The AI Empire: Value, Power, and the Race for Compute の全文書き起こし

know, former uh CEO briefly of, well not that briefly, of Google, and then has a lot of investments in AI and is very involved in the AI field. 00:02:26.000 --> 00:02:42.400 Anyway, it's his sort of like take alongside I think a geopolitical historian, if I remember rightly, but about the looking looking forward and looking at what we might be entering into and how it might be a little different to to what many of us think. 00:02:42.639 --> 00:02:45.680 And so we're gonna we're gonna talk about that. 00:02:45.919 --> 00:03:03.439 Talk about the idea that you know maybe the inexorable growth of AI and of the countries that are funding AI, um maybe those that thought of inexorable progress may not be quite what where we're actually headed. 00:03:03.520 --> 00:03:05.599 So really, really interested to talk about that. 00:03:05.840 --> 00:03:09.919 But let's kick off with the discussion around ROI and value. 00:03:10.159 --> 00:03:17.680 You know, guys, when we finished talking last week or last time we were on, um I was kind of really taken, we sort of had half a conversation about it. 00:03:17.759 --> 00:03:36.800 I don't know whether we're gonna have the other half or whether we're gonna have the next quarter or what we're gonna do here, but I found myself thinking quite a lot about the fact that when you think about investment in AI, there's obviously the investment that that investors are making in AI companies. 00:03:36.879 --> 00:03:47.599 There's the investment that governments are making in this, and then of course there's the investment that we make as individuals and the investment that we make as companies. 00:03:47.840 --> 00:03:50.240 As many people will know, we might have even mentioned it last week. 00:03:51.199 --> 00:04:00.800 The number of individual humans who pay for AI services is still a very low percentage of overall AI users. 00:04:00.960 --> 00:04:06.400 It's a very minor revenue stream for them compared to some other things, uh, other ways in which they make money. 00:04:06.639 --> 00:04:10.000 But businesses are spending a lot of money on AI. 00:04:10.960 --> 00:04:24.720 And if those AI companies are gonna thrive, then you know the businesses, governments, and other parties are gonna need to continue to want to spend money with them. 00:04:24.879 --> 00:04:31.120 So that kind of got me to this thought about value more broadly versus just return on investment. 00:04:31.199 --> 00:04:54.240 I want to start with you, Matt, and ask you as somebody who works in this field, as somebody whose company I know has invested in AI solutions, when you think about the value that you're looking for from an investment, how do you go about even contemplating what value means to you? 00:04:54.800 --> 00:04:55.439 Hmm. 00:04:56.480 --> 00:05:08.800 I'm actually struggling with this right now, making decisions on you know where to kind of place emphasis and priority with AI, whether you're building internal or utilizing other models, et cetera. 00:05:09.360 --> 00:05:32.000 I feel like these organizations, anthropic, no different, um, but even the smaller, more targeted ones that exist within industries that start to bubble up, whether it's a Salesforce type, something like that, or Viva, those types of organizations, they're sort of painting themselves into a collective good corner. 00:05:33.199 --> 00:05:37.600 Like it's in our best interest for them to be successful. 00:05:37.759 --> 00:05:39.439 We know that's gonna cost money. 00:05:41.279 --> 00:05:46.240 Can we get X value out of it? 00:05:46.399 --> 00:05:48.319 Let's say double the value of what you spend. 00:05:48.480 --> 00:05:53.439 Can you get that out of the system through either reduction of time? 00:05:54.160 --> 00:06:00.879 Um, you know, I don't know about people necessarily right out of the gate, but is there a value being returned? 00:06:00.959 --> 00:06:07.920 Maybe you're exponentially empowering your people to the tune of making them 2x more productive rather than reducing the numbers. 00:06:08.160 --> 00:06:21.040 Um so they've kind of boxed themselves into if they are going to charge this much for something that you subscribe to, for example, it has to be good or no one will subscribe. 00:06:21.120 --> 00:06:25.199 It has to keep getting better, or no one will subscribe, or continue to subscribe. 00:06:25.279 --> 00:06:26.480 It'll hit a point. 00:06:26.959 --> 00:06:45.600 And I wonder if it's almost like the dawn of a way to set a new type of paradigm for companies to exist because maybe infinite growth quarter on quarter isn't the goal here, but sustainability across the board. 00:06:45.839 --> 00:06:51.839 So it's a tough one, Paul, because I I think it's not going to give us much value yet. 00:06:52.399 --> 00:06:57.920 We'll find the value the more we use it, and then once we find the value, we won't want to give it up. 00:06:59.199 --> 00:06:59.600 Yeah. 00:07:00.480 --> 00:07:01.199 You know what I mean? 00:07:01.360 --> 00:07:08.480 Like we're not there for it yet, but we're having to pay for it in advance with the hope that it'll get there. 00:07:09.040 --> 00:07:16.959 And if they don't get there, I think that it is somewhat existential not only for them, but for a lot of things. 00:07:17.120 --> 00:07:24.720 So we kind of collectively have to make sure it works and it's valuable as a society, as a species at this point. 00:07:24.959 --> 00:07:27.040 It seems like it's being forced upon us. 00:07:27.199 --> 00:07:41.680 I mean, we can get into those other articles that we talked about, they all touch on this, but it's very interesting if we're gonna figure out the value, we have to measure it in much larger timescales. 00:07:42.160 --> 00:07:56.079 Well, I I found myself wondering whether the goal, if you were to get more behind closed doors, whether the goal is even to provide value or is the goal to get us hooked, right? 00:07:56.399 --> 00:08:02.720 I mean I don't know what the discussion

