Latest / In the Loop AI / Cathie Wood: Inside ARK's $20 Trillion Dollar Bets In AI, Elon Musk, Robotaxies, Brain Implants, & Crypto

Transcript

Cathie Wood: Inside ARK's $20 Trillion Dollar Bets In AI, Elon Musk, Robotaxies, Brain Implants, & Crypto · In the Loop AI

• Cathie: [0:00] We believe that real GDP growth is going to accelerate in the years ahead. We think in the next five years to 7%. Now, we have five innovation platforms evolving with AI turbocharging, all of them, robotics, energy storage, AI, blockchain technology, and multi-omic sequencing. So we should see a much more rapid acceleration this time around than last time. We're being conservative when we say 7%. Ryan: [0:31] Welcome to Bankless, where we explore the frontier of AI, robotics, brain implants, and crypto. This is Ryan Sean Adams. I'm here with Josh Cale today, so not David. Josh is the co-host from our new Limitless podcast. And today we've got an exciting episode. We are chatting with ARK Invest founder and CEO, Kathy Wood, along with Brett Winton. These are two people whose job it is to predict tomorrow and to buy the future. They got a message for us, and I think it's one that bankless listeners need to hear. It's not enough to be a siloed investor. You can't just be an AI investor or a crypto investor, because frontier technologies, they're converging. Ryan: [1:11] And to invest well, you need to identify all of the intersection points. That's what we do in today's episode. Josh: [1:16] So in this episode, you're about to hear a bunch of topics. How Elon's empire forms history's biggest AI data flywheel, Wright's law, which is the model that allows us to project the collapsing cost of AI, robo-taxis, humanoid robots. There's a $20 trillion battle happening between OpenAI, XAI, and Anthropic, and ArcHap and Stone, all three. The trillion-dollar markets that are birthed by Starship's ultra-cheap launches, we have Neuralink mind-control robots, and then finally, stablecoins and how they ignite a bullish 2026 productivity super boom. Ryan: [1:44] If that literally sounds like all the things, it's because it is. Now, usually we publish these episodes on Limitless, our frontier technology podcast, but we wanted you to hear this on Bankless too. But if you're not subscribed to Limitless, go do that now so you don't miss episodes like these. There's a link in the description. You can subscribe on Spotify, YouTube, or wherever else you listen. So let's get right to the episode with Kathy Wood and Brett Whitten. Very excited to introduce you once again to Kathy Wood. She's one of innovation's loudest champions. Kathy, welcome back to the podcast. Cathie: [2:11] Happy to be here, Ryan. Thank you. Thank you, Josh, as well. Ryan: [2:15] We also have Brett Witten. He's a chief futurist at ARK Investment Management. Brett, it's great to have you on as well. Brett: [2:21] It's great to be here. Ryan: [2:22] All right. So we're going to be covering a bunch of the biggest tech bets, I think, in 2025, the current era, how investors get exposure to these things. Of course, we'll be talking about AI, robotics, maybe some brain chips, maybe space. We'll end by talking a bit about stable coins and crypto. So, and before we get in, and I think at the beginning of this, one thing as Josh and I were preparing this agenda, we noticed is there was one person that seemed to be at the intersection of many of these technologies, at least four out of five, and maybe I want to say five out of five, and that person is Elon Musk. Yeah. Okay. Kathy, you've been a huge supporter of Elon Musk, and in this technology era, it's impossible to escape his influence on the space. What is it about Elon? Cathie: [3:05] Well, I'll start, but I'll tell you with Brett and all of our analysts, everyone has contributed to this discussion because you're right, it's crossing every sector. And I think in his brilliance, and this is more than serendipity, this is just brilliance, he saw that there was a convergence among technologies that was going to take place with AI at the center, which puts proprietary data at the center center, I guess I should say. You know, that's the competitive advantage. So, you know, when you look at Tesla, for example, and we were talking a bit about this before, you know, most analysts following it, they thought they were following an auto company, but they weren't. They're following the largest AI project on Earth. And they didn't know what to do with that, right? And they were also following the convergence among robotics, energy storage, and AI. Cathie: [4:16] So you have three separate S-curves for each of those technologies. And then they start feeding each other as we're seeing, as we believe we will Cathie: [4:27] see, I should say, in robotaxis. and you have explosive growth. And, you know, we began to think, okay, what are, as we were moving into the AI age and we understood proprietary data is the name of the game, Cathie: [4:43] Key, key, key, key to competitive advantage. Look at all of the different data that Elon has amassed. Nobody has his autonomous robo-taxi data or his robot data. We've all been driving robots. We didn't know it. I didn't think about it that way with a Model 3 and a Model Y for me, but we were gathering data for him every day. X, nobody has those data feeds. Speaker0: [5:15] If you look at even Neuralink. Cathie: [5:18] What is the biggest data explosion out there emerging? It is what we call multiomics data, genomic revolution data, you know, DNA, RNA, proteins. And guess what? He's doing Neuralink and neural networks, think AI, are patterned after the brain. He understood all of this, I think. And then, of course, you've got SpaceX, which is, you know, another new orbit of data. So I think that's the secret. He understood this. And I, you know, Brett has been totally obsessed with AI since, well, even before, probably since we began the firm in 2014, but especially after deep learning and transformer architecture in 2012 and two, well, 2017, specifically to ARC. And so he, I'm sure, has another very original take on this. Yeah, what would you say to this, Ryan: [6:17] Brett? So do you agree with Kathy that Elon is the person who can just see the convergence? Is that the secret sauce here? Cathie: [6:24] Yeah, part of that. Brett: [6:24] I mean, I think it's he is intolerant of sloth and delay. And when you have a rapidly changing technological landscape where it's actually it's unclear exactly how the tools can be used, the most important thing you can do is to do a lot of experimentation. Brett: [6:42] And so he drives his team hard to get to basically like an end result. And the end result could be the rocket exploding, you know, and, and, but when the technological landscape is, is changing so quickly, then you actually know what next move to make with better information on the backend, if you're kind of driving to that velocity. And so in times of technological transition, vertical integration is a dominant strategy. He clearly embraces that. and then having a very rapid iteration loop is really critical. Brett: [7:15] And so, of course, we have exposure across Neuralink. We own in our venture fund. We have SpaceX. We're obviously big investors in Tesla, as everybody knows. Brett: [7:25] X and XAI, now a combined entity, we have exposure to them through the venture fund. We really believe in his philosophy for developing product in an iterative way that ultimately gets to things that'll transform the world. I think his other superpower is that he's willing