All right, everybody. Welcome back to what is of course the number one podcast in the world, tweaking all the mid podcasters out there. Four or five years running, the All-In podcast. It's episode 295, and Sachs is back. Sachs is back. Sachs is back. How you doing, brother?
Good, good. I saw some fake news that you guys had stopped using the term superintelligence. You're getting some shade for that.
Oh. Listen, we're here for this week in superintelligence right now. I mean, the whole document is superintelligence.
It was kind of ridiculous.
Oh, are they... And maybe I shouldn't Streisand it, but they were trying to say that we were using the term AI still instead of SI.
This is literally what the press is spending their time on?
No, this is why it's such fake news. They even had a photo of me. I wasn't even on the October 23rd episode.
Though you do look very presidential there. Man, if only... It's fake news. I mean, they're desperate for influential tech podcast. Well, that's good. But actually, we're not just a tech podcast. We're the number one podcast in the world. Give us our due.
The press has cooled on the use of the term superintelligence. And then they tried to say that I somehow was defying the president's instructions. All right. But I want to clarify, I was not on the October 23rd episode, so I'll be keeping you guys in line on this from now on.
Absolutely. Well, we will get focused on superintelligence. With us again this week, your favorite organic farmer, David Friedberg. How's O'Halleran doing? You've been pretty busy yourself.
O'Halleran Kenya this week. Lots of excitement. Lines of farmers trying out true potato seed.
Your energy looks pretty good. You look like you got some good energy. And speaking of good energy, man, we had a fun time playing cards last week. That was a blast. A great dinner.
You didn't stop talking for 7 hours.
It's sort of like the podcast.
Yeah, it's like the podcast.
It's unbelievable. It's unbelievable. So much time. Come on. That was a lot of time.
No. Affleck texted me after, I never want J Cal on the game again.
You got to get out Affleck as a member of the group?
It's going to be Ben Affleck or you, and we pick him. So you're out.
That's fine with me. That's fine with me. I mean, it's an upgrade.
I mean, you just kept chirping for 8 hours straight. It's unbelievable. He could not stop talking. I haven't seen him at the game for a long time.
I had a lot of material I had stored up. It was a lot of... It was really fun though. It was really fun.
And Chamath Palihapitiya, everybody's favorite enterprise sales executive from the 80/90 corporation. How'd you do this week? Did you hit? Did you ring the gong? Did you hit your quota? Or are you going to have to work this weekend?
Q4 is off to a good start, but I never want to count the chickens.
Okay, that's good considering you whiffed Q3.
No, no, no. Q3 was okay. We whiffed Q2.
That's why when I invested... All my fault too, because I forced these poor sales people, I forced these three poor sales people to just only focus on the gigantic deals.
You're learning. You're learning.
You want to get those singles and doubles and the people who pay on time and who start the projects on time.
We'll see in Ant S-1, but like, you know, you can build a ginormous business with singles and doubles.
Absolutely you can. Speaking of gigantic businesses, Anthropic, Ant dropped their S-1. That's the document that outlines the nature of the business for an IPO, an initial public offering, and got to ask the question: this seems to be the greatest business in the history of entrepreneurship. Goldman and Morgan Stanley are going to lead the round, and they are targeting November 16th, right before Thanksgiving, for this IPO. We've been waiting for a long time. It's been delayed. The ticker will be Ant, A-N-T. And they're aiming to raise, wait for it, $100 billion. In other words, like 3 months of compute. And that would make it the largest IPO ever, surpassing SpaceX. They raised $85 billion back in June, if you remember. And Polymarket says, our great partners, that Polymarket say 78% chance Anthropic raises at least $95 billion. The valuation, 1.5 to 2 trillion. Here's your Polymarket. Right now, it's saying between 1.75 and 2 trillion. Market cap 31%. So, yeah, assume it's going to be slightly under 2 trillion. The revenue is the story, Sachs. 16.3 billion in the first half of 2026. That is 14x year-over-year. So, in other words, the first 6 months of 2025, 1/14 of what they did in the first 6 months of this year. Q2 is going to be a whopping 11.6 billion. A 30x jump over the last five quarters, and almost as much as they made in the first half of the year. They say the TAM, the total addressable market for their tokens and services, is 40 trillion. That's extraordinary. I don't know how they backed into that, but we'll talk about it. And they're projecting 100 billion of ARR by the end of the year. Now, unspecified if that's annual recurring revenue or their annual revenue run rate, which is ARR, arr, like a pirate. They've got $112 billion in cash and a very unique ownership structure. Seven founders, yes, they have seven, are going to hold 50.1% of the voting power through a single class F share. Sachs, is this the greatest business in the history of capitalism?
