ban Chinese access to American models, not American access to Chinese models.

Jason Calacanis: 5:41 Yes, they could and that is that is achievable, they could block it, so…

David Sacks: 5:45 They’re the ones in the best position to block it. If distillation, if industrial-scale distillation is a national security threat, they’re the ones who need to stop it because that is the place where distillation occurs. You have to stop it at the source. Once you allow Chinese companies to distill… The horse is out of the barn. Yeah. And the reality is that I think that what’s happening here is that in their lust for growth, Anthropic has done a very poor job at stopping distillation. I mean, they’re saying that distillation is occurring at industrial scale. Okay, if it’s industrial scale, it must be pretty obvious to see. So why aren’t you stopping it?

Jason Calacanis: 6:21 The way it’s happening is waves of accounts being created by students rolled up and sold on the dark web, you know, in those kind of channels. So you got people in Manila, in the Philippines I understand, and India, signing up for all those accounts and then sending them to the dark web and selling them using IP addresses from America. Yeah, so that’s how it occurs.

David Sacks: 6:42 Sure, but the more, the more industrial scale it is, the more obvious it is to see. Absolutely. And Chamath has been saying for a while, why don’t you KYC your customers? Exactly. Well, they know that if they KYC their customers, it’ll slow their growth. So instead, what they’re saying is, ‘Hey, ban our competitors.’ Well, that’s ridiculous. I mean, they’re in the best position to stop the distillation. I think that they’re negligent about doing that. I mean, if they really think it’s that big a threat, they should use a few points of their 90% gross margins to do that. What you don’t do is then say that American developers cannot use everything that’s in the public domain. So it seems, it seems to me that this debate is all backwards, that the question should be on Anthropic to explain why it’s doing such a bad job, not on the whole American open source ecosystem to be punished for Anthropic’s failure.

Jason Calacanis: 7:34 Alright, Freeberg, I have a really good question for you, but before we do that, Chamath, can you give me maybe a little bit of what you’re hearing on your sales calls for 80/90? You’re talking to enterprises. They’re hearing all these reports, whether it’s you, Dario, this pod, other places talking about, ‘Hey, open source is ready. This is the moment. Get control. AI sovereignty, etc.’ They must be calling you up and saying, ‘Hey, okay, we’re ready. Like, how do we get these things on-prem? How do we do it?’ So what’s the, what’s happening on those calls you’re doing with the enterprise? And then can you maybe give people an idea of what distillation is, just and why it’s so important here?

Chamath Palihapitiya: 8:13 Let’s start with the second thing. Distillation is when you fire up a model and you ask it a question and you observe it and you take its output and you use that in training of your own model. Now multiply that behavior by tens of millions and what you exfiltrate is essentially trillions of questions and answers. And Sacks is right. If you really care about distillation, you implement KYC. You force people to make an account, not just with a username and a password, but with some form of identification, maybe with a bounded credit card. There’s all kinds of steps that you can take that would frankly slow things down in terms of revenue traction but would solve the distillation problem on its face. So that isn’t really a thing. It’s a bit of a red herring. The other thing on distillation is everybody has at some point distilled. The question is who is distilling from whom? And it looks like that funny meme where there’s like nine Spiderman all pointing at each other. That’s what this is.

Jason Calacanis: 9:31 Because Anthropic has distilled from all of these publishers. They just pay a $1.5 billion fine. Apparently, OpenAI distilled from the New York Times. There’s still an ongoing lawsuit. The Chinese labs have distilled from Anthropic. Wholesale stealing everywhere. Yeah.

Chamath Palihapitiya: 9:48 Well, I don’t want to call it stealing because it’s not clear who actually owns the copyright in the first place. But here should be the important observation. These models are getting commoditized much faster than anybody thought. And how do we know this? Because there is no meaningful sustained advantage once a model publishes their performance criteria. What you see is literally within weeks other models, some open, some closed, some open weight, who are able to match and in some case exceed the performance. So I think what’s happening here is a handful of American companies have realized whoa, this value that we are seeing today may not be sustainable in a five and ten year period. And when you go and present a business model to Wall Street, you need to have that certainty, otherwise it impacts your valuation. And so I think a lot of what’s happening right now, Jason, is a valuation preservation game by the closed frontier labs. Because if you actually understood how commoditized these things are becoming and the velocity at which it’s happening, you see that the real business model is not in the foundational modeling anymore. It’s in the application layer above and it’s in the infrastructure below, whether that’s the cloud or whether that’s chips. And so I think in the absence of regulatory intervention and in the absence of the United States government stepping in to put their thumb on the scale, what will happen is that as people learn about how value is changing, they’re going to put more value in the application layer and more value in the infrastructure layer. That is bad for the closed frontier labs, especially when they’re mispriced 25 to 50x the open alternative. And so this is an attempt to stop a competitor that is of the same quality but just much cheaper. Now, there’s another important thing here, which is if the United States government intervenes, it will tank the stock market.

Jason Calacanis: 11:46 Okay.

Chamath Palihapitiya: 11:47 Period. Not debatable. Now you can debate which companies get tanked and we can probably play that scenario. …out. But for example, if they said, ‘No more open source, American companies cannot use open source.’ Okay, let’s just look at it from a stock perspective, let’s take an average normal company, Coca-Cola. Hey Coca-Cola, you’re trying to use AI to improve your business. You know what? You can only use these two options, and those things cost 50 to 100 times more than your other best alternative that you may use otherwise. That’ll eventually show up in your costs because AI is supposed to be this incredible thing that just kind of solves every problem and does everything for you. And so this incredibly important input into your cost model is now orders of magnitude, multiples greater than your competitors that are outside the United States simply because you’re in the United States. So what would the capital markets do? They’re going to say, ‘Wow, you have a crazy cost structure here. This doesn’t make sense. You’re forced to absorb costs that aren’t rational nor market driven.’ So then Coca-Cola has to get re-rated. But then you look at the people who are selling those tokens, and this is where Anthropic and OpenAI need to understand: if the government comes in and actually tells you that there’s no open source, their valuation will crater. Why? Because all of that revenue is artificially being propped up. It’s not being driven by market demand where you’re being forced to compete. It’s because of regulatory capture where you now get an artificial constraint, but it only works in one market. And so, anyways, all roads lead to market chaos if anybody gets involved. So we should just not get involved.

David Sacks: 13:35 Love it. Yeah. What sounds like is you’re saying that American enterprises will pay a token tax if the government gives Anthropic and OpenAI a government-enforced duopoly and enterprises are no longer free to use open source like the rest of the world.

Chamath Palihapitiya: 13:54 Dosvidanya, yeah.

David Sacks: 13:56 Yes, now the market’s decapitate.

Jason Calacanis: 13:58 Yeah, we will put ourselves on an island. We’ll be on an island of overly expensive AI. You can have Coca-Cola or Pepsi, or Coca-Cola or Pepsi.

Chamath Palihapitiya: 14:07 Well, it’s not just beverage choice. You have two beverage choices, but they cost 50 times more than Coke outside of America. This is the point. We already have Coke everywhere. So you can buy 50 cent Coke or $50 Coke. Why would you buy $50 Coke when you can buy 50 cent Coke?

Jason Calacanis: 14:24 All right, let’s get—let me get Friedberg here. Yeah.

David Friedberg: 14:26 Okay, I was going to say just one thing on this. This goes back to my point of these proposals to ban open source that are coming from Anthropic. They don’t solve the distillation problem. If distillation’s a problem, you have to stop it at the source. In other words, if you want to ban open source in America, the rest of the world will still be using Chinese open models. We won’t have solved that problem.

Jason Calacanis: 14:46 Yes, and that means they’ll get the data and they’ll get the reinforcement learning and then we lose the AI race, guaranteed. Friedberg, let’s take it from a Graham Allison and level up the discussion here. It would be quite provocative to… Then, the Chinese models had a Xi Jinping respond to that, how does the CCP respond to that? This is a crazy chessboard. It would seem like a pretty escalatory and we’d be going up the ladder, yeah, Friedberg?

David Friedberg: 15:12 Yeah, look, I don’t—I don’t think it’s as relevant that the open source model is published by China. I get that there may be security risks, but those can be estimated and addressed. I think on the three things that are worth highlighting on this issue, I’m very much aligned with Sacks and Chamath. And I think the three things are really around distillation. I don’t think distillation is just about AI. You know, distillation is a process whereby you look at the end product that someone else has produced in thinking about and learning about how to engineer your product. It is a common technique that is used across every product category and every industry. One carmaker will look at how the other carmaker’s car operates and they will use that to help them design a better car. You know, at Google in the early days, we would submit millions of search queries to Yahoo and Microsoft’s search engines to see what the result sets were, and we would compare our results against their results as a way of improving our search engine rankings and our algorithm. It was a very common technique. It doesn’t mean we were stealing their algorithm. We didn’t go into their servers and steal their software. We looked at the output of their software and used that to improve our software.

Jason Calacanis: 16:23 Which was called benchmarking, right? It was benchmarking.

David Friedberg: 16:26 You can call it—there’s been a million terms, exactly right. And so I don’t think that this matters as much. I think the question around copyright infringement or IP infringement—

Jason Calacanis: 16:34 That’s our second story.

David Friedberg: 16:35 —and Sacks is right, there’s a terms of service question here, but that’s on the service providers to fix. The terms of service blocking people from doing this if they so choose. But the copyright argument, the IP argument is really about, ‘did they steal the software?’. They didn’t steal the software and they just looked at the output. That’s not IP infringement, that’s not copyright infringement in the classical sense. And so I do think from a distillation IP argument perspective, it’s the output, not the process that matters. So the question is: are they taking copies of copyrighted software and using it, or are they looking at the output? And so I think output not process of engineering is what you really need to assess here.

David Sacks: 17:13 Can I insert something, Friedberg, and you can comment on it? Is I think a lot of people in the, let’s say, policy-making community don’t understand this distinction, but I think everybody in tech does, and I think it’s a big part of why there’s a disconnect on this. Is there’s a huge difference between model weights and outputs, right? So the weights are the—it’s the file of numbers, it’s the numerical parameters in a model.