to take on short-term pain for a long-term gain and gain of mission for his companies and, you know, gain of return for investors. And so that's characteristic across all, you know, whether it's crashing rockets or basically, you know, laying out the assets that eventually will be RoboTaxi capable. And now they are, we think, on a sensor stack that everybody said was not going to be viable for RoboTaxi. And just being willing to like lay the foundation that he sees is going to, you know, provide monumental returns over a medium time horizon, which aligns with us, despite the fact that everybody likes to make fun of him, you know, about kind of like the incremental step as if it's not going to work. Cathie: [8:25] You know, I'll just tell a story. You know, we come from a world, the traditional asset management world. This is not venture. We're very much involved and excited about venture with our venture fund. But the traditional world, you know, failure was not allowed. I mean, it was frowned on. And it was even you miss your operating income margin by 30 basis points. That can be a disaster. That's ridiculous. What do we see Elon do with Falcon 9? Falcon 9, one failure, two failures, three, four. I think he failed eight or seven to nine times. And then he got it right. And he's had hardly any failure since. And now he has 90%, maybe more than that, Brad, of all of the satellites that in the universe. Brett: [9:24] Yeah. Well, maybe not the universe. We don't know what's happening. Cathie: [9:27] Oh, yes, we do. Brett: [9:28] It's for solar leave. Let's not over claim here. But certainly in the solar system, I think that's a fair statement. Yeah. Josh: [9:38] And we're kind of seeing it again with Starship. So this is this repeating pattern that we've seen. And we're going to get into all these companies, which I am very excited to dive into. Josh: [9:44] But I kind of want to take a step back and start by defining these guardrails. We kind of have these canonical laws to help provide guidance in understanding what the future is going to look like. So we had Metcalfe's law, which is more users on a network, relates to its value in an exponential sense. We've had Moore's law, which is the number of transistors on a microchip, which doubles about every two years, which leads to this exponential rise in processing power. And now we have this law called Wright's law, which says as production volume doubles, the cost of production decreases by a constant percentage, and it creates this really pretty inverse chart. So Kathy or Brett, you and the team at ARC have actually gone as far as to publish a full page on your website about Wright's law. So what I would love for you to do is just explain the significance of it and why you think it's so important as we look forward over the next 10 years through these new paradigm shifts. Cathie: [10:28] We were looking at semiconductors hitting up against the laws of physics. And Brett was saying, oh, wait a minute. Okay. So this is all a function of time. Every 18 months, two years, twice the power, same cost or some derivative of that. And he was casting around saying, no, there has to be a more universal law here. You know, it can't be a function of time. You know, things, you know, stuff happens, right? And to use the nice version of that, stuff happens. And so it can't be. And he went back to the early days of civil aviation. So, Brett, why don't you take it from there? Brett: [11:14] Sure. So Theodore Wright was looking at airplane manufacture and realized that not for every cumulative kind of like number of airplanes produced, you had a consistent percent cost decline. But also you could look at the underlying cost components, like the cost of kind of the raw material, the refined raw materials that went into the stack, the cost of the motor. And it also followed this similar pattern. He's like, oh, wow, that seems like a good way to forecast kind of cost declines. And actually, it turns out there's a study from the Santa Fe Risk Institute that demonstrates that it's the best kind of way to do cost declines. And a really great example is in lithium ion batteries, where Moore's law is magically every however many years you get a cost decline. So that fails on first principles. Like if you stop investing in a technology, it's clearly not going to magically get less expensive. And what was happening with batteries is actually we were going from laptops with big honking batteries to cell phones with much smaller batteries. So the cumulative doublings of battery demand were diminishing. And so it looked as if on a Morse-Loft-style cost decline, Speaker0: [12:22] Battery costs were mature. Brett: [12:24] They were declining single digit percent per year. And so people were taking that assumption and then plugging it into their assumptions for how performant electric vehicles could be. So literally, you can go back 2014, 2015, EIA, which is the Government Forecasting Agency, thought they were going to be on the order of 100,000. It was, I think, 200,000 electric vehicle sales per year forever. Brett: [12:47] Like to 2050, it would never be more than like this niche category that appealed to people who wanted to feel good about their relationship to the earth and, you know, were willing to waste money to get a less performant product. But if you look at what happens to batteries, once you bring a car model to market that works, that's using batteries, there's so many kilowatt hours in that car model, it actually explodes the demand for the underlying batteries. So Elon pushing the Model S into a spot where it was like going to do a reasonable volume of sales and be a very performant vehicle, actually single-handedly bent the battery production curve, which then caused costs to begin to decline again. So people on Moore's Law were like, well, this is a mature technology forever. Whereas with using Wright's Law, you could see, hey, with this production requirement for this vehicle, it'll carry us across the entire cost segmentation stack of kind of like the motor vehicle transport. And so it's an illustration of how like actually paying attention to the unit economics of the future case and how much demand that unlocks can inform you as to how costly a future technology is going to become. Brett: [13:59] And so we generalize that across all the technologies we study. We have cost decline curves for all the technologies. And it's like you could go back to the Model T and look at the first 10 years of Model T production and predict on a cost per horsepower basis the cost of internal combustion engine vehicle to within 20% today. Over 100, like you take the first 10 years of data and you can get really close, if you think about like the error range there as to how much does a car cost and so it's really good if you have a medium to long term point of view understanding and underwriting, like, is this technology going to be meaningful? And to what future buyers will it be meaningful? And so that's how we apply it. Cathie: [14:42] And Josh, you mentioned that, I just want to clarify, and Brett did in terms of what he said, but you mentioned for every doubling, it's every cumulative doubling. So what does that mean? One to two, two to four, four to eight. And so what we're really looking for is very early technologies that have lots of cumulative doublings ahead of them. And at that time, electric vehicles did. It was for every cumulative doubling in the number of electric drive trains produced, so battery pack systems, et cetera, cost