was like with the cigarette companies, right? 00:08:02.800 --> 00:08:08.160 But I kind of doubt that they were talking about how do we provide value to our smokers. 00:08:08.319 --> 00:08:09.680 Maybe they were, I don't know. 00:08:10.160 --> 00:08:18.639 Um, but what they were talking about was making sure that you know the smokers were choosing Marlboro over camel. 00:08:19.040 --> 00:08:25.279 Um and if they were, everything else kind of took care of itself because the product was addictive. 00:08:25.519 --> 00:08:25.759 Right? 00:08:25.839 --> 00:08:34.159 You hear a lot in companies, startups, a lot about delighting their customers, becoming indispensable to their customers. 00:08:34.320 --> 00:08:42.720 I wonder if there's a difference, Bill, between making yourself indispensable in some way, right? 00:08:42.879 --> 00:09:00.559 Which could be you're indispensable in the same it's indispensable in the same way that a cigarette is to a cigarette smoker, or whether the primary goal is to actually provide more business value than the cost of the thing. 00:09:00.720 --> 00:09:03.120 I mean, we're speculating, but what do you think? 00:09:03.679 --> 00:09:05.279 Well, a couple of things come to mind. 00:09:05.519 --> 00:09:09.600 One, last week there was a lot of discussion around the company Netflix. 00:09:09.759 --> 00:09:13.039 So I think all of us probably know Netflix, most people do. 00:09:13.360 --> 00:09:18.720 And what was really interesting was there was this whole discussion around what is the value of Netflix. 00:09:18.799 --> 00:09:26.000 And what's interesting is Netflix, as we know, they sort of flooded their catalog with a lot of content. 00:09:26.639 --> 00:09:33.120 And there's this whole debate with people saying, okay, well, now it's like every time I go in, I have nothing to watch, right? 00:09:33.200 --> 00:09:43.440 And so we've got this huge volume of possibility, and yet we're sort of immobilized and we find it a lot of work, and we say, you know what, I'm just gonna go to something else, right? 00:09:43.519 --> 00:09:44.720 And that's been noticeable. 00:09:45.279 --> 00:09:46.480 The tyranny of choice thing. 00:09:46.799 --> 00:09:48.320 Yeah, tyranny of choice. 00:09:48.559 --> 00:09:53.200 That's kind of interesting to me because what it tells me is we keep thin slicing our attention. 00:09:53.519 --> 00:09:59.600 And if you zoom out to this AI paradigm that we're in, I think you can look at it two ways. 00:09:59.759 --> 00:10:05.200 One is additive, like you're talking about, Matt, which is like, hey, how can we add to the bottom line? 00:10:05.360 --> 00:10:07.360 How can we add to the productivity? 00:10:07.679 --> 00:10:16.559 But the other side of it that I'm seeing, even today, I saw the news that OpenAI listed its new, you know, 6.1 Tera is what it's called model. 00:10:16.960 --> 00:10:23.039 And the big distinction there is it's reduced its API pricing in that model by 50%. 00:10:23.759 --> 00:10:25.360 So that's discount. 00:10:25.600 --> 00:10:33.039 So I think we have some people that are trying to get our attention in an additive way, meaning here's more you can do. 00:10:33.200 --> 00:10:39.120 And there's other people trying to get our attention saying, here's the things you can do for less the price. 00:10:39.360 --> 00:10:44.399 And that brings me back to your point around cigarettes or things that are addictive. 00:10:44.879 --> 00:10:46.320 How are you going to hook people? 00:10:46.559 --> 00:10:53.279 You're eventually going to get to a certain point to say, hey, this vape pen or this cigarette or whatever is cheaper here. 00:10:54.240 --> 00:10:56.399 Is that more attractive to us? 00:10:56.720 --> 00:11:00.720 Or is the concept of brand and value and what it stands for more attractive to us? 00:11:00.879 --> 00:11:01.200 I don't know. 00:11:01.279 --> 00:11:05.360 That's a human behavior anthropological, sociological question. 00:11:05.600 --> 00:11:13.519 What I do worry about is we don't have, to your point around stamina, I don't know if we have the ability as humans to overcome some of these things. 00:11:13.759 --> 00:11:21.840 And I think that's being used against us in economic models, in industry models, and certainly Paul, as we'll talk about with Empire. 00:11:22.159 --> 00:11:22.720 Yeah. 00:11:22.960 --> 00:11:43.120 You know, the funny thing about what you just mentioned there is that I wonder if the goal is to get us to a point where as employees and as teams, we kind of forget how to do our work without it. 00:11:43.440 --> 00:11:45.759 And that doesn't take very long, right? 00:11:46.080 --> 00:11:49.440 And so this is not just a question of being additive. 