Ha. It could very well be. I mean, I think this is probably going to be the biggest venture outcome of all time. Especially given the time in which this was created. I mean, this is only a 6-year-old company. The exit here is going to be on par with SpaceX's IPO, but SpaceX has been around for like 20-something years. So, this is probably the fastest value creation of any venture-backed company in history. And I don't know if you saw, but there was a tweet this morning about the distribution of wealth this is going to create among the employees. Hundreds and hundreds of employees are going to be making hundreds of millions of dollars, and then thousands of employees are going to be making $10 million plus. I mean, this is going to be an enormous wealth creation event like San Francisco and maybe all of Silicon Valley has never seen before.
Now, there's a few other data points in this S-1 that I think reflect the strength here. So, number one is that this $80 billion loss number that we've heard so much about, that the media has constantly been pumping and harping on as proof that Anthropic was either a fake business or that we were in this sort of whole bubble, that narrative basically just got completely refuted because it turns out that that $80 billion loss is a phantom number. It resulted from the fact that Anthropic had issued convertible notes to basically their infrastructure partners like Amazon and so forth. And as Anthropic's valuation increased, those convertible notes got recorded as a loss on the balance sheet and that flowed through into the income statement, but it's not a real loss. If they had just issued equity, that loss wouldn't be there. So, it was not a cash loss. It's basically a fake number. And in fact, what was revealed in this S-1 is that Anthropic has been an extraordinarily capital-efficient business. So, they've raised something like, I think it was around $110 billion roughly, and they've only spent about 9 or 10 billion throughout the entire course of this business, and they still have, like you said Jay Cal, about $100 billion left of cash or cash equivalents on the balance sheet. So, everyone thought, based on the previous reporting, that this was a huge loss-making business. In fact, it ended up being a very capital-efficient business. And then the other thing that you saw was that their margins have been improving. I think their gross margin is now something like 85%, and what you see is that for a given task or capability, they're showing roughly 10x improvement year-over-year in their ability to deliver that task, the cost-effectiveness of that task. It gets 10x cheaper year-over-year. And that allows them to cut the cost of that model or that task by let's call it 50%. And so the consumer recognizes a huge cost savings, but they are able to realize an even greater savings because they're able to capture the rest of that efficiency in their margin. And that's what we basically saw is that again, they're cutting the cost of their tokens significantly, but their margins are going up even faster because they're able to produce these models that are more and more efficient. And they call them compute multipliers. It's very interesting the way that Anthropic sees their business. The framework that Dario has is that there's raw compute, which are scaling very rapidly as well, as a result of better and better chips and more and more of those chips. And then there's the compute multipliers, which are things like algorithmic improvements or better curation of the data or it can be things like better architectures, more, you know, MoE architectures, things that make all these models more efficient. So they're able to give you the same answer for 1/10 the cost or they can take those efficiencies and create greater capabilities. So this is what we're seeing is that that business is actually improving, and I think it is a little bit of a repudiation to all the people who've been saying over the past year that they're going to commoditize away, that open models are basically putting them on a hamster wheel where they're basically going to zero gross margin, or that it's a business that only works because it's heavily subsidized by venture capital. In fact, I think there's a lot of reasons to believe that Anthropic has a pretty big lead. And if you believe that recursive self-improvement has kicked in, as they are claiming it has, then their lead could actually grow.
Freeberg, your thoughts on the impact of this massive deluge of capital coming into Silicon Valley and then this corporate structure, 51% ownership, and what impact this might have on the technology industry, the venture capital industry, and then on democracy here in the United States. We've never seen this much capital flow through to a nonprofit organization, or whatever they're calling this Anthropic nonprofit entity, but 80% of the seven founder shares are going into this nonprofit entity of some type. Benefit corporation, who knows what it's going to be structured as. It's optional, but that's another interesting shoe to drop. So, take it whichever way you like, in whichever order you like.