declined at a 28% rate. If you apply it to industrial robots, it's 50%. If you apply it to DNA, short read DNA, it's 40%. If you apply it to AI, and our analysts did with both, when they combined the hardware and the software, and Brett, you can explain, and especially on the AI side, this is important for this audience, Cathie: [15:46] The metric we used, but that combination was 48%. But the astonishing thing, and this is when we were like gobsmacked, which was, wait a minute, this cumulative doubling is taking place in less than a year's time. That's why training costs are dropping 70% per year. Inference costs, we're getting information now that they could be dropping 98% Cathie: [16:12] for every cumulative doubling. It's astonishing. And maybe, Brett, you want to talk about the unit metric we are using. Brett: [16:22] Sure. I mean, for AI, it's total, it's like a index compute unit devoted to AI training. And so, you know, the fact that, and it makes sense intrinsically, like the fact that Elon and Sam Altman and Google are pushing huge dollars into building these training clusters, they figure out all kinds of different engineering ways to make them more efficient. And then there's software architectural advances on top that result in a performance metric, you know, like the output performance that we care about costs much less much less to deliver to the end customer. Ryan: [16:58] Let's talk about AI some more then. So back into that metric for us, just to make sure that the audience really understands this, Brett. So Kathy is making the case that if we're looking at something like Rice Law, we're really looking at sectors and areas where we're going to see a cumulative Ryan: [17:13] doubling in terms of demand. Cathie: [17:15] And I think- Technologies. Technologies. Ryan: [17:19] Okay, technologies. Okay. And so I think that that's not quite clear in terms of what to look for. To a lot of investors, I think the demand seems to come out of nowhere. So let's start to apply this to something like OpenAI and ChatGPT certainly seem to come out of nowhere. Suddenly, 100 million users in an AI chat box? How does that happen? So applying this to AI, how do you sort of forecast that cumulative demand of the technology? And what is the metric that you can see in terms of cost reduction? Brett: [17:53] One, I'm going to caveat this heavily because AI of all technologies is what we would think of as the most convergent technology. As in, I think it's pretty easy and accurate to say all of the advance we're seeing in the language model space and the multimodal model space is actually then applying at some delay to kind of like embodied AI, including robo taxis. And then it's some delay further into like more of the humanoid robot space. And then, you know, because it's so freaking complex to the biology space as well. And so, so actually we, the, the AI demand function that we currently have defined is really limited to kind of like knowledge worker productivity and the more strict language model space. And then there's, you really should almost have a separate expectation for what's going to happen in consumer and with kind of like generated media and entertainment content for people. But just sticking to the language model and enterprise space, which actually gets us our prospective cost decline, generally we look at how much productivity these will deliver to knowledge workers. So you ask, what's tangibly going on here? Well, I mean, I as an analyst need to underwrite a biotech asset. And a good first start is to actually ask the O3 Pro model to give me like a report on that biotech asset. And so it generates a report that Brett: [19:21] You know, I could have had an associate do it over the course of a week. Instead, it's in 20 minutes and it's ready for me. And so it's kind of like I can get an associate's week's worth of work in 20 minutes for, you know, much less cost. So I become more productive as somebody analyzing information. Or, you know, you call a customer support line and, you know, there are products in market today where it's like, instead of having a customer support agent handle this, let's triage upstream with an AI voice agent. Brett: [19:50] And, you know, the agent will hopefully be less painful than the annoying voice trees you have to go through where you just yell operator, operator, operator until they give you a human. And so you provide more customer support for less money. And so at a high level, you know, literally 10 plus trillion dollars are expected to be spent on knowledge work wages over the next, through 2030 and incremental knowledge work wages. We And enterprises tend to pay for the productivity they get from software. They don't pay full freight. They pay like 10% of the value they get. So meaning they have an ROI of like 10x on their investment in software. If you make those same assumptions here, you end up in a high single digit trillion spent on AI software. And what does that get spent on? It gets spent on generating these tokens that kind of like provide better customer support, better analysis, better kind of like copy, you know, more customized advertisements, the whole kind of like productivity Brett: [20:49] advance and administration. And so you take all that and then you can translate it into a cost decline and say, hey, these things are going to become wildly more performant over the course of five years, probably a hundred to a thousand times. Josh: [21:00] You and Brett are both describing this huge productivity unlock where there will be a lot of wealth generated. Some of these numbers are in the trillions of dollars and that money is going to flow into some companies. And Ryan and I were snooping around your portfolio and we were looking at, hmm, what are they invested in? And we found XAI, OpenAI, Anthropic. And I was kind of looking to understand why you chose those companies. Is this a hedge against AGI? Is one going to be faster than the other? What is the general role that they play in your portfolio and how do you see them all playing out? Cathie: [21:31] You're picking up on something that another question we're asked is, wait a minute, typically in the VC world, you know, you're not going to see all three of those in the same venture fund. And that's because many venture funds end up sitting on the board. They have a tremendous amount of information about a company and no other company in the same space would want information shared. So we don't sit on boards. And we do think this is winner take most and, you know, with distribution and speed and data, proprietary data being critical. So that's why we own all three. We think they are in the pole position. So it's winner take most. There will be somewhat more specialization, different language models, better at different things. And Brett, I can see you have your itching to say something here. Brett: [22:33] Well, I think just at a very high level of the money that's going to go into AI software, some of that goes into like software as a service type companies. Then, you know, some of their costs is going to be to trigger these underlying foundation models and platform as a service type companies. And then that'll go into the infrastructure as a service layer. For that foundation model layer, we think there's a $1.5 trillion revenue opportunity. And, you know, I actually think it's a conservative model, but I can debate it with the team, but call it, you know, approaching $2 trillion. And so then, you know, at least given our expectations