00:11:49.600 --> 00:12:04.320 I mean, I know, for example, if you took my AI out of my cold dead hands today, right, there would be things that I would be like, okay, well, I don't really remember what I did or how I did it before I had AI. 00:12:04.720 --> 00:12:22.399 And so in that sense, it can be indispensable, but that again is different, I would suspect, from being valuable in a truly holistic, almost like measurable way. 00:12:26.399 --> 00:12:32.000 I also want to touch on narrowness versus broadness. 00:12:32.639 --> 00:12:52.480 So when we had Jeff Moore on the show, and Jeff was talking about uh how these organizations should be run, you guys will recall that he was discussing basically, I would say it was almost like first principle stuff. 00:12:52.559 --> 00:13:06.799 It was like, um, okay, go industry by industry or problem by problem in each industry, and say, okay, what are the specific uh problems that these uh organizations face? 00:13:08.080 --> 00:13:11.360 Much of it probably industry specific, right? 00:13:11.679 --> 00:13:23.679 And what would be the highest value thing I could do in that industry, and what is the AI-powered solution to build that thing. 00:13:25.279 --> 00:13:31.200 And what we're seeing right now is a bit different to that,

right? 00:13:31.279 --> 00:13:41.679 What we're seeing is almost as aren't we clever, aren't we cool, you know, look at us do this mathematical proof. 00:13:42.240 --> 00:13:47.759 Or, you know, it's I mean, my degrees in math, I love that, right? 00:13:47.840 --> 00:13:55.120 I love the disruption, and I love the challenge, and I love you know sort of how this has all uh gone about and I love the challenge to the field. 00:13:55.200 --> 00:13:58.799 I think that's interesting, and I think it's interesting that it does show the power of it. 00:13:59.039 --> 00:14:07.440 But that is I think quite different to verticalizing rapidly. 00:14:08.320 --> 00:14:11.279 And this is where I come back to you, Matt. 00:14:11.919 --> 00:14:22.320 Tell me if I'm wrong, but I think when when most organizations think of value and when most of us as individuals think of value, right? 00:14:23.039 --> 00:14:37.120 We're thinking about relatively specific, relatively narrow things that equate to specific problems that we're looking to solve in our industry. 00:14:37.759 --> 00:14:55.600 And so these companies have got to figure out are they gonna like start very rapidly, very fast building proper vertical practices to go after these problems? 00:14:55.759 --> 00:15:14.000 Are they going to build out a, you know, an oncology division, for example, that maybe is going literally into different forms of cancer, or are they gonna look at it in terms of a set of processes within life sciences and go after these individual processes? 00:15:14.240 --> 00:15:16.559 Or are they gonna say, you know what, we're none of those things. 00:15:16.639 --> 00:15:25.679 We're a platform, and the actual value will be provided by other companies on top of us that sell to companies like your company, Matt. 00:15:25.840 --> 00:15:41.519 But it seems to me they gotta do one of those two things, because otherwise they are literally just a they're not gonna in medium term, they're not gonna be seen anymore in as a value creator any more than say a gas company is seen as a value creator. 00:15:41.600 --> 00:15:50.639 So what when you when you think of value, are you thinking about it, the sort of broad value that an anthropic can provide, or are you thinking of a narrow, I need to solve this problem? 00:15:50.799 --> 00:15:55.519 And if you do, where does a company like an anthropic or an open AI fit? 00:15:56.159 --> 00:15:56.799 Hmm. 00:15:57.360 --> 00:16:10.799 I guess the way I'm thinking of them is more like utility, generic utility, and I want to use the best model for whatever I need it to do. 00:16:13.360 --> 00:16:18.000 I think a straight competition between them the way it exists now. 00:16:18.240 --> 00:16:28.879 I mean, I've certainly felt this is that the two models, if you just take OpenAI and Claude, for example, have very different personalities, and I use them for very different things. 00:16:29.679 --> 00:16:35.279 So as a business, I kind of need to have them both if I want to get the full value of AI. 00:16:35.840 --> 00:16:38.879 If you're kind of interactively conversationally with them, right? 00:16:39.440 --> 00:16:40.159 Yeah, yeah. 00:16:40.559 --> 00:16:45.200 Yeah, I'm not sure how they can get around staying broad. 00:16:45.600 --> 00:16:51.519 I mean, at the same time though, like you remember Google started a healthcare organization. 00:16:51.600 --> 00:16:57.440 They figured, hey, we're gonna have the data, we're gonna have a you know, this amazing uh potential. 00:16:59.039 --> 00:17:09.200 It's pretty volatile, and to throw out a$1.5 to$2 trillion valuation in such an unknown is pretty bold. 00:17:11.200 --> 00:17:15.759 So yeah, I think like I said, uh we have to get the value out of it. 00:17:15.839 --> 00:17:19.039 Otherwise, what are we all doing? 00:17:21.519 --> 00:17:23.279 It's not I don't think it goes away. 00:17:23.680 --> 00:17:31.759 Well, we might just be becoming dependent upon it without you know, potentially without defined or real value, I suppose, is the answer. 00:17:32.079 --> 00:17:35.359 Well, early, early, before we know what the value is, right? 00:17:35.440 --> 00:17:43.680 Like to bring back the cigarette thing, I don't know if anybody ever said there was anything beyond looking amazingly cool that cigarettes provide value. 