Can I just comment on the economic value that's arising from AI that's kind of perhaps revealed a little bit in this? So, in the S-1, Anthropic shows how many customers they have that are spending 100k or more. If we take a look at the image here, it's a pretty remarkable ramp in customers spending 100k or more: 200 customers in 2024, end of year, 1,500 end of 2025, and at the end of September, 11,900 customers spending 100k or more. So, I asked the question, you know, how many customers are there for which a $100,000 annual spend is less than 1% of their total annual spend? And the answer is, in the US, it's about 200 to 300,000 businesses. Globally, it's a bit over a million. So, you know, if you just say, 'Hey, look, Anthropic's customers that could spend 100k or more and have it be less than 1% of their total spend, the addressable customer base is about a million customers.' So, they're only at call it 1% of total customer penetration potential. Now, the question is, is there real money to be made here? So, I took a look at the kind of AI value stack to try and identify who's making money where. And it turns out everyone's making money. So, and I think we've talked about this quite a bit and I know Thomas got some views on this as it relates to the Neo quads.
What is this chart? Is this like AI output? Or that, or you made this?
I had Claude make this for me.
Or I should just say my analyst made this for me in my shop.
You know, I basically said like start with Nvidia, how much does Nvidia spend to make the chips? And then how much do they sell the chips for? So, for every dollar of COGS, every dollar that Nvidia spends, that turns into $4 of revenue for Nvidia, 75% margin. And then that gets sold to SpaceX and then SpaceX turns around and they're making, you know, an 83% margin on that chip spend because they're kind of looking at a five-year depreciation. So, they spent 10 billion on Colossus 1, five-year life, 2 billion year depreciation. And so, that's kind of the cost per year, but then they're making 1.25 billion a month on Colossus. And then Anthropic is showing in this S-1, pretty remarkably to Sax's point, that they're making about 85% incremental gross margin on every incremental dollar coming in. There's obviously a baseline of compute spend, but 85% is kind of the incremental gross margin. So, at the end of the day, a dollar that Nvidia spends turns into $18 that Anthropic is making selling to customers. And the question is, are customers making money? So, we're seeing the customer ramp as it is and the spend ramp as it is, and the customers are making call it on the order of two to 10 times the money they're spending because otherwise there would be probably a retention issue and a decline in spend. Maybe I'm wrong, maybe it's early days of exuberance and people are losing money and we could debate that point a bit, but generally speaking if AI is delivering value to the customers, I think this shows that there really is true productivity arising from AI. That across the whole value chain, there is such margin to be had, there's so much value being created all the way through to the customers, and we're only 1% or Anthropic is only 1% penetrated on their customer potential and probably less than 10% of the spend potential per customer. That this really is what I think Eric Schmidt, a long time ago in the early days of Google, used to say: 'Google's total addressable market is global commerce.' And I think, for the first time ever, we're seeing a business actually potentially starting to realize that theory: that if you can create enough productivity gain for your customers, that your TAM is commerce. And it really highlights that maybe, you know, we're almost growing the equivalency of the human working population by a 100X, or 1,000X, or a million X with this technology stack. So, I just thought like to me it was so striking how early we are, but also how incredibly profitable everyone is, which indicates there's quite a bit of productivity being produced across the whole value stack. Those are my biggest takeaways kind of from the S-1. I don't know if you guys agree with this, but you know, it all felt pretty remarkable to me. I'd be curious to know if you guys think I'm a little loopy on being too exuberant in my reading of this stuff.
I think the thing that is probably the most important is the last box that you said where you wrote it in the assumption.
Where you said assumption. Right.
Well, how did they measure that? By the way, I agree with you, but...
That box I made up. I said assume two to 10X. So, I'm just, I have no idea. But I'm a user of Claude. I think we get a lot of value for Claude. I think our people are able to do more because of Claude than they would otherwise be able to do. So, our business grows faster, we can get more done in a shorter period of time. I think I tangibly experience that pretty regularly. I've talked a lot about moving to open source and we have our own cluster installed with our own kind of source now, but I think what Claude is showing is that they have such a high margin, they actually have good elasticity on pricing. So, they could push price down and you could start to see them start to segment the market a little bit where they could start to be more competitive perhaps with cheaper tokens for cheaper easier tasks and then premium tasks charge a higher price for intelligence and start to stratify the value of the intelligence that they're providing by changing the price points. And I envision that's where everyone's going to go.