for the margin structure, Brett: [23:11] we think that's a $15 to $20 trillion enterprise value opportunity. Cathie: [23:15] The space is so fluid right now, Brett: [23:18] You can't say for sure that one company is going to win over another. You can say chat GPT clearly has a huge distribution advantage and OpenAI is delivering more tokens to end users than probably any company in the world. And that gives them a lot of data and a lot of ability to iterate on product that delivers kind of magical experiences to end users. You can say XAI has... Speaker0: [23:40] Basically, distribution through all of the. Brett: [23:42] Like, what is going to be the default AI model on Starlink? Well, I would wager it's going to be XAI. You know, what's going to happen as kind of like prediction markets and capital markets get fully integrated into X, the platform? Well, probably XAI-powered agents are going to become an important part of providing liquidity across those markets. And what's going to happen to the value of the information on X as a platform as XAI's underlying model improves? It's going to get better. So there's an interesting distribution advantage XAI has. And if you are making a video game that has a character that has to do violent things to another character, you don't want to ask the API to threaten to kill somebody and the API says, as a large language model, I can't do this. It ruins your video game experience. And so XAI also probably has a set of APIs that will be able to deliver to end users the actual kind of thing they want as opposed to kind of a very guarded kind of like politically correct output. And then Anthropic has the best coding model in the world right now and is being built in on the back end of Cursor in a profound way. And so they all have like different angles that they're playing where ultimately we think there's this 15 to $20 trillion bucket people are competing after. And there's probably two or three winners there. And hopefully we own like two or three of those winners Speaker0: [25:06] And, you know, the space is so dynamic. Like Kathy said, you know. Brett: [25:12] China's coming in with these open source models. How do these teams respond? How can they, like, they're all trying to vertically integrate. OpenAI is, you know, building out an endpoint hardware. I would wager that Elon Musk is also going to ask for an endpoint device. He's hinted that it's going to, that XAI is already vertically integrated through its data center. OpenAI is moving in that direction. You know, X has the X platform. Open AI is lacking on the kind of consumer in-facing distribution, except for chat GPT, but they don't have like the social network angle. So there's a lot of kind of, you know, positioning that's going on right now. I will say that we think exposures there as opposed to mega cap tech, where everybody else in the world is exposed, are a much better risk adjusted spot to be. Because like Google should have a credible play here. And they're interesting. They're like, you know, they have a lot of talent. And they also have this core search franchise that's under severe turmoil because of what's happening. Like the marginal attention is going to an AI bot to answer questions. Can Google cross that bridge? Maybe. Apple should be great at this, but their intelligence, Apple intelligence is almost a joke. Siri is horrific. Cathie: [26:19] And we knew that. I just want to focus on those two right now because we were fighting the Mag-6, you know, for the last four years in terms of, you know, We didn't think that they were necessarily going to be the big winners. And we learned a long time ago from Apple, we saw them in the autonomous, you know, the robo-taxi field. Like, if you think about the ultimate mobile device, it's a robo-taxi. And so Apple should have been all over it. And they were trying. And they had one management team turnover after another after another. And we finally said, they don't know what they're doing on AI. And anyone using Siri, still, I can't believe it. I mean, there's no way I would use it. Had to know. They didn't know what they were doing. Same with Google. Cathie: [27:05] It was very interesting. They have the best AI researchers in the world. And so everyone said, sure, they're going to kill it. Speaker0: [27:13] They're not. Cathie: [27:15] And one of the things we learned, and Brett, I think we did this on X, but it's very interesting that Anthropic is able to figure out, this was Frank's post on X, I think. He, Anthropic, was able to figure out how to do something for Frank, very personalized, that Google couldn't, even though Google had all of Frank's information. And I don't even remember exactly why, but these are the nuances that matter now. Maybe you remember why. Brett: [27:48] Yeah, well, and I think there's another. So the foundation model space for enterprises, we think, is this $15 to $20 trillion opportunity. And it's also extremely clear, given the way in which the input-output into computation has changed, that we're undergoing an operating system platform transition. As in, these AI models are the new operating system. Kind of like Cursor is an application that was built on top of it. You know, ChatGPT is with Deep Research. These are other applications that OpenAI is building. And so when you have an operating system transition, it's almost universally true that the prior operating system just does cannot make the adjustments. Ryan: [28:30] Operating system transition in the way we went from kind of desktop to mobile, like that sort of. Yeah. Brett: [28:34] Yeah. So you go from keyboard to mouse. That's where you went from IBM to Microsoft. And Microsoft basically, like, if you look at a stock price to stock price basis, as that transition was occurring, IBM fell in half and Microsoft 10x'd. You went from mouse to multi-touch. That's when you went from Microsoft into iOS and Apple and where Apple began to dominate the landscape. It's pretty clear we're going from multi-touch to natural language, you know, primarily voice and typed. And so like a way to think about how Apple is weak is right now, If I want to control my iPad for my kid and be like, okay, this should only be available to this kid on Tuesdays for half an hour. And during that half an hour, he should be able to browse the web, but only like, you know, selected, you know, a kid safe website. And he can use either of these two applications. And if he doesn't use it, he gets to use it on Wednesday for a half an hour. And this is the half hour block at all other times. He should not be able to open this thing, except open the calculator app, right? Okay, that's like, why can't I tell Apple's device that that's what I want? And it just does the software transformation. Ryan: [29:48] You're speaking my language. I've got three kids and the way I figure out those problems, in order to understand all of the Apple iOS toggles and switches and parental controls, I actually go and I ask ChatGPT. I'm like, where do I find this in the Apple user interface? And it tells me and then I go and like navigate it. I mean, that shouldn't be happening Brett: [30:06] In this day and age. And even when you do that, But it's like impossible to. And so the reason Apple has trouble making this transition because they actually have to re-architect their operating system for this new user interface, right? They've designed and made like all of these broke additional settings menus all based on multi-touch. And you have to start from scratch. Like that's