00:17:44.000 --> 00:17:58.559 Umas there's tyranny of choice, there's so many things you could potentially use AI to make better and add value for it's almost like we don't even know where to start. 00:17:58.720 --> 00:18:10.480 Most people aren't building code, most people are you know asking questions, chatting, asking the Google, and it happens to be AI answering them, but yeah, I think we're stuck with it. 00:18:11.039 --> 00:18:13.039 And we better figure out how to make it work. 00:18:13.119 --> 00:18:15.200 But 1.5 trillion yikes. 00:18:15.519 --> 00:18:16.319 Where's it going, Bill? 00:18:16.480 --> 00:18:19.839 What are we gonna what are we gonna how are we gonna be defining value? 00:18:20.000 --> 00:18:29.119 And is it gonna be is it gonna be the open AIs and the anthropics of the world that are actually defining the value, or is it gonna be solutions that are built on top of them? 00:18:30.000 --> 00:18:34.559 I I think that the question could be answered by what's left on the shelf. 00:18:34.880 --> 00:18:36.880 And I don't think we know that yet. 00:18:37.119 --> 00:18:48.640 And you know, to the point that you were making, Matt, like it feels a little bit like we're paying for the trade routes, but we don't know what the goods are on the trade routes, right? 00:18:48.720 --> 00:18:51.279 So it's like, okay, am I doing the Silk Road? 00:18:51.599 --> 00:18:53.759 Or am I doing like what am I paying for? 00:18:53.920 --> 00:18:58.640 I'm paying for sort of this guaranteed experience. 00:18:58.960 --> 00:19:09.519 And as you said, Matt, we're kind of having it in more like curiosity challenges versus actual physical assets. 00:19:10.000

--> 00:19:20.960 But we're at a point now where I think part of the magic, it all just feels so like, well, whatever gets whatever's on that ship is worthy of my money because it's come this trade route. 00:19:21.039 --> 00:19:25.119 Like I was told it came from OpenAI, or I told it came from Anthropic. 00:19:25.279 --> 00:19:32.319 And I I do wonder if at some point people are gonna stop and say the thing that's coming off that ship is not of use to me anymore. 00:19:32.640 --> 00:19:41.440 Um I'm not benefiting, I'm not changing, I'm not growing, therefore I stop paying, or it's just stop showing up to the dock, so to speak. 00:19:41.680 --> 00:19:42.720 What happens then? 00:19:42.880 --> 00:19:47.119 And that's where this valuation thing has me really curious. 00:19:47.279 --> 00:19:58.000 If we don't start taking some of this and applying it to local relatable problems, I don't know how the house of cards doesn't fall apart. 00:19:58.079 --> 00:19:59.200 I really don't. 00:19:59.920 --> 00:20:00.400 Yeah. 00:20:00.640 --> 00:20:14.960 Well, I think that's a good segue into say the news of the news of the day, which is the fact that Anthropic has done its what I think is its final filing, that it needs to not be able to go public. 00:20:15.440 --> 00:20:29.759 For those who haven't seen it yet, the stock ticker, all important, what three letters have they chosen, or three or four letters have they chosen for their stock ticker, they've gone with ANT, Ant, which I think which I think is very, very interesting in its own right. 00:20:29.920 --> 00:20:31.759 And can't believe that wasn't ticking. 00:20:32.000 --> 00:20:32.400 Yeah. 00:20:32.640 --> 00:20:39.920 Well, I it makes me wonder if when they named the company they already thought of that, because it kind of is apropos, isn't it? 00:20:40.000 --> 00:20:44.720 If you think about the sort of collective intelligence that it sits within ant colonies and all that. 00:20:45.039 --> 00:20:45.920 For all ants, yeah. 00:20:46.160 --> 00:20:50.000 Yeah, and then yes, and then turning us into the ants in the picture. 00:20:50.160 --> 00:20:52.160 But that's part of the filing. 00:20:52.319 --> 00:20:58.799 Existential, when they're pointing out risks to their business, there is a bunch of stuff about existential risk. 00:20:58.880 --> 00:21:03.200 I don't think they were as direct as saying, well the world can end, in which case we wouldn't exist. 00:21:03.920 --> 00:21:09.279 Because I suppose if the world ends, then no company exists, so it's probably not a specific risk to them. 00:21:09.599 --> 00:21:18.799 But they did list a bunch of potential headwinds, including, you know, shifts in government regulations and government attitudes towards them and all that kind of stuff. 00:21:19.119 --> 00:21:29.599 And then, of course, the other thing is the number that is believed to be associated with it, which is looking right now between a valuation between 1.5 and 2 trillion. 00:21:29.759 --> 00:21:38.160 Which, ironically, if you actually were to look back a few months ago, that's slightly on the low end of numbers that were being thrown about. 00:21:38.319 --> 00:21:41.359 There were numbers as high as 30 trillion. 00:21:41.519 --> 00:21:44.240 That was right on the screen, but there were numbers as as high as that. 00:21:45.200 --> 00:21:56.160 I had a conversation with a couple of friends of mine relatively recently, and they wanted to know what would be the number where I would think it was, you know, of value. 00:21:57.119 --> 00:21:59.279 And I kind of gave an answer. 00:22:00.240 --> 00:22:06.960 But I I don't think it's I think it's so difficult to know to both of your points, right? 