David Friedberg: 17:35 That’s the software code. That’s the code.

David Sacks: 17:37 That’s the software, that’s the code. And if Chinese companies were to steal weights, proprietary weights from Anthropic or OpenAI, that would be theft. Okay? But that’s not what we’re talking about here because no one’s accused that. What we’re talking about here is taking model outputs and then trying to learn from them.

David Friedberg: 17:56 Using other people’s software to learn.

David Sacks: 17:58 Yes, and it’s exactly the situation you said with like the—the Google searches or whatever. Whatever. And here’s the thing that’s so hypocritical is that OpenAI and Anthropic have both argued that they are free to train on all the world’s output, regardless of whether the creator wants them to or not.

Jason Calacanis: 18:12 That’s right.

David Sacks: 18:14 That is their current position. That’s like Chamath mentioned the New York Times lawsuit.

Jason Calacanis: 18:16 We’re going to get to that.

David Sacks: 18:17 The New York Times is suing OpenAI right now for going onto the New York Times website in violation of the New York Times terms of service, scraping all the information and training on it. And OpenAI’s argument is, look we’re not stealing anything, we’re taking the output of the New York Times and we are deriving our own model weights.

Chamath Palihapitiya: 18:37 100 percent.

David Sacks: 18:40 100 percent. And that is exactly what these Chinese models are doing here is they are taking the output of American models and then they’re deriving their own weights, they’re learning from it.

Chamath Palihapitiya: 18:43 And they’re learning from it.

Jason Calacanis: 18:46 Yeah. And so the accusation though, just so we’re clear here, is that there is ethical issues around the industrial scale covert breaking of terms of service. That’s what the White House has been talking about as well. Just so people understand.

Chamath Palihapitiya: 19:00 The terms of service, yes. 100 percent.

David Sacks: 19:03 Yeah, look, let me be clear that I’m not defending China in this. In fact, you know, look at my bona fides, I was the first administration official to even talk about distillation. I did it in January of 2025 when DeepSeek came out. I went on Laura Ingraham and I think I was probably the first person in the government to even explain this concept publicly to people. And moreover, you know, I was a co-author of the winning the AI race report, in which the whole premise of it was that we want to win, we want to beat China. So, you know, I’m definitely not someone in this camp that doesn’t want the US to win. I want the US to win. The question is how, and if we shoot ourselves in the foot by banning open source, which is to say not letting all of our American companies take advantage of open source when the rest of the world is able to, then that is a huge problem. Now, I am fine with Anthropic and OpenAI enforcing their terms of service. They need to do a better job, stop the distillation from occurring in the first place.

Jason Calacanis: 20:00 If there’s things that the government can do to help them, okay.

David Sacks: 20:04 But I’m not sure what those things are. What needs to happen is those companies need to do a better job enforcing their terms of service.

Jason Calacanis: 20:11 Okay. Now Friedberg, you had… I think you said you had three points to make, I think you made one, I want to get the other two out of you.

David Friedberg: 20:15 Yeah, so the other one was just on the free speech argument, which is look, what is an open source model? And I think the audience needs to understand this if you’re not from the software industry, but open source is a downloadable package of software. You can think about it as downloading a text, a book. You’ve just got all the code. Once you get the code, you’ve got it on your computer, you don’t have to be connected to the internet, you can just run it and use it. And I think part of the challenge that’s going to be faced here if there is any attempt at restricting open source, it’s going to open a can of worms on how do you actually enforce restrictions on open source? Because you’re basically telling people once they’ve downloaded and gotten a copy of this free publicly available software, they’re not allowed to use it. And that becomes a real challenge. I don’t think we have a lot of great precedent for that. It’s going to be very ugly to try and enforce. and stop open source. I do think there’s a question on like copyright action, but if there is, there’s a legal due process to go through to make that case and stop that open source from being available. And then my third point is just open source is better for the world. To Chamath’s point, this is a hundred times cheaper. That is better for the industry, for enterprise, the beneficiaries are going to be the economy, the consumer. Fundamentally, if you look back on the internet, in the early days, Netscape made a proprietary browser and a proprietary server software, the Netscape software, and they went public and they were the first to do this and it was super, it was super valuable and profitable. That company ended up getting crushed because of open source. The Mozilla Foundation was formed to create an open source web browser called Firefox, and then Google ended up hiring everyone and made it Chrome, but it was still open source. The Apache Foundation set up the first HTTP server as an open source product rather than having to pay Netscape or Microsoft or Oracle for their server software. Anyone with a computer could download the Apache software and make a web server and be on the internet and create a website. And what ended up happening is the value accrued to the internet. It didn’t accrue to the small number of software providers that controlled the gate and the portal of the internet. Basically everything got open sourced, and Google took off, and eBay took off, and Etsy and Amazon and all the millions of small websites and all the millions of small businesses and everyone that benefited from an openly accessible open sourced internet. If the internet was closed and there was proprietary software gates and portals throughout the internet that everyone had to pay to get through, the internet would not have taken off and the economy wouldn’t have grown and all these jobs wouldn’t have been created.

Jason Calacanis: 23:58 It was called AOL and CompuServe. Like literally, we had that. Don’t need to have a wealth tax. Chamath, you were going to add to this.

David Friedberg: 24:01 Exactly. And now, now when we look at this analogy, the analogy here is if open source AI takes off, then all the worries that Bernie Sanders and Elizabeth Warren and all the socialists are harping and larking about are not going to be the case anymore because you’re not going to see all the value of AI accrue to two or three or four companies and their small group of billionaire shareholders. What will happen is AI proliferates and a million AI integrated enterprises all over the world will benefit. Everyone will benefit, the economy will grow, jobs will be created and AI becomes a power for good for creating an open economy. So the, I know that some people that are listening to this in the government and that are on the other side are going to say, ‘But but but Chinese open source versus American open source.’ Let it all proliferate. And frankly, if the Chinese are violating copyright, stealing software, go after them for that. Put in place trade sanctions. Do all the shit you can do to stop that from happening. But fundamentally open source AI will transform the global economy and it will ensure that the economic value of AI will diffuse to everyone and not be held captive by a small number of companies.

Chamath Palihapitiya: 24:01 No. We have to acknowledge that it’s incredible how fast the value capture in this segment of the market has basically evaporated. I’ve never seen it in my 25 years in Silicon Valley where a sector of the economy can absorb hundreds and hundreds of billions of dollars, and then you think that there’s going to be economic pricing power many decades into the future, and it effectively evaporates in months. Months. It took off in months, remember? Remember how big Anthropic was?

David Sacks: 24:13 This is why the 1% that controls open source, if they have it that the oligarchs are going to get their clocks run and… we’ve got to…

Jason Calacanis: 24:34 It took off in months and it’s evaporated in months.

David Sacks: 24:37 No, I got… this is an area where I disagree with you guys. Evaporated is a strong term. It does look like it could slow down or it could or it could plateau.

Chamath Palihapitiya: 24:48 Go ahead Sacks, make your argument and I’ll give you my argument.

Jason Calacanis: 24:51 Yeah, but let me pull up the Anthropic revenue chart before you go there Sacks, because we can… this will help mitigate here and educate the audience. So as you can see here, we got a little bit of a stall in Anthropic’s revenue in the last couple of months and it seems… and this is third-party tracking, so it’s not perfect data, but we do see that this is a clear headwind. And then do you have the second chart I had? I talked on the pod just… it was last week or the week before when we were having the discussion about open source and like, I think it was Brad Gerstner from Altimeter was talking about, ‘Hey, tokens are still growing.’ Well, there are routers that track open router. The better an open source model does and the easier it is to implement on your own servers and take it in-house, etc., those are dark tokens. They’re not recorded… you’re not going to see them show up on a revenue chart anywhere because they’re free, essentially. You just need to have servers and energy to do them. And here you see that now well over 50% and a lot of these are coming from Chinese models.

Chamath Palihapitiya: 25:54 By the way Sacks, I want to be clear before you give the counter: I’m not saying that these companies won’t make money. That’s not what I’m saying. But what I am saying is that markets are very savvy in looking through current earnings and asking a very specific question, which is: what does this revenue look like 10 years from now? Does it go up, does it go down? Is there more competition or is there less competition? Is it effectively monopolistic or is it more of a commodity? If it’s a commodity, how many people can price this good? At what price is the market clearing price of that good in 10 years? And all I’m observing is, normally those variables get exposed, those cards get turned over relatively slowly. And so you have five, 10-year cycles to transition from being an exclusive provider of a good to effectively a commodity provider of a good. And all I’m observing is, it’s so unique that only technology could create a market where that cycle could get compressed into a few years because it is very hard if you’re an allocator of money to sit there and… Look at this data and not wonder to yourself why it’s not a commodity in five to seven to ten years. And when they get to that conclusion, which every capital allocator will because it’ll be pretty negligent to not, it’s very hard to assign huge future premiums. And where the real money is going, by the way, and we saw it in Google’s earnings which I’m sure we’ll talk about, it’s going to the cloud, it’s going to the infrastructure. So, by the way, there’s another cohort of people that don’t want to see the end of open source because they want to serve the cheapest models possible because they know that’s where all the margin capture is.

Jason Calacanis: 27:37 What’s your… where’s this going Sacks? Is it both open… we’re going to just see a proliferation?

David Friedberg: 27:39 I mean, as far as I’m concerned, there is an unlimited appetite for on-demand intelligence and there’s going to be… I don’t think there’s an upper bound for how much intelligence you can tap as long as it’s continuously getting better, which then the thesis would be, yes, it’s a commodity and then the prices keep going down, but consumption keeps going up. What’s your thoughts on…?

Chamath Palihapitiya: 28:01 You have to assume that this is exactly the example that you guys used. It’s a web browser. And in a web browser, it’s a mechanism to get to a place. And so the apps that are actually the places are where the value is captured. And I think if you assume for a second that intelligence becomes completely ubiquitous, it’s widely available, the marginal cost of it is effectively zero, the energy to generate it is effectively zero, which I think is an accurate assumption, it’s very hard to make the case of why this isn’t like the browser.