Nick, I'm going to show you this is to Friedberg's point about what is the potential downstream productivity gains that everybody can have theoretically with AI. Just put that on the screen. This is from Microsoft's earnings. My research team put this together. It's really quite incredible. What this shows you is that Microsoft's revenue went from 198 billion to 332 billion in the last 4 years while staff only increased by 2,000 people. I mean, isn't that stunning? Like you would have thought before AI that if somebody said great, we're going to, you know, get close to doubling revenues at this kind of scale, right? You probably would have thought that head count would have doubled too. And the reality is head count stayed flat and we don't know how many people will take this offer that Microsoft just gave for, you know, to leave because I think they offered it up to 60,000 people, but this number could be under 200,000 depending on how many take it. And so, just the scale and the impact on the productivity to your point. This idea that it creates job loss, I don't think it's true. It's more that you're seeing productivity gains at the same number of human input. And I think that's interesting.
I went through the S-1. You know, I'm going to give you my little S-1 breakdown. This is the most incredible business I've ever seen. I hope that they go public after November 11th because it will force them to update with Q3 and then we'll get even more insight into... This is like some contraption that I've never seen before. I've never seen a business like this. It's really, really, really incredible. So, let's cross our fingers because post November 11th, you'll see Q3 drop. What people have been whispering is that the revenue growth kind of moderated to sort of like the 50% range. Q4 is TBD. But let's break this down into a bunch of charts because I think it's interesting. Okay, so Nick, start with chart number one and we'll just go through them one by one. Okay. So, the first thing that I always read in an S-1 is what are the risk factors? And what I had my team do is I said, 'Go put together a word cloud.' This is crazy. This is really cool. First tell me what are they afraid of? Okay? And when you look at it, the share of words in the risk factors they're allocated to different risks are very important. And what this tells you is the single biggest risk to the Ant business is compute. They mention it more than anything else. And specifically inside of compute, they talk about data centers, power, chips, grid, electricity, hardware, etc. Then, a close second, but still second, is regulation. Okay? They care about government policy. They care about sanctions. They care about export controls. Then, it's this idea of zero data retention and privacy and IP. Then it's AI safety. Then it's just, you know, making money. Then there's their competitors. Then there's governance. What you notice is what are they not afraid of? They're not afraid of customer traction because they rarely mention it. They're not really afraid of employees leaving because you'd be an idiot to leave if you're inside. And they don't really care about taxes and any of these other things. So, this is a business that is very much on the forward foot that understands the binding constraint. So, now, if you know that the compute is their biggest risk factor, then you double-click into that and I had my team run a further analysis, which is to say, okay, how do they describe these risks, right? How many times do they mention them? How many words are you spilling talking about this? And what you find is that inside of the compute problem, their biggest issue is really data centers. And then further, it's the financing commitments that they have, and it's the power. All of these are really important because again, it helps us highlight what the binding constraint is to this business. So, as goes data centers, as goes power, and by the way, this is a great setup for the politicians because it's very obvious if you're a Democrat as you read this, and I'll speak about it in a second, what you should be trying to do to win in '28 versus what Republicans have to do in '28, right? Basically, the takeaway from all of this is I think, and it's emblematic of the entire AI complex, which is to say, it's all about data centers, which means it's all about power. If you're a Democrat, the best thing you want to do is slow all of this down so that there's a recession going into '28. The only thing you need to do is slow down power and slow down data centers. On the other side, if you're a Republican, and you want to have the best chance of winning in '28, this is the one thing that must be done right. You must get the power, electricity, data center thing right.
Okay, so based on this, your premise is they need data centers. If they don't get data centers, it can't keep growing. However, Sachs, the truth is these are the liberal of all liberals in Silicon Valley. It is the largest amount of money that has ever been made by founders outside of Elon Musk. It would make sense to me perhaps the opposite of what Chamath's saying. Shouldn't the Democrats come in and say, 'We are pro data center and energy because this is going to help society,' because they want to see Dario and his group win because they are going to be the equivalent on the left of Elon Musk on the right. Wouldn't it be better for them to steal the data center charge and make Anthropic even more profitable so that 51% of all of their, or 80% of all their money just flows into Democratic super PACs?
