Is There an A.I. Bubble? A Debate. | The Ezra Klein Show

Artificial Artificial intelligence is now the driving intelligence is now the driving intelligence is now the driving force behind the growth of the force behind the growth of the force behind the growth of the US stock market, US stock market, US economic growth. It US economic growth. It absorbs so many absorbs so many absorbs so many construction resources construction resources that it now seems to be that it now seems to be that it now seems to be crowding out housing, crowding out housing, crowding out housing, manufacturing facilities manufacturing facilities , and virtually everything , and virtually everything , and virtually everything else that could be else that could be else that could be needed for needed for needed for construction. And construction. And construction. And the question, even from a the question, even from a the question, even from a financial financial financial perspective, is it perspective, is it perspective, is it all worth it? all worth it? all worth it? Will all these investments pay off Will all these investments pay off Will all these investments pay off , or is AI , or is AI , or is AI just a bubble? And just a bubble? And just a bubble? And one of these companies will one of these companies will one of these companies will miss its miss its miss its profit target profit target profit target or its rating will or its rating will or its rating will drop, and the whole thing will drop, and the whole thing will drop, and the whole thing will implode, taking implode, taking implode, taking all of our all of our all of our retirement portfolios with it. retirement portfolios with it. retirement portfolios with it. So I would like to have So I would like to have So I would like to have two different perspectives two different perspectives two different perspectives on this. Dr. Paul on this. Dr. Paul on this. Dr. Paul Kedrosky is a partner at Kedrosky is a partner at Kedrosky is a partner at venture capital firm SK venture capital firm SK venture capital firm SK Ventures. He is a fellow at Ventures. He is a fellow at MIT's MIT's Digital Economy Initiative, Digital Economy Initiative, Digital Economy Initiative, his popular his popular his popular newsletter is at paulkedrosky.com, newsletter is at paulkedrosky.com, newsletter is at paulkedrosky.com, and he and he and he has been very vocal about has been very vocal about has been very vocal about his belief that we his belief that we his belief that we are in an are in an are in an AI bubble. Azim AI bubble. Azim AI bubble. Azim Azhar is a British Azhar is a British technology entrepreneur, author, and technology entrepreneur, author, and analyst. He's the analyst. He's the analyst. He's the creator of creator of creator of Exponential Exponential Exponential Insight, which is a Insight, which is a Insight, which is a great Substack that I great Substack that I great Substack that I read often and a read often and a read often and a research research research platform where, among platform where, among platform where, among other things, they other things, they other things, they create these create these create these dashboards dashboards dashboards that track whether there's a that track whether there's a that track whether there's a financial crisis in the financial crisis in the financial crisis in the AI ​​industry AI ​​industry AI ​​industry . And his . And his . And his view is somewhat more view is somewhat more view is somewhat more optimistic, I optimistic, I optimistic, I think it's fair to think it's fair to think it's fair to say, than say, than say, than Paul's view. So I'd like to Paul's view. So I'd like to Paul's view. So I'd like to talk to the talk to the talk to the two of them to try to two of them to try to two of them to try to understand why you understand why you understand why you might think, and why might think, and why might think, and why you might not think, you might not think, you might not think, that we're in an that we're in an that we're in an AI bubble. Paul AI bubble. Paul AI bubble. Paul Kedroski, Azime Azhar Kedroski, Azime Azhar , welcome to the , welcome to the , welcome to the show.

show. show. Hello, Azim, nice Hello, Azim, nice Hello, Azim, nice to be here. to be here. to be here. It's good to be here, thank you. It's good to be here, thank you. It's good to be here, thank you. Paul, prove to me that Paul, prove to me that Paul, prove to me that this is a bubble. I this is a bubble. I this is a bubble. I mean, that's a big, mean, that's a big, mean, that's a big, big statement, big statement, big statement, right? You say right? You say right? You say all these investors, all all these investors, all all these investors, all this money this money this money coming in, all these coming in, all these coming in, all these experienced analysts experienced analysts experienced analysts at BlackRock and Goldman and so on, and at BlackRock and Goldman and so on, and at BlackRock and Goldman and so on, and you know you know you know hyperscalers hyperscalers hyperscalers like Google and Meta, like Google and Meta, like Google and Meta, but they're all wrong about something but they're all wrong about something but they're all wrong about something and you're not. and you're not. and you're not. So why do you think it's So why do you think it's So why do you think it's a bubble and they're a bubble and they're a bubble and they're all idiots? all idiots? all idiots? Well, I wouldn't go that Well, I wouldn't go that Well, I wouldn't go that far, since far, since far, since they're all just idiots. they're all just idiots. they're all just idiots. While that's true, I'm While that's true, I'm While that's true, I'm just...I'm...I'm going to digress a little just...I'm...I'm going to digress a little just...I'm...I'm going to digress a little from that from that from that question. So, what's question. So, what's question. So, what's unusual here? unusual here? unusual here? Let's quickly Let's quickly Let's quickly talk about what a talk about what a talk about what a bubble is. A bubble is. A bubble is. A bubble is when bubble is when bubble is when asset prices asset prices asset prices break away from break away from break away from any semblance of any semblance of any semblance of economic economic economic reality. We have reality. We have reality. We have one of the largest one of the largest one of the largest waves of competitive waves of competitive waves of competitive capital spending in capital spending in capital spending in history. So, in history. So, in history. So, in Western economic Western economic Western economic history, there were about history, there were about history, there were about eight or nine eight or nine eight or nine such moments when such moments when such moments when we were building some we were building some we were building some giant giant general- general- purpose technology, GPT, but purpose technology, GPT, but purpose technology, GPT, but not the kind that prepares GPT, but not the kind that prepares GPT, but general- general- purpose technology. And so purpose technology. And so purpose technology. And so when we build this, we've when we build this, we've when we build this, we've been railroads, been railroads, been railroads, electrification, electrification, electrification, canals, canals, canals, interstate interstate interstate highways, and highways, and highways, and each of those moments has each of those moments has each of those moments has led to led to overbuilding. There has n't been n't been a single major a single major a single major episode of capital episode of capital episode of capital spending in the history of the Western economy that hasn't spending in the history of the Western economy that hasn't spending in the history of the Western economy that hasn't ended in a series of ended in a series of ended in a series of crises, vastly crises, vastly crises, vastly inflated inflated inflated asset prices, and massive asset prices, and massive overcapacity, right overcapacity, right ? Railways, ? Railways, ? Railways, fiber optics, fiber optics, fiber optics, take your pick. So, take your pick. So, take your pick. So, it's a bubble.

it's a bubble. it's a bubble. The bubble is that The bubble is that The bubble is that we once again we once again we once again have a have a have a huge surplus of huge surplus of huge surplus of something due to something due to something due to the availability of capital the availability of capital the availability of capital and the desire of and the desire of capital providers capital providers to continue to continue to continue building in building in building in anticipation of anticipation of anticipation of future demand. future demand. future demand. This doesn't mean that This doesn't mean that This doesn't mean that artificial intelligence is artificial intelligence is artificial intelligence is frivolous, or that frivolous, or that frivolous, or that it doesn't work, or that there it doesn't work, or that there it doesn't work, or that there aren't aren't aren't incredibly useful incredibly useful incredibly useful arguments for it. It's not that arguments for it. It's not that arguments for it. It's not that important, nor was it important, nor was it important, nor was it the purpose of the railroads, is the purpose of the railroads, is the purpose of the railroads, is it? Railroads it? Railroads it? Railroads were incredibly were incredibly were incredibly useful, but still useful, but still useful, but still ended up in ended up in ended up in a situation where a situation where a situation where about 50% of the about 50% of the tracks laid tracks laid tracks laid were useless, were useless, were useless, right? So, right? So, right? So, a bubble is a a bubble is a a bubble is a huge surplus of huge surplus of huge surplus of something, something, something, largely fueled by largely fueled by largely fueled by credit, which in credit, which in credit, which in turn will lead to turn will lead to turn will lead to massive massive massive business bankruptcies and asset write-offs business bankruptcies and asset write-offs business bankruptcies and asset write-offs . And all this . And all this . And all this credit spread throughout the credit spread throughout the credit spread throughout the economy. This is economy. This is economy. This is within within within insurers, insurers, insurers, within private within private within private lending, lending, lending, within sovereign wealth within sovereign wealth within sovereign wealth funds. And we're now at funds. And we're now at funds. And we're now at about a about a about a trillion dollars, and we'll trillion dollars, and we'll probably hit probably hit probably hit 10 trillion before 10 trillion before 10 trillion before it's done. it's done. it's done. So all of these So all of these So all of these investments are at investments are at investments are at risk, and they risk, and they risk, and they span the entire span the entire span the entire economy and economy and economy and are are are inside all sorts of things, inside all sorts of things, inside all sorts of things, but even but even inside, you know, inside, you know, your your your individual, individual, individual, you know, your 401(k) you know, your 401(k) you know, your 401(k) retirement plan, right retirement plan, right ? I mean, we've ? I mean, we've ? I mean, we've all seen the data that all seen the data that all seen the data that something like the something like the something like the top 10 stocks in the S&P top 10 stocks in the S&P top 10 stocks in the S&P 500 account for about 500 account for about 500 account for about 40% of the index's market 40% of the index's market 40% of the index's market cap cap cap . So if you . So if you . So if you hold the S&P 500 index, hold the S&P 500 index, hold the S&P 500 index, my friend, you are very, very my friend, you are very, very my friend, you are very, very long AI.

long AI. long AI. Get out of my briefcase, Get out of my briefcase, Get out of my briefcase, Paul. Paul. Paul. Yeah, get that [ __ ] out of Yeah, get that [ __ ] out of Yeah, get that [ __ ] out of my briefcase. So my briefcase. So my briefcase. So you're a very, very long you're a very, very long you're a very, very long AI, and here's the AI, and here's the AI, and here's the problem: it's an expression problem: it's an expression problem: it's an expression you hear you hear you hear all the time. This is a all the time. This is a all the time. This is a one-factor world; one-factor world; one-factor world; everything is involved in AI in everything is involved in AI in everything is involved in AI in every every every asset class, in all asset class, in all asset class, in all retail retail retail investments. And the investments. And the investments. And the last statistic last statistic last statistic I'll give you, and the I'll give you, and the I'll give you, and the reason why it's reason why it's reason why it's important, is important, is important, is that we've reached that we've reached that we've reached record highs record highs record highs in the percentage of in the percentage of in the percentage of people's net worth people's net worth people's net worth tied up in stocks. tied up in stocks. tied up in stocks. 50% of people's net worth is 50% of people's net worth is 50% of people's net worth is now now now in stocks. in stocks. in stocks. So analyze these So analyze these So analyze these mathematical mathematical mathematical calculations, if 50% of calculations, if 50% of net worth net worth net worth is in stocks, is in stocks, is in stocks, and it's driven by and it's driven by and it's driven by AI trading AI trading AI trading , people are even , people are even , people are even more dependent on more dependent on more dependent on that single trade that single trade that single trade than they ever were on than they ever were on than they ever were on dotcoms, and that dotcoms, and that dotcoms, and that creates this creates this creates this incredible fragility incredible fragility . So this whole, this . So this whole, this . So this whole, this kind of nexus of kind of nexus of kind of nexus of factors is creating this factors is creating this factors is creating this bullish moment that bullish moment that bullish moment that has nothing to has nothing to has nothing to do with whether do with whether do with whether AI is useful, but has AI is useful, but has AI is useful, but has everything to do with everything to do with everything to do with why the other side of this is why the other side of this is why the other side of this is going to see, you know, a going to see, you know, a going to see, you know, a series of crises that series of crises that series of crises that could potentially could potentially could potentially become really become really become really catastrophic. catastrophic.

I want to focus for a moment I want to focus for a moment on the on the on the rationality of the rationality of the rationality of the major players, major players, major players, your metas, your your metas, your your metas, your alphabets, your alphabets, your alphabets, your open AIs, and the open AIs, and the open AIs, and the anthropes, all of whom anthropes, all of whom anthropes, all of whom believe that there will believe that there will believe that there will only be a few only be a few only be a few AI winners. I mean, AI winners. I mean, Zuckerberg gave this Zuckerberg gave this amazing quote where amazing quote where amazing quote where he said that if we he said that if we he said that if we end up spending end up spending end up spending a couple hundred a couple hundred a couple hundred billion dollars for something else billion dollars for something else billion dollars for something else , that , that , that would be unfortunate, but the would be unfortunate, but the would be unfortunate, but the risk of not risk of not risk of not acting aggressively enough is acting aggressively enough is acting aggressively enough is much higher. And so there much higher. And so there much higher. And so there 's this dynamic 's this dynamic 's this dynamic where, I think, they where, I think, they where, I think, they believe that this is going to be a believe that this is going to be a believe that this is going to be a highly concentrated highly concentrated highly concentrated or oligopolistic or oligopolistic or oligopolistic market, where maybe market, where maybe market, where maybe one model one model one model will rule will rule will rule them all, or three models them all, or three models them all, or three models will rule will rule will rule them all. And so there's this them all. And so there's this them all. And so there's this race to be race to be race to be one of the three, but it one of the three, but it one of the three, but it creates a kind of creates a kind of creates a kind of pressure to invest pressure to invest pressure to invest far beyond what the far beyond what the far beyond what the market can market can market can collectively withstand collectively withstand . . Right. This is a Right. This is a classic competitive classic competitive classic competitive dynamic that dynamic that dynamic that existed in each of existed in each of existed in each of these capital these capital these capital investments. It investments. It investments. It existed with existed with existed with the railways. You the railways. You the railways. You couldn't afford to be left couldn't afford to be left couldn't afford to be left behind. behind. behind. You had to You had to You had to invest, and that's why invest, and that's why invest, and that's why we got we got we got duplicate routes duplicate routes duplicate routes between cities. I between cities. I between cities. I mean, it's the essence of a mean, it's the essence of a mean, it's the essence of a competitive competitive competitive dynamic dynamic dynamic crossed with a crossed with a crossed with a kind of kind of kind of prisoner's dilemma that I would like to prisoner's dilemma that I would like to prisoner's dilemma that I would like to slow down, but I can't slow down, but I can't slow down, but I can't because no one else because no one else because no one else wants to. And so we all have to wants to. And so we all have to wants to. And so we all have to continue continue continue this fierce this fierce this fierce competitive dynamic competitive dynamic . And so, this is what we . And so, this is what we . And so, this is what we see embodied there.

see embodied there. see embodied there. And we see that even And we see that even And we see that even when I don't know, a couple of when I don't know, a couple of when I don't know, a couple of weeks ago Dario weeks ago Dario weeks ago Dario Amodei from Anthropic Amodei from Anthropic Amodei from Anthropic was suggesting that we was suggesting that we was suggesting that we need, and this is a need, and this is a need, and this is a nonsensical statement, that nonsensical statement, that nonsensical statement, that we need to we need to we need to step on the border, step on the border, step on the border, which sounds like which sounds like which sounds like training for the training for the training for the Boston Boston Boston Marathon. But it's the idea Marathon. But it's the idea Marathon. But it's the idea that we need to that we need to that we need to slow down. Well, slow down. Well, slow down. Well, of course, everyone in the of course, everyone in the of course, everyone in the classic classic classic prisoner's dilemma said, "Come on. prisoner's dilemma said, "Come on. prisoner's dilemma said, "Come on. You slow down. I'm not You slow down. I'm not You slow down. I'm not slowing down." And slowing down." And slowing down." And worse, they worse, they worse, they said, "Even if said, "Even if said, "Even if all of us here at all of us here at all of us here at American American American border border border companies companies companies agree to agree to agree to slow down, China slow down, China slow down, China won't." And won't." And won't." And then you get into then you get into then you get into this terrible debate, this terrible debate, this terrible debate, well, with what's well, with what's well, with what's going on, about what going on, about what China is going to do. And here's the problem China is going to do. And here's the problem , there's a , there's a , there's a racial dynamic that's created, that racial dynamic that's created, that racial dynamic that's created, that if you stop, there's a if you stop, there's a if you stop, there's a fear that either my fear that either my fear that either my competitors will gain an competitors will gain an competitors will gain an advantage or my advantage or my advantage or my shareholders will punish shareholders will punish shareholders will punish me. And so, as me. And so, as me. And so, as a result, the racial a result, the racial a result, the racial dynamic dynamic dynamic continues, and that continues, and that continues, and that guarantees guarantees guarantees overbuilding. As long as overbuilding. As long as overbuilding. As long as you have the capital stock you have the capital stock you have the capital stock and the momentum of and the momentum of and the momentum of the race, you are the race, you are the race, you are guaranteed excess guaranteed excess guaranteed excess capacity, capacity, capacity, e.g. canals, e.g. canals, e.g. canals, railways, railways, railways, electrification – electrification – take your pick. take your pick. take your pick. Azim, you and your Azim, you and your Azim, you and your team have been building this dashboard for the team have been building this dashboard for the team have been building this dashboard for the last year, last year, last year, year and a half, year and a half, trying trying to track to track to track financial stress in the financial stress in the financial stress in the AI ​​industry AI ​​industry AI ​​industry , , , trying trying trying to think through what it to think through what it to think through what it looks like when it's looks like when it's looks like when it's on a on a on a healthy healthy healthy financial path, and what financial path, and what financial path, and what it looks like when it looks like when it looks like when it becomes a bubble.