Jason Calacanis: 28:40 And it becomes… just gets subsumed into the Amazon Web Services cloud businesses, Elon’s web service, etc. Sacks, where do you think this is going? Then we’re going to go to our second story, which is the IP story, third story is going to be the markets and Google specifically and Tesla and space-x.

David Sacks: 28:54 Look, there’s been a lot of violent agreement on this show so far. So let me just make the counterargument. I think that both open source and closed source will be big winners in this. I think the market is huge and they each serve their purpose. What happened was with the introduction of Kimi K3, there was a little bit of a panic in which everybody said, oh my god, all the Chinese models have caught up and they’re giving them away for free and they’re going to destroy our leading American frontier labs. And Anthropic and OpenAI are running around saying, listen, we can’t continue to invest billions of dollars if Chinese companies can just steal our weights, right? So that’s the argument that they’re making that government officials are responding to. The truth of the matter is that, look, when Kimi K3 first launched, I was concerned about it because I was like, oh, has China caught up? Are they now able to produce a much cheaper frontier model? Then the details started coming out. So Ben Thompson on his blog went through some of the cost numbers and it turns out that Kimi K3 is not that much cheaper to run. There’s not a significant cost advantage to it. So that’s point number one. Point number two is that China has not caught up. It’s true that Kimi K3 scored really well…

Chamath Palihapitiya: 30:00 on the arena battleground for front-end coding, for web development, but that’s just one test, that’s just one dimension. There are areas where it scores really well, but there’s lots of other areas where it doesn’t score that well. So it is not a clear advance or a clear catch-up to the leading American models. Moreover, you still have stuff in the labs by Anthropic and OpenAI that is way ahead of this and the reports are that Sam is going to Washington over the next week to go talk about GPT-6.0, which is blowing the doors off. So I don’t believe that China has really caught up. I think there’s

Jason Calacanis: 30:38 Sam’s going to go see Daddy? And what is he going to ask for?

Chamath Palihapitiya: 30:41 I think there’s incredible unreleased models in the pipeline and we need to let our horses run here and not slow them down with a bunch of unnecessary hoops. And if we do that, I think we’re going to be just fine. I don’t believe that China has caught up. I still think we are six months ahead. And then just the final point on this, if you look at revenue, which is the test of real usage in the real world, Anthropic and OpenAI are blowing the doors off. They are by far the fastest growing tech companies at scale that we’ve ever seen. Jason, you showed this chart that supposedly shows a hiccup in Anthropic. Listen, their internal forecast was to 10x this year from 100 to a billion. We’re in the middle of the year, they’re already over 700 million of ARR. They’re doing great. They’re easily going to get to a billion. And we don’t really know what this little blip is here. I think one thing it might be is that OpenAI, if you superimpose the OpenAI chart on this, they have re-accelerated over the past month. So I think that if you were to add OpenAI

David Sacks: 32:40 Yeah, Codex is excellent.

Chamath Palihapitiya: 32:41 Codex is excellent and I think they’re taking a little bit of share and Sam is out there tweeting that we’ve got our mojo back and they’re taking their forecasts up. I think they were expecting to end the year at 4 billion of ARR and I think they’re forecasting more like 6 billion of exit ARR. So my guess is that if you were to superimpose OpenAI and Anthropic and looked at them together and you were basically just to say that, you know, let’s call it the frontier model duopoly in the US, you do not see any slowdown, you don’t see any blip. They’re taking their forecasts up. And the reality is that if distillation is going on, it’s been a thing for, you know, again, I pointed it out back in January of 2024 with DeepSeek. So it’s been a thing this entire time that they’ve been growing exponentially. So I just don’t believe that these guys are actually suffering in any way. I don’t think they need government protection. I think they’re growing exponentially still. This is a little bit of a case of, you know, what do you call it in basketball when a player flops, you know, that you-

Jason Calacanis: 32:48 You’re flopping. You’re trying to draw a foul.

David Sacks: 32:51 Yeah, it’s a flop.

Jason Calacanis: 32:52 Well, or you basically act super dramatic after a foul in order to draw the charge. You’re foul baiting when you’re trying to get a foul and then you’re flopping, which is just about making like an exaggerated one like LeBron does.

Chamath Palihapitiya: 33:00 And that’s exactly- yes, it’s a LeBron, it’s a flair flop, you know, where you-

David Sacks: 33:00 Flip flop. These guys are trying to—they’re trying to draw the foul, they’re trying to get the government to intervene. Now why are they doing this? Because they’re in the middle of roadshows right now and Chamath, to your point, I do think they get the legitimate question about why won’t you be commoditized over time by open source? And by far the best response to that would be that if they can lure the government into giving them a government-protected duopoly, then that would be incredible.

Chamath Palihapitiya: 33:24 The best answer is what you gave and Anthropic and OpenAI should own this because they’re good at it, which is they’re gonna go up the stack to the application layer.

Jason Calacanis: 33:33 Well they already have like—

Chamath Palihapitiya: 33:34 I know, but they’ve done it in this way which is a little ham-handed in some cases, but they’re excellent at it. These end-user apps are really good and they should just own that. And that should be their answer to Wall Street, which is, guys, we have the best model, we will eventually go up the stack. And if I were them, I’d actually practice the following answer: There may be a version of a model that I don’t release and just keep for myself and I’ll just use in my own applications.

Jason Calacanis: 33:53 How about that?

David Sacks: 33:54 Yeah well, that would be—

Jason Calacanis: 33:55 That’s the real answer to their question.

David Sacks: 33:56 This—no, it’s not—it would be super anti-competitive.

Chamath Palihapitiya: 33:59 No, it’s not. They’re allowed to build a model and not release it and use it for themselves.

David Sacks: 34:00 Hold up, hold on, let me answer that. It would be anti-competitive in the eyes of their customers. It might not be in the government’s, but if you are using them as a customer and you’re Loveable, which I talk to, and they’re paying a ton of money, or you’re ElevenLabs and you’re paying them a ton of money and they say, ‘Hey, we got our latest and greatest, you can’t use it because we’re going to compete with your company,’ they would stop using them and they’d go to open source.

David Friedberg: 34:16 I’ll tell you why Sacks, I think you’re wrong on this issue. I think open source is having its moment. I work with startups, they are all moving off of these and they’re using open source and they’re using it at much cheaper rates because a lot of the jobs don’t need the latest models. They can use these—the last generation’s models and people are moving them local, they’re hosting them themselves. This is, I think, a non-zero chance that this is going to derail Anthropic and OpenAI’s IPOs and their—

Chamath Palihapitiya: 34:31 No, they’re going to be fine. They’re going to be fine as long as everybody stays in their lane.

David Friedberg: 34:33 No, no, hold on, let me finish guys, I’m making my point. Shut the fuck up for 30 seconds. I believe that this is going to derail their IPOs, I’m taking it from the top. It’s going to derail their IPOs. It’s going to be headwinds against it because I think that they’re going to have massive margin compression. I believe they’re spending so much money that I think they’re going to get caught in a trap. I think this could be a trap for them. They overspend, they don’t have the same profitability and it doesn’t pencil out. And startups are the future. The startups are what eventually the enterprise copies. I think you’re wrong, Sacks, go ahead.

Jason Calacanis: 35:37 Ah!

David Sacks: 35:38 Well, I’ll come back but let Chamath respond because he’s just raising his hand.

David Friedberg: 35:42 You’re wrong, you’re totally wrong.

Chamath Palihapitiya: 35:44 Sacks, I think you’re wrong. I think they have caught up for 95% of the jobs. This is a crazy habit—I’m not saying the IPOs are off, but I think their market caps and the headwinds are coming. It’s a non-zero chance that they’re going to get slowed down. I think the answer is slightly different— 95% of the bleeding-edge tasks can be done by everybody. It’s that 95% of the tasks can be done by many different models. That’s the actual answer. And that’s okay.

Jason Calacanis: 36:10 That’s exactly right, Chamath. I agree.

Chamath Palihapitiya: 36:12 It’s also okay for Anthropic to say, “You know what, I have this next-generation class of model. I’m going to instantiate, I don’t know, a life sciences program, a cybersecurity business.” That’s where they can capture, as Sacks was saying before, over time, trillions and trillions of enterprise value… …because I do think they have excellent models and they have excellent engineers and they have momentum. But if you’re going to build a business model that tries to ascribe this layer as having a lot of terminal value, I think that that is a mathematical mistake. That’s it. You can’t do it.

David Friedberg: 36:45 Open source is going to be a specific headwind to these companies specifically because, because all of their best customers—and I talk to them, whether it’s Luvable or 11 Labs or the startups that were spending hundreds of thousands of dollars with them every quarter just last, you know, six months ago—they have all moved en masse, Llama 3.1, making their own models. The cat’s out of the bag. They are going to start losing a lot of customers to open source, and Google and AWS and Elon Web Services are going to host them. And how do I know this? Every time I do a job, I’m using Perplexity Compute—not a paid partnership or anything, I just happen to be the best harness that I’ve found—and I start with Grok, Nemotron, Llama 3.1, and I also put it into Claude, and the results are as good or better. In other words, I can’t even tell the difference between these.

Jason Calacanis: 37:38 Let’s superimpose the OpenAI numbers on top of the Anthropic numbers because I think it supports the point I’m trying to make.

David Friedberg: 37:45 Yeah, and these are estimates, by the way. This is not, like, literally from the companies. Just want to make sure people know.

Jason Calacanis: 37:51 So, according to this company that—I mean, look, who knows how they derived this, we don’t know that they’re totally true—they’re basically showing that OpenAI’s run rate, which is ARR, it rose from 33 billion in May to 41.3 billion in July. So they’re seeing acceleration. Like I said, I mean, Sam is out there saying they got their mojo back, they’re going to have their best 12 months forward-looking ever. And look, Anthropic is still growing really fast.

Chamath Palihapitiya: 38:13 We’ve talked about this on a previous show. It’s not physically possible to grow 10x year-over-year forever. You’ll run out of compute, you’ll run out of energy, you’ll run out of everything…

David Sacks: 38:24 People.