Maybe, but I don't think that's the Democratic Party that we have today. I mean, the old Clinton Democrats, I mean, Bill Clinton Democrats are long gone. I don't think the abundance Democrats are picking up a lot of momentum. I mean, the Democratic Party is, as Gavin Newsom said, is becoming DSA-dominated. So, what is their agenda with respect to superintelligence? Either to ban it outright, as Bernie Sanders and Elizabeth Warren and AOC want to do. They basically want 20-year jail sentences for developers of superintelligence and they want corporate death penalty for companies that build superintelligence. Or they want to have some extremely draconian regulatory regime that would feature a federal department of AI, or federal department of SI, I guess, that would review and approve every single model before its release. Basically, they want to have a DMV for models. So, that look, that is the Democratic agenda on this. Now, could that be shifted through oligarchs like, I don't know, Reid Hoffman or maybe some of these new Anthropic oligarchs like maybe a Dustin Moskovitz or something like that. Maybe they can, through their massive influence in Democrat circles, effectuate a change. But, I tend to think that Chamath is right that Democrats want to do anything they can to sabotage the Trump economy. And this whole SI boom is what is driving the Trump economy right now. So, I tend to think that's going to be the partisan impetus.
So, you don't need to do much to... I think Friedberg just scuttled the Trump economy since the war, tariffs, and all that other stuff have driven inflation and all those problems. Wouldn't it be smarter for the Democrats...
Hold on. I mean, just hold on a second. I mean, look. J. Cal, I think the latest numbers from like Fed GDP now, I think it was 3.7% in Q3, and it was 2.5% in the first half of the year. And you can also look at all the CapEx that's being spent right now on this buildout, and it's roughly 3% of GDP. So, you could say that all of the economic growth in the United States right now is coming from this superintelligence buildout.
Concur with that. We can have fun with numbers, but the Trump economy is a disaster, according to Trump's own supporters. So, let's just call balls and strikes here. His own voters, the independents who flipped over and voted for him, they are very displeased with the economy, and that can be seen very clearly in him having the worst approval rating of his entire career, and there's been some doozies in that career. Friedberg, what should the Democrats...
I don't think even those critics... First of all, I don't know what Trump's supporters you're talking about.
I'm talking about the America First people who feel that he didn't support the 20 million people that they still have... I'm just talking about that cohort. You could put Tucker in there. You could put Megan in there. You know, all that sphere.
Yes, there are definitely people who don't like the war, and they recognize that there is some inflation being caused by the war, but I don't think they've said the economy is a disaster. Anyway, whatever. I don't want to quibble with you about words.
But if you're not an equity holder, I think you look at this economy and say both parties, Biden and this Trump 2.0, I think the bottom half of the economy believes that both parties don't care about them. That's what all the feedback that you're getting, and we'll confirm it all next Tuesday and see what happens with the actual voting.
Chamath, let's continue your teardown.
So, that was section one, which is you always read the risk factors first, get a sense of what is management concerned about, right? So, again, the takeaway is management is concerned really only about compute. This is a compute problem. Okay, the next thing that I always look at is I look at the revenue side of the ledger and I try to understand what kind of business they have. And this is an incredible, incredible business. Exactly what Freeburg said. Why? Because at the growth that they have and the revenue that they have, they have barely scratched the surface of the commercial working world. And so this is a little bit more of a nuanced chart because what it does is it starts to really break down what we think the revenue stratification is by customers inside of Anthropic. So there's about 40 million individuals and small businesses that are spending under 100k. 40 million. Then there's about 10,000 that spend between 100 and a million. There's about 1,650 that spend between 1 and 10 million. And only 250 customers spend 10 million and more. So to your point, I think what Freeburg said is right. The setup is that these folks can negatively churn into these existing customers and grow and attract new customers at an incredible rate, which I think is just crazy. And so this allows me to think about what is the upper bound of revenue. And you know, you don't need to get to 100% of what's possible. You need to get to 25 or 30% and then you need to have a reasonable distribution of negative churn assumptions. And what you start to see is that this can be a very large revenue business in the hundreds of billions of dollars in sort of this three to seven year time frame. Now, what slows that down? As far as I can tell, and this is more of me putting the 80/90 hat on, the reason why you have so much traction in the million and under is when a company gets large enough, they allocate responsibility and budgets down. And at the director, manager, VP level, there are millions of humans running around with kind of million dollar budgets. And so there's a lot of experimentation that can happen inside of that bucket. They're willing to sign up for three, four, five, six, seven, eight, nine