what's going to win. And so I just don't think, you know. Ryan: [30:30] Well, that's right. So this is why it sounds like it's a very exciting time. And it's like a couple of things that I'm hearing you guys say, both Kathy and Brett is like, so number one, this is an absolutely massive market, 15 to $20 trillion. It's absolutely huge. And what's also interesting is we have some strong competition here. I don't think I've ever seen a market where you have so many strong competitors all vying for that chunk of a massive prize here. And then the other thing you're saying is you're not betting on the incumbents here. You're betting on the pure play, the natives. In the crypto world, we call these the crypto native companies, right? The coin bases of the world rather than a bank trying to convert into a, you know, a quote unquote blockchain company. Cathie: [31:13] Or, Ryan, a company like Tesla understands what embodied AI is in terms of robo taxis and humanoid robots. So that's an existing company, but the strategic vision was so long term that nobody understood really who or what this company was. And can I just say one other thing? Since Brett mentioned Microsoft, now Microsoft did not get cut in half. That has been probably the biggest surprise to me over the years, that Satya has done a magnificent job, magnificent job now. Brett: [31:58] But it did spend a decade basically in the penalty box when Balmer was running it. Cathie: [32:05] That's true. But I never expected them to resurrect in this new world. Now, what is their challenge now? They're very enterprise driven, right? And there's a lot of competition there. In AI, their weak spot is consumer. And I think they know that. And half of the solution is understanding the problem. So we'll see what happens. Satya has been the real in the mobile, of course, since I lived through that. You know, it was Motorola, Ericsson, Nokia, they own the market. Of course, the smartphone opportunity was theirs. No, no, no, no. They They were not defining the market correctly. That's where the big mistake was, not understanding the problem and therefore not even thinking about a solution. Ryan: [32:57] So you guys are putting some chips on OpenAI, Anthropic, and XAI. One here that maybe Josh and I could use some understanding on is XAI and X. Actually, so there was this merger that happened, I believe, in April. X, the formerly Twitter social media platform merged with XAI. What's the AI strategy here? I mean, get into Elon Musk's head around this whole confluence. We'll talk about robotics and Tesla a little bit later, but just isolating the AI piece. What's the XAI and X thing doing? Brett: [33:31] I think you can think of X as a data platform and then a real-time execution environment for building very good AI. So if you look at, even if you look at the rate at which Reddit is monetizing its data, the value of X's data annually should be around $1.5 billion. So there's like, in integrating, they get less, so XAI had some access to that data, but it had to be an arm's length deal that like tied in distribution. And you can imagine if they want on an incremental basis, do something else, then you have to renegotiate the deal. So it makes It's more efficient to combine the entities so you're not having to face the friction of what effectively becomes an internal contract. Cathie: [34:14] We were thrilled. We were thrilled because we own both. Brett: [34:19] And X of itself is a profoundly powerful platform in terms of like, instead, it's all the world's newspapers and it's the particular newspaper for you. And so the direction of travel for them is to become more of a financial super app, being able to buy, sell and hold anything, being able to transfer resources. I think information transmission is actually not that far apart from money transmission, particularly as you digitize monies. And so that serves as like a execution environment where then XAI can have its agents operating. And so like one of the examples, they signed this deal with Polymarket. I'm sure you're familiar with Polymarket. Like the high level thesis for us of what's going to happen with prediction markets is that you're going to have, you know, more of the world is going to get financialized into more and more granular contracts that gives you a better read on kind of what's going on in the world. And the reason that hasn't happened to date is because you can't get enough liquidity into any individual contract unless you attract a lot of attention. But if you have AI agents that are operating in these markets and can do a reasonable job underwriting, then you'll get much more efficient markets. And so you'll have a much better kind of like market informed view of the state of the world, you know, on all kinds of particulars that matter to companies, Brett: [35:41] Particulars that matter to individuals, you know, which then will allow kind of risk transformation functions to take place and in a DeFi way, as in you'll be able to insure an individual house on an individual block through kind of like DeFi, because you'll be able to index back to kind of the prediction market on the fire risk for that particular area, you know, in a way that you couldn't try her. Cathie: [36:03] So, and, and, you know, at, at, at, at, X, well, now XAI, has partnered with Polymarket. And what I think we both like about it, what I love about Grok3, myself, is, you know, what was the starting point? The starting point from Elon's point of view was, you know, not censored, right? Truth, seeking the truth. And, you know, as an investor, you know, seeking the truth and watching how the election was, you know, the proof of concept, you know, the poly market got it so much closer to the mark than the traditional pollsters. And that's because there was so much of a political dynamic that was clouding those, whereas you put real money on these bets or you offer the opportunity to do that, you're going to get a better outcome. Brett: [37:07] And there's an alignment of mission where it's like X is trying to basically be the world's kind of information platform, surfacing truth. And then XAI is also trying to, you know, laser focused on surfacing truth. So you can think of it as data, distribution, execution environment, and then just like eliminating friction within kind of like these two closely partnered entities. So I think that Like if you imagine XAI is clearly in competition with OpenAI and OpenAI has its chat GPT platform where it's doing hundreds of millions of active users. Well, by XAI partnering with X, they have more unfettered access to hundreds of millions of active users. So it gets them kind of a data advantage as well. Cathie: [37:56] Can I just say something about this? because we probably should have talked about meta platforms and the drama taking place there, the scale AI and paying, you know, a hundred million dollars supposedly for researchers. They got the entire Zurich office, I think. And now they've gotten another four we hear today. And so we track and maybe we'll send you this chart so you can provide it to your audience. We track how these different models are doing over time, and we can see what's happened here with meta platforms. They've really lost their slope. Their rate of increase hasn't changed in terms of performance. And you look at Grok and there was a hockey stick in February. It got in February to where O3 Pro is now. And so we've been asking questions, why did that happen? How did that happen so quickly with Grok? It was so far behind. It