Jason Calacanis: 38:25 Yeah, there’s just no way.

David Sacks: 38:27 The economy, customers to pay.

Jason Calacanis: 38:29 But Anthropic—I mean look, if anybody had said Anthropic would be at over a billion at the midpoint of the year, back in January they’re at 100 million…

Chamath Palihapitiya: 38:37 Nobody would have said it.

Jason Calacanis: 38:39 You would have said this is the fastest growing tech company of all time. The idea that they’re at risk of getting their entire franchise destroyed, it’s not in the data yet, is what I’m trying to say. And moreover, Chamath, to your point, it’s not only about the model, it’s also about the harness, it’s about the connectors, it’s about the enterprise agreements. There’s a lot of things here that you need in order…

David Sacks: 39:00 to grow a business like this. Yeah. And you know, this idea that they need to race to get government protection because of a competitive risk that might happen in the future, I think is just kind of unseemly and gross. I mean, this is literally the most successful tech company of all time, and they’re racing to the government to basically say you need to protect us against our competitors.

Chamath Palihapitiya: 39:23 That’s a great argument.

David Sacks: 39:24 Not just our Chinese competitors, our potential future competitors, our American competitors.

Chamath Palihapitiya: 39:28 They should hire Lina Khan.

David Sacks: 39:29 Okay. Frankly, it’s gross. And this — let me give a couple of examples because I know people don’t have a lot of sympathy for Chinese companies. That’s fine. I’m not defending Chinese companies. I’m defending American developers who need to be able to use everything in the public domain. And let me give an example. Mira Murati, her new model —

Chamath Palihapitiya: 39:44 Thinking Machines.

David Sacks: 39:46 Thinking Machines. They currently have the best American open source model. You know how it was trained? It was bootstrapped. It was distilled off a Chinese model, Qwen 2.5. Now, if you say that Chinese model is based on IP theft, what is Thinky? It’s a derivative work off a model that Anthropic is trying to taint as being IP — by the way, there’s been no evidence of this, there’s no evidentiary process. They are simply trying to paint with a very broad brush here and say that now that model is tainted. Let me give you another example. So Cursor rolled out its new product, Composer 2. They were able to post-train that model using Qwen 2.5 on their own proprietary coding data. Okay, so think about this. They started with a Chinese open source model and then they used their own data and they came up with a new derivative product. This is the way that open source works. You take things that are in the public domain, you fork them, you make them your own. And by the way, once it’s in the public domain, it’s not a Chinese model anymore. It is open weights that are freely available to anyone. It’s a file, okay? And you take that, you fork it, you run it on your own hardware in an American data center. No packets are going back to China. No data’s going back to China. Nothing’s going back to China. An American company has taken open source contributions in the public domain, made it their own and then developed their own model. And if you say that American companies can’t do that or that somehow it’s tainted with IP theft, you are going to basically put a dagger through the heart of the entire American open source ecosystem, and that is exactly what Anthropic wants because they do not want to have the competition.

Jason Calacanis: 41:31 Friedberg, maybe you can close this out here and then I’ll put my final —

David Friedberg: 41:36 I’ll just zoom out for a second and I’ll say think about the strategy as well for China. If you think about the global economy of the last 50 years, the US has accrued so much value by being at the core of the knowledge economy and effectively a services economy and in that sense through the development of intellectual property of — key of knowledge and then the conversion of one bit to another bit, we’ve been able to drive trillions of dollars in GDP. Meanwhile, we outsourced manufacturing and created a sleeping giant in China, where they have this incredible manufacturing capacity. And at the end of the day, if you think about the course of like human technology evolution and human prosperity, it’s largely driven by our capacity to convert molecules from one form to another and use the least amount of energy possible to do that. That’s all of technology ultimately leads to that simple equation: molecule conversion. Making that beautiful couch behind you at the lowest cost possible, making materials, making semiconductors, making all this stuff. Everything in our world is driven by molecule conversion. So at the end of the day, if the knowledge economy and the services economy gets compressed, much like AI and open-source AI in particular effectively flattens that value because all of that value is now open-source, it’s free, and it’s simply a function of turning on a switch and running it. The value of the US and the Western economy has been largely degraded. And what’s left is the value of the molecule economy: the ability to convert molecules and use energy to do that. When you look at the juxtaposition of China versus the United States today, we have one terawatt of electricity production capacity in the US, and they’re on their way to having eight. We have about 10 billion square feet of manufacturing capacity, they have 200 billion square feet of manufacturing capacity. So they have 20x the manufacturing capacity, 8x the electricity production plus all of their other sources of energy. And I think that’s the long game for China over a decade, two, three-decade process is by compressing the knowledge economy and the services economy, commoditizing it completely, they are left holding all the value in the global economy because they can make stuff and they can make it cheaper than anyone because they have the most electricity production. That’s a very simple rubric for kind of how I look at the long game that they’re trying to play here.

Jason Calacanis: 44:10 You believe they’re trying to commoditize this very important space, just like they did for solar panels and cars, etc.?

David Friedberg: 44:18 Global knowledge and global services. The creation and movement of bits gets commoditized. And what’s left over is the creation of electricity and the creation of molecules, both of which they have this very difficult-to-surmount advantage that’s going to make them the core dependency for the world. That’s what I think is kind of the long game here.

Jason Calacanis: 44:36 And just so you know, this data comes from reports in places like The Information or other sources or leaked numbers, and they try to just make charts based on it.

David Sacks: 44:54 Yeah, and by the way, I, you know, I have my own sources too. I’ve talked to investors in these companies and I’m just telling you that both Anthropic and OpenAI are taking their estimates up right now. Their forecasts up. So I mean, look, I think in the future it may be the case that open source takes share. Fine. It’s because the market’s so big. And there is always a market for open because open is more controllable, it’s more customizable, you can own it, you get the data sovereignty. You get the sovereignty, but it’s also more work. So there are different use cases, there’s different parts of the market.

Jason Calacanis: 45:21 That’s actually very interesting. The more work part there, Sacks, is super interesting. Like 3 to 6 months ago it was so much work to stand these up and now there’s so many intermediaries building the harnesses that default to it that that is getting worked out. But that has always been the issue with open source for sure is the amount of work it takes to implement. If you do want to implement this inside of your organization, please make a call to 80-90 and try the software factory. Use the promo code JCAL to get a free consultation at 80-90. Alright.

David Friedberg: 45:54 Can they get a free consultation?

Jason Calacanis: 45:55 Maybe! Everyone’s talking their books.

Chamath Palihapitiya: 45:57 Everyone’s talking their books. Absolutely including, I’d say, the Anthropic and OpenAI investors. It’s amazing how many of these investors…

Jason Calacanis: 46:07 Brad was on the show two weeks ago! Were you on the episode? Brad was like, ‘Let me tell you why this is gonna’, he’s holding on, he’s holding on to it white-knuckled. White-knuckling it! He’s like, ‘Argh’!

Chamath Palihapitiya: 46:21 I actually give Brad a lot of credit because I do think that he’s objective about public policy or as objective as you can be given that he does own all these companies. But look, let me just tell you that, you know, I see a lot of folks who are suddenly China hawks and saying we need to stop China, we need to stop open source. Like, okay.

Jason Calacanis: 46:41 Yeah another China hawk, they definitely wanted us to send their ships there.

Chamath Palihapitiya: 46:43 Yeah, like how about disclosing first whether you’re on the cap table of Anthropic, okay?

David Sacks: 46:46 Absolutely.

Jason Calacanis: 46:47 Are you guys on any of these cap tables?

David Friedberg: 46:49 No, I’m not.

David Sacks: 46:50 Not directly.

Chamath Palihapitiya: 46:51 No, not directly, yeah.

David Friedberg: 46:52 Indirectly, maybe a little access.

Jason Calacanis: 46:53 You know what it’s like, Chamath, it’s like when you got that great flush or straight and you’re like, ‘Please don’t pair the board’. Brad’s like, ‘Don’t pair the board, please don’t pair the board’! Alright, here we go. I love open source. I love this open source AI stuff. I love it. I think it’s so awesome. The value, so punk rock. It’s just like there’s so much to be done with it. It’s just exciting and awesome and yeah. Chamath’s point is exactly right. Most of the models can do 95% of the tasks. And if that’s the case, then it’s not like everyone needs to scramble to get the best open source model, you just need open source to do 95% of what you want to do with AI, and then the other 5% you get specialized or high value or you pay a premium. And by the way, if you’re a big enterprise and you need to have wrappers and support and all these other tools for your employees, buy Anthropic or OpenAI or Grok’s tools or Gemini, like just make sure…

David Friedberg: 47:44 Make sure you have a good partner, use the promo code JCAL you get a Zoom call with Chamath. Okay.

Jason Calacanis: 47:50 You just have a Zoom call with Chamath and take a selfie with him on Zoom. I’m sorry, yeah.

Chamath Palihapitiya: 47:53 Anyone who’s gonna sign a 10 million dollar contract with 80-90 you get to have three of the four of us for a five minute Zoom call.

Jason Calacanis: 47:59 At the same time!

Chamath Palihapitiya: 51:00 So, okay look. And, you know, obviously I’m not a huge fan of Anthropic because I think they’re potentially destroying the whole ecosystem for their own purposes or regulatory capture. But let’s just be very clear about what, let’s just be very clear…

Jason Calacanis: 51:11 Come on the show anytime, Dario. Yeah. Let’s just be very clear about what this judgment was and was not. Okay.