things. The minute you sort of transition from a million to like multi-million dollar deals, and this data sort of confirms it, it's a much more thoughtful pitch. It's much more cross-functional inside of a company. And you have to show some form of ROI or total cost of ownership advantage, right? So what's interesting about this, Jason, is 1,900 companies have been able to make the case. That's what I would say. I think there's a ton of experimentation. And I think about 1,900 companies can clearly see an ROI where they can now start to rip in big checks. Now, of that 1,900, a big portion of those are going to be AI natives, right? Folks that are venture funded. So you have to strip those folks away. And then what's left over is probably a couple of hundred global 2000 type companies. So I think the ROI case is still a little speculative, but the potential is there. And as long as the potential is there, I think the market's going to price it up. So now that's revenue composition. So then the next thing I think about is capitalization. And I said this before. I said Ant needs to have 200 billion. And I really hope that they actually get close to 250. First of all, let me just say the CFO of Ant, Krishna Rao, has done an unbelievable job. The fact that he, entering June, already had 125 billion dollars is unbelievable. And there's a chart I didn't include, but their total net burn to get there has been like 9 billion dollars or 10 billion dollars. It's just infinitesimally low relative to the market cap. So I thought that they had a lot less cash than this. So when I said they need to raise between 200 to 250 and they need to price closer to a trillion, I was assuming they had 30, 40, 50 billion on the balance sheet. I was wrong. So these guys need to raise between, I would say, 80 to 120 billion to get to that quarter trillion. And I think in order to do that, the price is probably now going to come in between 1.3 on the low end to 1.7 on the high end. So, call 1.5 the midpoint. So, if I had to price the IPO just eyeballing this, I kind of think it'll come in around 1.5. If I had to just kind of guess. And the other thing that I noticed is how much cash Alphabet has and also how much cash Meta does not have. I was shocked to see that Meta had negative 1.2 billion of cash. This must be a mistake, but if you look at the 10-Qs as of the end of September, this is what it shows, which net cash means.
Maybe it doesn't include short-term investments or something.
I'll take a look at Meta's numbers right now.
Yeah, I'll tell you in 1 second.
Your larger point is right that Anthropic, like you said, over 100 billion of cash they're sitting on. They've burnt very little of the money that they've raised. Again, their company's been a model of capital efficiency. They're about to raise 100 billion in this IPO. I think they've got about a debt facility with another 80 billion. So, you add it all up and this is a company with over 300 billion of dry powder or it will soon have that. So, then you've got all these, you know, Anthropic bears who are saying they've made all these CapEx commitments in the future that they're never going to be able to live up to. Well, actually, this secures those commitments because they've got the capital now to basically back up all of those compute orders, all that forecasted CapEx spending. So, I don't know, 2030?
You're exactly right. And because they only have to raise 80, right? Or 90 or 100, that's a lot different than trying to raise 150 or 175 or 200, which is where I thought they needed to be. And so, the demands they're going to put on the capital markets into the IPO are going to be less, which means that they'll have more pricing power. They'll be able to manage the float better. All of these dynamics, I think, give you a constructive read going into the IPO. Okay, now let's move on because I think here's my biggest takeaway.
Meta's net cash is -2. It is 80 billion of cash and short terms and 82 billion of debt.
Wow. Okay, now go back to this. What is my biggest takeaway? It's bigger than Ant. I'm going to build on what Zach said. The entire United States economy is levered to the AI complex. So, very narrowly, we're talking about seven or eight companies that sets the tone for the entire country. And whether we like it or not, that is where we are. Why do I say that? One, the most important companies in the world are in a massive spending binge. Right? These were highly profitable, highly cash accretive, highly cash flow generative businesses that have said, 'We're going to put the foot down on spending because we believe in what's coming into the future.' That's what drives GDP, and it's what drives growth. Okay. Next chart. Now, as they do that though, they deplete their cash, and they have to raise debt. Okay? And this is what you see. All but Microsoft, by the way, what an incredible job Satya's done. He's halved his debt, while everybody else has, you know, just gone to an unbelievably crazy high number. So, the AI complex now owes 340, 400 billion dollars to the market. And you'd say, 'Okay, why is that important?' The problem is this. These companies are raising so much money. It starts to crowd out the capital markets. And the most important part of the capital market that it starts to crowd out is the appetite for government treasuries. And what you can see here is that when you look at these AI firms, the amount of money that they need to raise through the end of '27, it's 30% of all of what Treasury is about to issue in that same period. So, there, whether we like it or not, they are going to start competing for demand. Isn't this nuts? It's like you have seven companies that are crowding out the treasury market.