started so much later. I mean, with XAI and so forth. And one of the answers we think is clustering. They're all in Memphis. So latency, And see, I don't know, Brett, if you had any more thoughts about this, but we were in one of our brainstorms. And I said, how did Grok get there so much faster than O3 Pro? This was on a certain benchmark. Cathie: [39:23] And the answer was clustering. Well, it was like, okay, I guess that's important. And maybe you want to take it from there, Brett. Brett: [39:31] Yeah. I mean, I think generally, like I said, with Elon, it's like we need 100,000 GPUs all in the same spot to most efficiently train this thing. Therefore, I will build it and motivate the power resources and to get the data center spun up in four months as opposed to the 18 to 24 that third party providers were promising. Brett: [39:49] And when like if you imagine the difference between starting in four months versus 18 to 24, when you have the cost decline you see in AI, it's in old school tech. It's like the difference between starting in two years versus 10. You know, so there's like a, there's a real, real kind of velocity urgency in the AI space. I'd also say like, so Kathy mentioned Tesla and like, so you focused on kind of the direct language models exposures we have in the venture portfolio, but also within like our ARCW portfolio, which is an AI focused portfolio in public equities. It's like Palantir is an interesting position where it's a platform as a service company. So you can imagine that. One thing I like to say is the cost to learn software is basically still flat. Like if I have to spin up on a new enterprise, I don't know, employee management software, it's like, oh gosh, I have to figure out how this user interface works. And, you know, and it's staying flat, but the cost to write your own software is falling like a knife. And so for a lot of traditional software as a service applications, it's literally like more costly for me to learn it than it is to just write the feature I want and have it happen. And so, yeah. Josh: [41:03] Well, a lot of this happens in a box. So what I want to move to next is kind of AI out of a box, the physical manifestation of this AI. And you both have mentioned it a little bit earlier, but that kind of comes in the earlier form. We're seeing this in robo taxis, which are basically robots, but with four wheels and in the form factor of a car. And what I'd like for you to do is kind of just describe the landscape for the people and the opportunity? Because I was talking to a lot of friends prior to recording the show, and they're actually not even aware that Tesla has a robo-taxi network that was rolled out last week. And they made their first delivery without a driver sitting in the driver's seat, right from the factory to the delivery. So could you just explain to us not only the landscape, but the opportunity of this physical manifestation of AI through the form of a robo-taxi? Cathie: [41:43] So can I just set it up much more broadly? You're talking about- Please do, yeah. Yes, yes. So embodied AI, robotaxis, yes, humanoid robots. So Sam Khoris and Daniel McGuire on our team. And we should always give credit to our analysts on the AI front. That's Frank Downing, Joseph Soja. And well, I guess everyone is participating in this, of course. Brett Winton, like obsessed with AI for so long, but also Charlie Roberts, who heads up our venture effort and founded a unicorn, which is taking advantage of the convergence between sequencing technologies and AI to diagnose cancer with a blood test. Cathie: [42:33] In stage one or before stage one, if it's colorectal cancer and polyps are shedding into the blood. The biggest AI project on earth near term is the Robotaxi project. The most profound is in healthcare, curing disease, cutting the time to discover new drugs and get them to market, you know, perhaps in half or over the next five to 10 years. And as I mentioned, diagnosing disease in its earliest stages when we can do something about it, or curing disease, which is the convergence, sequencing technologies, Cathie: [43:18] Artificial intelligence, and CRISPR gene editing. And just to put a fine point on the kind of data project we're dealing with in healthcare, each one of our genomes has between 35 and 40 trillion cells. Think about that. And now there's something called single cell sequencing. So we can sequence the 35 to 40 trillion cells in each of our body and get a lot of data out to inform healthcare decisions. That's the ultimate data project and the most profound application of AI we believe. Brett: [44:01] Yeah, and I described it a little earlier, but I do think it's like, yes, we talk about language model space, but then the rate of advance here also applies in other areas. And in some of those other areas, like Robotaxi is an example, People don't yet, they haven't, because they haven't like seen the large scale rollout yet, which you have seen with language models. These are undoubtedly, like it's still a shockingly small number of people are using language models, but it's very clear that, hey, we're going to go from 10% penetration to, or 5%, I guess, to 75 over the course of a few years here. RoboTaxi, I think you'll have a similar or, you know, similarly, at least dramatic uptake trajectory where the high level way to think about it, if you buy a new car in the US, you are spending, you're basically buying a bundle of miles and it's for a dollar plus per mile that you're buying that bundle of miles. Basically, you're buying the asset to have the option value to travel for a dollar plus per mile. We think Tesla's CyberCab will be able to, you know, deliver miles at less than 50 cents a mile where you won't even have to drive. Brett: [45:12] And like even the Model Y and Model 3 that they're rolling out on now, you know, could deliver very profitable to Tesla miles at a dollar per mile or less. So you'll either be like, I'm going to buy a new car that gets me miles at more than a dollar per mile, or I can be driven around all the time for a dollar or less. Like who would choose to buy the new car? You have to be a very niche, user, like where you have esoteric requirements, like, oh, I need to haul around stuff in the back of my truck. Well, then maybe you still buy the truck. But for most people, kind of like their default mode of getting from point to point will transform. And so if you look at the number of miles driven in the US or globally, like it's trillions and, you know, a Tesla that was once sold to an end user for, you know, call it $5,000 in operating profit or so can now do potentially 100,000 miles a year at a dollar per mile on some platform fee to Tesla, which nets to, you know, tens of thousands in potential operating earnings per vehicle in fleet for as long as the vehicle lasts. So it transforms the business model of Tesla. I'm most excited because it transforms my life. I live in LA. Brett: [46:27] Getting from place to place in LA is a huge pain. I'm lucky I have a Tesla. I use FSC all the time, but I still have to pay attention to it, paying attention to the road. I'd rather be able to just work and have the option of going from place to place inexpensively at the same time and have the option of sending my kids off somewhere Brett: [46:47] without having to drive. Cathie: [46:48] I remember Brett saying, as he was having children himself, mine had already passed this stage, He said, you know, this is one of the most important things that can happen for my children, because