Chamath Palihapitiya: 51:19 So, what Anthropic did is they pirated all these books from Libgen and they trained on them. And the reason why they got in trouble is because they basically took stolen books. They didn’t even pay for one copy of them. But if they had paid for just one copy of each book, they could not have been nailed for piracy. They would have been potentially under fair use, which I understand J-Cal is still being litigated in the courts, but that would have been their defense. So, the reason why they got nailed with this $1.5 billion judgment is they wouldn’t even buy one copy. It is still Anthropic’s position, and it’s OpenAI’s position, that they should be able to train on all of these books under fair use if they buy one copy. And that issue has not been resolved yet. Now, you should be able to see the total hypocrisy of their point of view relative to the previous issue, which is they believe they should be able to train on every creator’s output in the world, you know, as long as I guess they bought one copy of it, against the will of those creators, whether those creators like it or not. They believe it is fair use to train their models and derive their own weights based on fair use. However, they say that the one type of content that you should never be able to train on is their output. That is currently their position. It’s completely hypocritical. And actually, if you go back to Anthropic’s blog post in February where they define this concept of industrial scale distillation attacks for the first time, they coined that expression. And this is, you know, I worry that people in the government policy makers don’t understand that this is all part of an Anthropic op. No one used the terms distillation and attack together until Anthropic wrote that blog post. Distillation was simply an industry standard practice. But then Anthropic coined this idea of industrial scale distillation attacks. In any event, if you go to that blog post, search for the words IP theft. It’s not in there. Anthropic did not claim, even though they were trying to coin this new concept and brand this idea of industrial scale distillation attacks, they did not have the chutzpah to claim that it was IP theft.

David Sacks: 53:31 The Cahones.

Chamath Palihapitiya: 53:33 The Cahones, the hypocrisy, the chutzpah to claim that it was IP theft. Why? Because they maintain that it is their right to train their models on all the world’s output, even if the creators don’t want them to.

Jason Calacanis: 53:46 IP for me, but not for thee.

Chamath Palihapitiya: 53:49 Exactly. So J-Cal, I don’t even want to get into whether you’re right or not on the fair use question. Maybe you are right. I don’t know, okay? But my point is about the hypocrisy. And they themselves never claim

David Sacks: 54:00 that this was IP theft by the Chinese companies. What they tried to claim was that it was a national security threat because what would happen is these Chinese companies would distill off them, create their own models and those models would not have guardrails. So they were making a different kind of argument. That argument never found purchase with policymakers because I think that they could see that, look, guardrails are important but you know, it never really found purchase until Anthropic started claiming, ‘Oh, this is IP theft.’ But they have not been willing to make that argument publicly because they know that it would poison all of their fair use lawsuits that are happening and Chamath, like you mentioned, the New York Times is currently suing OpenAI for basically an industrial-scale distillation attack. I mean, OpenAI went on the New York Times website, used scrapers, slurped up all of their information at a scale that no human could achieve, and then they used that as training data and reverse-engineered the model weights effectively. So my point is that even Anthropic and OpenAI won’t publicly admit that what China is doing is IP theft because they are doing it themselves. It’s totally hypocritical, and I don’t think policymakers should be making arguments that these companies themselves won’t make because they know that they will lose all these court cases.

Jason Calacanis: 55:22 Friedberg, any thoughts here on—and obviously this is still being litigated as we’ve discussed, it’s 150 major cases, New York Times is one of them, music industry—

Chamath Palihapitiya: 55:36 Let me ask you a question, J-Cal. Let’s say you wrote—let’s say your book, your book, I don’t know, let’s call it Genuflect the novel, the Great American Novel.

Jason Calacanis: 55:42 Okay.

Chamath Palihapitiya: 55:43 It could be Angel in 11 different languages. Okay, so you write this book, okay? At HarperCollins. And you opt to not submit it to AI because there’s a restrict—you keep it closed. No one can read your book.

Jason Calacanis: 55:52 Like we did for Google Search, you can’t be on Google.

Chamath Palihapitiya: 55:54 Yeah, no one can read your book, you’re not letting anyone read your book, you have to pay for it. You want to pay for your book if you want to read it. Ten bucks. And someone pays thirty dollars for Genuflect, the Great American Novel, and they read it, and then they write a review. And they publish their review on the internet. Now, on the internet, the reviewer talks about your book and describes your book, gets web-crawled by an AI engine. The AI engine learns from that, learns about your book, and now there’s some commentary made about your book when someone asks a question about your book in the AI engine. Do you feel like your copyright was violated in that sense? So the book, the AI never ingested your book, it ingested metadata about your book, it ingested reviews about your book, it ingested third-party analysis about your book, all of which was on the open internet, and it didn’t just copy that stuff, but it used it to learn about your book.

Jason Calacanis: 56:44 So you’re saying these 1500 reviews for Angel: How to Invest in Technology Startups—Timeless Advice from an Angel Investor Who Turned 100,000 into 100 Million, these reviews would then be the basis of the AI. So I guess I would have to be okay, but this like, you know, this eloquently brash blueprint for angel investing that very— Five-star review?

David Sacks: 57:01 Yes, I would be fine with that five-star review being in there.

Jason Calacanis: 57:03 Okay, so now you have no knowledge—you now have no knowledge of the process. JCal? Absolutely JCal.

David Sacks: 57:09 JCal, how many of those 1498 reviews did you not write? Three?

David Friedberg: 57:15 Okay, no. Actually, I’ll tell you the secret when you guys actually get asked to write a book or any of you have the capability of completing a book—

Jason Calacanis: 57:21 Yeah, ‘cause I’m 97 years old living in the 19th century.

David Sacks: 57:24 What AI engine did you use to—what bot did you suppose that?

David Friedberg: 57:26 AI didn’t exist in 2017!

Chamath Palihapitiya: 57:28 Now we know where he pointed that stupid open source AI slop cannon that he’s built.

David Friedberg: 57:33 Actually, I verified the purchase too!

Chamath Palihapitiya: 57:35 Friedberg, you make a great point. You make a—oh, there it is. Genuflect—a great American novel. Rascal Wisdom: For mastering ambition, building resilience, and winning at life. Work, love, and innovation.

David Sacks: 57:51 That’s a master virtue signaling.

David Friedberg: 57:53 Absolute—

Chamath Palihapitiya: 57:54 A master virtue signaling. Incredible. Beautiful. That is beautiful. But JCal, I mean, this is my point. Knowledge can’t be contained. It’s diffuse. And so the form of copyright is very clear, the case law on copyright is very clear. I cannot lift text out of your book, reprint it, and claim it as my own. That is a violation of copyright. But my reading of your book, my reading of the reviews of your book, the diffusion of the knowledge that arises from your book, that is ultimately going to lead to some abstract transformation of knowledge into a new output that someone might read, and I think it is very unlikely that we will find ourselves in a place where the idea that knowledge transferred digitally, processed digitally, and turned into other content is going to end up violating copyright.

David Friedberg: 58:40 I understand—I understand your position. There are workarounds. Obviously, we’ve always had Cliff Notes, right? So if a book became good enough, somebody could write the Cliff Notes of it, you can’t stop that. There’s a four-part test for this, we’ve talked about this for three years here on the pod. What I’d say is American companies should take 10% of their revenue if they’re building these models and do splashy, catchy, and do settlements. And that’s exactly what’s happening. If you’re a copyright owner, you should study what the music industry did. They are rabid dogs, and they will fight tooth and nail and keep you in the courts until you submit and make a settlement and agree that you’re licensing it, and then that gives them that case law and that settlement to go to the next person and the next person and the next person, and that’s why they’ve been able to successfully defend it. And then, if you’re competing with me, that becomes the issue. So if you said, ‘Hey, what’s this book about? And what do people think about it? What are the best parts of it?’ And that comes from the reviews, okay, fine, fair enough. The problem is, and I’ll share with you, there’s been some other lawsuits here that are making their way through the courts. As they go into the application layer that you talked about, Chamath, as they go into the application layer and they use this, there’s a big court case here, obviously you know about some of these, but there are now other cases, there are music cases, there’s the New York Times case. The one that’s kind of interesting is Thomson Reuters versus Ross.

Jason Calacanis: 60:00 This is a final judgment on AI training copyright. There’s a company called Westlaw, they’re like LexisNexis, and people have been trying to claim that they can train on the outputs of something like Westlaw and when you’re in the same business as me, that has a special place in copyright law because I have your your infringing on my ability to use my copyright and your argument I think Friedberg holds up that it wouldn’t be… it wouldn’t stop somebody from buying the book, but the second you are actually competing with me like directly, that’s when these things have problems and that’s why I think the music industry’s going to win and some other places are going to win. But listen, this is brand new territory. These lawsuits are brand new territory and IP law does not actually have the nuance yet, so we’re going to as a society have to make a decision here on what is fair.

David Sacks: 60:44 Let me ask you a question.

Jason Calacanis: 60:45 I agree. I suggest just like the self-regulatory group that we talked about last week, all the AI companies should get together, take 10% of your revenue, put it in a pool and keep paying the people and getting permission from them so you can get updates on the content, so you can get the next book, so you get the next New York Times story, the next Reuters story.

Chamath Palihapitiya: 61:13 10% is not going to satisfy the rabid dogs. Let me tell you, they’re going to go for 100%. Now JCal, let me… here’s the question I want to ask you. Given that the fair use doctrine is not a decided matter yet, given that Anthropic and OpenAI are embroiled in huge lawsuits against very well-financed content creators and those communities and the outcome is indeterminate and there’s billions of dollars at stake, maybe even their entire product at stake, do you think that they have potentially made a fatal mistake by arguing that distilling content against the wishes of its creator is IP theft? Do you see what I’m saying?

Jason Calacanis: 61:52 Yes, like is this potentially a fatal mistake. See, what you would bring that to the Supreme Court and you’d say, hey listen, you are saying…

Chamath Palihapitiya: 61:57 Wait, but here’s what they could have done. What Anthropic could have done is it could have said listen, we have these Chinese companies are creating fake accounts and they’re using proxies to basically use our product in violation of our terms of service. Now, using those model outputs is not IP theft because it’s fair use. However, it’s a deceptive business practice for these guys to lie about who they are when they set up accounts at scale. And so they’re engaged in a deceptive business practice and we’re going to do everything we can to stop that, but we’d like the government’s help in stopping that too. However, we’re not saying anything about IP theft. Wouldn’t that be the more nuanced approach? Because I think that they’re on the verge of being hoisted on their own petard here.