00:22:07.759 --> 00:22:23.519 And of course there's a company that you can sort of think of as a as an AI company that recently IPO'd, which was SpaceX, and that came in I think at one point eight trillion, if I get that right, if I remember that right, but it's right around that range. 00:22:25.279 --> 00:22:33.680 But the interesting thing about SpaceX, which is sort of different, is yes, X AI is part of it is considered to be part of that. 00:22:35.200 --> 00:22:38.160 But there's so many other factors. 00:22:38.319 --> 00:22:44.480 I mean SpaceX is you know, it shoves physical stuff up in sp in space. 00:22:44.640 --> 00:22:56.720 It's got a ton of established government contracts, it's kind of at a spot where it's where there are no real alternatives to some of the work that it does. 00:22:56.960 --> 00:23:06.799 Um how we've let ourselves get to that point is a very difficult and troubling question, and I think kind of does apply to AI as well. 00:23:07.359 --> 00:23:22.720 Um, the idea that you've got the defense of nations enthralled to the whims of one company and a potentially unstable leader of that one company doesn't seem to me to make a ton of sense. 00:23:23.359 --> 00:23:47.119 But the point here is that there's an AI company, but also an AI company that that does specific stuff that we know we will continue to need into the far distant future, and right now at least, is we know that we'll need more of it, and we know that this company is sort of uniquely positioned to do it. 00:23:47.200 --> 00:23:50.559 So you can sort of work your way into evaluation from that. 00:23:50.640 --> 00:23:53.680 But this is different to that, and this is potentially a higher value. 00:23:54.559 --> 00:23:56.480 We'll throw something in the uh go ahead, Bill. 00:23:56.720 --> 00:23:59.759 I think the other thing to keep in mind here, and Matt, you had mentioned this. 00:23:59.839 --> 00:24:02.079 We've talked a little bit around like utility. 00:24:02.559 --> 00:24:10.000 One of the things that I thought was really interesting as I looked into some of their filings was these commitments, and I put in quotes, right? 00:24:10.559 --> 00:24:16.559 Meaning their commitment to cloud spend in order to deliver these trade routes of whatever to us.

00:24:16.720 --> 00:24:19.039 Some of these numbers are really huge. 00:24:19.200 --> 00:24:25.200 So they have committed, and I think these are non-cancelable, meaning you commit to make them. 00:24:25.519 --> 00:24:55.440 So Google, they've committed 111 billion to Amazon 110 billion to Microsoft 31 billion to Broadcom 161 billion to XAI 84 billion to, and AMD 20 billion to so I'm not a math guy, I didn't add all those things up, but like you're sitting here and you're like, okay, this valuation, but also commitment of spend, but also revenue coming in, subscriptions, like it all gets very congealed. 00:24:55.920 --> 00:25:02.240 And that's what I don't understand how people are making, sort of throwing a dart at the wall and saying, makes sense to me. 00:25:03.200 --> 00:25:07.759 Yeah, because those companies that you mentioned are all gonna add this to their balance sheet. 00:25:07.920 --> 00:25:08.079 Yeah. 00:25:08.400 --> 00:25:12.319 And if it doesn't come in, they're gonna report a loss. 00:25:12.559 --> 00:25:13.680 Because oh man. 00:25:13.920 --> 00:25:14.319 Yeah. 00:25:14.640 --> 00:25:34.319 And we are in a position where, as was pointed out in an article that I shared with you guys and we'll pop in the show notes, these astronomical valuations, they kind of require astronomical performance associated with them. 00:25:34.799 --> 00:25:42.480 Um and the FD got a hold of some reporting on the numbers that that Anthropic were getting. 00:25:42.559 --> 00:25:49.279 And they and you can look at those numbers through two separate lenses. 00:25:49.759 --> 00:26:02.400 You know, one lens through which you could look through it is oh my goodness, how on earth did they get to where they are from when they started? 00:26:02.559 --> 00:26:24.319 I mean, how do you how do you get to a point where your annualized revenue run rate is close to$80 billion when you didn't have any run rate in January of 2024? 00:26:24.400 --> 00:26:34.079 I mean, that's just kind of like mind-blowing, and it gets you to a point where you think, oh my goodness, I suppose this is a big, big, big, big deal. 00:26:34.880 --> 00:26:41.119 But and it is, you know, truly there's never been anything like this. 00:26:41.839 --> 00:26:47.039 But at the same time, the valuation sort of reflects that. 00:26:47.440 --> 00:26:57.119 And as the article that I shared with you guys talks about, there's a line in it which I love, which is like, why are investors complaining about the goat? 00:26:57.440 --> 00:27:03.680 Because everything is unprecedented in AI, and so unprecedented revenue is not enough. 00:27:04.480 --> 00:27:23.039 And that's the thing that probably is gonna make a lot of people very, very rich, but it's also the thing that if you fail ever so slightly, or if you just don't quite make what you were supposed to make, then you've potentially got a problem. 00:27:23.279 --> 00:27:29.200 And I don't know how well, I don't know what you guys think, but you know, how does Elon get around this? 00:27:29.279 --> 00:27:36.559 Well, Elon gets around this by like, so far anyway, he gets around it by having other companies that he just merges them with, or something like that. 00:27:36.720 --> 00:27:41.519 You know, so right now he's talking about merging Tesla with sp you know with SpaceX, right? 