it becomes a bubble. it becomes a bubble. So, why don't you So, why don't you So, why don't you tell me about tell me about tell me about your perspective, the your perspective, the your perspective, the metrics metrics metrics you're following most you're following most you're following most closely, and why you're closely, and why you're closely, and why you're a little more a little more a little more optimistic than optimistic than optimistic than Paul right now. Paul right now. Well, when we Well, when we look at look at look at the technology boom, the the technology boom, the the technology boom, the technology cycle, technology cycle, technology cycle, we need to we need to we need to consider two consider two consider two different things. First, do different things. First, do different things. First, do customers find customers find customers find it useful, do it useful, do it useful, do they pay for it, or do they pay for it, or do they pay for it, or do they pay more for it they pay more for it they pay more for it ? In other ? In other ? In other words, is the demand words, is the demand words, is the demand real, and does real, and does real, and does demand exist? And when you demand exist? And when you demand exist? And when you look back at look back at look back at bubbles that bubbles that bubbles that burst, it's almost burst, it's almost burst, it's almost always related to a always related to a always related to a problem in the problem in the financing mechanism, financing mechanism, the refinancing of the refinancing of the refinancing of short-term short-term short-term debt. It hits a debt. It hits a debt. It hits a problem, capital problem, capital problem, capital disappears, and you see disappears, and you see disappears, and you see all the things we all the things we all the things we associate with associate with associate with bubbles. The question is bubbles. The question is bubbles. The question is where are we today where are we today ? And where we are today is that ? And where we are today is that ? And where we are today is that demand is very demand is very demand is very strong from strong from strong from American and American and American and global businesses global businesses . There is no . There is no . There is no surplus yet. There is a surplus yet. There is a surplus yet. There is a huge shortage of huge shortage of huge shortage of computing computing computing hardware needed hardware needed hardware needed for AI. And the race for AI. And the race for AI. And the race seems to be seems to be seems to be whether we can whether we can whether we can get enough get enough get enough chips and powerful chips and powerful chips and powerful chips in the chips in the chips in the next two to three next two to three next two to three years to keep up years to keep up years to keep up with demand. And on the with demand. And on the with demand. And on the funding side, we are funding side, we are funding side, we are certainly certainly certainly giving up the giving up the giving up the cash that cash that cash that hyperscalers hyperscalers hyperscalers have earned over the last 20 have earned over the last 20 have earned over the last 20 years or so years or so years or so , and we are raising , and we are raising , and we are raising debt, but it's not yet at debt, but it's not yet at debt, but it's not yet at levels where you would levels where you would levels where you would say it's say it's say it's historically historically historically high and high and high and challenging. So for me challenging. So for me challenging. So for me today it's, you know, today it's, you know, today it's, you know, this productive boom this productive boom this productive boom with hints of maybe, with hints of maybe,

with hints of maybe, excuse excuse excuse my French, a little my French, a little my French, a little stinky. stinky. stinky. So, I want to So, I want to So, I want to focus on focus on supply and demand at supply and demand at this very this very this very simple level. Paul simple level. Paul simple level. Paul essentially implies essentially implies essentially implies that there is a strong that there is a strong that there is a strong incentive to create incentive to create incentive to create too much too much too much supply. And you mean supply. And you mean supply. And you mean that that that the demand is actually the demand is actually the demand is actually high. So I want to high. So I want to high. So I want to break down what we mean break down what we mean break down what we mean when when when we talk about demand. we talk about demand. So this is a demand for So this is a demand for the use of AI, the use of AI, the use of AI, which is tokens, which, which is tokens, which, which is tokens, which, again, for again, for again, for people who are not used people who are not used people who are not used to this, when you to this, when you to this, when you ask a question, the ask a question, the ask a question, the AI ​​model consumes AI ​​model consumes AI ​​model consumes tokens to tokens to tokens to answer it. answer it. When you ask an When you ask an AI developer AI developer AI developer to write your to write your to write your program code, they program code, they program code, they consume tokens to consume tokens to consume tokens to write the program. So write the program. So write the program. So functionally, with functionally, with functionally, with companies like companies like companies like OpenAI or Anthropic, people OpenAI or Anthropic, people OpenAI or Anthropic, people buy tokens from them buy tokens from them buy tokens from them to use, to use, to use, right? They're not right? They're not right? They're not selling you, Claude, selling you, Claude, selling you, Claude, they're selling you they're selling you they're selling you tokens at high tokens at high tokens at high levels. This is why levels. This is why levels. This is why tokens are an important tokens are an important tokens are an important concept here. And so, firstly, concept here. And so, firstly, the question arises, how many the question arises, how many tokens tokens tokens will people want, and how will people want, and how will people want, and how strong will the demand for strong will the demand for strong will the demand for tokens be? And then tokens be? And then tokens be? And then there is the secondary there is the secondary there is the secondary question: how much question: how much question: how much will they pay for them?

will they pay for them? will they pay for them? Because the price per token Because the price per token Because the price per token drops sharply. So drops sharply. So drops sharply. So the question arises: the question arises: the question arises: even if there is a even if there is a even if there is a high demand, will there high demand, will there high demand, will there be a demand for those very be a demand for those very be a demand for those very expensive models that expensive models that American American companies are investing in or not? So I'd like you to companies are investing in or not? So I'd like you to tell me what this tell me what this demand picture demand picture demand picture looks like for you. looks like for you. looks like for you. Well, we need to Well, we need to Well, we need to start at a level start at a level start at a level below below below tokens. The asset tokens. The asset tokens. The asset we invest in is we invest in is we invest in is data centers, data centers, data centers, 65% of the cost of which 65% of the cost of which 65% of the cost of which is a chip, is a chip, is a chip, GPU, GPU, GPU, or or or artificial intelligence accelerator. artificial intelligence accelerator. artificial intelligence accelerator. So, in my opinion, it So, in my opinion, it So, in my opinion, it 's important to pay 's important to pay 's important to pay attention to what attention to what attention to what the demand is for these chips to the demand is for these chips to the demand is for these chips to perform the computations perform the computations perform the computations that in some cases that in some cases , and in most , and in most , and in most cases, produce cases, produce cases, produce tokens. And who are the tokens. And who are the tokens. And who are the customers of these customers of these customers of these GPUs GPUs ? Uh, when we look ? Uh, when we look ? Uh, when we look at this, we see these at this, we see these at this, we see these hyperscalers hyperscalers hyperscalers like Amazon, Microsoft, and Google, and like Amazon, Microsoft, and Google, and like Amazon, Microsoft, and Google, and then new then new then new labs labs labs called Neo Cloud. These people called Neo Cloud. These people called Neo Cloud. These people are buying millions of these are buying millions of these are buying millions of these GPUs GPUs . They put them . They put them . They put them together in together in together in data centers, which data centers, which data centers, which will now start will now start will now start costing billions of costing billions of costing billions of dollars, and rent out these dollars, and rent out these dollars, and rent out these GPUs GPUs GPUs . And what I . And what I . And what I think about a lot is think about a lot is think about a lot is how the capital how the capital how the capital expenditure on the actual expenditure on the actual expenditure on the actual production asset production asset production asset is converted into is converted into is converted into revenue with enough revenue with enough ROI to ROI to pay for itself. And now the pay for itself. And now the pay for itself. And now the reason we reason we reason we know we're know we're know we're in a in a compute-constrained environment is because people because people are getting an are getting an are getting an extremely good extremely good extremely good return on the asset return on the asset return on the asset after 18 to 24 months

after 18 to 24 months after 18 to 24 months because they can because they can because they can charge essentially charge essentially charge essentially extremely high extremely high extremely high prices for their prices for their prices for their GPUs. And about GPUs. And about GPUs. And about 50% of their revenue goes to 50% of their revenue goes to 50% of their revenue goes to two major two major AI companies, AI companies, Anthropic and OpenAI, for their Anthropic and OpenAI, for their Anthropic and OpenAI, for their expenses. The other 50% is a expenses. The other 50% is a expenses. The other 50% is a mix of other mix of other mix of other companies and AI firms companies and AI firms that are directly that are directly using these using these using these open open open source models that you source models that you source models that you mentioned or mentioned or mentioned or open source models on open source models on hyperscaler infrastructure. And hyperscaler infrastructure. And then I guess we then I guess we then I guess we move on to your move on to your move on to your other question, which is, other question, which is, other question, which is, if 50 if 50 % of this is coming from % of this is coming from % of this is coming from Anthropic and OpenAI, what of Anthropic and OpenAI, what of Anthropic and OpenAI, what of that is actually the that is actually the that is actually the real real real customer spend? And I would say customer spend? And I would say that if these that if these that if these ratios ratios ratios remained remained remained the same for the the same for the the same for the next three or next three or four years, we would four years, we would four years, we would end up in a world where end up in a world where end up in a world where someone is actually someone is actually someone is actually paying themselves through the paying themselves through the paying themselves through the back door. But what back door. But what back door. But what 's happening is that 's happening is that 's happening is that this ratio is this ratio is this ratio is going down because, going down because, going down because, essentially, essentially, essentially, hyperscalers are hyperscalers are hyperscalers are getting more getting more getting more and more business from and more business from and more business from real customers. And real customers. And real customers. And all of this, to me, is a all of this, to me, is a all of this, to me, is a sign of a market that is sign of a market that is sign of a market that is maturing week after maturing week after maturing week after week, quarter after week, quarter after week, quarter after quarter.

quarter. quarter. Paul, how do you look at Paul, how do you look at Paul, how do you look at this differently? this differently? this differently? I don't know where to I don't know where to I don't know where to start. Um, let start. Um, let start. Um, let me just say me just say me just say that a lot of the discussion, and that a lot of the discussion, and that a lot of the discussion, and I use that I use that I use that word intentionally word intentionally word intentionally in the way in the way in the way people talk about these people talk about these people talk about these things, reminds me of things, reminds me of things, reminds me of that old joke that old joke that old joke about the mother about the mother about the mother coming down in the morning coming down in the morning coming down in the morning , finding her , finding her , finding her daughter under the Christmas tree, and daughter under the Christmas tree, and daughter under the Christmas tree, and there's a huge pile of there's a huge pile of there's a huge pile of horse manure, and the horse manure, and the horse manure, and the daughter is digging in the pile of daughter is digging in the pile of daughter is digging in the pile of horse manure, and horse manure, and horse manure, and the mother says, "But what are you the mother says, "But what are you the mother says, "But what are you doing?" What are you watching? doing?" What are you watching? doing?" What are you watching? ""Well, there's a huge ""Well, there's a huge ""Well, there's a huge pile of horse manure there. There pile of horse manure there. There pile of horse manure there. There must be a pony here somewhere must be a pony here somewhere must be a pony here somewhere . “And that...To me, . “And that...To me, . “And that...To me, a lot of this a lot of this a lot of this discussion about the discussion about the discussion about the giant scale of giant scale of giant scale of tokens is tokens is tokens is a lot like digging a lot like digging a lot like digging in a pile of manure, because there in a pile of manure, because there in a pile of manure, because there has to be a pony in there somewhere. Yes has to be a pony in there somewhere. Yes , because it is a confusion , because it is a confusion , because it is a confusion of scale, a huge of scale, a huge of scale, a huge number of tokens with number of tokens with number of tokens with their value. So their value. So their value. So I'll agree with I'll agree with I'll agree with Asim's point about, um, this Asim's point about, um, this Asim's point about, um, this huge huge huge use of tokens use of tokens . Tokens alone don't . Tokens alone don't . Tokens alone don't matter, right matter, right matter, right ? This is a pile of manure under the Christmas ? This is a pile of manure under the Christmas ? This is a pile of manure under the Christmas tree. What is important is tree. What is important is tree. What is important is the multiplication of tokens the multiplication of tokens the multiplication of tokens by price, and you by price, and you by price, and you mentioned earlier that mentioned earlier that mentioned earlier that over the last few over the last few over the last few years we have seen years we have seen years we have seen about a 13x about a 13x about a 13x annual decline in annual decline in annual decline in token price. This is token price. This is token price. This is adjusted for adjusted for adjusted for performance. So performance. So performance. So the response from companies the response from companies the response from companies developing developing developing models on the front lines models on the front lines models on the front lines will be: "Hey, hey, will be: "Hey, hey, will be: "Hey, hey, hey, hey, hey."

hey, hey, hey." hey, hey, hey." Of course, older Of course, older Of course, older models fall like this, models fall like this, models fall like this, but not new ones. The new but not new ones. The new but not new ones. The new models are the ones that models are the ones that models are the ones that can break through...These are can break through...These are can break through...These are still much more still much more still much more expensive." “So they expensive." “So they expensive." “So they hold their hold their hold their value, but the value, but the value, but the problem, of course, is problem, of course, is problem, of course, is that for that for that for most people, for most people, for most people, for most of what most of what most of what they do, they don’t they do, they don’t they do, they don’t need the much more need the much more need the much more expensive models. And expensive models. And expensive models. And in fact, most in fact, most in fact, most normal people normal people get by quite adequately with relatively get by quite adequately with relatively low-performance low-performance low-performance models. A good models. A good models. A good example is this example is this example is this incredible incredible incredible craze for craze for craze for Meta's Muse product, which Meta's Muse product, which Meta's Muse product, which dominated the dominated the dominated the app charts and everything app charts and everything app charts and everything else that else that else that uses a uses a uses a relatively primitive relatively primitive relatively primitive model. He doesn't model. He doesn't model. He doesn't use a use a use a cutting-edge model, and no cutting-edge model, and no cutting-edge model, and no one cares. one cares. one cares. It just It just It just works great for what works great for what works great for what most people want most people want most people want to do, which allows to do, which allows to do, which allows me to go back to the me to go back to the me to go back to the beginning. One of the beginning. One of the beginning. One of the reasons people reasons people reasons people really mischaracterize really mischaracterize this whole market, and it's like the this whole market, and it's like the old old old venture capital adage, is to venture capital adage, is to never confuse never confuse your early your early your early clients about what the clients about what the real market will look like. A real market will look like. A lot of what lot of what lot of what happened with the advanced happened with the advanced happened with the advanced models is that models is that models is that people confused the early people confused the early people confused the early advanced advanced advanced model customers with what you and I model customers with what you and I model customers with what you and I want. So want. So who were the early who were the early who were the early customers of customers of customers of artificial intelligence models, artificial intelligence models, artificial intelligence models, right? Who was right? Who was right? Who was actually in charge of actually in charge of actually in charge of this? Programmers, this? Programmers, this? Programmers, right? Something like right? Something like right? Something like 80% of 80% of 80% of token usage and..." The revenues token usage and..." The revenues token usage and..." The revenues from the biggest AI, from the biggest AI, from the biggest AI, the biggest companies the biggest companies the biggest companies developing developing developing models on the front lines models on the front lines models on the front lines are coders, are coders, are coders, developers. And developers. And developers. And developers are truly developers are truly developers are truly unique for at least unique for at least unique for at least two reasons. Why?

two reasons. Why? two reasons. Why? First, they want First, they want First, they want advanced models, advanced models, advanced models, right? They are willing right? They are willing right? They are willing to pay for them because they to pay for them because they to pay for them because they have really have really have really complex problems. complex problems. Secondly, they seem to me to be Secondly, they seem to me to be Secondly, they seem to me to be quite quite quite extensive in terms of extensive in terms of extensive in terms of using tokens using tokens , that is, I enter a , that is, I enter a , that is, I enter a small query and say small query and say : "You know, bro, : "You know, bro, : "You know, bro, create a fitness create a fitness app for me." And suddenly app for me." And suddenly app for me." And suddenly seven words seven words seven words turn into a turn into a turn into a million lines of code. million lines of code. million lines of code. This is not at all like what you and I This is not at all like what you and I do most of the time. do most of the time. Much of what Much of what Much of what ordinary ordinary ordinary people and ordinary people and ordinary people and ordinary businesses do is closer businesses do is closer businesses do is closer to one-on- to one-on- to one-on- one relationships or even one relationships or even one relationships or even succinctly. As I succinctly. As I succinctly. As I said, I got 25 said, I got 25 said, I got 25 emails emails emails from Goldman Sachs from Goldman Sachs from Goldman Sachs talking about something. Is talking about something. Is talking about something. Is any of any of any of this worth reading? And that's this worth reading? And that's this worth reading? And that's concise, isn't it? concise, isn't it? concise, isn't it? Because I take a bunch of Because I take a bunch of Because I take a bunch of things, create them, and things, create them, and things, create them, and turn them into something turn them into something turn them into something small. And that small. And that small. And that seems seems seems effective to me. Or effective to me. Or effective to me. Or similarly, some similarly, some similarly, some other other other email, calendar, or email, calendar, or email, calendar, or something like that. So, the bottom line is something like that. So, the bottom line is something like that. So, the bottom line is that the everyday that the everyday that the everyday people these people these people these frontier models frontier models frontier models want to sell want to sell want to sell to now are not very well to now are not very well to now are not very well represented by what represented by what represented by what people did in the early people did in the early people did in the early stages of the market. Because stages of the market. Because stages of the market. Because decoders are truly decoders are truly decoders are truly unusual. And you're already unusual. And you're already unusual. And you're already seeing some of this, uh seeing some of this, uh , meta is , meta is , meta is halving the number of halving the number of halving the number of in-house developers in-house developers in-house developers who are allowed to who are allowed to who are allowed to use use use external frontier external frontier external frontier tools. Microsoft, same thing tools. Microsoft, same thing tools. Microsoft, same thing . You know, the days of . You know, the days of . You know, the days of token maximization token maximization , when, you know, there were , when, you know, there were , when, you know, there were internal charts internal charts internal charts that showed who was that showed who was that showed who was winning by winning by winning by using using using the most tokens. It's the most tokens. It's the most tokens. It's long gone.