Jason Calacanis: 62:40 Yeah, I mean it’s a… I think the kids call it a self-own, right? Like you basically… it’s just a classic self-own and, you know, in most of these cases settlements happen. So again, if you just look at the music industry… The newspapers, the magazines, and some of those folks and book authors, they’ve tended to be very meek. There’s a new trend happening now, Friedberg. They’re… a lot of content providers are saying… to Google, take us out of the index because Google has one bot and that bot does the Google crawl and that bot also does the AI crawl. And what the industry is saying is, hey, split that up. I want to be in Google, but I don’t want to be indexed in AI. So people are now saying, ‘Hey, we’ll take you out of the index,’ which Rupert Murdoch got right. If all the newspapers said collectively, ‘Do not index us Google, we’re no index,’ that would have made Google come to the table and give them a royalty and give them some money for being indexed, I believe.

David Sacks: 63:28 They did, dude. There was a deal. That happened, but it didn’t go the way you’re describing.

Jason Calacanis: 63:31 Well, ‘cause they didn’t have a united front. Now I think the New York Times and others—

David Sacks: 63:34 No, it’s because these guys wanted Google’s user base, so they ended up doing a deal where they had this like paywall like exclusion rule, where it was like you could show a certain number of free articles. There was a whole negotiated settlement.

Jason Calacanis: 63:46 I’m talking even long before that when the first index happened. But here is, you know, I think we are on the cusp of a some type of a settlement getting done here, and I think that would be good for America to take a leadership position in that. Because you do want to keep getting that, but all of these content companies, they should be meeting with each other—the music industry, the New York Times, all of them—to stop their content from getting used without their permission and from competing with them. That’s my belief.

David Friedberg: 64:14 Well, I mean, you know, you only get a settlement when both sides can agree. And it seems to me that if you’re one of these content creator lobbies and you see that Anthropic has just told the government that training on a creator’s output without their consent is IP theft—

Chamath Palihapitiya: 64:31 Yeah, they’re on your side.

David Friedberg: 64:33 They have now basically confessed to their entire product being stolen. And it seems to me that why wouldn’t the content creators now assert that they’re entitled to own 100% of Anthropic’s revenue? It seems to me that this could be a bridge too far. You know, that they’re so good at regulatory capture, they’re so good at making these arguments and getting the government involved to create new regulations to protect them. But I wonder if this was just a little bit too cute. And again, if they had just positioned it slightly differently, if they said, ‘Look, these Chinese companies are creating fake accounts, that’s a deceptive business practice, we’re not saying this is IP theft,’ right? But instead, they said IP theft, and now the whole startup community is activated. You saw that Garry Tan and 200 startups wrote that letter. Why? Because they know that if this IP theft thing sticks, that all derivative works of training models are tainted now too. So that means the whole startup ecosystem is now at risk. So I just wonder if these guys have just gone—it’s just all a bridge too far—and they’d be better off rolling it back.

Jason Calacanis: 65:40 I mean, if you go to Washington and you lay down with the dogs, don’t be surprised if you wake up with the fleas. Like, you… they decided to engage in this, you know, so they may have poked the tiger. I think… I think it’s pretty accurate. Oh, by the way, the publishing schedule for 2027 was released. We have some new books coming. You heard mine, ‘Genuflecting’… Coming to you now, Ferrari on my wrist, how to win at life and afford a $250,000 watch by David Sacks.

David Sacks: 66:07 Oh my god.

Jason Calacanis: 66:09 This is Pre-Ozempic Sacks. We’re going to get you to get that done. Here it is: the fight to save America, destroying socialism, shrinking the debt and winning AI, by David Friedberg.

David Friedberg: 66:17 I like that book.

Chamath Palihapitiya: 66:18 I’m in.

David Sacks: 66:19 Yeah, that’s pretty good one.

Jason Calacanis: 66:20 That’s a good book. I love that. By the way, this is our new All-In… here it is: Chamath Palihapitiya, a sexual Italian summer. Wooo! It’s a romance novel.

David Sacks: 66:30 This is the only person who decided to do fiction. This is fiction.

Jason Calacanis: 66:33 It’s fiction.

David Sacks: 66:34 It’s fiction, Chamath.

David Friedberg: 66:34 I got Chamath’s non-fiction novel also.

Jason Calacanis: 66:37 Oh, you have his non-fiction as well. Here it is. Enterprise sales, Chamath, Dystopia: how my software startups bought my Italian summer. There you go! That’s great! It’s great! Oh my god! Wow!

Chamath Palihapitiya: 66:46 The IT guy in Milan.

Jason Calacanis: 66:47 Absolutely. This is coming from All-In books, it’s our new publishing label coming in 2027. We’ll also have the Brad Gerstner, Bill Gurley, Elon Musk and other titles coming, so we’ll have those on future episodes. More titles coming. Breaking topic here, Google and Tesla shared their results today. Had a lot of talk about capital expenditures. Google blew the doors off of their…

David Sacks: 67:06 Blew the doors off, I mean…

Jason Calacanis: 67:08 It was insane, outrageous, outrageous, and they were down like 7 to 10%.

Chamath Palihapitiya: 67:12 Google Cloud, growing 82% year-over-year.

Jason Calacanis: 67:15 Yeah, because of their cash flow numbers, because of the cash flow numbers. And now is on a $100 billion runway. That’s but one business. Tesla’s CapEx surged 140% year-over-year and they expect 25 billion in CapEx. Google’s CapEx forecast from 195 to 205 billion this year, so next year it’ll be even higher. Tesla down 14, Google down 7% our taping, who knows, but both reported negative free cash flow. In other words, the amount of cash in the bank went down instead of up. And for Google, that was the first time ever. IPO update: SpaceX down 30% from its day-one closing price, now trading 1.5 trillion, a lot of pressure on the stock, we’ll talk about that as well. Obviously they went public at 2 trillion, got a big pop, the Elon pop, and there could be more downward pressure or it could have found a bottom, you know, you never know with these things. This is unprecedented territory. We’ve never had an IPO this big, but some lock-ups… here’s the chart. Bunch of lock-ups are happening at different staged intervals. So let’s, Chamath, talk a little bit… here’s your SpaceX. I guess CapEx, Chamath, is being built out obviously for AI.

Chamath Palihapitiya: 68:11 And there is the case of OpenAI spending on CapEx in order to provide their service, but then there’s also Google, which is making these CapEx investments to resell it as part of Google Cloud, an incredible product.

Jason Calacanis: 69:00 And then on the other side they’re using it for their own obviously infrastructure and their business is just growing like crazy whether it’s YouTube or Google Cloud or even search is still growing. So I looked at this and I said, well this seems like a really good use of capital instead of just giving dividends like building out this infrastructure to me sounds like an investment in the future seems like a buy signal to me but the market is obviously disappointed. Why is the market disappointed? Is it a buy signal for you? Does it make you more excited about management and what Sundar and Sergey are doing over there or does it make you concerned? Chamath.

Chamath Palihapitiya: 69:30 Go back to… I’m more bullish. Do you know what Google’s 25-year average return on invested capital has been since going public? Take a guess.

Jason Calacanis: 69:42 17 percent.

Chamath Palihapitiya: 69:44 No.

Jason Calacanis: 69:45 21 percent.

Chamath Palihapitiya: 69:46 No.

Jason Calacanis: 69:47 23 percent.

Chamath Palihapitiya: 69:48 No.

Jason Calacanis: 69:49 29 percent.

Chamath Palihapitiya: 69:51 Keep going. Price is right.

Jason Calacanis: 69:53 35 percent.

Chamath Palihapitiya: 69:55 32 percent.

Jason Calacanis: 69:57 Jesus.

Chamath Palihapitiya: 69:58 Okay? This is when you are a machine and a group of people and a business model that compounds money at 32 percent over 20-year average, you give these guys the benefit of the doubt. These are not people that are flying fast and loose, they are methodically investing in their edge. And this is… I go back to the first conversation. They are going to get massively rewarded. You know, there was a lot of Twitter chatter or X chatter about Gemini usage and was it real or was it not real and is their revenue growth really coming from AI-enabled workflows? It’s all malarkey. Google has an incredible search experience. They’ve seemed to be navigating this transition to use AI—it’s been done very well. They have an incredible cloud business and they have an incredible silicon business. The best thing that can happen to them is 500 different models proliferate and they support all of them because they will make so much money at the silicon layer, they’ll make so much money as the cloud provider, and they’ll find a bunch of apps including YouTube and other things to make money from because you use the AI to target ads better or to help make better content etcetera etcetera.

Jason Calacanis: 70:25 Fragmentation’s good for them.

Chamath Palihapitiya: 70:27 Oh it’s great for them. It’s a compounding machine. I think the reaction by the way is because Jason I saw a tweet from Ryan Petersen. I don’t know if it’s true but he said Google will be spending 20% of this year’s military budget in CapEx. So maybe what people are reacting to is just the scale of the investment they haven’t seen. Their free cash flow negative for the first time since going public, so that obviously takes people by surprise.

David Friedberg: 70:42 I think the reaction by the way is because Jason I saw a tweet from Ryan Petersen. I don’t know if it’s true but he said Google will be spending 20% of this year’s military budget in CapEx. So maybe what people are reacting to is just the scale of the investment they haven’t seen. Their free cash flow negative for the first time since going public, so that obviously takes people by surprise.

Chamath Palihapitiya: 71:01 But they’re in a huge investment period and I think it’ll pay off dividends even if they, as Friedberg’s first guess was, half the number.

David Friedberg: 71:04 So even if they did half the number, they’d still be overachieving massively, doubling…

Jason Calacanis: 71:09 Still yum yum, yeah.

David Friedberg: 71:11 Yeah, it’s still good.

Jason Calacanis: 71:13 I mean Friedberg if you look at Apple, I think they bought back half their stock, they’ve given hundreds of billions of dollars back in profits and gosh it seems to me giving all this back to… money back, buying back in the form of buying back your shares, you know, or giving tons of dividends. I think it’s great that tech companies are now saying ‘wait, we have something to invest in’. The next big thing is on-demand intelligence and there is no upper bound for intelligence. Or I don’t think anybody, I certainly don’t see, you know, any time in the next 10 years people saying, ‘ah, I got enough intelligence, I’ve solved all the problems in the world’. I think they’re going to keep wanting it. So what do you think about this incredible change in, you know, basically 100 billion dollars, 200 billion dollars, you know, depending on the company, just going into CAPEX? This seems like it’s a savvy move, right? This is, all of us think this is a good thing.