00:27:41.599 --> 00:27:47.920 And there's always a kind of look over here, a look over here, a look over here that keeps people excited. 00:27:48.079 --> 00:28:01.440 But if you're the comp if you're the everything company, you know, if you're the everything AI company, which they are right now, what is the look over here that you can do if you already say that your AI can do everything, right? 00:28:01.680 --> 00:28:04.640 And I find that part fascinating. 00:28:07.759 --> 00:28:19.200 I wonder if you look on the balance bill of those commitments, if there's an equal outgoing licensing cost going to anthropic from them, and it's just as easy. 00:28:19.440 --> 00:28:20.240 That's a great question, right? 00:28:20.319 --> 00:28:21.279 That's a really good question. 00:28:21.599 --> 00:28:30.160 Well, and this is where it's almost like compared to like, are we talking the equivalent of like fulfilling airplane orders, right? 00:28:30.319 --> 00:28:32.720 Are you actually going to get something on the other end of that? 00:28:32.880 --> 00:28:37.039 Or is it just like uh, hey, I mark this down here, you mark this down here. 00:28:37.119 --> 00:28:39.279 This was always my issue with cloud computing, right? 00:28:39.359 --> 00:28:44.160 And a lot of these companies doing business together because you mark the revenue over here. 00:28:44.400 --> 00:28:46.559 This was what happened with OpenAI and Microsoft. 00:28:46.640 --> 00:28:52.559 They said you count the revenue in Azure spend, and then we just count it because you're giving us Azure Cloud Compute. 00:28:52.720 --> 00:28:55.119 I'm like, so are we talking funny money or what? 00:28:55.279 --> 00:28:58.400 And apparently it just doesn't seem to be a big deal to a lot of people. 00:28:58.960 --> 00:29:10.640 Well, it was probably less than a year ago that we were talking, you know, collectively, the community was talking about this sort of shell game that at least appeared to be going on, right? 00:29:10.799 --> 00:29:18.400 I mean, those companies making commitments to NVIDIA because, you know, they were the major spends. 00:29:18.640 --> 00:29:34.400 I've sort of related, you might have seen that a lot of the revenue that is being reported by SpaceX because it encompasses the XAI thing, is around data center builds. 00:29:34.559 --> 00:29:34.799 Right. 00:29:35.359 --> 00:29:56.720 And those data center the theory is, and investors seem to like it, is they build out a bunch of data center capacity, they lease elements of that data center capacity out to some of the other AI companies, and then when demand for XAI increases, then they pull that back and make use of it

themselves, right? 00:29:56.799 --> 00:30:00.640 So it's sort of almost like an arbitrage type thing going on. 00:30:01.200 --> 00:30:04.400 But more and more and more that sounds like oil companies. 00:30:05.039 --> 00:30:05.599 Exactly. 00:30:05.920 --> 00:30:06.559 Utilities. 00:30:07.039 --> 00:30:07.599 Yeah. 00:30:07.839 --> 00:30:12.799 I mean, it's a very strange It's a very strange thing. 00:30:13.119 --> 00:30:20.640 And if it is oil companies, I suppose, you know, that gets us through the next until it ultimately settles. 00:30:20.720 --> 00:30:37.920 I suppose that gets us to a a certain point, and then they do become these sort of de facto large companies that provide a whole bunch of compute and some models that other people can layer value on top of, and they are the utility provider, maybe. 00:30:38.480 --> 00:30:41.359 But again, everything is maybe, right? 00:30:41.599 --> 00:30:43.039 It's the wildest thing. 00:30:43.200 --> 00:30:55.440 A one point a 1.5 to$2 trillion valuation on a thing where you can ask, you know, 20 experts, what does the future look like associated with these companies? 00:30:55.680 --> 00:31:06.799 And you'll either you well, because we live in the world of you know instant and strong opinions, you're probably gonna get, well, it's definitely gonna be this, but in reality, underneath it, it's like, I don't know, right? 00:31:07.839 --> 00:31:11.599 And and investors are working their way through this. 00:31:12.160 --> 00:31:15.359 I got a theory I want to throw past you guys. 00:31:15.839 --> 00:31:18.640 There is venture capital money exists, right? 00:31:18.799 --> 00:31:28.400 And until it doesn't, right, and it's kind of like the primary pretty much primary source of net new business growth in the American system. 00:31:30.240 --> 00:31:44.640 I think we spoke a little while ago about the idea that venture capital is weird right now, because if you're building out software, you know, you can't build a competitive moat on software, basically. 00:31:44.960 --> 00:31:56.319 And so I wouldn't want to particularly be in VC right now because how would I value a software company that had an AI component, right? 00:31:56.480 --> 00:32:02.799 That provided software that people, you know, that had end users and had an exp and had a user experience. 00:32:02.880 --> 00:32:10.559 I'd be like, I don't know how to value that thing because I don't know whether this company can be sustained, period. 