long gone. long gone. Normal companies Normal companies Normal companies cannot tolerate this. cannot tolerate this. cannot tolerate this. People spent People spent People spent their entire budgets for their entire budgets for their entire budgets for months. So my point is months. So my point is months. So my point is that we are already that we are already that we are already moving away from those moving away from those moving away from those early days of early days of early days of wasteful wasteful wasteful token use token use , when everyone was looking for ponies , when everyone was looking for ponies , when everyone was looking for ponies inside inside inside giant piles of dung. giant piles of dung. giant piles of dung. Azim? Azim? Azim? Yes. In his Yes. In his Yes. In his giant pile of giant pile of giant pile of dung, Paul also had dung, Paul also had dung, Paul also had some sleight of hand. some sleight of hand. some sleight of hand. Uh, what Paul Uh, what Paul Uh, what Paul was talking about is frontier was talking about is frontier was talking about is frontier models and tokens. Uh, models and tokens. Uh, models and tokens. Uh, and what I was talking about and what I was talking about and what I was talking about is the capital is the capital is the capital expenditure and the expenditure and the expenditure and the power that power that power that we were talking about, namely the we were talking about, namely the we were talking about, namely the GPUs. And GPUs. And GPUs. And I think these are two I think these are two I think these are two different things that different things that different things that deserve two deserve two deserve two different discussions. I different discussions. I different discussions. I mean, what mean, what mean, what happens to the happens to the cutting-edge labs, OpenAI cutting-edge labs, OpenAI and Anthropic, with this and Anthropic, with this and Anthropic, with this business model of business model of business model of increasingly powerful increasingly powerful increasingly powerful AI systems AI systems AI systems that have a that have a that have a smaller and smaller smaller and smaller smaller and smaller set of users, is set of users, is set of users, is quite different quite different quite different from the question of what from the question of what from the question of what happens to the happens to the happens to the trillions of dollars of trillions of dollars of capital that will be capital that will be invested in invested in invested in building data centers building data centers where American and where American and global companies global companies global companies will start will start will start using more and more using more and more AI. And AI. And I think we I think we I think we need to distinguish between need to distinguish between need to distinguish between these two issues. One is these two issues. One is these two issues. One is about the about the about the fundamental fundamental fundamental viability of Anthropic viability of Anthropic viability of Anthropic or Open AI in the or Open AI in the or Open AI in the long long long term and what term and what term and what they can they can they can do in do in do in the face of the pressures the face of the pressures the face of the pressures that Paul talked about. And the that Paul talked about. And the that Paul talked about. And the other question is other question is other question is whether whether whether GPUs GPUs GPUs and this new digital and this new digital and this new digital infrastructure that's infrastructure that's infrastructure that's being built will find being built will find being built will find application application application primarily among primarily among primarily among enterprises, but enterprises, but enterprises, but of course, as we of course, as we of course, as we heard correctly, among heard correctly, among heard correctly, among consumers, and also consumers, and also consumers, and also through things like, through things like, through things like, you know, news and you know, news and you know, news and other agents. I think these are

other agents. I think these are other agents. I think these are different issues different issues different issues when when when you separate these you separate these you separate these issues. And I understand issues. And I understand issues. And I understand that maybe this is an that maybe this is an that maybe this is an important point of important point of important point of confusion that confusion that confusion that needs to be disentangled, needs to be disentangled, needs to be disentangled, because people can because people can because people can mean different mean different mean different things when they... things when they... talk about whether talk about whether talk about whether there is a bubble, and what it there is a bubble, and what it there is a bubble, and what it would mean if it would mean if it would mean if it bursts or bursts or bursts or deflates. As far as deflates. As far as deflates. As far as I can tell, if it I can tell, if it I can tell, if it turns out that Open AI, turns out that Open AI, turns out that Open AI, Anthropic, or for that Anthropic, or for that Anthropic, or for that matter Alphabet or anyone matter Alphabet or anyone matter Alphabet or anyone else, that their business else, that their business model doesn't work here, model doesn't work here, model doesn't work here, then people are actually then people are actually then people are actually quite happy quite happy quite happy using Deep Seek from using Deep Seek from using Deep Seek from China for a lot China for a lot China for a lot less money, and Anthropic is essentially less money, and Anthropic is essentially less money, and Anthropic is essentially collapsing on its collapsing on its collapsing on its own, which own, which own, which will lead to the implosion of will lead to the implosion of will lead to the implosion of the economy. I the economy. I the economy. I think this will become a think this will become a think this will become a contagion in the stock contagion in the stock contagion in the stock market. I don't market. I don't market. I don't necessarily think necessarily think necessarily think it's going to lead to a it's going to lead to a it's going to lead to a depression, but I think depression, but I think depression, but I think it's going to have a very, it's going to have a very, very big impact on very big impact on very big impact on ratings, on what people ratings, on what people ratings, on what people are willing to borrow in the are willing to borrow in the are willing to borrow in the future. Do future. Do future. Do n't you think so? n't you think so? Well, I think you Well, I think you need to step back need to step back need to step back and ask what the and ask what the and ask what the fundamental fundamental fundamental risk is with risk is with risk is with building out building out building out infrastructure, which is that infrastructure, which is that infrastructure, which is that you don't have enough you don't have enough you don't have enough customers customers who demand the who demand the who demand the product and pay product and pay product and pay enough for that enough for that enough for that underlying asset. And if underlying asset. And if underlying asset. And if people can transition, people can transition, people can transition, businesses can businesses can businesses can transition to transition to transition to open-source models.

open-source models. open-source models. which are driven which are driven which are driven directly by directly by directly by the hyperscalar, i.e. the hyperscalar, i.e. the hyperscalar, i.e. simply the movement of demand. This simply the movement of demand. This simply the movement of demand. This may hurt may hurt may hurt Anthropic's bottom line, but it Anthropic's bottom line, but it Anthropic's bottom line, but it still monetizes still monetizes still monetizes major capital major capital major capital expenditures. And then, what is expenditures. And then, what is expenditures. And then, what is one of the things one of the things one of the things that these chips are, essentially? that these chips are, essentially? that these chips are, essentially? Data centers. Data centers. Data centers. Nvidia Anthropic may Nvidia Anthropic may Nvidia Anthropic may fail, but fail, but fail, but Nvidia will be fine. Nvidia will be fine. Nvidia will be fine. But, but even so, But, but even so, But, but even so, Anthropic has many Anthropic has many Anthropic has many degrees of freedom. I degrees of freedom. I degrees of freedom. I mean, in the mean, in the mean, in the last 24 hours or last 24 hours or last 24 hours or 48 hours since we 48 hours since we 48 hours since we wrote this, it has wrote this, it has wrote this, it has lowered prices and lowered prices and lowered prices and released even cheaper, more released even cheaper, more released even cheaper, more efficient lightweight efficient lightweight efficient lightweight models. These are still models. These are still models. These are still models around which models around which models around which they have a pretty they have a pretty they have a pretty decent margin. They decent margin. They decent margin. They will do what will do what software companies have done software companies have done for decades, which is for decades, which is for decades, which is try to create try to create try to create reasons to reasons to reasons to use their use their software beyond its software beyond its core core core capabilities. That's capabilities. That's capabilities. That's why they have why they have why they have Claude as a colleague, and they have Claude Claude as a colleague, and they have Claude Claude as a colleague, and they have Claude as a lawyer. And so I as a lawyer. And so I as a lawyer. And so I think the question is, think the question is, think the question is, if Anthropic disappears if Anthropic disappears if Anthropic disappears tomorrow night, well, I can't tomorrow night, well, I can't tomorrow night, well, I can't imagine the imagine the imagine the circumstances under which that circumstances under which that circumstances under which that would happen. But what I would happen. But what I would happen. But what I mean is that mean is that mean is that you see the demand from you see the demand from you see the demand from the end customers, and that's something the end customers, and that's something the end customers, and that's something without which it will without which it will without which it will always end always end always end horribly.

horribly. horribly. Can I just Can I just Can I just ask you too, ask you too, ask you too, because I want to because I want to because I want to actually ask actually ask actually ask you both this question, you both this question, you both this question, what is considered a what is considered a bubble bursting for you? bubble bursting for you? Because there's a world where, you Because there's a world where, you Because there's a world where, you know, you have, let's say know, you have, let's say , a giant , a giant , a giant stock market correction, but stock market correction, but stock market correction, but data centers are data centers are data centers are still still still being used, and being used, and being used, and I think you're I think you're I think you're saying that this is not a saying that this is not a saying that this is not a bubble bursting. It's bubble bursting. It's bubble bursting. It's just a correction or just a correction or just a correction or something. something. something. Well, historically, Well, historically, Well, historically, bubbles, when they bubbles, when they bubbles, when they burst, resulted burst, resulted burst, resulted in two things. in two things. First, there is a huge First, there is a huge First, there is a huge decline in the level of decline in the level of decline in the level of investment in this investment in this investment in this particular asset, and particular asset, and particular asset, and it was it was it was accompanied by a accompanied by a accompanied by a significant crash in the significant crash in the significant crash in the relevant relevant relevant asset market, whether it be the asset market, whether it be the asset market, whether it be the stock market or stock market or stock market or something else. And this is the test something else. And this is the test something else. And this is the test I would use. I I would use. I I would use. I think we could think we could think we could end up in a world where there will end up in a world where there will end up in a world where there will be extremely be extremely be extremely low and shameful low and shameful low and shameful returns for the returns for the returns for the credit funds and credit funds and credit funds and for the hyperscalers for the hyperscalers for the hyperscalers because of the investments because of the investments because of the investments they made, but they made, but they made, but that level of pain is that level of pain is that level of pain is not a bubble bursting.

not a bubble bursting. It's just someone It's just someone made a few made a few made a few bad investments and bad investments and bad investments and bought expensively when... they bought expensively when... they should have bought at a lower should have bought at a lower should have bought at a lower price. I mean, price. I mean, price. I mean, I would use a very I would use a very I would use a very classic historical classic historical classic historical definition, which is that definition, which is that definition, which is that we we we need to see a need to see a 50% drop in investment levels, and 50% drop in investment levels, and we need to we need to we need to see a really see a really see a really significant, usually 50% significant, usually 50% or more, correction or more, correction or more, correction in any in any in any relevant market. relevant market. relevant market. Paul, the same question Paul, the same question Paul, the same question to you. to you. I'm fine. So, I...I I'm fine. So, I...I disagree to some extent disagree to some extent disagree to some extent with almost with almost with almost every point that every point that every point that Azim makes. These are Azim makes. These are Azim makes. These are arbitrary definitions, arbitrary definitions, arbitrary definitions, and I think let's and I think let's and I think let's just say it like this just say it like this . Okay, back . Okay, back . Okay, back to my original to my original to my original point. Something like point. Something like point. Something like 50% of the current net 50% of the current net 50% of the current net worth of people in the worth of people in the worth of people in the United States is United States is United States is tied up in stocks. tied up in stocks. tied up in stocks. Most stocks Most stocks Most stocks right now are technology. right now are technology. right now are technology. Most technology Most technology Most technology today is artificial today is artificial today is artificial intelligence. Things are intelligence. Things are intelligence. Things are only going to get worse only going to get worse only going to get worse as these two as these two as these two monstrous IPOs monstrous IPOs monstrous IPOs hit the market hit the market hit the market in the next, I in the next, I in the next, I don't know, 2 months or don't know, 2 months or don't know, 2 months or so. so. so. OpenAI and Anthropic? OpenAI and Anthropic? OpenAI and Anthropic? OpenAI and Anthropic. And just OpenAI and Anthropic. And just OpenAI and Anthropic. And just to put that in to put that in to put that in context, by the way, OpenAI context, by the way, OpenAI , Anthropic, and SpaceX, if , Anthropic, and SpaceX, if , Anthropic, and SpaceX, if you want to include... you want to include... Together, those three IPOs are bigger Together, those three IPOs are bigger than all the IPOs since than all the IPOs since than all the IPOs since World World World War II combined. This is a War II combined. This is a War II combined. This is a truly extraordinary truly extraordinary truly extraordinary moment.

moment. moment. This is a crazy This is a crazy This is a crazy statistic. statistic. statistic. This is a crazy This is a crazy This is a crazy statistic. Not statistic. Not statistic. Not only that, the combined only that, the combined only that, the combined value of these value of these value of these companies is greater than companies is greater than companies is greater than the value of the the value of the the value of the Nasdaq market capitalization at Nasdaq market capitalization at Nasdaq market capitalization at its peak. Bubble its peak. Bubble its peak. Bubble is sitting right in front of is sitting right in front of is sitting right in front of people, and they want people, and they want people, and they want to say, "Well, I expect to say, "Well, I expect you to come with you to come with you to come with blonde hair, blonde hair, blonde hair, not brown." " not brown." " Dude, just because it doesn't Dude, just because it doesn't Dude, just because it doesn't look the way you look the way you look the way you want it to doesn't make it any want it to doesn't make it any want it to doesn't make it any less fragile." So, less fragile." So, less fragile." So, you can imagine, let's say you can imagine, let's say , just for the sake of , just for the sake of , just for the sake of argument, the Anthropic IPO argument, the Anthropic IPO argument, the Anthropic IPO would be delayed. This would be delayed. This would be delayed. This will not happen in will not happen in will not happen in November, but November, but November, but next year. next year. next year. There will be a drastic There will be a drastic There will be a drastic reassessment of their reassessment of their reassessment of their ability to raise ability to raise ability to raise funds for the scammers, as funds for the scammers, as funds for the scammers, as well as the companies that well as the companies that well as the companies that supply them, supply them, supply them, because everything depends on the because everything depends on the because everything depends on the valuation of these companies, valuation of these companies, valuation of these companies, which which which are now in the are now in the are now in the private markets. So private markets. So , I like to say that , I like to say that , I like to say that this failure here is this failure here is this failure here is overrated. There are so overrated. There are so overrated. There are so many ways things many ways things many ways things can break, and can break, and can break, and break really break really break really catastrophically, catastrophically, catastrophically, because I'll give you another because I'll give you another because I'll give you another example. What's example. What's example. What's called the disease is the called the disease is the Shiller cyclically adjusted price-earnings multiple, Shiller cyclically adjusted price-earnings multiple, which is a kind of which is a kind of which is a kind of long-term long-term long-term indicator of where indicator of where U.S. stocks are. We are U.S. stocks are. We are at the levels we at the levels we at the levels we were at just were at just were at just before the 2029 crash.

before the 2029 crash. before the 2029 crash. We are at We are at We are at absolutely historically absolutely historically absolutely historically unprecedented unprecedented unprecedented levels in terms of levels in terms of levels in terms of valuation of the entire market. valuation of the entire market. valuation of the entire market. And the problem is that And the problem is that And the problem is that people lack imagination and people lack imagination and people lack imagination and they just they just they just refuse refuse refuse to engage with the to engage with the to engage with the broader data about..." broader data about..." broader data about..." what's really what's really what's really going on in the world, going on in the world, going on in the world, because they say, "Well, because they say, "Well, because they say, "Well, this doesn't look like the this doesn't look like the this doesn't look like the last bubble, last bubble, last bubble, so we must be so we must be so we must be fine." " fine." " Well, let me... Well, let me... But I think I'll But I think I'll But I think I'll just get just get just get straight to the straight to the straight to the point." I mean point." I mean , I think it's very , I think it's very , I think it's very easy easy easy to use very to use very to use very colorful language. I colorful language. I colorful language. I always remember the always remember the always remember the data. This is not colorful data. This is not colorful data. This is not colorful language. These are actually language. These are actually language. These are actually numbers. These are numbers, like numbers. These are numbers, like numbers. These are numbers, like your answer, we your answer, we your answer, we are dealing with are dealing with are dealing with historically historically historically unprecedented unprecedented unprecedented numbers, like numbers, like numbers, like cyclically cyclically cyclically adjusted earnings for prices. adjusted earnings for prices. adjusted earnings for prices. Like, is this good? Is that Like, is this good? Is that Like, is this good? Is that good in your book? good in your book? good in your book? Well, one of the answers Well, one of the answers Well, one of the answers to the Cape to the Cape Shiller index is Shiller index is Shiller index is that the Cape that the Cape Shiller index doesn't Shiller index doesn't Shiller index doesn't say anything about time. say anything about time. say anything about time. No, that's because it's an No, that's because it's an No, that's because it's an adjusted adjusted adjusted long-term long-term long-term horizon. This is not horizon. This is not horizon. This is not just an annual, just an annual, it is a historical 10- year look back, year look back, year look back, which can sometimes which can sometimes which can sometimes underestimate fundamental underestimate fundamental underestimate fundamental changes in the changes in the changes in the economic economic and market regime. But there's and market regime. But there's something else that I think something else that I think something else that I think a lot of what you're a lot of what you're a lot of what you're describing, Paul, has already been describing, Paul, has already been describing, Paul, has already been priced in... by priced in... by priced in... by equity investors, and its price equity investors, and its price equity investors, and its price is determined by the is determined by the is determined by the credit markets.