Chamath Palihapitiya: 72:41 I mean, if you want to bet against Google’s deployment of capital into infrastructure because you’d rather have them give you cash for your shares today, you shouldn’t own the stock and someone else will buy it. I think Google wins in a lot of different ways. There’s just so much to Google, there’s the consumer business, which is a lot of stuff. There’s also YouTube, there’s also GCP, there’s also this portfolio of other bets, which by the way includes 10% of SpaceX and a good chunk of Anthropic that they own and so on.

Jason Calacanis: 73:10 And Waymo worth a hundred…

David Friedberg: 73:11 120 billion.

Chamath Palihapitiya: 73:13 I think they just took like a 100 billion dollar write up on Anthropic in the quarter. So they had a 100 billion dollar mark-to-market on Anthropic, just in one quarter. And they own a piece of all these businesses. So there’s a lot to like about Google. But just on GCP…

David Sacks: 73:32 I think that there’s probably no better suited enterprise layer than GCP to take advantage of capturing value with AI for that enterprise setting.

Jason Calacanis: 73:43 Why? Why is that?

David Sacks: 73:45 I think you’re just so much better because you have so much of your enterprise data, all your email, your drive, a lot of information that you would want to have AI have knowledge of and have AI have access to to improve workplace productivity. And then they’re model agnostic. I mean, you can run any model you want and you can run any workflow you want and you don’t have to be tied in. A lot of other cloud service providers, cloud SAS, they’re kind of model dependent to run in a certain way, with Google you can better tune your system how you want to tune it.

Jason Calacanis: 74:14 And what’s the worst, worst, worst case scenario? The worst, worst, worst case scenario is they have the lowest cost infrastructure in the world to run other people’s models as a service like Elon did with Grok with the Colossus.

David Friedberg: 74:24 Exactly.

Jason Calacanis: 74:25 Elon Web Services seems to be doing pretty great.

Chamath Palihapitiya: 74:27 I look at the return Elon’s making on the Colossus install. So I think if Google in the worst case scenario, none of their application layer stuff works, none of their network effects work and they don’t have any good models, they’re still going to have the world’s best infrastructure they can print cash on for years if you believe in AI. So if you want to bet on AI, I think the best public market stock to own is Google. And by the way, you also get YouTube, you also get the consumer, you also get everything else. The multiple is kind of ridiculous right now.

Jason Calacanis: 74:56 We talked about this on a previous issue, I think Chamath, you and I were talking about like what Apple should do next and… I think we both came to the conclusion of like why not have Apple web services? They have such great relationships with developers, they have the App Store, they have this deep developer relationship.

Chamath Palihapitiya: 75:08 It’s not so easy because you have to- to build a cloud service provider, you have to build some critical infrastructure. It’s taken Amazon, call it 17 years to perfect it. It’s taken Google, call it 12 or 13 years to mostly catch up. But the minute that you sign up to be a web provider, Jason, and a cloud service provider, what you’re really signing up for is five nines of reliability and uptime. And that is just extremely expensive. Getting to the first two nines, you know, 99% uptime, if you’re hosting something for a pharma company or a defense company, you can probably do it for relatively cheaply. Getting to the third nine, 99.9, probably costs you in the billions. Getting to the fourth nine costs the tens of billions. But getting to that fifth nine costs hundreds of billions. And that takes a real investment and real technical skill. And there’s only three games in town.

Jason Calacanis: 75:59 Yeah, I think Tim Cook’s not the guy to do it, but this new CEO might be since he’s an engineer. But they’ve returned in the last decade approximately 900 billion. 755 in buybacks and 140 billion, Sacks, in dividends. Just let that sink in. 900 billion. If they had invested that in Anthropic and SpaceX and other things, they just could have found some good uses for that money.

David Sacks: 76:28 Who’s to say, sorry, but who’s to say that the investors that got that money didn’t find a good use for it?

Jason Calacanis: 76:33 Oh, yes. So on a society level. But I just think Apple could have been more ambitious if they had spent half that money instead of on buybacks and dividends on creating new products and actually releasing their car, maybe buying some interesting companies. I think they would be a better- better off.

Chamath Palihapitiya: 76:45 I think it was very much a do-no-harm capital allocation strategy, which worked for the stock. It’s going to be really interesting to see if John Ternus flips the script.

Jason Calacanis: 76:54 I think he does. I think he’s going to be like engineer guy and yeah. Alright, enough on the markets. The markets are going to do what markets do. We will talk about Iran and other stuff like that when there’s more news for venture capitalists to comment on. Right now it’s just on and on. I know everybody keeps asking. I don’t think there’s much for us to say on it. I do think there’s a lot for us to say, uh, in Socialism Corner, our new recurring, uh, theme here, Friedberg. Every week there’s more news coming out of Socialism Corner. Producer Nick?

David Friedberg: 77:24 Why don’t you show my- my videos on socialism going back six years.

Jason Calacanis: 77:27 Let’s add to the Friedberg socialism rants earlier this month. New York City dictator slash mayor Zohran Mamdani…

David Sacks: 77:35 Jason, don’t- don’t misuse that term please, come on.

Jason Calacanis: 77:38 Posted a rental rip-off fury at New York City’s Tenement Museum after the hearing he introduced a rental rip-off report. It’s passed. Bars landlords from charging applications for credit checks. It’s going to let landlords require a credit check or the 40x rent income standard, but not both. Legally recognizes tenant unions and more. He’s obviously- So frozen the rent for a year. At the hearing, an activist wore a COVID mask and referred to evictions as the violence of evictions. Here’s your 20-second clip. The Mamdani administration is emboldening us so that we no longer tolerate the violence of evictions as a matter of business as usual. What the? What were we just watching? Is that Mushmouth? Do you remember the guy from Fat Albert who had the hat like that? Can we pull that guy up? From… do you remember the guy from Fat Albert who had the hat? What’s his name? Oh my god.

Chamath Palihapitiya: 78:37 Is COVID still happening? I thought COVID was over.

David Friedberg: 78:40 No, no, she just got a booster. I mean, that was like a… that was like a super duper mask. That was a super duper mask. That just wasn’t like a little cloth one. That was one of those big ones.

Jason Calacanis: 78:48 Whoa. I mean, combined with the hat, it was… that was pretty… there it is. I remember that guy from… what is going on?

Chamath Palihapitiya: 78:57 Wait, let’s give Friedberg a chance to rant because… I mean, I could do a rant on this.

Jason Calacanis: 79:00 No, no, let’s give Friedberg his… this is Friedberg rare meat. Friedberg, what do you think?

David Friedberg: 79:04 In 1787, in 1787, John Quincy Adams published a work called A Defense of the Constitutions of Government of the United States of America. And in that work, he had a comment: ‘The moment the idea is admitted into society that property is not as sacred as the laws of God, and that there is not a force of law and public justice to protect it, anarchy and tyranny commence.’ And then in 1791, he made the statement publicly: ‘Property must be secured or liberty cannot exist’ in an essay series.

David Sacks: 79:44 Can you unpack it and explain why he said that, you think?

David Friedberg: 79:46 Fundamental to the foundation of the United States of America was this idea of private property rights. Because if you think about where everyone that came to America was coming from, there were these tyrannical governments, monarchies or whatever, where some overlord or some cabal could decide at any point to take the things that you have. You had no private property rights as an individual. They could come in, they’re like, ‘That farm is my farm, you’re actually a serf, I’m the lord. That thing is my thing, you have a right to use it because I vest you that right to use it. I am the all-powerful, I am the tyrannical overseer of these lands.’ And it was that stasis that drove so many to come to the United States and say, ‘We want a place where individuals, one person, can say I own something and no one can take it from me.’ Private property rights are the foundations of liberty in America. This idea that you can then claim acts of violence, that you can then claim circumstances of extraordinary extravagant wealth and say to that individual, ‘I now have a right, the government now has a right to take your private property,’ ultimately leads to this tyrannical form. And it starts out as being anarchic because, remember, anarchy is a temporary state. It’s always in between one state and another. All anarchies end up in tyranny. Groups of people fight each other, they’re all stealing shit from each other. Everyone just goes and takes and gets what they want, and eventually people coalesce, they form groups. And those groups become the more powerful groups and the powerful groups end up winning, and they become the tyranny over the mass. And that is why all anarchies eventually evolve into tyranny. So anarchy and tyranny are one and the same. And fundamentally what’s going on with these socialist principles is that we are taking your private property and you no longer have rights on your private property. Whether you are a landlord or whether you are a wealthy person that we’ve deemed to have too much wealth, we now will have the right to come in and take your property and control it and take it from you. And it always starts with this framing of moralistic intent. We are good, you are bad for the following reasons: you have committed violence against the people that live in your building, you have taken too much wealth and none of us have wealth, we have a right to go and take your wealth from you. It has always started from this point of view that you are bad. So the first framing is that the private property owner is evil, and that the private property owner has committed an act of injustice against those who don’t have the property, and that is the justification for taking away their private property rights. And it is the beginning of this transition towards a tyrannical system, which will ultimately be what I call this kind of Great American Politburo or whatever socialist framework gets set up by the cabal of the socialists and what they’re trying to put together. So all of these little acts, while seeming ridiculous and insane and inappropriate, in aggregate are the same thing. They’re a transition away from private property rights, which is the foundation of the United States of America. That’s why I think we should all be so shocked. And to John Quincy Adams’ point, we need to vehemently defend those rights. As soon as those rights start to slip away, even in the tiniest way, it is a cascading effect and everything will become tyrannical and it will be very ugly in the United States of America.

Jason Calacanis: 82:56 Sacks, uh, what are your thoughts here on comrade Mondale?