00:32:11.279 --> 00:32:36.640 And actually, even in cases where it doesn't have an AI and a particular end user component, there have been you know major releases of software over the last couple of months whereby somebody's come out with a big splash, and a week later other parties have reverse engineered it and taken this highly valuable thing and made it potentially valueless. 00:32:37.039 --> 00:32:38.880 So I wouldn't want to be those companies. 00:32:39.839 --> 00:32:54.880 So maybe this is just like them looking at this going, well, I don't know what it is, but I got money and they've at least got something that I can point to and say that's valuable, right? 00:32:55.039 --> 00:32:58.640 These frontier AI models have value. 00:32:59.039 --> 00:33:03.599 The data centers upon which they run have value. 00:33:04.160 --> 00:33:41.759 I know that there's gonna be a need for a ton of compute capability, and very there's gonna be a need for a massive load of GPUs running a massive load of cycles doing stuff, and these companies know how to build those things, so I'm gonna push all my money into that because I know that's at least something, and it's so big that you can't have, you know, the three of us can't say, you know what, we're gonna do our version of Anthropic tomorrow. 00:33:42.640 --> 00:33:48.640 We could build an app that runs on it, but we couldn't say we're gonna we're gonna take out Anthropic tomorrow, right? 00:33:48.880 --> 00:33:57.039 So in that sense, there is value to it, even though you don't know what exactly what that value is, it has value in the way that all these other things do not. 00:33:57.119 --> 00:34:03.279 So maybe they just say, Oh, we'll throw all our chips in on that, because at least it has something definable as value. 00:34:04.720 --> 00:34:05.279 I don't know. 00:34:05.359 --> 00:34:06.160 What do you guys think? 00:34:06.480 --> 00:34:07.599 For right now, yeah. 00:34:07.759 --> 00:34:15.119 But like again, you know, Paul, you and I have kicked around this idea of like at some point, does this become standard oil? 00:34:15.440 --> 00:34:18.000 Does this become the Dutch East India Company? 00:34:18.159 --> 00:34:19.920 Does this become like I don't know? 00:34:20.880 --> 00:34:33.119 Well, let's take that exact analogy of the Dutch East India Company, because that gets us to a final thing, and we're gonna have to chat about it very briefly. 00:34:33.440 --> 00:34:38.719 But boy, you send it my way, Bill. 00:34:39.119 --> 00:34:45.599 Talk about the remarkable article in Foreign Affairs magazine that I'd urge everybody to read. 00:34:45.920 --> 00:34:51.280 Yeah, so as people may or may not know, this was co-written by Eric Schmidt and Niall Ferguson. 00:34:51.360 --> 00:34:56.400 Niall Ferguson's a historian who's written a lot about empire, including the British Empire. 00:34:56.639 --> 00:35:13.920 And essentially, if you boil this down, it basically says that as we talk about AI as industry and how it's reshaping industry, both Eric and Niall Ferguson and a few others have kind of taken this approach on it, which is it's more akin to empire than it is to industry. 00:35:14.239 --> 00:35:28.320 And the reason being, if we look back at empires, now empires are built and how they're sustained, it's about, as we talked about before, it's about trade routes, it's about assets, it's about resources, it's about holding things close and charging high rates. 00:35:28.400 --> 00:35:34.000 It's about, as we were talking about,

utility and gas and food and access. 00:35:34.480 --> 00:35:45.119 And so this article sort of said if you look at it behaviorally and habitually, some of the things we might see are more in the realm of empire than they are in industry. 00:35:45.280 --> 00:35:53.440 And then zoom out even further and this sort of war between geopolitical US, China, as we talked about before. 00:35:53.599 --> 00:35:56.559 And it really comes down to whoever has the resources. 00:35:56.800 --> 00:36:09.760 And if you look at the tech stack as an analogy to the resources that we need to run, quote unquote, an empire, it's analogous and it's quite upsetting and it's quite disturbing. 00:36:09.920 --> 00:36:23.840 And I hope people read that article and don't see the world the same because I feel like you, Paul, you and I sat down the first day we met and we were having this conversation and saying this is why we need to sort of slow and pull the plug a little bit. 00:36:24.079 --> 00:36:32.000 But I think it's gonna take people once again looking at history in the past to understand where we're going. 00:36:32.239 --> 00:36:33.440 So that's what I would say. 00:36:33.519 --> 00:36:34.800 I would love your thoughts on it, though. 00:36:35.119 --> 00:36:38.159 Well, I got multiple interesting takeaways from it. 00:36:38.320 --> 00:36:46.320 One, of course, is just a very, very simple thing, which is that when you are in a position of power, right, when you are building an empire, right? 00:36:46.559 --> 00:36:49.440 In a sense, you're building the empire because you're the only ones who can, right? 00:36:49.519 --> 00:36:56.719 You've got the financial resources to be able to build the thing, and technology advancement comes alongside that, right? 00:36:56.800 --> 00:37:01.760 You see a new way of developing technology, you build the new technology. 00:37:02.000 --> 00:37:06.320 And you have control over that at the time at which you build the technology. 00:37:06.400 --> 00:37:14.400 What you don't have control over is what do people fully do with it once that has been built. 