credit markets. credit markets. We've only seen this in the We've only seen this in the We've only seen this in the last few last few last few weeks. We saw an weeks. We saw an weeks. We saw an IPO of an Australian IPO of an Australian data center company get data center company get canceled because canceled because canceled because investors were investors were investors were unhappy with what unhappy with what unhappy with what they were promised. they were promised. they were promised. That's not the reason That's not the reason That's not the reason it was canceled. it was canceled. it was canceled. Well, why Well, why Well, why was it canceled then? was it canceled then? was it canceled then? It was canceled because of It was canceled because of It was canceled because of this unprecedented this unprecedented this unprecedented moment where two moment where two moment where two trillion dollar IPOs are trillion dollar IPOs are trillion dollar IPOs are swirling around swirling around swirling around us, waiting to go us, waiting to go us, waiting to go to market and to market and to market and sucking all the sucking all the sucking all the venture capital out of the venture capital out of the venture capital out of the market, as people say, market, as people say, market, as people say, "Yeah, this looks "Yeah, this looks "Yeah, this looks really nice, this is the IPO really nice, this is the IPO really nice, this is the IPO we have, but we have, but we have, but we don't have the ability to we don't have the ability to we don't have the ability to do it right now." " do it right now." " Okay, but I think that Okay, but I think that Okay, but I think that 's somewhat 's somewhat 's somewhat in line with in line with in line with Asim's point." When you talk Asim's point." When you talk Asim's point." When you talk about these massive IPOs that are about these massive IPOs that are about these massive IPOs that are happening right now, and happening right now, and happening right now, and we're talking about Open AI we're talking about Open AI we're talking about Open AI and Anthropic, and Open AI has been pushed back a little bit on and Anthropic, and Open AI has been pushed back a little bit on and Anthropic, and Open AI has been pushed back a little bit on when it when it when it looks like they're going to looks like they're going to looks like they're going to go go go public. What's public. What's interesting, from the interesting, from the perspective of this not perspective of this not perspective of this not just being a bubble, is that just being a bubble, is that just being a bubble, is that experienced experienced experienced investors in large investors in large investors in large numbers, who really numbers, who really numbers, who really have a stake in this have a stake in this , are looking at this and , are looking at this and , are looking at this and saying it makes sense saying it makes sense . So when you hear this . So when you hear this . So when you hear this from other people, Paul, from other people, Paul, from other people, Paul, and you're an investor, and you and you're an investor, and you and you're an investor, and you think about these things, think about these things, think about these things, what mistake do you what mistake do you what mistake do you think all these think all these think all these people are making? When you say it's people are making? When you say it's people are making? When you say it's just just just math, it's clearly math, it's clearly math, it's clearly not just math, not just math, not just math, because everyone else can because everyone else can because everyone else can do math. This is do math. This is do math. This is your interpretation of your interpretation of your interpretation of the math. So how is the math. So how is the math. So how is your interpretation of your interpretation of your interpretation of mathematics mathematics mathematics different from, different from, different from, you know, the market, the you know, the market, the you know, the market, the usual one. So you usual one. So you usual one. So you need to think about need to think about need to think about why why why these IPOs are happening, and why people…” these IPOs are happening, and why people…” these IPOs are happening, and why people…” pile on them.

pile on them. pile on them. Part of it is in Part of it is in Part of it is in anticipation of what anticipation of what anticipation of what they think could they think could they think could be, you know, a be, you know, a be, you know, a fantastic fantastic fantastic business with a business with a business with a long-term long-term long-term franchise, yeah yeah yeah yeah franchise, yeah yeah yeah yeah ; but it's also because ; but it's also because ; but it's also because they think that when the they think that when the they think that when the time comes and when the time comes and when the lock-up expires, they'll lock-up expires, they'll be able to sell be able to sell be able to sell the shares to someone else the shares to someone else the shares to someone else who thinks that way, who thinks that way, who thinks that way, right? This is the essence of right? This is the essence of right? This is the essence of going back to the very going back to the very going back to the very beginning of why beginning of why beginning of why IPOs exist? IPOs used to IPOs exist? IPOs used to IPOs exist? IPOs used to exist to exist to exist to allow you to allow you to allow you to raise money for a raise money for a raise money for a company that company that company that needed to needed to needed to grow. That's not true, is grow. That's not true, is grow. That's not true, is it? These companies it? These companies it? These companies have completed their have completed their have completed their growth. They'll probably growth. They'll probably growth. They'll probably actually actually actually have, I don't know, have, I don't know, have, I don't know, $100 billion in $100 billion in $100 billion in revenue over time revenue over time revenue over time by the time they exit, by the time they exit, by the time they exit, or $70 billion or $70 billion or $70 billion , or something , or something , or something like that. So, these are the like that. So, these are the like that. So, these are the outputs. These are people outputs. These are people outputs. These are people inside the company inside the company inside the company who say, "I who say, "I who say, "I don't need don't need don't need my shares anymore, you can my shares anymore, you can my shares anymore, you can have them." “So they have them." “So they have them." “So they sell their shares to sell their shares to sell their shares to outsiders because at outsiders because at outsiders because at that valuation they that valuation they that valuation they would rather have cash would rather have cash would rather have cash than Anthropic shares, or than Anthropic shares, or than Anthropic shares, or cash than Open cash than Open cash than Open AI shares. AI shares. AI shares. At that price, I'd At that price, I'd At that price, I'd like to have like to have like to have cash too. Yes, cash too. Yes, cash too. Yes, of course, absolutely, of course, absolutely, of course, absolutely, yes, I would give them away. But yes, I would give them away. But yes, I would give them away. But the point is, it's not a the point is, it's not a the point is, it's not a gesture when they gesture when they gesture when they say, "I really say, "I really say, "I really like the like the like the company's prospects company's prospects ." There will be ." There will be selling shareholders in all selling shareholders in all selling shareholders in all directions because this is an directions because this is an directions because this is an output event, not a output event, not a output event, not a financial event. It's a financial event. It's a financial event. It's a gesture where other people gesture where other people gesture where other people say, "I say, "I say, "I like your like your like your prospects as a prospects as a prospects as a company, and I want to company, and I want to company, and I want to own it." The own it." The own it." The same thing was said in same thing was said in same thing was said in anticipation of the SpaceX IPO, and anticipation of the SpaceX IPO, and anticipation of the SpaceX IPO, and we saw what happened we saw what happened we saw what happened after that, all that after that, all that after that, all that enthusiasm before it enthusiasm before it enthusiasm before it turned into, turned into, turned into, you know, a slight you know, a slight roller coaster-like decline.

roller coaster-like decline. So I expect you'll So I expect you'll So I expect you'll see the same thing because from a see the same thing because from a see the same thing because from a market perspective, these are market perspective, these are market perspective, these are weekend events. You weekend events. You weekend events. You will be a wall of will be a wall of will be a wall of selling shareholders selling shareholders .” showing up after .” showing up after .” showing up after the lockup, selling the lockup, selling the lockup, selling shares to anyone who shares to anyone who shares to anyone who shows up and wants them, and shows up and wants them, and shows up and wants them, and wants to take the shares away from them wants to take the shares away from them wants to take the shares away from them . And that is why this . And that is why this . And that is why this should be understood as an should be understood as an should be understood as an event of exit. People event of exit. People event of exit. People buy expecting to be buy expecting to be buy expecting to be sold to. If sold to. If sold to. If you want...It's like the you want...It's like the you want...It's like the old phrase about the old phrase about the old phrase about the poker table. If poker table. If poker table. If you don't know who the you don't know who the you don't know who the bigger fool at the bigger fool at the bigger fool at the table is, it's you. That's it table is, it's you. That's it table is, it's you. That's it , and I don't understand , and I don't understand , and I don't understand why people don't see why people don't see why people don't see it that way, because that's what it that way, because that's what it that way, because that's what makes it so makes it so makes it so unprecedented. unprecedented. unprecedented. I'll just...just I'll just...just I'll just...just finish on this finish on this finish on this . I . I . I mean, what mean, what mean, what Paul describes Paul describes Paul describes is actually what many IPOs is actually what many IPOs is actually what many IPOs have been for have been for have been for many, many many, many many, many years. And when we years. And when we years. And when we look at Anthropic and look at Anthropic and look at Anthropic and whether it can be a whether it can be a whether it can be a sustainable business, we sustainable business, we sustainable business, we probably need to probably need to probably need to look at what look at what look at what it will show in its S it will show in its S -1, when it's...the S-1 is a -1, when it's...the S-1 is a -1, when it's...the S-1 is a document. This is a document document. This is a document , essentially, about what , essentially, about what , essentially, about what this company is like when you're this company is like when you're this company is like when you're going to buy it going to buy it going to buy it at an IPO, and look at at an IPO, and look at at an IPO, and look at what's happening with what's happening with what's happening with its its its revenue trajectory and its revenue trajectory and its revenue trajectory and its margin trajectory, and margin trajectory, and ... the breadth of its customer ... the breadth of its customer ... the breadth of its customer base, and whether base, and whether base, and whether customers are staying customers are staying customers are staying with it, whether those with it, whether those with it, whether those accounts are growing, accounts are growing, accounts are growing, except that, frankly except that, frankly except that, frankly , there have been a lot of , there have been a lot of , there have been a lot of mixed reports mixed reports mixed reports and leaks, some of and leaks, some of and leaks, some of which are accurate and some of which are which are accurate and some of which are which are accurate and some of which are inaccurate, but not a inaccurate, but not a inaccurate, but not a lot of them give us a lot of them give us a lot of them give us a clearer picture of clearer picture of clearer picture of the company. And I think that the company. And I think that the company. And I think that highlights one highlights one highlights one of the really big of the really big of the really big challenges that we've challenges that we've challenges that we've all had to all had to all had to face over the face over the face over the last two or three last two or three last two or three years, which is that these years, which is that these years, which is that these companies have gotten very companies have gotten very companies have gotten very big, but they're big, but they're big, but they're still private still private still private companies. They companies. They companies. They don't really don't really don't really need to reveal need to reveal need to reveal much information, much information, much information, if anything at all if anything at all if anything at all . And what . And what . And what comes out are

comes out are comes out are small rumors, small rumors, small rumors, gossip, and occasional gossip, and occasional gossip, and occasional leaks from well- leaks from well- leaks from well- confirmed sources confirmed sources . And because they have . And because they have . And because they have essentially been essentially been essentially been structural to the structural to the structural to the AI ​​trade AI ​​trade AI ​​trade so far, and as a so far, and as a so far, and as a result, they have been what result, they have been what result, they have been what has influenced the has influenced the has influenced the valuations of the broader market, valuations of the broader market, valuations of the broader market, whatever it may be whatever it may be whatever it may be now, 38% of the technology. now, 38% of the technology. But we couldn't But we couldn't verify, as we verify, as we verify, as we usually can with usually can with usually can with public public public companies, what was companies, what was companies, what was actually actually actually happening. And I happening. And I happening. And I think what happens with IPOs is think what happens with IPOs is we won't get all we won't get all the answers, but we will the answers, but we will the answers, but we will get some of the get some of the get some of the answers that we answers that we answers that we need about these need about these need about these companies. companies. companies. However, we do have a However, we do have a However, we do have a partial answer partial answer partial answer to this question, to this question, to this question, which I think is which I think is which I think is amazing because it amazing because it amazing because it originated from originated from originated from horror movies, where are the killers in the horror movies, where are the killers in the horror movies, where are the killers in the house? This is one of house? This is one of house? This is one of those cases where those cases where those cases where Anthropic accidentally Anthropic accidentally Anthropic accidentally revealed information revealed information revealed information about one of their about one of their about one of their largest clients. largest clients. largest clients. They said we They said we They said we needed...what was it needed...what was it needed...what was it called? Mega called? Mega customer service manager or something . They tried to . They tried to . They tried to hire one, and hire one, and hire one, and then everyone quickly then everyone quickly then everyone quickly realized that realized that realized that by naming their by naming their by naming their biggest client, they could biggest client, they could triangulate and triangulate and say it was say it was say it was for Meta. So this was for Meta. So this was for Meta. So this was their their customer service manager who customer service manager who was in sales for was in sales for was in sales for one company. Then one company. Then one company. Then it turned out that Meta it turned out that Meta it turned out that Meta was responsible for about was responsible for about was responsible for about 10% of their revenue. And this 10% of their revenue. And this 10% of their revenue. And this goes back to the fact that goes back to the fact that Meta accounts for 10% of Meta accounts for 10% of Anthropic's revenue.

Anthropic's revenue. Anthropic's revenue. That's true. That's true. That's true. But that was in 2025, But that was in 2025, But that was in 2025, right? Yes. right? Yes. right? Yes. And let's just And let's just And let's just clarify. Let's clarify. Let's clarify. Let's just clarify just clarify just clarify why this is important to them why this is important to them why this is important to them . . . No, but the important thing is that No, but the important thing is that No, but the important thing is that Anthropic's revenue in the first Anthropic's revenue in the first Anthropic's revenue in the first quarter...of this year quarter...of this year quarter...of this year was more than in the was more than in the was more than in the entire last entire last entire last calendar year. So calendar year. So calendar year. So we look in the we look in the we look in the rearview mirror rearview mirror rearview mirror and say, "Oh, we and say, "Oh, we and say, "Oh, we hit a squirrel 700 yards hit a squirrel 700 yards hit a squirrel 700 yards back." “We need to back." “We need to back." “We need to look through the look through the look through the windshield. windshield. windshield. That's why these companies are so That's why these companies are so That's why these companies are so unusual. We have unusual. We have unusual. We have never seen never seen never seen companies of this companies of this companies of this size go size go size go public with such a public with such a public with such a small number of small number of small number of customers. customers. customers. The concentration of their The concentration of their The concentration of their client list, client list, client list, perhaps eight perhaps eight perhaps eight companies account for 80% of companies account for 80% of their revenue. This their revenue. This their revenue. This does not happen. And so, as an does not happen. And so, as an does not happen. And so, as an equity analyst equity analyst equity analyst many years ago, I many years ago, I many years ago, I considered this to be a considered this to be a considered this to be a colossal source of colossal source of colossal source of risk. I look at risk. I look at risk. I look at it and say that it and say that it and say that reliance on such a reliance on such a reliance on such a small number of small number of small number of customers who customers who customers who mostly mostly mostly use it use it use it for the same for the same for the same purpose is historically purpose is historically purpose is historically unprecedented for unprecedented for unprecedented for such a concentration of such a concentration of such a concentration of customers. So, customers. So, customers. So, by chance, by chance, by chance, information leaked about the information leaked about the information leaked about the instability of instability of instability of the business in terms of the business in terms of the business in terms of its impact not just its impact not just its impact not just on one client, but on one client, but on one client, but on one client on one client on one client who was doing the same who was doing the same who was doing the same old old old task – task – writing code. And that's exactly what writing code. And that's exactly what writing code. And that's exactly what I want to I want to I want to say. That's what say. That's what say. That's what makes them so makes them so makes them so unusual. The unusual. The unusual. The beauty of what beauty of what beauty of what Paul just said, which is Paul just said, which is Paul just said, which is eight customers – 80% eight customers – 80% of the business, is that it’s of the business, is that it’s of the business, is that it’s 10% per 10% per 10% per customer. This should be customer. This should be customer. This should be disclosed in the S-1. So disclosed in the S-1. So disclosed in the S-1. So we actually have a…” A we actually have a…” A we actually have a…” A very, very very, very very, very disproven disproven disproven prediction that we will prediction that we will prediction that we will be able to check be able to check be able to check when the S-1 comes out if when the S-1 comes out if when the S-1 comes out if Anthropic successfully goes Anthropic successfully goes Anthropic successfully goes public and say public and say public and say whether it is really 80% and whether it is really 80% and whether it is really 80% and eight companies eight companies eight companies represent 80% represent 80% of the business. And that's really of the business. And that's really of the business. And that's really helpful, because if helpful, because if helpful, because if Paul is right, that's a little Paul is right, that's a little Paul is right, that's a little exotic, isn't it?

exotic, isn't it? exotic, isn't it? This is a little spicy. This is a little spicy. If you don't subscribe If you don't subscribe to The New York Times, we have to The New York Times, we have to The New York Times, we have news for you. news for you. news for you. You can now You can now You can now browse the Times browse the Times browse the Times for free without for free without any paid any paid any paid subscriptions for the subscriptions for the subscriptions for the first month on first month on first month on The New York Times app. I The New York Times app. I The New York Times app. I think Apple is a think Apple is a think Apple is a really interesting really interesting really interesting company here. Because company here. Because company here. Because unlike Alphabet, Google, unlike Alphabet, Google, unlike Alphabet, Google, unlike Meta, unlike Meta, unlike Meta, it's clear that OpenAI and it's clear that OpenAI and it's clear that OpenAI and Anthropic are Anthropic are Anthropic are artificial artificial artificial intelligence companies. Microsoft intelligence companies. Microsoft intelligence companies. Microsoft was a major was a major was a major investor in OpenAI and is investor in OpenAI and is investor in OpenAI and is now also now also now also trying to create trying to create trying to create its own advanced AI. its own advanced AI. its own advanced AI. Oracle Oracle Oracle has invested heavily here and has invested heavily here and has invested heavily here and is now is now is now just above just above just above junk bond status. junk bond status. Apple simply decided Apple simply decided not to compete in not to compete in not to compete in creating advanced creating advanced creating advanced AI. Whereas their AI. Whereas their AI. Whereas their competitors are now competitors are now competitors are now spending, you know, $ spending, you know, $ spending, you know, $ 100 billion to $200 billion to 100 billion to $200 billion to 100 billion to $200 billion to $300 billion in $300 billion in $300 billion in capital expenditures capital expenditures capital expenditures every year. Apple every year. Apple every year. Apple spends how much per year, spends how much per year, spends how much per year, approximately 15 approximately 15 approximately 15 billion in billion in billion in capital expenditures capital expenditures capital expenditures per year? Their shares have performed per year? Their shares have performed better this year than better this year than any other any other any other company's shares. Does company's shares. Does company's shares. Does this mean that Apple, being this mean that Apple, being this mean that Apple, being one of the major one of the major one of the major players, is not players, is not players, is not trying trying trying to compete in the to compete in the Frontier Model Olympics? It seems to Frontier Model Olympics? It seems to win her over.

win her over. win her over. Oh, Apple doesn't do that. Oh, Apple doesn't do that. Oh, Apple doesn't do that. Let me Let me Let me start with the topic. start with the topic. start with the topic. Yes, yes, of course. Yes, yes, of course. Yes, yes, of course. Yes. I mean, Yes. I mean, Yes. I mean, Apple has a business model Apple has a business model Apple has a business model that means they don't that means they don't that means they don't need to launch a need to launch a need to launch a Frontier model. We Frontier model. We Frontier model. We buy their buy their buy their phones, we buy phones, we buy phones, we buy their laptops, and their laptops, and their laptops, and then we might then we might then we might subscribe to subscribe to subscribe to some movie or something some movie or something some movie or something like that. So they like that. So they like that. So they make money make money make money regardless of regardless of regardless of artificial intelligence. artificial intelligence. artificial intelligence. But, but I also think there But, but I also think there 's a really important 's a really important 's a really important cultural dynamic, cultural dynamic, cultural dynamic, which is that which is that which is that Apple cares a lot about Apple cares a lot about Apple cares a lot about detail. I remember detail. I remember detail. I remember the story of how the story of how the story of how Steve Jobs, when he Steve Jobs, when he was launching the iPhone many, many years ago, was launching the iPhone many, many years ago, called the called the called the then then then -SVP of Google, -SVP of Google, -SVP of Google, who was in charge of who was in charge of who was in charge of partnerships, and partnerships, and partnerships, and criticized him criticized him criticized him because they used the because they used the because they used the wrong wrong wrong color yellow, uh, on the color yellow, uh, on the color yellow, uh, on the Google logo, uh, Google logo, uh, Google logo, uh, then Google. Apple really then Google. Apple really then Google. Apple really cares about the details, and cares about the details, and cares about the details, and they care about they care about they care about control. And big control. And big control. And big language models are powerful language models are powerful language models are powerful in many in many in many ways, but they're ways, but they're ways, but they're not about control. They not about control. They not about control. They can be very, can be very, can be very, very unsuccessful. And I very unsuccessful. And I very unsuccessful. And I think one of the think one of the think one of the cultural reasons cultural reasons cultural reasons why Apple would hold back, why Apple would hold back, why Apple would hold back, even though even though even though it has such a it has such a it has such a good opportunity good opportunity good opportunity to be a personal to be a personal to be a personal agent, it has a agent, it has a agent, it has a device, it has device, it has device, it has privacy, is that privacy, is that it doesn't want it doesn't want the mess, uh, that a the mess, uh, that a big language model would eventually create big language model would eventually create if you have a billion if you have a billion if you have a billion people using it, which is what people using it, which is what Apple's property looks like.