You know, Chamath, a lot of people are looking at that number and they're trading these stocks down based on that. I kind of view it the opposite, which is I don't... These are smart companies. I don't think they'd be making those kinds of investments unless they thought there was ROI there. And so...
Well, there obviously is ROI based on the token usage and the fact that only 1 or 2% of people in corporate America are even using these tools. We are greatly underestimating the uptick here. Claude hasn't even launched a bot yet, a Grok bot. And OpenAI just launched theirs and it's in like a small use case. When 100 million or 200 million white-collar workers subscribe to, say, the office suite, right? The similar Office 365 thing happens in corporate America and everybody's got five agents running, I think the utilization of these things is going 500x. The token usage is going 500x. That means we're going to have not enough compute and everybody's got to build out... completely wrong. There won't be enough and people are going to look back on this moment and say, 'Why didn't you build more compute? Why didn't you spend more on CapEx?'
You're absolutely right, Jason. The problem is the regulatory pushback. The amount of approvals and the ability for utilities to find the staff and the ability to actually turn these things on and then the ability to get the capital equipment you need, it just isn't there right now. So, we as a country are sort of clogged up. We can't do it even if we wanted to do it.
Furthermore, you know, the other issue here is: if we are not spending enough building this out and a large percentage of the country believes in socialism and is against the oligarchs, both Republican, independent, and Democratic oligarchs, this is what they think of us and of the leaders in our industry and other industries. This is really dangerous because we're not going to be able to build out the compute we need. And what does that do to the entire industry in the next three to five years? If we can't build out this, and how do we turn that narrative around, so people are like, yeah, we should have data centers.
I have a view that humans try to organize into groups, organizations to try and create power that benefits the members of that group. So, you know, we form tribes, we form communities. And in a sense, governments are a form of an organization. It's an organizational coalition of individuals to achieve an objective for, in the case of a democracy, the majority of the voters. And as a result, organizations in a world of, you know, call it pseudo limited resources, compete with each other for those resources. Nonprofits, for-profits, governments, they're all effectively competing for resources because the individuals that they represent the interests of are going to be served if they get more resources, get more power, get more influence. So, to some extent, the power of AI and the influence it enables in terms of media, in terms of capital accumulation, in terms of changing people's minds, in terms of aggregating resources through its competitive capacity, productivity, etc., makes it a bit of a threat to government organizations. And so, to Chamath's point, if one of the manifestations of that is that AI is now accruing more interest from capital than governments, governments aren't going to want to let that happen. And the governments, whether they're right or left, are going to fight for the AI to become either part of the government and the government to subsume the AI, or control the AI organizations. And this is why I've kind of argued the case. I think that you're going to end up seeing compute get regulated in the future where compute capacity is going to be assigned by government and they're going to pass laws to try and do that. Post-production that is, like meaning they're not going to control the chips, they're going to control the actual compute layer and they're going to determine who's going to get access to what amount of compute. I think for governments to compete. And I think one of the ways this is going to play out to Chamath's point is in the near term in capital. There's more capital going to the AI organizations and the governments aren't going to want to let that happen. So they're going to try and stop the data centers because then...
I've never seen something in my adult working life where a complex of a handful of companies are effectively competing with governments now for capital. Because you've never seen companies in an organized manner have one very narrow thing that they need money for that is a one-way bet on something that attracts the same amount of money that would otherwise go to a traditional two, three, four, five-year, 10-year bond auction. It's kind of nutty actually. And you didn't see it...
What you're going to hear from both sides is a story of which side benefits the people more. And that's going to be the narrative argument which is going to be every side's going to say the other side's lying. Because remember if one side's lying and another side's not lying, the side that's not lying says, 'Hey, those guys are lying.' And they're correct. The side that is lying de facto says that the other side is lying because they're liars. And so you'll hear both sides screaming, 'We are the better path for society. We are the better path for humanity.' It's either government or private market AI and that'll be the huge battle that's going to brew over the next half decade. Wow. And we'll see which one, you know...
Politicians are for sale and Dario and Anthropic should start buying. Just like the right did.
I don't see them with champions. Where are their champions? Do they have champions yet?