teenagers, when you put them in a car for the first time, we end up with weapons of mass destruction of all kinds, right? I mean, if you think about how you first, I mean, even myself, you know, I was... I was very careful, but that was a problem. You know, I was a menace to other drivers on the road. So, you know, I think from, as you say, it saves you time and you don't have to be a chauffeur and that's great. But it's also going to save 40,000 lives in the United States, maybe 1.2 million lives globally, because more than 80% of the accidents and fatalities out there in the road are caused by human error. And we're just now getting to the point where both Waymo and Tesla, we think Tesla will get there first, but it's a race, Cathie: [47:58] Will be able to show that their robo-taxis are safer than human drivers. Human drivers in the United States have an accident once every 700,000 miles. And both Waymo and Tesla are close to that point right now. We think they'll certainly Tesla will cross over this fall and maybe in a year or two will be two to three to four times safer than human drivers. And then there's no doubt, there will be no doubt in regulators mind what the right thing to do is. Brett: [48:34] And another point I'd make here is that if you look at technological history, Speaker0: [48:40] The robo-taxi, we think. Brett: [48:42] Is going to be more productive on a productivity delivery to the economy basis than the steam engine was because of how quickly it can concentrate in terms of its uptake and because everybody is wasting their precious, beautiful minds doing bad amateur driving right now. Like, we shouldn't – what a waste of my mind that I have to drive. What a waste of your mind. You know, not only are we not great at it. I mean, we're relatively good, but we kill a lot of people. but we could be doing other stuff. And so like, you know, the dishwasher got invented and got taken up by households, which then freed up a bunch of unpaid labor, mostly housewives who are like scrubbing clothes to do other stuff. And so actually many of them entered the workforce. And so here we have an hour commute block a day that we're not getting paid for, but it's clearly labor that we're performing. And so it doesn't get recognized in the GDP statistics because it's amateur hour, right? But once you're paying a robo-taxi provider to do it on your behalf, then it gets recognized in as actually production in the economy. And then you free up that time to, you know, hopefully I'm doing work, but maybe I'm binge-watching things on Netflix to do other stuff that's economically productive. Listen to the podcast, Cathie: [50:01] For example, which you should be on a doing it. Of course. Brett: [50:04] Well, no, you can watch the podcast now. You don't just have to listen. You can watch. Cathie: [50:08] I just want to add an exclamation point here to something that Brett just said. We believe that real GDP growth is going to accelerate in the years ahead. We think in the next five years to 7%. Now, if you're an economist, you think that's absolutely crazy. It hasn't happened in anyone's lifetime that we've seen that kind of secular acceleration. But if you and this is a study that Brett did, if you look at what happened around the turn of the last century. Cathie: [50:44] Telephone, electricity, internal combustion engine. We went from roughly a half a percent per year of real GDP growth to 3% for the last 125 years. That's a five-fold increase. So three-ish percent, this is a global real GDP growth number, to seven. We have five innovation platforms evolving with AI turbocharging, all of them, robotics, energy storage, AI, blockchain technology, and multi-omic sequencing. So we should see a much more rapid acceleration this time around than last time. We're being conservative when we say 7%. But the other thing to note is that all of these technologies, as we've been talking about, are very deflationary, highly deflationary. And so we could, the trade-off for the 7% real GDP growth is a healthy one, falling prices, increasing purchasing power for consumers and businesses. So it's going to be a pretty wild, we think, and wonderful world. Josh: [51:56] And that's the trend we're kind of seeing, right? With the cost declining for each category. So with AI that we started with, cost per token is declining. Now Brett was talking about cost per mile, cost per kilometer. Then we're going to get into SpaceX soon, where it's cost per kilogram. And there's this general trend towards decreasing costs per whatever it is that we're measuring. In the case of robo-taxis, I think that cost per mile is something that's really interesting because I'd love to compare that to Waymo. A lot of the conversation was around Tesla. And Tesla has a very low cost per mile, but also isn't Waymo kind of doing the same thing? They have no drivers in the car. They're actively being rolled out in cities like Texas and California. What is the difference and how would you compare the two when you're viewing the two, really the biggest competitors in the RoboTaxi network right now? Cathie: [52:37] What's interesting is Waymo commercialized, started commercializing in 2018. Tesla went about it differently with FSD. And here we are now, they've launched, which is great. And Brett, why don't you talk about the cost advantages, which this is the key question. Brett: [52:59] I mean, I think that, one, I love Waymo. It's a great product. It costs roughly as much as an Uber today, and you have to wait a little longer. But people still take it because one thing about Robotaxi, it's going to be like spot on the average. Whereas you get an Uber driver, you might get a good Uber driver, but you also might get a bad Uber driver. And so you have a lot of vol in terms of your experience. Like, I don't know if you've ever been in a smelly Ubermobile, but it's not fun. And so, or somebody who's driving super erratically. Brett: [53:28] The issue Waymo has is they committed very early on, even before deep learning came into play to a very hefty sensor stack, which then raises the cost of the computations they have to do and raises the cost of manufacturing and servicing the vehicle. So it looks like the cost of Waymo's sensors alone is roughly as much as Tesla is going to need to spend to manufacture the entire CyberCab. So on the one hand, you have a bundle of sensors that you still have to attach to and manufacture a vehicle to support. On the other hand, you have a fully functioning robo-taxi, same cost. The issue is that that is the most sensitive factor to the cost per mile you can deliver to your end customer. Because the cost of your manufacturer correlates with the cost of maintenance. It also means you're hypersensitive to utilization of the network. So the other issue Waymo has is basically there are way too many Waymos at night and not nearly enough during the middle of the day because they have to strike this middle ground where they have to have the assets sitting around. Whereas compared to Uber where there are some people who Uber drive full-time, like they Uber drive all the time, they provide the baseload. There's some people who are like, I can't be bothered to try to do it unless they're offering me a higher rate. And so they get pulled in with their vehicles. Brett: [54:53] Tesla's advantage of having a mass manufactured set of vehicles that can all operate as robo-taxis is they can have their own owned and operated baseload vehicles that go all the time. And then they can literally could