Apple's property looks like. So, yeah, I think So, yeah, I think So, yeah, I think Apple didn't do a very smart thing Apple didn't do a very smart thing Apple didn't do a very smart thing about this. I about this. I about this. I think they think they think they mostly mostly mostly failed and got failed and got failed and got lucky. Um, they lucky. Um, they tried to tried to raid AI a few times. I raid AI a few times. I raid AI a few times. I mean, they had mean, they had mean, they had teams, they were teams, they were teams, they were disbanding teams, disbanding teams, disbanding teams, they were going to they were going to they were going to integrate it with Siri integrate it with Siri , they weren't going to , they weren't going to , they weren't going to integrate it with Siri integrate it with Siri , and they never got anywhere , and they never got anywhere , and they never got anywhere . And what . And what . And what they gained from that was that by they gained from that was that by they gained from that was that by their their their inaction they inaction they inaction they were a classic were a classic were a classic free rider, free rider, free rider, right? They got right? They got right? They got free travel free travel free travel for everyone. the gigantic for everyone. the gigantic for everyone. the gigantic expenses of others, and on the expenses of others, and on the expenses of others, and on the other hand they other hand they other hand they smell like roses because smell like roses because smell like roses because they didn't actually they didn't actually they didn't actually have to make those have to make those have to make those expenses, they can expenses, they can expenses, they can use use use what others are building and what others are building and what others are building and sit on it. But, sit on it. But, sit on it. But, I think overall, I think overall, I think overall, we're going to see a lot of we're going to see a lot of we're going to see a lot of that because people that because people that because people are burning money at the are burning money at the are burning money at the stake, which can stake, which can stake, which can benefit benefit benefit a lot of other a lot of other a lot of other companies, you know, companies, you know, companies, you know, so Apple, sitting on so Apple, sitting on so Apple, sitting on that, not having to that, not having to that, not having to build build build models. Uh, models. Uh, models. Uh, look at look at look at people like Cloudflare, people like Cloudflare, people like Cloudflare, for example. I for example. I for example. I mean, here's a company mean, here's a company mean, here's a company that's that's that's been largely forgotten, that was been largely forgotten, that was been largely forgotten, that was kind of a kind of a kind of a basic internet basic internet infrastructure, infrastructure, infrastructure, whose stock whose stock whose stock has tripled in the has tripled in the has tripled in the last 18 months last 18 months last 18 months because suddenly people are because suddenly people are because suddenly people are like, " like, " Wait a minute. Wait a minute. Wait a minute. Wait a minute. Wait a minute. Wait a minute. Wait a minute. Wait a minute. Wait a minute. If we have all these If we have all these If we have all these bots running around bots running around bots running around the internet, this is kind of a the internet, this is kind of a the internet, this is kind of a new form of new form of new form of attack. We need to attack. We need to attack. We need to make sure that make sure that make sure that we have the ones that can we have the ones that can we have the ones that can handle all this handle all this handle all this new traffic." "And new traffic." "And new traffic." "And so a lot of so a lot of so a lot of this investment that this investment that this investment that we've seen is going to be the we've seen is going to be the we've seen is going to be the beneficiary of a whole beneficiary of a whole beneficiary of a whole bunch of companies whose bunch of companies whose bunch of companies whose names aren't OpenAI, Anthropic, Grok names aren't OpenAI, Anthropic, Grok names aren't OpenAI, Anthropic, Grok or whatever. And I or whatever. And I or whatever. And I think that's a really think that's a really think that's a really important point. And this goes back to the important point. And this goes back to the electrification of the 1920s electrification of the 1920s . Who was the . Who was the . Who was the main main main beneficiary when beneficiary when utilities began to utilities began to collapse during the collapse during the collapse during the 1929 crisis?

1929 crisis? 1929 crisis? Of course, these were Of course, these were Of course, these were people who created people who created people who created things that required things that required things that required electricity, electricity, air conditioner manufacturers and air conditioner manufacturers and others. And yes, we're definitely others. And yes, we're definitely others. And yes, we're definitely going to see a lot of going to see a lot of going to see a lot of , you know, , you know, , you know, random random random interception and interception and interception and free free free rides on top of rides on top of rides on top of that. that. that. This, but not a random This, but not a random This, but not a random interception and a interception and a interception and a free ride. free ride. free ride. This is This is This is not what Apple planned not what Apple planned not what Apple planned . . . This is called This is called This is called additional additional additional services and services and services and additional additional additional businesses. This is what businesses. This is what businesses. This is what economists, when they're economists, when they're , you know, very , you know, very , you know, very moderate in their moderate in their moderate in their language, just describe language, just describe language, just describe the process of a the process of a general- general- purpose technology coming along, it... it's purpose technology coming along, it... it's purpose technology coming along, it... it's constantly constantly constantly improving, and improving, and improving, and then it starts then it starts then it starts to create a need for to create a need for to create a need for additional services, additional services, additional services, whether it's whether it's whether it's maintenance, or maintenance, or maintenance, or repair, or whatever it is that repair, or whatever it is that repair, or whatever it is that can now be done with can now be done with can now be done with that that that technology. Eh, it's a technology. Eh, it's a technology. Eh, it's a normal evolution of normal evolution of general- general- purpose technology purpose technology purpose technology finding its place, finding its place, finding its place, and in some cases and in some cases and in some cases it's a bit of a free it's a bit of a free it's a bit of a free ride. In other ride. In other ride. In other cases, what's happened with Cloudflare is cases, what's happened with Cloudflare is cases, what's happened with Cloudflare is that we've that we've that we've been able to been able to been able to automate so automate so automate so many bots and many bots and many bots and AI agents AI agents AI agents across the across the across the economy that economy that economy that companies on the internet companies on the internet companies on the internet are turning to Cloudflare are turning to Cloudflare are turning to Cloudflare to try to to try to to try to protect them if protect them if protect them if they can.

they can. they can. Let me Let me Let me ask you something about these ask you something about these ask you something about these big companies that have big companies that have big companies that have come up a lot in the come up a lot in the bubble debates I've been bubble debates I've been following. These are the big following. These are the big following. These are the big companies we're companies we're companies we're talking about here, your talking about here, your talking about here, your Alphabets, your Metas, etc. Alphabets, your Metas, etc. Alphabets, your Metas, etc. They just have a lot of They just have a lot of They just have a lot of money. And so they money. And so they money. And so they do it from a very do it from a very do it from a very healthy balance sheet, which is healthy balance sheet, which is healthy balance sheet, which is sometimes not the case for these sometimes not the case for these sometimes not the case for these railroad companies and railroad companies and railroad companies and so on. When Zuckerberg so on. When Zuckerberg so on. When Zuckerberg says, you know, it would be says, you know, it would be says, you know, it would be sad if we sad if we sad if we spent spent spent hundreds of billions of hundreds of billions of hundreds of billions of dollars for nothing, that's what he also means dollars for nothing, that's what he also means dollars for nothing, that's what he also means . if we . if we . if we can can can spend hundreds of spend hundreds of spend hundreds of billions of dollars in an unintended way, billions of dollars in an unintended way, billions of dollars in an unintended way, everything will be fine. And one everything will be fine. And one everything will be fine. And one argument I've argument I've argument I've seen is that seen is that seen is that one of the reasons one of the reasons one of the reasons you don't really you don't really you don't really need to worry need to worry need to worry about this is because about this is because it's happening it's happening it's happening because of companies because of companies because of companies that can absorb that can absorb that can absorb losses on their own losses on their own losses on their own balance sheets without balance sheets without balance sheets without spreading to the rest of the spreading to the rest of the spreading to the rest of the economy. You're not economy. You're not economy. You're not dealing with something dealing with something dealing with something like, you know, like, you know, like, you know, derivatives during the derivatives during the derivatives during the 2008 financial crisis 2008 financial crisis 2008 financial crisis . Um, Paul, you've . Um, Paul, you've . Um, Paul, you've been thinking a lot about the been thinking a lot about the been thinking a lot about the private private private lending aspects. I lending aspects. I lending aspects. I mean, there is a mean, there is a mean, there is a movement movement movement towards debt in some towards debt in some towards debt in some aspects of aspects of aspects of financing right now. What do you financing right now. What do you financing right now. What do you think about the question of think about the question of think about the question of whether we are protected from whether we are protected from whether we are protected from this as a society this as a society this as a society in general, because these in general, because these in general, because these companies have companies have companies have incredible amounts of incredible amounts of incredible amounts of money to money to money to lose if they lose if they lose if they choose to?

choose to? No, we are not protected. Em. No, we are not protected. Em. So, this is a funny thing. So, this is a funny thing. So, this is a funny thing. So, a year ago, people So, a year ago, people So, a year ago, people were telling me were telling me were telling me to stop talking to stop talking to stop talking about it because these are about it because these are about it because these are smart companies, smart companies, smart companies, they have huge they have huge they have huge cash flows, who are you to cash flows, who are you to tell them what tell them what to do? And I thought, "Oh to do? And I thought, "Oh , damn, you , damn, you , damn, you caught me." So I'll caught me." So I'll caught me." So I'll stop saying... stop saying... this. And then, of course, a this. And then, of course, a this. And then, of course, a year passed. What year passed. What year passed. What does the data show? does the data show? does the data show? Expenses move Expenses move Expenses move from cash flow from cash flow from cash flow to debt. So, you know, to debt. So, you know, to debt. So, you know, you go up and you go up and you go up and say, "Hey, say, "Hey, say, "Hey, wait a minute, wait a minute, wait a minute, wait a minute." wait a minute." wait a minute." Remember what you Remember what you Remember what you said last said last said last year? This is not true. year? This is not true. year? This is not true. So, should I be So, should I be So, should I be worried now? And worried now? And worried now? And it's like: no, you don't it's like: no, you don't it's like: no, you don't need to worry right now need to worry right now need to worry right now because because because they have impeccable they have impeccable they have impeccable balance sheets. These are great balance sheets. These are great balance sheets. These are great loans. And I'm like, "Oh, loans. And I'm like, "Oh, loans. And I'm like, "Oh, [ __ ], maybe I [ __ ], maybe I [ __ ], maybe I just got the just got the just got the time wrong twice." And time wrong twice." And time wrong twice." And then, of course, then, of course, then, of course, seriously, what seriously, what seriously, what happens when happens when happens when you have perfect you have perfect you have perfect balance? Wall Street is balance? Wall Street is balance? Wall Street is very good at very good at very good at making it less than making it less than making it less than perfect because perfect because perfect because they they they will keep will keep will keep raising money for you raising money for you raising money for you , and raising , and raising , and raising, and , and raising, and , and raising, and raising for you, raising for you, raising for you, because the fees are because the fees are because the fees are huge and that, and once huge and that, and once huge and that, and once they they they do that, your do that, your do that, your credit rating credit rating credit rating goes down, and we goes down, and we goes down, and we see that with Oracle. We see that with Oracle. We see that with Oracle. We see that we are on the verge of see that we are on the verge of see that we are on the verge of this with Microsoft. We're on the this with Microsoft. We're on the this with Microsoft. We're on the verge of this with some verge of this with some verge of this with some other other other hyperscalers hyperscalers hyperscalers because that's what because that's what because that's what Wall Street is doing.

Wall Street is doing. Wall Street is doing. They see a They see a They see a spotless balance sheet, a spotless balance sheet, a spotless balance sheet, a company with no company with no company with no debt on debt on debt on industry instruments industry instruments , and they , and they , and they will continue will continue will continue to burden it with to burden it with to burden it with loans. So the way loans. So the way these companies these companies these companies try to get around try to get around try to get around this, Meta is probably the this, Meta is probably the this, Meta is probably the best example, is best example, is best example, is that that that they use they use they use slightly more exotic slightly more exotic slightly more exotic structures so that the debt structures so that the debt structures so that the debt doesn't actually doesn't actually doesn't actually stay on their stay on their stay on their balance sheet. They balance sheet. They balance sheet. They use things use things use things called called called purpose- purpose- purpose- built built built companies. They companies. They companies. They use use use off-balance sheet off-balance sheet off-balance sheet financing. They financing. They financing. They say, “Hey, hey, hey, hey say, “Hey, hey, hey, hey say, “Hey, hey, hey, hey , hey, it’s not my , hey, it’s not my , hey, it’s not my debt. I know I debt. I know I debt. I know I control it, but there it is control it, but there it is control it, but there it is , with that , with that , with that funny name that funny name that funny name that sounds like a sounds like a sounds like a New Orleans New Orleans New Orleans bagel. "Hmm, that's someone's bagel. "Hmm, that's someone's bagel. "Hmm, that's someone's debt." So this is all debt." So this is all debt." So this is all just a hoax, and the just a hoax, and the just a hoax, and the reality is that, when reality is that, when reality is that, when you put it all together, you put it all together, you put it all together, we've now reached a we've now reached a we've now reached a point where point where point where technologies, technologies, technologies, whose main advantage has whose main advantage has whose main advantage has historically been historically been historically been rapid growth, rapid growth, rapid growth, have no debt, have no debt, have no debt, right? Technology right? Technology right? Technology is now the largest is now the largest is now the largest issuer of issuer of issuer of investment-grade investment-grade investment-grade bonds in the bonds in the bonds in the United States. This is an United States. This is an United States. This is an industry whose claim industry whose claim industry whose claim to fame, besides to fame, besides to fame, besides growth, was that growth, was that growth, was that they were they were they were relatively clean. relatively clean. relatively clean. Now they work the Now they work the Now they work the other way around. These are companies other way around. These are companies other way around. These are companies burdened with huge burdened with huge burdened with huge debts that increasingly debts that increasingly debts that increasingly dominate the dominate the dominate the emission market to the emission market to the emission market to the point that it even point that it even point that it even affects sovereign affects sovereign affects sovereign states, right? states, right?

states, right? So, one of the reasons So, one of the reasons So, one of the reasons why the bond market why the bond market why the bond market panicked over the panicked over the panicked over the last two to three last two to three last two to three weeks was the role of weeks was the role of weeks was the role of hyperscalers, who hyperscalers, who hyperscalers, who issue loans with the issue loans with the issue loans with the same duration same duration same duration as the 10-year as the 10-year as the 10-year bond in the bond in the bond in the United States, which United States, which United States, which actually outperformed actually outperformed actually outperformed the US, which led to a the US, which led to a the US, which led to a rise in the yield on the rise in the yield on the rise in the yield on the 10-year bond. 10-year bond. That's how That's how crazy things have gotten. crazy things have gotten. crazy things have gotten. Yes, I know this is a Yes, I know this is a Yes, I know this is a completely transitional completely transitional completely transitional period that will have period that will have period that will have huge implications in huge implications in huge implications in the future in the future in the future in terms of this debt terms of this debt terms of this debt overhang, which is overhang, which is overhang, which is starting to look more and more like starting to look more and more like starting to look more and more like Japan Japan Japan circa 1990. It's circa 1990. It's circa 1990. It's all these corporations with all these corporations with all these corporations with giant giant giant balance sheets, balance sheets, balance sheets, loaded with loaded with loaded with debt, debt, debt, trying to trying to trying to work them off. work them off. work them off. Well, the other side of this, Well, the other side of this, Well, the other side of this, Azim, is that it's Azim, is that it's Azim, is that it's not just debt that's not just debt that's not just debt that's being displaced when these being displaced when these being displaced when these hyperscalar hyperscalar hyperscalar companies are competing companies are competing companies are competing with everyone else for with everyone else for with everyone else for money, it's also money, it's also money, it's also construction, it's also construction, it's also construction, it's also real materials. I real materials. I real materials. I mean, I'm mean, I'm mean, I'm sure you've sure you've sure you've seen these graphs seen these graphs seen these graphs that show that there's a that show that there's a real pushback happening. real pushback happening. Yes, we are building, we are Yes, we are building, we are Yes, we are building, we are investing money in investing money in investing money in capital expenditures capital expenditures capital expenditures for artificial intelligence, for artificial intelligence, for artificial intelligence, but but but house construction in a country where there is house construction in a country where there is house construction in a country where there is not enough housing is not enough housing is not enough housing is declining.