00:37:14.719 --> 00:37:25.360 And so what it means is rather perversely and ironically, is that in many cases, and they go into this with regard to the British Empire as as well, right? 00:37:25.519 --> 00:37:35.199 You get to a point where you build the thing, and the act of building the thing is the thing that accelerates your undermise. 00:37:36.159 --> 00:37:39.760 Because then other people take advantage of the thing of which you've built, right? 00:37:40.000 --> 00:37:41.760 And so I think that is fascinating. 00:37:41.920 --> 00:37:52.960 I think the other thing is fascinating is that, and I love this because I've been sort of having this, he doesn't know I'm having a fight with him, but I've been having a fight with Peter Diamandis for the last six months, right? 00:37:53.519 --> 00:37:56.639 As in, I'm just commenting on his stuff and he's not reading it, right? 00:37:57.039 --> 00:38:03.280 But this whole idea of the age of superabundance and the age of superabundance is upon us. 00:38:03.440 --> 00:38:07.599 I've always thought blind in in a bunch of different ways. 00:38:07.760 --> 00:38:12.559 It's convenient for people that are promoting what they're promoting and then that are investing and all that kind of stuff, right? 00:38:12.639 --> 00:38:13.760 It's convenient for them. 00:38:14.000 --> 00:38:41.039 But as this article points out, actually, as it always has been, this type of development is really scarcity, not abundance, because you get to a point where you're growing stuff so much or the demands of what you're of what you're building are so great that you start to hit upon um challenges that you've not hit on before. 00:38:41.199 --> 00:38:47.199 Like, have we got enough rare earth minerals, for example, would be would be one example, right? 00:38:47.440 --> 00:39:01.760 And so now you're uncovering these heretofore unexpected barriers that are then hit that potentially then arrest things in ways that we haven't thought about before. 00:39:02.079 --> 00:39:03.280 Matt, I know you've read it. 00:39:03.519 --> 00:39:11.760 Brief thoughts before we get out of here, we'll continue hopefully the discussion on this because it's a big, big deal, and we will, and as I say, we'll link to it in the show notes. 00:39:12.239 --> 00:39:13.199 Yeah, two things. 00:39:13.360 --> 00:39:20.239 One is that it was interesting to me that you know we're building this much faster than previous imperial empires took to build. 00:39:20.320 --> 00:39:22.719 This was within you know a year or two, as you mentioned. 00:39:22.960 --> 00:39:25.440 Anthropic didn't exist and now it does. 00:39:25.599 --> 00:39:39.679 Um, but the thing that really struck me in this article was the idea of the imperial trilemen, what they called it, where you can defend your dominion, have tech primacy, or the welfare of the masses, and you gotta pick two, right? 00:39:40.159 --> 00:39:54.079 So that does put us in somewhat of a tough spot, and you know, certainly would hope that we maybe choose the welfare and the tech primacy and come together and don't have to defend dominion, but that's the thing. 00:39:54.320 --> 00:39:56.800 History doesn't show us that fact, Matt. 00:39:57.440 --> 00:39:58.239 Does not at all. 00:39:58.400 --> 00:39:59.440 Yeah, it does not at all. 00:39:59.760 --> 00:40:21.280 I mean, it is it is it is utterly, utterly wild to think that here we when you sit here in the present and you're and you assume that the countries, the United States and China, will have primacy forevermore because of the actions that they're taking right now. 00:40:22.079 --> 00:40:31.519 And actually history, time and time and time again, actually tells us the exact opposite of that. 00:40:32.159 --> 00:40:35.280 And that is that's fascinating. 00:40:36.320 --> 00:40:38.320 So, you know, who knows? 00:40:38.480 --> 00:40:39.920 But it's an interesting thing. 00:40:40.079 --> 00:40:43.280 We almost certainly I would say we won't be alive to see it, but we might. 00:40:43.360 --> 00:40:45.440 It's all this whole thing is moving so fast. 00:40:45.760 --> 00:40:47.519 Yeah, we should know by next year, I think. 00:40:48.400 --> 00:40:51.199 By next year, we'll know it's like, are we doing something really stupid here? 00:40:51.360 --> 00:40:51.760 I don't know. 00:40:51.920 --> 00:40:54.400 Anyway, all right, we'll leave it there today for that. 00:40:54.559 --> 00:41:01.039 I know I I think we should probably have an extra episode where we're talking about nothing other than this. 00:41:01.119 --> 00:41:02.320 I think it's so important. 00:41:02.559 --> 00:41:05.199 So we'll look into I think we just agreed on that. 00:41:05.280 --> 00:41:06.719 So we'll we'll do exactly that. 00:41:06.880 --> 00:41:10.320 We'll uh record a special extra edition on that at some point. 00:41:10.480 --> 00:41:14.480 Uh in the meantime, guys, thank you so much as ever, and we'll see you again soon. 00:41:14.719 --> 00:41:15.519 Thanks, all. 00:41:17.599 --> 00:41:20.559 Humanity Working is produced by Matt Neal. 00:41:20.880 --> 00:41:26.800 If you liked this episode, please take a moment to rate and review us on your favorite podcast platform. 00:41:26.960 --> 00:41:29.920 It makes a huge difference in helping others find us. 00:41:30.079 --> 00:41:36.800 And don't forget, you can find all episodes and get our weekly newsletter at Humanity Working.