For sure. Listen, Anthropic even as a startup prioritized government affairs. They were very active during the Biden administration. And all those former Biden officials who were using AI in the Biden administration immediately joined Anthropic at the end of 2024.
My point, Sachs, is that you don't... I get that, but you don't see them doing what you guys did, hosting a fundraiser for Trump, being out there publicly in front of it. You don't see them...
Oh, sure, but they have spent, they've created PACs and all sorts of entities and nonprofits, all NGOs, all that kind of stuff. And they have been massively funding it. They have a huge government affairs operation.
What did you guys do like in, you know, getting out there in front of it?
I don't know what you mean by you guys. I did...
You were public about your support for Trump.
I'm just a person publicly coming out for... I'm not a corporation. Yeah, exactly. I'm not... Can you let me explain? I'm not a corporation. I'm just one person who has views and yeah, I support certain candidates and I do fundraisers for candidates. I'm not like seeking regulatory capture. I'm just one person.
I'm saying why hasn't Dario or the other founders done that? Why haven't they said, 'We choose this candidate. We choose these candidates.'
We don't know that they haven't.
We know they haven't done it publicly because we've never heard them do it.
That's my point, but they have given huge amounts of money to PACs which is public information and those PACs have supported candidates.
If they're going to make a run at this, they should do it publicly.
There's a huge revolving door where Anthropic has been snapping up every former official. First it was the Biden administration. Now, frankly, they're hiring a lot of former Trump administration officials.
And yeah, yeah, this is all public information. But they are building a massive government affairs operation. And look, I think you can expect that as they power up through all of this capital having a $2 trillion market cap, they're going to have even more influence in Washington or they're going to be capable of buying a lot more influence. And I think that is concerning. One thing I will say though is that as they go from being a startup to being a public company that's one of the top handful of companies by market cap on Earth, it's going to be very obvious that they are a big tech company and therefore they should be like other big tech companies, which is to say with suspicion, as opposed to this being some poor little startup that, you know, has innocuous views or is not the 800-lb gorilla they actually are. So, I do think it'll be clarifying on that level.
for them to be seen correctly as what I've been saying is that they are the most powerful monopoly or duopoly ever in human history, which is they're creating AGI, right? I've been saying that for a long time. So, I do think that them going public will actually right-size them in the public consciousness. And one thing you're seeing already actually is there's starting to be a doomer-Anthropic split. There's a very interesting article in Vox, which is this huge doomer publication, over the past week, and it was quoting a bunch of these NGOs, kind of doomer NGOs, that were basically pointing out that, you know, like Anthropic actually, if you separate their rhetoric from what they actually do, they are the ones leading the charge on recursive self-improvement. There's no single company in the world that is doing more to race towards full RSI than Anthropic. They are hell-bent for leather on getting to full recursive self-improvement. In fact, a lot of the reason why they are becoming so efficient with their compute, the reason they're able to cut costs, offer the same capability at 1/10 the price, is because they've increasingly been using their own models to generate these algorithmic improvements, to create these compute multipliers. If you go back to August, Anthropic published a very interesting blog talking about how they were using RSI and they had a chart showing that 26% of their tasks that are involved in basically trying to create the next generation of models, they were already using RSI to do that. Now, that was back in August, so who knows what they're at now, being in October. So, look, they are pursuing RSI very, very quickly, and, you know, one way to think about that whole pacing letter that Dario published a few months ago is it's almost like they're creating an alibi for themselves where they are acknowledging that, 'Hey, we are pursuing RSI as quickly as anyone.' And it's almost like they're saying to the world, 'Well, we wanted it to stop, right? Like we wanted it to slow down. We tried to get you all to agree to pace, but you guys wouldn't agree to do that. So, therefore, we kept going.' You know, and it's almost like creating an alibi in advance for themselves if this RSI thing goes really bad.
But I think that they had to know when they published that pacing letter that they're not going to get the whole world on board with pacing. So, the whole thing was sort of an exercise in virtue signaling; but, in any event, I think what it shows again is I do think that it's possible that an Anthropic-doomer split is coming because there's no way for them to maintain their lead as the number one superintelligence company without continuing to advance the frontier very rapidly and to continue to pursue recursive self-improvement. And that is something that the doomers are most afraid of. Of all the various risks, right? They are most worried about a breakaway superintelligence. So, I do think that this is the beginning maybe of a divorce between Anthropic and the doomers, but we'll see.