declining. declining. Costs for manufacturing Costs for manufacturing Costs for manufacturing plants and factories plants and factories plants and factories are being reduced, or are being reduced, or are being reduced, or at least at least reduced by about the same amount. And so, it reduced by about the same amount. And so, it seems that seems that seems that there is a movement there is a movement there is a movement from investing in from investing in from investing in other sectors of other sectors of other sectors of the economy at the level of the economy at the level of the economy at the level of construction and construction and construction and capital expenditures capital expenditures capital expenditures to investing in to investing in to investing in artificial intelligence. I artificial intelligence. I artificial intelligence. I think if think if think if AI AI AI works well, that's works well, that's works well, that's good, but if it doesn't, good, but if it doesn't, good, but if it doesn't, or even if you or even if you or even if you just think just think just think these other things are, you know, these other things are, you know, these other things are, you know, undervalued, as I undervalued, as I undervalued, as I said, maybe I said, maybe I said, maybe I think for think for think for homes, homes are a homes, homes are a homes, homes are a more stable more stable more stable investment. They don't investment. They don't investment. They don't give the same give the same give the same returns as returns as returns as investing in Nvidia, investing in Nvidia, investing in Nvidia, but they are quite but they are quite but they are quite important, and I have a important, and I have a important, and I have a clearer way clearer way clearer way to explain why more to explain why more to explain why more homes will improve homes will improve homes will improve the lives of more the lives of more the lives of more people. How do people. How do people. How do you feel about the you feel about the you feel about the issue of displacement? issue of displacement? issue of displacement? Um, when we Um, when we Um, when we look at, uh, look at, uh, look at, uh, investment rates investment rates investment rates a year ago, back in the summer of a year ago, back in the summer of a year ago, back in the summer of 2025, I wrote about 2025, I wrote about 2025, I wrote about how I assessed whether we how I assessed whether we how I assessed whether we were in a boom were in a boom were in a boom or a bubble, and one or a bubble, and one or a bubble, and one of my observations of my observations of my observations was that was that was that right now, right now, right now, hyperscalers hyperscalers hyperscalers are spending money from are spending money from are spending money from their own cash their own cash their own cash flows and their own flows and their own flows and their own balance sheets. Once balance sheets. Once balance sheets. Once that starts to that starts to that starts to change and they change and they change and they move to move to move to external external external financing, you financing, you financing, you introduce a lot introduce a lot introduce a lot more risk, and my more risk, and my more risk, and my threshold for more threshold for more threshold for more risk was risk was risk was about 30%, er, 30%, which is about 30%, er, 30%, which is about 30%, er, 30%, which is roughly where roughly where roughly where we're we're we're approaching now, approaching now, approaching now, at the end of 2026.

at the end of 2026. These things are not These things are not final. final. final. The problem of displacement The problem of displacement is a real problem is a real problem , and it is a truly complex , and it is a truly complex , and it is a truly complex and delicate issue. and delicate issue. I mean, it's I mean, it's worth thinking about worth thinking about what happened in what happened in what happened in the Netherlands in the the Netherlands in the the Netherlands in the 1970s, when they 1970s, when they 1970s, when they discovered a lot of discovered a lot of discovered a lot of natural gas near the natural gas near the natural gas near the North Sea, and the North Sea, and the North Sea, and the entire economy was entire economy was entire economy was focused on focused on focused on gas extraction; gas extraction; gas extraction; There was such a thing There was such a thing There was such a thing as Dutch as Dutch as Dutch disease, when disease, when disease, when this part of this part of this part of the economy actually absorbed the economy actually absorbed the economy actually absorbed all activity. And, all activity. And, all activity. And, you know, there may be a you know, there may be a you know, there may be a risk that this risk that this risk that this will happen in the US as well, will happen in the US as well, will happen in the US as well, if it's not already if it's not already if it's not already happening now. And happening now. And happening now. And then the question is then the question is then the question is what what what to do with it, and how to to do with it, and how to to do with it, and how to actually actually actually adapt it? And I'm afraid adapt it? And I'm afraid adapt it? And I'm afraid my only answer is my only answer is my only answer is that it that it that it requires some requires some requires some policy, some policy, some policy, some political incentive, political incentive, political incentive, to provide these to provide these to provide these sectors with the activity they need sectors with the activity they need sectors with the activity they need , because I'm , because I'm , because I'm not convinced that not convinced that not convinced that market forces on their market forces on their market forces on their own would do it. This own would do it. This own would do it. This seems like a difficult seems like a difficult seems like a difficult recipe at a time recipe at a time recipe at a time when in both cases when in both cases when in both cases we have very we have very we have very unstable unstable unstable leadership, but our leadership, but our leadership, but our borrowing costs are borrowing costs are borrowing costs are quite high against the backdrop of quite high against the backdrop of quite high against the backdrop of incredible levels of incredible levels of incredible levels of debt.

debt. debt. Well, listen, your Well, listen, your Well, listen, your politics are your politics are your politics are your politics, and you know, you politics, and you know, you politics, and you know, you can vote can vote can vote every 4 years and make the every 4 years and make the every 4 years and make the choices that you choices that you choices that you make. make. make. Um, and then live with Um, and then live with Um, and then live with them. That them. That them. That 's right, and then 's right, and then 's right, and then live with them. But that doesn't live with them. But that doesn't live with them. But that doesn't necessarily have to necessarily have to necessarily have to include subsidies, include subsidies, include subsidies, right? This could right? This could right? This could include other types of, include other types of, include other types of, you know, political you know, political you know, political interference. So, I'm not interference. So, I'm not interference. So, I'm not saying that this is, this is a saying that this is, this is a saying that this is, this is a really difficult really difficult really difficult problem, and it problem, and it problem, and it will continue as long as will continue as long as will continue as long as the returns the returns the returns on other investments on other investments on other investments are higher, and as long as are higher, and as long as are higher, and as long as this creates this creates this creates supply shock for supply shock for supply shock for electricians, electricians, electricians, construction workers, HVAC construction workers, HVAC construction workers, HVAC technicians, technicians, and everything else and everything else that's needed in that's needed in data centers. But, but, and there's data centers. But, but, and there's one "but" in all of this, one "but" in all of this, one "but" in all of this, which is that which is that which is that , of course, there's , of course, there's , of course, there's a lot of a lot of upskilling going on. Many upskilling going on. Many electricians undergo electricians undergo electricians undergo training. Many training. Many training. Many heating, heating, heating, ventilation, and ventilation, and air conditioning engineers undergo air conditioning engineers undergo training. Many training. Many training. Many construction workers construction workers construction workers are undergoing training. There are are undergoing training. There are are undergoing training. There are construction companies that construction companies that construction companies that are developing and are developing and are developing and will begin to have free will begin to have free will begin to have free capacity. And I don't capacity. And I don't capacity. And I don't know how it know how it know how it will end, but will end, but will end, but in a year, two or three in a year, two or three in a year, two or three years, this ability, these years, this ability, these years, this ability, these skills will be in skills will be in skills will be in the economy and they will be the economy and they will be the economy and they will be usable usable .

. So I want to ask So I want to ask this not from the perspective of this not from the perspective of this not from the perspective of stock markets and stock markets and stock markets and financial flows. I financial flows. I financial flows. I want to try to want to try to want to try to ask what does ask what does ask what does this economy look like where this economy look like where this economy look like where companies, people, and companies, people, and companies, people, and governments use governments use governments use enough enough enough AI to make this AI to make this AI to make this investment pay off investment pay off investment pay off in large numbers? in large numbers? in large numbers? Even if, you know, Even if, you know, Even if, you know, some companies some companies some companies may not end up in the may not end up in the may not end up in the winner's circle and winner's circle and winner's circle and may may may fail. So, Azim, fail. So, Azim, fail. So, Azim, you spend a lot of you spend a lot of you spend a lot of time thinking about this. time thinking about this. If you were to try to If you were to try to paint me a paint me a paint me a picture of what picture of what picture of what things look like in things look like in things look like in regular companies, regular companies, regular companies, your your your CEOs, your CEOs, your CEOs, your Coca-Cola, um, in this Coca-Cola, um, in this Coca-Cola, um, in this world where the AI ​​bet world where the AI ​​bet world where the AI ​​bet is paying off, what does that is paying off, what does that is paying off, what does that look like? What's look like? What's look like? What's different? What are different? What are different? What are we not spending money on we not spending money on we not spending money on if so much money if so much money if so much money is being spent on is being spent on is being spent on AI tokens? What is AI tokens? What is AI tokens? What is going on here?

going on here? You know, when I You know, when I talk to talk to talk to executives, and I executives, and I executives, and I talk to them all over talk to them all over talk to them all over Europe, the US and the Europe, the US and the Europe, the US and the US, dozens of people US, dozens of people US, dozens of people every year, they hear the same every year, they hear the same every year, they hear the same thing: we're thing: we're thing: we're making progress making progress making progress with artificial with artificial with artificial intelligence. This is more intelligence. This is more intelligence. This is more complicated than we complicated than we complicated than we thought. We thought. We thought. We will continue to will continue to will continue to do this and we do this and we do this and we are starting to get are starting to get are starting to get results. And those who are more results. And those who are more results. And those who are more confident say confident say confident say that we are getting that we are getting that we are getting more and more more and more more and more results. But results. But results. But a lot of that is a lot of that is a lot of that is coming now, I coming now, I coming now, I think it will come if the think it will come if the think it will come if the business doesn't grow business doesn't grow business doesn't grow very, very quickly at very, very quickly at very, very quickly at the expense of slowly the expense of slowly the expense of slowly reducing reducing reducing employee replacement. These employee replacement. These employee replacement. These companies, which become companies, which become companies, which become more productive more productive more productive thanks to artificial thanks to artificial thanks to artificial intelligence, intelligence, intelligence, will provide will provide will provide more money per more money per more money per employee. And employee. And employee. And if they don't if they don't if they don't grow faster grow faster grow faster than this than this than this productivity growth, productivity growth, productivity growth, their their their workforce will start to workforce will start to workforce will start to decline. But we decline. But we decline. But we must add that there must add that there must add that there are very few are very few are very few companies companies companies based on based on based on artificial intelligence. artificial intelligence. artificial intelligence. There are so many... There are so many... When we think about When we think about When we think about electricity, electricity, electricity, the companies the companies the companies based on based on based on electricity were electricity were electricity were like Ford Motor Company like Ford Motor Company like Ford Motor Company with its end-to-end with its end-to-end with its end-to-end production line. production line. production line. But most of But most of But most of America wasn't America wasn't America wasn't fully fully fully electrified for another 15 electrified for another 15 electrified for another 15 or 20 years after or 20 years after or 20 years after that. And I don't think I that. And I don't think I that. And I don't think I can point to any can point to any can point to any Main Street firm Main Street firm outside of outside of Silicon Silicon Silicon Valley that you could Valley that you could Valley that you could say has say has say has gone AI native gone AI native . What you . What you can see are can see are can see are parts of the business that parts of the business that parts of the business that look like look like look like they're using they're using they're using a lot of artificial a lot of artificial a lot of artificial intelligence.

intelligence. intelligence. Good. Okay, but if the Good. Okay, but if the AI-native company AI-native company you're talking about you're talking about you're talking about is a company where you've is a company where you've is a company where you've managed to replace managed to replace managed to replace humans with humans with humans with AI... AI... AI... Yes. Yes. Yes. So now you So now you So now you seem to be moving very quickly seem to be moving very quickly seem to be moving very quickly to a world to a world to a world that is very that is very that is very concerned about concerned about concerned about AI AI AI negatively impacting negatively impacting negatively impacting jobs across the jobs across the jobs across the economy, unless you economy, unless you economy, unless you believe that all these new believe that all these new believe that all these new companies that companies that companies that are being created are are being created are are being created are somehow very somehow very somehow very labor-intensive, even though labor-intensive, even though labor-intensive, even though theoretically the new theoretically the new theoretically the new companies that companies that companies that are being created will be are being created will be are being created will be more more more AI-native, AI-native, AI-native, meaning less meaning less meaning less labor-intensive. So is labor-intensive. So is labor-intensive. So is this actually a good this actually a good this actually a good world, and if it is a world, world, and if it is a world, world, and if it is a world, who spends more who spends more who spends more money in this world money in this world money in this world where people are losing their where people are losing their where people are losing their jobs? jobs? jobs? Well, I think there are a Well, I think there are a Well, I think there are a few few few different things going on. You know, first of all different things going on. You know, first of all different things going on. You know, first of all , there , there , there are still things that are still things that are still things that AI won't AI won't AI won't be able to do, and that be able to do, and that be able to do, and that will be the levels of will be the levels of will be the levels of human resources human resources human resources that do that do that do the verification, the final the verification, the final the verification, the final validation, validation, validation, do the things that models do the things that models do the things that models can't do can't do can't do well. And what usually well. And what usually well. And what usually happens is that happens is that happens is that these people get paid these people get paid these people get paid more and more, and more and more, and more and more, and better, simply because better, simply because they're they're they're in this in this in this narrow zone. I think the narrow zone. I think the narrow zone. I think the second thing you're second thing you're second thing you're going to start seeing is a going to start seeing is a going to start seeing is a whole class of roles that whole class of roles that whole class of roles that we can't we can't we can't imagine in companies.

imagine in companies. You know, like a webmaster who didn't webmaster who didn't webmaster who didn't exist in 1993. And exist in 1993. And exist in 1993. And thirdly, it's really a thirdly, it's really a thirdly, it's really a cultural dimension of cultural dimension of general- general- purpose technology. It was here purpose technology. It was here , you know, Paul was describing , you know, Paul was describing , you know, Paul was describing GPTs, that they tended to GPTs, that they tended to GPTs, that they tended to create entirely create entirely create entirely new classes of work and new classes of work and new classes of work and behavior that we behavior that we behavior that we couldn't understand. And couldn't understand. And couldn't understand. And you know, if you you know, if you you know, if you just think about just think about just think about Henry Ford: well, it's Henry Ford: well, it's Henry Ford: well, it's because of Henry Ford that because of Henry Ford that because of Henry Ford that we got we got we got cars; Since cars; Since cars; Since we had we had we had cars, people had cars, people had cars, people had suburbs, suburbs, suburbs, and suburbs had and suburbs had vacations, and vacations, and vacations, and vacations had all the vacations had all the vacations had all the services that are available there. I am services that are available there. I am services that are available there. I am not indifferent to this. not indifferent to this. not indifferent to this. I mean, the I mean, the I mean, the problem we problem we problem we might run into is might run into is might run into is that that that creating these new creating these new creating these new roles takes more roles takes more roles takes more time than anything else time than anything else . But the other . But the other . But the other observation I observation I observation I would make is, you would make is, you know, my time know, my time horizon for this horizon for this horizon for this is 5, 7, 10 years, is 5, 7, 10 years, is 5, 7, 10 years, not one or two, not two not one or two, not two not one or two, not two or three, because I or three, because I or three, because I think think think management management management teams just teams just teams just need time to need time to need time to learn what they learn what they learn what they need to do with this need to do with this need to do with this technology, and it's not like technology, and it's not like technology, and it's not like you get the you get the you get the best model from Open best model from Open best model from Open AI and say, "Fix AI and say, "Fix AI and say, "Fix my company," and it my company," and it my company," and it fixes it. I fixes it. I fixes it. I mean, I just think mean, I just think mean, I just think these things take these things take these things take time, and that's what I time, and that's what I time, and that's what I see when I talk see when I talk see when I talk to bosses. No matter how to bosses. No matter how to bosses. No matter how fast they want fast they want fast they want to move, it just to move, it just to move, it just takes time.

takes time. takes time. Paul, how do you look at this Paul, how do you look at this Paul, how do you look at this ? ? ? Uh, we're essentially Uh, we're essentially Uh, we're essentially overestimating labor, overestimating labor, overestimating labor, aren't we? I aren't we? I aren't we? I mean, it's the mean, it's the mean, it's the fastest- fastest- fastest- growing growing growing technology in history, technology in history, technology in history, and its fundamental and its fundamental and its fundamental function is to function is to function is to reimagine reimagine reimagine cognitive work. And cognitive work. And cognitive work. And I mean, it's a I mean, it's a I mean, it's a deep struggle deep struggle deep struggle because the essence of because the essence of because the essence of West West West Coast ideology is that Coast ideology is that Coast ideology is that technology technology technology always creates always creates always creates more jobs more jobs more jobs than it destroys. And so it than it destroys. And so it than it destroys. And so it becomes just something becomes just something becomes just something people repeat like a people repeat like a people repeat like a mantra and mantra and mantra and hope it hope it hope it becomes true. And becomes true. And becomes true. And the reality is that the reality is that the reality is that we have never seen a we have never seen a we have never seen a universal universal universal technology that would have technology that would have technology that would have such widespread such widespread such widespread application, would be application, would be application, would be adopted so quickly, and adopted so quickly, and adopted so quickly, and at the same time at the same time at the same time would revalue would revalue would revalue labor. Because when I labor. Because when I labor. Because when I talk about revaluing talk about revaluing talk about revaluing labor, I mean labor, I mean labor, I mean that it significantly that it significantly that it significantly reduces the cost of things like reduces the cost of things like reduces the cost of things like call call centers. And as centers. And as centers. And as a result, these things, a result, these things, a result, these things, this change is going to be this change is going to be this change is going to be destabilizing destabilizing destabilizing long before we long before we long before we figure out how figure out how figure out how to turn them all into to turn them all into , I don't know what the , I don't know what the , I don't know what the equivalent of equivalent of webmasters is. And that's the webmasters is. And that's the webmasters is. And that's the problem, isn't it? problem, isn't it? problem, isn't it? Is this happening Is this happening Is this happening much faster than much faster than much faster than any previous any previous any previous GPT, which is, you know, GPT, which is, you know, GPT, which is, you know, hand-waving hand-waving hand-waving about creating about creating about creating more more more jobs than jobs than jobs than destroying them, contradicting the destroying them, contradicting the destroying them, contradicting the lived experiences of lived experiences of lived experiences of the people who have the people who have the people who have been affected by been affected by been affected by this. And so it can this. And so it can this. And so it can quickly turn quickly turn quickly turn into some kind of strange mantra into some kind of strange mantra , like: don't worry, , like: don't worry, , like: don't worry, your sweet little head, your sweet little head, your sweet little head, everything will be fine. So no everything will be fine. So no , I think these disruptions will , I think these disruptions will , I think these disruptions will far outpace far outpace far outpace our ability our ability our ability to create new to create new to create new jobs.