David Sacks: 82:59 Yeah, I just want to add a layer to this idea, you know, these DSA socialists say that evictions are violence and they basically want to stop ‘em. I think Friedberg’s making the point that this deprives the landlord of their property and that’s true, but I think we also have to stop and consider what this means for the other residents in these buildings. I mean, first of all, if the landlords aren’t making income because they’ve got a bunch of delinquent tenants in the building, they can’t now pay for upkeep and maintenance and so these buildings become more dilapidated and that affects the other tenants. But also, I have a friend who manages these apartment buildings and he makes the point that it’s often these squatters, these delinquent tenants who should be evicted but can’t, who make the worst neighbors. So when you think about the problems in a, in an apartment complex where you’ve got people creating noise at night, maybe they’re playing loud music or you’ve had people punching walls or… There’s disgusting smells coming from apartments or they’re misusing common areas, or you have drunken and disorderly behavior. I mean, this is all the kind of stuff that happens in these rent-controlled apartment buildings. And you have to remember that there’s long-standing residents, a lot of old people who can’t afford to find a new place. They depend on the rent control. And when you can’t evict that unruly tenant, yes, it affects the landlord, but it affects the neighbors even more, because now they’re stuck in a downward spiral. And I think this is a problem with the progressive mindset across the board, is that these DSA types are always these, like, highly educated and often affluent types, and they can afford to have luxury beliefs about public spaces because they never use them, right? They don’t use the bus or the subway or parks.

Jason Calacanis: 84:51 Yes.

David Sacks: 84:53 Right, exactly. And so when they get taken over by homeless drug addicts, they always defend the addicts as opposed to the middle-class and working people. Easy thing to do if you don’t live in the Tenderloin. Right, exactly. And it falls hardest on the working class because they actually need these amenities. They need the parks for their kids, or they need to use the subway, right? And it’s really a problem when you get people shooting up or doing drugs or, you know, defecating in a subway. These are disgusting.

Jason Calacanis: 85:19 Or you’re walking your kids to school. Like, the person in Marin County who has these luxury beliefs, or on the Upper East Side, they just don’t—they’re abstracted from this. They don’t need to deal with it.

David Sacks: 85:28 We talked about this on—and it’s—it’s no different with these rent-controlled apartments. I think that’s the important point here is that if over a period of several years you can’t evict anyone, no matter how problematic they are, you effectively turn these apartment buildings into the equivalent of housing projects. And, you know, that really affects decent people of modest income who have nowhere else to go. And their quality of life suffers. And look, you know, the private equity wives in their gated communities will still feel good about themselves because, you know, they prevented these evictions. But it’s the people in the building who are going to suffer the most.

Jason Calacanis: 86:05 And at least, and Chamath, these people are not just thinking from first principles. If you want to solve—

David Friedberg: 86:11 If you want to solve the housing problem, anybody with any basic understanding of economics would just say, well, increase the supply and the price will go down. And it actually doesn’t matter which supply you add. It doesn’t matter if it’s luxury units or multi-family or single-family. As long as there’s more housing and there’s transportation to get to it and to move people in and out, it’ll be fine. As I—as I’m sitting here in Tokyo, like, they figured this out a long time ago. Just build up and put more units in. They figured it out in Texas, Florida, Nevada. The only people who can’t seem to figure this out, you know, is like New York, LA, and San Francisco. Just happen to be liberal elite enclaves. And it’s not hard on a conceptual basis. Just allow people to build some more units and different types of units, and then the price will go down.

Jason Calacanis: 87:00 Go down. Chamath, any thoughts here?

Chamath Palihapitiya: 87:02 The data from Austin says once you relax the permitting constraint, you’ll get more units built, and for every unit that comes online, it literally drives down the rent. So if you want low rent, you need to have more units. If you want more units, you need to permit more aggressively. That’s it. So it’s just a decision. The same political will that it would take Mamdani to pass this law, he could actually pass some permitting reform and it would do a lot more good.

David Sacks: 87:31 The thing on private property, the reason I asked Friedberg to explain it is I really believe what he’s saying is really important. It’s like funny to me, this idea that at the limit, let’s just say you’re driving your car and all of a sudden somebody jumps in it and they’re like, “Well, you know, now I’m here. You can’t kick me out.” Or, you know, you go outside, you leave your door open to go get a FedEx package or Amazon box and somebody runs in and sits on your couch and now all of a sudden they… you can’t kick them out. And it sounds so ludicrously dumb.

David Friedberg: 88:01 If you force the owners of physical property to not be able to credit check and differentiate who they rent their apartments to, what’s going to happen is rents will go up even more. So I suspect that what they should do is they should pass this law and they should observe what the outcome is, and then you can do a pretty scientific A/B comparison between New York City and Austin and you’ll know what works and what doesn’t work.

Jason Calacanis: 88:27 Yeah.

David Sacks: 88:29 Oof. I mean, here’s the problem is that the socialists never learn. I mean, we already had the—

Chamath Palihapitiya: 88:34 They go down to rent control in Argentina, rents went down.

Jason Calacanis: 88:36 Yeah.

David Sacks: 88:37 Yeah, I mean, this is the problem with Chamath is that, look, if the socialists ever learned from their failed experiments, you wouldn’t have Chicago, you know? It’s just you… like, we don’t need New York to go down the tubes to know this isn’t going to work because it’s happened in so many other places already. But somehow it just never seems to stop.

David Friedberg: 88:54 They never seem to learn, so they should run the experiment and learn. I mean, what’s crazy is you’ll be learning and it’ll be a failure on the grandest stage possible.

Jason Calacanis: 89:04 New York City, yeah. My goodness.

David Friedberg: 89:06 I mean, and Sacks, what happens if a landlord cannot raise the rent reasonably? Well, they have no incentive to invest in new units, they have no incentive to upgrade new units. And there’s this trap in New York City specifically, they have made the regulations for apartments such that if you do a renovation, it has to hit certain codes, and there are a ton of codes. Okay, so that means it’s incredibly expensive. So now they have… you have ghost apartments in New York.

David Sacks: 89:40 Yes, now the housing stock becomes dilapidated. And look, they’re doing something even worse now, I think, which is they are banning landlords from doing credit and background checks on potential tenants. And they say that you can’t look at their income, so you can’t vet whether they can actually pay the rent.

Jason Calacanis: 89:59 It’s tough to be a landlord. No more landlords.

David Sacks: 90:00 No, so they’re banning eviction, and then they’re preventing you from doing the diligence. to see if this is even a tenant who will pay the rent. We’ve lost the script.

David Friedberg: 90:03 So what are you supposed to do?

Chamath Palihapitiya: 90:04 The interesting question is, if you were forced to live under these rules as a landlord, what would you do? And the obvious answer is, you’d start rent three or four times higher, and you force people to sign up to a multi-month prepayment, and you’d slowly ease those conditions until you find a market clearing price. That’s the only way to do it. So, rents will not go down. Rents will go up. So, run the experiment, and let’s just observe what happens.

David Friedberg: 90:30 That’s a really good point, Chamath. Like, let’s say that, you know, X percent of tenants are going to become delinquent, right? Because—

David Sacks: 90:38 And by the way, what’s their incentive to pay when they know they can’t be evicted?

Chamath Palihapitiya: 90:41 Zero. Zero.

David Sacks: 90:42 So, like, actually a pretty significant percentage of people could just decide, eh, I’m going to make rent optional. And so now the landlord has to absorb those losses.

Chamath Palihapitiya: 90:52 Exactly.

David Sacks: 90:53 And that means they have to pass on a higher rent to everybody else.

Chamath Palihapitiya: 90:55 You’re going to set the rent 3X higher, and you’re going to slowly meander it down. And like I said, you’re going to have to wire in the first full year of rent.

David Sacks: 91:03 Well, and good luck. How is that more affordable?

David Friedberg: 91:05 It’s even worse, Sacks. Not only do the landlords keep the dilapidated apartments, in some cases, it’s better for them to just leave apartments, housing stock, empty. So somebody leaves, they are forced to renovate it, and it costs more, you know, hundreds of thousands in renovations. So they say, you know what? I’ll just leave it empty for now. And so you have 50,000 ghost apartments, according to reports in New York City.

Chamath Palihapitiya: 91:34 Well, that’s an Airbnb problem. That’s like a—

David Friedberg: 91:36 No, they banned Airbnb in New York. You cannot get an Airbnb.

Chamath Palihapitiya: 91:40 Oh, really?

David Friedberg: 91:41 Yes.

David Sacks: 91:41 That’s really interesting. Yeah, look, if you’re a landlord, okay? I mean, I guess one thing you’d do is just sell and go to another jurisdiction. Another thing you could do is just wait this out. Because it’s not profitable to run an apartment building. You can’t raise your rents, you can’t evict people, you can’t diligence the tenants. So maybe you just leave the building empty and you wait this out. I mean, that’s assuming you don’t have too much debt on it, right?

David Friedberg: 92:06 More ghost apartments, yeah.

David Sacks: 92:07 And then you have ghost apartments. Yeah.

Jason Calacanis: 92:09 All right, listen. You’ve been thinking about coming to the All-In Summit. This is your year. Speakers are world-class, the events, the parties, the networking. 80% of the people there, founders, investors or high-level operators. This year, greater focus on networking.

David Friedberg: 92:22 The musical performance.

Jason Calacanis: 92:24 Oh, yes. I mean, in the past we’ve had—

David Sacks: 92:26 The food, the drink.

Jason Calacanis: 92:27 Grimes, Diplo.

Chamath Palihapitiya: 92:29 Diplo.

Jason Calacanis: 92:30 We’ve had so many incredible people. Go to theallinsummit.com. All right everybody, it’s another amazing, all-time legendary episode of the All-In Podcast. Love you boys.

Chamath Palihapitiya: 92:44 Bye-bye.

David Sacks: 92:45 Bye-bye.

David Friedberg: 93:04 Besties are back. Go 13.

Jason Calacanis: 93:07 That’s why my dog takes a dump in your driveway, Sacks.

David Sacks: 93:12 Oh man.

David Friedberg: 93:13 My underwear room… meet me at the pool.

Jason Calacanis: 93:16 We should all just get a room and just have one big huge orgy because there’s all this just like this sexual tension that they just need to release somehow.

Chamath Palihapitiya: 93:27 We need to get merch.

Jason Calacanis: 93:28 Besties are back.