jobs. jobs. This also struck me This also struck me This also struck me as another form of as another form of as another form of economic risk economic risk economic risk not just for not just for not just for workers but workers but workers but for AI, and for, you know, for AI, and for, you know, for AI, and for, you know, some form of bursting some form of bursting some form of bursting of this bubble or of this bubble or of this bubble or deflation, if it is a deflation, if it is a deflation, if it is a bubble. Because you bubble. Because you bubble. Because you can imagine a can imagine a can imagine a world where the real world where the real world where the real problem isn't whether problem isn't whether problem isn't whether companies are ready companies are ready companies are ready to implement AI. What to implement AI. What to implement AI. What really happens, especially really happens, especially really happens, especially under greater under greater under greater economic pressure, economic pressure, economic pressure, interest rates interest rates interest rates continue to rise continue to rise , maybe some kind of , maybe some kind of , maybe some kind of recession, and companies recession, and companies recession, and companies start to very start to very start to very aggressively replace aggressively replace aggressively replace labor, human labor, human labor, human labor, which, you know, labor, which, you know, labor, which, you know, needs health care needs health care needs health care , is , is , is unionized, unionized, unionized, and has an and has an and has an opinion on AI. And opinion on AI. And opinion on AI. And then there is a then there is a then there is a huge huge huge backlash. There is already a backlash. There is already a backlash. There is already a huge huge huge backlash against AI. People backlash against AI. People backlash against AI. People are already organizing are already organizing are already organizing against the construction of against the construction of data centers. They don't want them where they live. But, you know, live. But, you know, if you know anything if you know anything if you know anything about the history of about the history of professional licensing professional licensing , or , or , or building construction, building construction, building construction, or whatever, it's or whatever, it's or whatever, it's not that hard for, say, teachers not that hard for, say, teachers to organize themselves and to organize themselves and say, "No AI say, "No AI say, "No AI in the classroom." In fact, in the classroom." In fact, in the classroom." In fact, education is off limits education is off limits education is off limits to AI. Doctors to AI. Doctors to AI. Doctors suddenly need to suddenly need to suddenly need to get organized.

get organized. get organized. There might be There might be There might be a few bad events a few bad events a few bad events where people where people where people get bad get bad get bad diagnoses from AI and diagnoses from AI and diagnoses from AI and then they get kind of then they get kind of then they get kind of locked out of locked out of locked out of hospitals. Anthropni in hospitals. Anthropni in hospitals. Anthropni in their risks for their risks for their risks for their IPO say: " their IPO say: " Existential risk Existential risk Existential risk to humanity to humanity to humanity is perhaps one of the risks of is perhaps one of the risks of is perhaps one of the risks of this IPO." If there this IPO." If there this IPO." If there are any are any are any major major major control incidents where instead of hacking a control incidents where instead of hacking a face hugger, OpenAI hacks face hugger, OpenAI hacks Wells Fargo. He's hacking Wells Fargo. He's hacking Wells Fargo. He's hacking something in China. something in China. something in China. Yes. Yes. Yes. And suddenly regulation And suddenly regulation And suddenly regulation will appear not gradually, will appear not gradually, will appear not gradually, but immediately. All of these things but immediately. All of these things but immediately. All of these things seem to me to be a seem to me to be a seem to me to be a real risk, that real risk, that real risk, that the possibility of the possibility of the possibility of intermittent intermittent intermittent regulation regulation regulation seems to me to be higher seems to me to be higher seems to me to be higher than elsewhere, and also, I than elsewhere, and also, I than elsewhere, and also, I don't want to say, don't want to say, don't want to say, inevitable, but quite inevitable, but quite inevitable, but quite likely. So I'm likely. So I'm likely. So I'm curious what you curious what you curious what you think about this. think about this. think about this. This has been my complaint This has been my complaint This has been my complaint from the very beginning of the from the very beginning of the from the very beginning of the process by process by process by which these companies which these companies which these companies launched models. The fact launched models. The fact launched models. The fact that they that they that they were essentially conducting were essentially conducting were essentially conducting unlicensed unlicensed unlicensed drug trials drug trials drug trials on the public, in on the public, in on the public, in the sense that these are the sense that these are the sense that these are extremely important extremely important extremely important technologies that are technologies that are technologies that are being launched in a being launched in a being launched in a way that if they way that if they way that if they were a pharmaceutical were a pharmaceutical were a pharmaceutical company, you would company, you would company, you would immediately have the FDA immediately have the FDA immediately have the FDA on your back. And yet, on your back. And yet, on your back. And yet, thanks to their thanks to their thanks to their technology technology technology companies, you can companies, you can companies, you can do these mass do these mass do these mass trials. By the way trials. By the way , we are going to , we are going to , we are going to transform all of transform all of transform all of education. We're going to education. We're going to education. We're going to launch these things launch these things launch these things that, as you know, that, as you know, that, as you know, can infiltrate can infiltrate can infiltrate government websites and government websites and government websites and other things. So this idea that other things. So this idea that other things. So this idea that you can you can you can do this with impunity in do this with impunity in do this with impunity in technology because technology because technology because we live in a we live in a we live in a timeline where these things have timeline where these things have timeline where these things have n't killed us yet is just n't killed us yet is just n't killed us yet is just ridiculous and ridiculous and ridiculous and unjustified. And I unjustified. And I unjustified. And I think we are inevitably think we are inevitably think we are inevitably heading towards a world where, heading towards a world where, heading towards a world where, after some after some after some terrible event,

terrible event, terrible event, some kind of ex some kind of ex some kind of ex post facto post facto post facto regulation will be introduced, but only regulation will be introduced, but only regulation will be introduced, but only after indirect after indirect after indirect damage. And I will give you a damage. And I will give you a damage. And I will give you a specific explanation. specific explanation. specific explanation. That's how I often That's how I often That's how I often try try try to describe it to describe it to describe it to people, because to people, because to people, because they say, "What they say, "What they say, "What if you if you if you just come up with just come up with just come up with something like that?" Like you, something like that?" Like you, something like that?" Like you, if you think about it, there are if you think about it, there are if you think about it, there are probably five 10 15 probably five 10 15 probably five 10 15 different scenarios that different scenarios that different scenarios that we can come up with we can come up with we can come up with that could lead that could lead that could lead to regulation. None to regulation. None to regulation. None of them probably of them probably of them probably have a probability of, have a probability of, have a probability of, say, less than, say, less than, say, less than, maybe more than 5%. maybe more than 5%. maybe more than 5%. So, let's say I So, let's say I So, let's say I have six events, each with a have six events, each with a have six events, each with a 5% chance of 5% chance of 5% chance of occurring. What is occurring. What is occurring. What is the probability that the probability that the probability that any of these any of these any of these will happen? This is will happen? This is will happen? This is approximately 20%. If approximately 20%. If approximately 20%. If I have a dozen I have a dozen I have a dozen possible scenarios, possible scenarios, possible scenarios, all of which are bad, and all of which are bad, and all of which are bad, and each has no each has no each has no more than a 5% chance more than a 5% chance more than a 5% chance of occurring, what is the of occurring, what is the of occurring, what is the probability that probability that probability that one of them will happen? one of them will happen? one of them will happen? It is over It is over It is over 50%. We have over a 50%. We have over a 50%. We have over a dozen ways dozen ways dozen ways these things can these things can these things can cause problems cause problems cause problems and eventually require and eventually require and eventually require regulation. So, regulation. So, regulation. So, if you look at it from if you look at it from if you look at it from that perspective, it's almost that perspective, it's almost that perspective, it's almost inevitable. So there's a inevitable. So there's a inevitable. So there's a 50-60% chance that 50-60% chance that 50-60% chance that something will happen that something will happen that something will happen that will lead to will lead to will lead to regulation, but people regulation, but people regulation, but people keep sitting back keep sitting back keep sitting back and and and saying, "Well, you know, saying, "Well, you know, saying, "Well, you know, it's not that likely, it's not that likely, it's not that likely, or it's not that or it's not that or it's not that likely." So no, likely." So no, likely." So no, regulation is regulation is regulation is inevitable here. It will take the inevitable here. It will take the inevitable here. It will take the form of something form of something form of something like the FDA, maybe like the FDA, maybe like the FDA, maybe some kind of atomic some kind of atomic some kind of atomic energy board, because it's just energy board, because it's just energy board, because it's just as important, and as important, and as important, and because there are so because there are so because there are so many different many different many different ways in which they ways in which they ways in which they can have broader can have broader can have broader implications for implications for implications for society. And it society. And it society. And it surprises me that people somehow surprises me that people somehow surprises me that people somehow frivolously ignore frivolously ignore frivolously ignore this. Azim, how do you this. Azim, how do you this. Azim, how do you see this dimension of see this dimension of see this dimension of political risk?

political risk? political risk? Well, you know, I think Well, you know, I think Well, you know, I think when you look at when you look at when you look at industries, industries industries, industries industries, industries do quite well do quite well do quite well when there's a little when there's a little when there's a little regulation, like, regulation, like, regulation, like, you know, the you know, the you know, the pharmaceutical pharmaceutical pharmaceutical industry, which is a industry, which is a industry, which is a great example. And great example. And great example. And I think when we I think when we I think when we think about who's going to think about who's going to think about who's going to pay for it, pay for it, pay for it, AI companies AI companies AI companies , from , from , from model model model labs to labs to labs to hyperscalers, are hyperscalers, are hyperscalers, are going to make a lot of going to make a lot of going to make a lot of money. So it's quite money. So it's quite money. So it's quite obvious that there obvious that there obvious that there is money in the system, and there are is money in the system, and there are is money in the system, and there are people who are firms people who are firms people who are firms that can cover that can cover that can cover some of the some of the some of the bill. And I think bill. And I think bill. And I think one of the things that's one of the things that's one of the things that's really fascinating about the really fascinating about the data center protests that you're seeing, data center protests that you're seeing, especially in the US, is that it's especially in the US, is that it's actually people actually people expressing their expressing their expressing their concerns about concerns about concerns about this technology, the way this technology, the way it's it's it's manifesting itself. And if manifesting itself. And if manifesting itself. And if you look at it that way, it you look at it that way, it you look at it that way, it 's a democratic 's a democratic 's a democratic expression, right? It is expression, right? It is expression, right? It is an expression of will, it is an expression of an expression of will, it is an expression of an expression of will, it is an expression of uncertainty. And what uncertainty. And what uncertainty. And what you hope to you hope to you hope to get as get as get as a result of something a result of something a result of something like this is a better like this is a better like this is a better social license social license social license that people that people that people feel feel feel increasingly comfortable with.

increasingly comfortable with. increasingly comfortable with. Now, you know, I Now, you know, I Now, you know, I suspect that in the US, suspect that in the US, suspect that in the US, because you have the because you have the because you have the governing authority governing authority governing authority at the state level, at the state level, at the state level, some of this could some of this could some of this could be done on a be done on a be done on a statewide statewide statewide basis. In the basis. In the basis. In the UK, there are currently UK, there are currently UK, there are currently attempts to attempts to attempts to encourage people who encourage people who encourage people who develop data centers develop data centers to reinvest their to reinvest their annuities into the local annuities into the local annuities into the local community. Because one community. Because one community. Because one of the things that we, of the things that we, of the things that we, you know, don't you know, don't you know, don't talk about very much is that a talk about very much is that a talk about very much is that a huge huge huge data center data center data center can employ very can employ very can employ very few people. I few people. I few people. I mean, it might mean, it might mean, it might be billions of be billions of be billions of dollars, but it's a dollars, but it's a dollars, but it's a handful of handful of handful of jobs, and it's not quite the jobs, and it's not quite the jobs, and it's not quite the same as same as same as two to six people for a two to six people for a two to six people for a multi-billion dollar multi-billion dollar multi-billion dollar facility. facility. facility. If you calculate If you calculate If you calculate the cost of a the cost of a the cost of a job per job per job per job created job created job created , I've done that often , I've done that often , I've done that often , looking at , looking at , looking at a center a center a center in New Mexico recently. That's in New Mexico recently. That's in New Mexico recently. That's $4 million in $4 million in $4 million in subsidies per subsidies per subsidies per job created job created job created , not including , not including , not including the cost of the center. When the cost of the center. When the cost of the center. When you look at the you look at the you look at the tax breaks and tax breaks and tax breaks and needed improvements to needed improvements to needed improvements to local local local utilities, the numbers are crazy. utilities, the numbers are crazy. utilities, the numbers are crazy. So So So none of this none of this none of this works in the works in the works in the long long long run. But it is run. But it is run. But it is useful that there useful that there useful that there is money in the system is money in the system is money in the system because profit because profit because profit is generated across the is generated across the is generated across the entire spectrum of entire spectrum of entire spectrum of issues. And, you know, issues. And, you know, issues. And, you know, you would think you would think that because of that, there's an that because of that, there's an that because of that, there's an obvious place where obvious place where obvious place where this bill should this bill should this bill should go, and this go, and this go, and this bill could bill could bill could go through you.

go through you. go through you. I don't think I don't think I don't think self-regulation self-regulation self-regulation works, and I works, and I works, and I agree that we will agree that we will agree that we will need to find, need to find, need to find, whether at the local, whether at the local, whether at the local, national or national or national or international level, international level, international level, real rules about real rules about real rules about how these how these how these systems work and the systems work and the systems work and the harm and damage harm and damage harm and damage they can cause. they can cause. they can cause. Well, what's Well, what's Well, what's interesting to me is that interesting to me is that interesting to me is that , of course, there will be , of course, there will be , of course, there will be regulation. This is regulation. This is regulation. This is clear to me. clear to me. clear to me. The range of how this The range of how this The range of how this could span could span could span from from from pharmaceutical pharmaceutical pharmaceutical regulation, where we regulation, where we regulation, where we want want want pharmaceutical pharmaceutical pharmaceutical companies to produce companies to produce companies to produce safe medicines. Although safe medicines. Although safe medicines. Although people might people might people might argue, you know, argue, you know, argue, you know, we also did it we also did it we also did it too slowly. Up too slowly. Up too slowly. Up to, say, to, say, to, say, the regulation of nuclear the regulation of nuclear the regulation of nuclear energy, where, in energy, where, in energy, where, in essence, we ultimately essence, we ultimately essence, we ultimately want no want no want no new nuclear new nuclear new nuclear power plants. And power plants. And power plants. And depending on what depending on what depending on what happens with artificial happens with artificial happens with artificial intelligence, and intelligence, and intelligence, and given given given how negative how negative how negative people are about it people are about it people are about it , one , one , one possibility here is that possibility here is that possibility here is that the situation could the situation could the situation could change in such a change in such a change in such a way that the answer way that the answer way that the answer to a bunch of questions will to a bunch of questions will to a bunch of questions will simply be "no." We don't simply be "no." We don't simply be "no." We don't want to be want to be want to be replaced by robots. Which replaced by robots. Which replaced by robots. Which actually, when you actually, when you actually, when you say it out loud, say it out loud, say it out loud, seems pretty seems pretty seems pretty straightforward, and that's straightforward, and that's straightforward, and that's probably what probably what probably what people people people would say, in a would say, in a would say, in a democracy or not, we democracy or not, we democracy or not, we don't want robot don't want robot agents crawling all over agents crawling all over agents crawling all over the economy. No, we don't the economy. No, we don't the economy. No, we don't want to be want to be want to be endlessly endlessly endlessly called by your called by your called by your Muse agents Muse agents Muse agents trying trying trying to book tables to book tables to book tables everywhere because you've everywhere because you've everywhere because you've just made it just made it just made it functionally functionally functionally impossible for a impossible for a impossible for a restaurant to restaurant to ever answer ever answer ever answer phone calls phone calls phone calls with a human, and this is kind of a with a human, and this is kind of a with a human, and this is kind of a collective collective collective extension of all of extension of all of extension of all of that. I mean, that. I mean, that. I mean, I know people who I know people who I know people who you know are you know are you know are managers of different managers of different managers of different firms, and they just firms, and they just firms, and they just say I'm just say I'm just say I'm just drowning in reports drowning in reports drowning in reports generated by artificial generated by artificial generated by artificial intelligence, all of a sudden.

intelligence, all of a sudden. My email is My email is simply filled with simply filled with simply filled with requests, apparently requests, apparently requests, apparently written by artificial written by artificial written by artificial intelligence, which reduces intelligence, which reduces intelligence, which reduces my efficiency in my efficiency in my efficiency in all areas. I'm not all areas. I'm not all areas. I'm not just saying whether there will be just saying whether there will be regulation or not, but I regulation or not, but I think one of the think one of the think one of the risks is risks is risks is that people that people that people will decide maybe will decide maybe will decide maybe they don't want it they don't want it they don't want it in large enough in large enough in large enough sectors of the economy sectors of the economy sectors of the economy or to use or to use or to use it in some of the ways it in some of the ways it in some of the ways that people expect. that people expect. that people expect. There's a very good article in New York magazine by There's a very good article in New York magazine by There's a very good article in New York magazine by John John John Herman about the Herman about the Herman about the initial pains of initial pains of initial pains of companies suddenly companies suddenly companies suddenly absorbing a huge absorbing a huge absorbing a huge number of requests from number of requests from number of requests from Muse agents. And we're not Muse agents. And we're not Muse agents. And we're not ready for ready for ready for everyone to have an agent everyone to have an agent everyone to have an agent calling you like a calling you like a calling you like a robot, trying to robot, trying to robot, trying to negotiate a negotiate a negotiate a price reduction, price reduction, price reduction, book a table, book a table, book a table, sell a sell a sell a concert ticket or whatever, concert ticket or whatever, concert ticket or whatever, and people might and people might and people might not really not really not really like that. like that. like that. Yes. Yes. Yes. They might not They might not They might not like it. like it. like it. The problem is that there The problem is that there The problem is that there seem to be some seem to be some seem to be some obvious preferences here, obvious preferences here, obvious preferences here, and this picture is not and this picture is not and this picture is not uniform. So only uniform. So only uniform. So only about 2.5% of about 2.5% of Americans pay Americans pay Americans pay for an AI subscription, which for an AI subscription, which for an AI subscription, which considering they've been considering they've been considering they've been available for 3 years is a available for 3 years is a available for 3 years is a very slow very slow very slow takeoff.

takeoff. takeoff. I like, I like, I like, excuse me, I excuse me, I excuse me, I like this like this like this statistic in one of statistic in one of statistic in one of your recent your recent newsletters, that more newsletters, that more people are paying for GLP-1, people are paying for GLP-1, people are paying for GLP-1, more more more Americans are paying Americans are paying Americans are paying for GLP-1 than for AI. for GLP-1 than for AI. for GLP-1 than for AI. Well, yes, because it's Well, yes, because it's Well, yes, because it's really quite really quite really quite beneficial and you beneficial and you beneficial and you feel the effects feel the effects feel the effects within a few within a few within a few weeks. But hundreds of weeks. But hundreds of weeks. But hundreds of millions of people in millions of people in millions of people in America and around the America and around the America and around the world use world use world use AI through the free AI through the free AI through the free version. And when version. And when version. And when companies like Meta companies like Meta companies like Meta start producing and start producing and start producing and giving you these giving you these giving you these tools, I'm tools, I'm tools, I'm sure people sure people will eventually use will eventually use them, even if they them, even if they them, even if they vote with vote with vote with their their their social media posts social media posts social media posts saying they don't want saying they don't want saying they don't want any AI. Either it any AI. Either it any AI. Either it will become inevitable, or will become inevitable, or will become inevitable, or they will still they will still they will still have it on have it on have it on their phones and their phones and their phones and use it. use it. use it. I think we'll just have to I think we'll just have to accept this accept this possibility. I think possibility. I think possibility. I think we also have to we also have to we also have to realize, and this is where realize, and this is where realize, and this is where we think of we think of we think of technology as this technology as this technology as this giant giant giant unauthorized unauthorized unauthorized experiment that it has experiment that it has experiment that it has been for 40 or 50 been for 40 or 50 been for 40 or 50 years, that what you're years, that what you're years, that what you're describing about describing about describing about robocalls robocalls robocalls from AI agents to from AI agents to from AI agents to hotels and hotels and hotels and restaurants, or as restaurants, or as restaurants, or as my my my orthopedic surgeon told me, he orthopedic surgeon told me, he orthopedic surgeon told me, he just gets long just gets long just gets long emails of a emails of a emails of a thousand words, 10,000 words thousand words, 10,000 words thousand words, 10,000 words from patients who from patients who from patients who have clearly checked their have clearly checked their have clearly checked their medical records in the chat medical records in the chat . Unfortunately, it . Unfortunately, it . Unfortunately, it will happen, and these will happen, and these will happen, and these costs are now costs are now costs are now being imposed on the system, being imposed on the system, being imposed on the system, and we will have to and we will have to and we will have to figure out how to deal with it figure out how to deal with it figure out how to deal with it . Now I . Now I . Now I pay, it's not a lot, I pay, it's not a lot, I pay, it's not a lot, I pay about a dollar pay about a dollar pay about a dollar a month, uh, to my, uh a month, uh, to my, uh a month, uh, to my, uh , phone , phone , phone company to company to company to filter out fake filter out fake filter out fake phone calls, phone calls, phone calls, which are essentially

which are essentially robocalls. Uh, and this is what will happen, and we will happen, and we will happen, and we will call it GDP. These are will call it GDP. These are will call it GDP. These are additional costs. additional costs. additional costs. I think one of the things I think one of the things I think one of the things that your comment, that your comment, that your comment, Ezra, really reminded Ezra, really reminded Ezra, really reminded me of is that me of is that me of is that in society, in in society, in in society, in certain Western certain Western certain Western capitalisms, there's this capitalisms, there's this capitalisms, there's this idea that it's always idea that it's always idea that it's always about about about efficiency. In a strange efficiency. In a strange efficiency. In a strange way, societies way, societies way, societies need friction. need friction. need friction. Hmm. Hmm. Hmm. And when you take away And when you take away And when you take away friction, even friction friction, even friction you didn't know existed, it you didn't know existed, it suddenly has consequences. suddenly has consequences. suddenly has consequences. You know, there have been examples of You know, there have been examples of You know, there have been examples of people people people using using using big language models, uh big language models, uh , to file , to file , to file lawsuits in lawsuits in lawsuits in small claims court, small claims court, small claims court, for example. Do for example. Do for example. Do we want there to be more, do we want there to we want there to be more, do we want there to we want there to be more, do we want there to be less friction? be less friction? be less friction? Do we want Do we want Do we want more of this? Do we want more of this? Do we want more of this? Do we want this, you know, your this, you know, your this, you know, your example of this with example of this with example of this with restaurants? And we have restaurants? And we have restaurants? And we have all these frictions that all these frictions that all these frictions that exist in exist in exist in society, and blindly society, and blindly society, and blindly removing them creates, in the removing them creates, in the removing them creates, in the language of language of language of technology, almost a technology, almost a technology, almost a denial-of- denial-of- denial-of- service attack on service attack on service attack on modernity, where modernity, where modernity, where suddenly we have different things coming at us from all suddenly we have different things coming at us from all suddenly we have different things coming at us from all sides sides sides , almost like, , almost like, , almost like, you know, people, you know, you know, people, you know, you know, people, you know, Russian bots Russian bots Russian bots attacking a website, attacking a website, attacking a website, only it's only it's only it's happening from all happening from all happening from all sides because of the sides because of the sides because of the growth of, you know, um, growth of, you know, um, growth of, you know, um, the number of bots and the number of bots and the number of bots and other things. And I think that's what other things. And I think that's what people react to, that people react to, that feeling of being feeling of being feeling of being out of control, out of control, out of control, where all the friction where all the friction where all the friction is gone, and suddenly, is gone, and suddenly, is gone, and suddenly, you know, I can't you know, I can't you know, I can't access access access things, or nothing things, or nothing things, or nothing works the way I works the way I works the way I expect. So simply expect. So simply expect. So simply reducing friction has reducing friction has reducing friction has huge implications huge implications huge implications for for for society as a whole. And I society as a whole. And I society as a whole. And I think people think people think people are responding to that. What are responding to that. What are responding to that. What will happen is that will happen is that will happen is that these two these two these two AIs will be AIs will be AIs will be throwing things back and forth throwing things back and forth and to and to and to the right.

the right. the right. And it's going to be an increasingly And it's going to be an increasingly And it's going to be an increasingly nasty nasty nasty economy, so I economy, so I economy, so I think those are going to be think those are going to be think those are going to be big areas where people big areas where people big areas where people are going to say, maybe not. are going to say, maybe not. So I want to So I want to close this question, close this question, close this question, which is, first of all, which is, first of all, which is, first of all, about you, about you, about you, Azim, what do you Azim, what do you Azim, what do you need to see need to see need to see to start thinking, "Oh to start thinking, "Oh to start thinking, "Oh no, it's getting pretty no, it's getting pretty no, it's getting pretty rough here?" And then, for rough here?" And then, for rough here?" And then, for you, Paul, it's the opposite. you, Paul, it's the opposite. you, Paul, it's the opposite. What do you need to What do you need to What do you need to see to see to see to think, "Oh, maybe think, "Oh, maybe think, "Oh, maybe I'm wrong, and I shouldn't be I'm wrong, and I shouldn't be I'm wrong, and I shouldn't be ... These investors ... These investors ... These investors are right, and there's are right, and there's a lot of money to be made here... a lot of money to be made here... Um... Well, there's so Um... Well, there's so Um... Well, there's so many risks many risks many risks that we've talked about, and there's that we've talked about, and there's that we've talked about, and there's all the complexity and all the complexity and all the complexity and fragility of fragility of fragility of financing. What financing. What financing. What keeps this on the other side of keeps this on the other side of keeps this on the other side of health is whether health is whether health is whether real customers real customers real customers are spending real are spending real are spending real money and getting money and getting money and getting real results. real results. So what would, uh, So what would, uh, worry me if I worry me if I worry me if I saw a couple of saw a couple of saw a couple of quarters where the amount quarters where the amount quarters where the amount that American that American that American businesses businesses businesses were spending on AI was, were spending on AI was, were spending on AI was, uh, the growth rate uh, the growth rate uh, the growth rate was halved was halved , let's say, two , let's say, two , let's say, two quarters in a row. That would quarters in a row. That would quarters in a row. That would mean they've mean they've mean they've satisfied their satisfied their satisfied their appetite, and there's not going to be the appetite, and there's not going to be the appetite, and there's not going to be the revenue to revenue to revenue to keep it on the keep it on the keep it on the right side of right side of right side of health.

health. Field. Field. Um, there's so many, I Um, there's so many, I Um, there's so many, I mean I'm mean I'm mean I'm convinced I'm convinced I'm convinced I'm wrong about wrong about wrong about most things, so most things, so most things, so I'm always happy to find I'm always happy to find I'm always happy to find more things that I'm more things that I'm more things that I'm wrong about. So I wrong about. So I wrong about. So I just I look at the just I look at the just I look at the economic data. So economic data. So economic data. So if we started to see if we started to see , I don't know, , I don't know, , I don't know, productivity growth that was way above productivity growth that was way above trend, that would be trend, that would be amazing, wouldn't it? amazing, wouldn't it? amazing, wouldn't it? So far, we have So far, we have So far, we have trend data on trend data on productivity growth productivity growth in the United States. There's no in the United States. There's no in the United States. There's no indication that AI is indication that AI is indication that AI is improving improving improving productivity. If productivity. If productivity. If we saw a dramatic we saw a dramatic we saw a dramatic increase in increase in increase in participation, that participation, that participation, that would help me say, " would help me say, " Okay, AI is an Okay, AI is an Okay, AI is an inclusive inclusive inclusive technology that's technology that's technology that's bringing people back into the bringing people back into the bringing people back into the economy." That would economy." That would economy." That would really help. really help. really help. If we saw If we saw If we saw GDP growth GDP growth GDP growth come back above come back above come back above trend, we're still trend, we're still at about, you at about, you know, post- know, post- know, post- World War II levels, at the World War II levels, at the World War II levels, at the lower end. Any of lower end. Any of lower end. Any of that would tell me that that would tell me that something something big was happening, like a big was happening, like a big was happening, like a real real real productivity shock in productivity shock in productivity shock in the economy that the economy that the economy that justified this kind of justified this kind of justified this kind of spending. So, you know, spending. So, you know, spending. So, you know, if any of those things, and if any of those things, and if any of those things, and probably three of them probably three of them probably three of them at the same time, happened, at the same time, happened, at the same time, happened, that would tell me that that would tell me that that would tell me that I was wrong. You know, I was wrong. You know, I was wrong. You know, if these productivity numbers if these productivity numbers come out, then come out, then well, that's going to change. well, that's going to change. well, that's going to change. But if not, you're But if not, you're But if not, you're looking at something that's looking at something that's looking at something that's more like the 1930s more like the 1930s more like the 1930s than, I don't know, the than, I don't know, the than, I don't know, the 2000s.

2000s. 2000s. Okay, I think that's a Okay, I think that's a Okay, I think that's a good place to good place to good place to end. So, end. So, end. So, our always last our always last our always last question is, what three question is, what three question is, what three books would you books would you books would you recommend to recommend to recommend to the audience? Paul, why do the audience? Paul, why do the audience? Paul, why do n't we start with you? n't we start with you? n't we start with you? Sure. Um, Sure. Um, Middlemarch is kind of Middlemarch is kind of Middlemarch is kind of my main my main my main book, which is book, which is book, which is about ordinary people being about ordinary people being about ordinary people being caught up in the winds of caught up in the winds of caught up in the winds of change when there's a change when there's a change when there's a big big big technological technological technological revolution going on. It was revolution going on. It was revolution going on. It was right when the right when the right when the Industrial Revolution Industrial Revolution Industrial Revolution was starting to take was starting to take was starting to take off, and I think off, and I think off, and I think it's an incredible, uh, it's an incredible, uh, it's an incredible, uh, incredible book about incredible book about incredible book about that. So the other book that. So the other book that. So the other book that I recommend is that I recommend is Emma Griffin's Dawn of Freedom, which is, Emma Griffin's Dawn of Freedom, which is, um, a collection of biographies um, a collection of biographies um, a collection of biographies and notes from and notes from and notes from ordinary ordinary ordinary workers during the workers during the workers during the Industrial Industrial Industrial Revolution. It's just a Revolution. It's just a Revolution. It's just a wonderful, wonderful wonderful, wonderful wonderful, wonderful look at what it look at what it look at what it was like in the middle of that was like in the middle of that was like in the middle of that painful transition. And painful transition. And painful transition. And when people talk when people talk when people talk about economic data about economic data about economic data and everything, like it and everything, like it and everything, like it was to experience. What was to experience. What was to experience. What were those feelings? What was the were those feelings? What was the were those feelings? What was the pain I was feeling? And, pain I was feeling? And, pain I was feeling? And, you know, what were my you know, what were my you know, what were my hopes for the future? It's hopes for the future? It's hopes for the future? It's just a great book just a great book just a great book about that. And the last one, um, about that. And the last one, um, about that. And the last one, um, and I was going to and I was going to and I was going to read Samuel read Samuel read Samuel Insull, a largely forgotten Insull, a largely forgotten Insull, a largely forgotten figure.

figure. figure. He was probably one He was probably one He was probably one of the most important of the most important of the most important entrepreneurs of the 1920s entrepreneurs of the 1920s entrepreneurs of the 1920s , whose empire in the , whose empire in the , whose empire in the electricity industry electricity industry electricity industry shows what's shows what's shows what's going on, that going on, that going on, that you can build an you can build an you can build an empire on GPT, empire on GPT, empire on GPT, and even if you and even if you and even if you do it successfully it do it successfully it do it successfully it can still lead to can still lead to can still lead to failure and disaster. failure and disaster. failure and disaster. But it's a great story But it's a great story . So, "The Insult, Rise and . So, "The Insult, Rise and . So, "The Insult, Rise and Fall of the Billionaire Fall of the Billionaire Utilities Tycoon Utilities Tycoon " by Forrest " by Forrest " by Forrest MacDonald. Azeem? Well, MacDonald. Azeem? Well, MacDonald. Azeem? Well, so my three books. so my three books. so my three books. The first one is "The Day The first one is "The Day The first one is "The Day the Universe Changed." It the Universe Changed." It the Universe Changed." It was written by a was written by a was written by a BBC science journalist back in the BBC science journalist back in the BBC science journalist back in the early '80s. I read it as a early '80s. I read it as a early '80s. I read it as a teenager, and I just teenager, and I just teenager, and I just reread it, and it's reread it, and it's reread it, and it's about how, as we about how, as we about how, as we develop new types of develop new types of develop new types of knowledge, we start knowledge, we start knowledge, we start to rethink our to rethink our to rethink our role in the world, what's role in the world, what's role in the world, what's important to us, and important to us, and important to us, and how how how disturbing that is. And it disturbing that is. And it disturbing that is. And it takes us through takes us through takes us through a thousand-plus years of a thousand-plus years of a thousand-plus years of Western and Western and Western and Asian history. Asian history. The second book is The second book is Turgenev's Fathers and Sons, Turgenev's Fathers and Sons, Turgenev's Fathers and Sons, written in the 19th century written in the 19th century . It's at that moment that . It's at that moment that . It's at that moment that the ideas of science and the ideas of science and the ideas of science and modernity are modernity are modernity are penetrating Russia, and there penetrating Russia, and there penetrating Russia, and there 's a 's a 's a generational conflict between the more generational conflict between the more generational conflict between the more traditional traditional traditional generation and the generation and the generation and the younger generation, younger generation, younger generation, and what remains as a and what remains as a and what remains as a link between them is something that link between them is something that even even the latest breakthroughs in the latest breakthroughs in science can't science can't science can't tell us about what it means tell us about what it means tell us about what it means to be human. I to be human. I to be human. I think that's very think that's very think that's very relevant. And the last one, relevant. And the last one, relevant. And the last one, which I'm just rereading which I'm just rereading , is , is , is Voltaire's Candide. Azime Azhar

Voltaire's Candide. Azime Azhar Voltaire's Candide. Azime Azhar Podgursky, Podgursky, Podgursky, thank you very much. Thank you. Thank you.