send to the screens of people that own Teslas. It's like, hey, do you want to send your vehicle out as a robo-taxi now for $120 for three hours. You know? And so they'll be able to demand-match or supply-match the demand curve for transportation. And so it Speaker0: [55:21] I think Waymo, one. Brett: [55:23] Has cut through the regulatory thicket to some degree, but their tech has built up over a period of time, which makes it difficult for them to scale. And then Tesla, assuming they can cross the threshold of commercialization, which they are just now doing with their sensor and technology stack, should be able to scale very rapidly. To give you a precise data point, Waymo is expected to produce 2,000 additional robo-taxis this year because they have to buy these Jaguar vehicles and then they have to fit all the sensors on and they use a company called Magna to do that. And next year, they might get to 4,000. That's their plan. Tesla produced 2,000 vehicles today before lunch. Cathie: [56:04] That's amazing. Yeah, yeah. There's another thing. I often say Tesla has had six or seven million robots roaming around the world collecting data. Again, we're back to the data conversation and pushing it back. Waymo, I mean, so that's orders of magnitude more data than everybody else combined. So I think that's a very, very powerful. Brett: [56:30] Yes. And so then also your rate, your ability to improve in market is conditioned on the number of assets you have that are operating in that market. So Tesla should also be able to have, not only should they be able to launch, they're launching in Austin now, but any market they enter, they should be able to ramp up in terms of capability. And you could think of capability as like number of places you can go, like how big is the map that you can address? You could think of capability as, can you go on highway? Waymo still doesn't commercially go on highways. Speaker0: [57:01] You could think of capability as. Brett: [57:02] Like, how fast can you get from place to place? Like, can I get into Tesla and be like, follow that car? Or, you know, I'm in a hurry. Like, you know, and you can think of capability of, like, the vehicles being able to communicate to each other and beginning to platoon on highways and do things that allow them to generate better throughput. And then you can also think of like the broader set of Elon enterprises and think about these private tunnels he's digging under cities and say, hey, I really need to like go from the valley to the west side in LA without going over the 405 overpass. And he might dig a tunnel that the Tesla cybercabs can go through that Waymo won't have access to. So, yeah, I think there's a yeah. I was I Josh: [57:43] Was thinking about that, too, is that the key advantages that Tesla has that Waymo doesn't is that Tesla is much more than just a car company. It's much more than an autonomous company. And what was a helpful reclassification for me was redefining the car as a robot. I think if you think of the car as a robot instead of a car, it kind of helps you come to conclusions that will kind of help people see the next product. So in this case, I'm talking about humanoid robotics. Humanoid robotics require a lot of the same sensors that these cars use. They have actuators and motors and batteries. And what I'd love for you guys to describe is kind of like the humanoid opportunity, because that's clearly another big avenue that Tesla's pursuing. They have the Optimus program. And that should give them a big advantage, right? Is having this RoboTaxi network, having the data that they're collecting, and kind of retrofitting it and applying it to this new form factor of robots. Speaker0: [58:30] Yeah. Cathie: [58:30] So it's the convergence of the same three technologies, effectively, right? robotics, obviously, energy storage, and artificial intelligence. So, you know, Elon says that he believes Tesla will be number one in humanoid robots around the world. And then number two through 10, he thinks, are going to be in China, from China, because they are so focused on and from a demographic point of view, they need to be. But also they have this government-driven initiative, you know, new productive forces. So, and again, Sam Khoris and Daniel have done a huge amount of work on humanoid robots that you, I know, Brett, have helped with this. Brett: [59:19] With humanoid robots, that's the technology that has actually moved Furthest to the left as a result of the acceleration in AI for us, we really thought it was an early 2030s type commercialization event that because of the rate pace change in AI has been pulled forward where now it looks like late 2020s is when you'll have these things useful enough to meaningfully commercialize. And right now they are not useful enough to meaningfully commercialize. Like they trundle around in factories, they pick up things. They're kind of like demonstrations of potential capability. Brett: [59:53] And, you know, as I described, like the acceleration in language model space is translating to the acceleration in robo-taxis. And it also translates at some delay to the acceleration in robots. At a very high level, humanoid robot we think is around 10,000x harder than a robo-taxi. So like there's some that go it's wildly hard like think about like think about how easy it is to drive you just have to go left and right and then like a brake or accelerator those are your only three options like the sensors are at the same spot in the vehicle every time with a robot you have all of these independently operating degrees of freedom plus when you move your body like your sensor set shifts and you're operating in an environment that's not designed necessarily to instruct you what to do. You don't have like lane changing markings in a hallway, you know? So there's like, it's a wildly unstructured, very, very challenging problem. And advantage in robots is that you are more error tolerant. If you get in an accident, that's a big deal in a car. If you bump your shin, I mean, you don't like it, but it doesn't, you know, cause death and destruction. And so we think that it's like 10,000 times more Brett: [1:01:02] Difficult, but maybe you're a hundred times more error tolerant and net it ends up in this spot where we think kind of real commercialization can happen in the late 2020s, given our expectation for the rate of pace of change in underlying AI. And for Tesla specifically, yes, they have the same chip in the robot that they have in their vehicles. That's an advantage. The real world data they get off their vehicles and the data center they're using to train their vehicles, they'll be able to shift into kind of the robotic space. So everything they're learning there translates. And so, you know, I think it's, they are, they have the best set of ingredients to win in this market. And it's still, and also they have the internal demand as in they can turn the robots on in their factories to like do stuff. But net, I would say it's kind of clearly a multi-trillion dollar potential opportunity where we're still very early on knowing exactly when these things will be capable enough that like you or I would want to buy them into the household. But I do think you could imagine the way in which the future is going to shift Speaker0: [1:02:11] Previously U.S. Brett: [1:02:13] Households devoted, call it $30,000 to a vehicle that they only used 5% of the time to drive them around. Well, you don't need that vehicle anymore because a robo-taxi will drive you around. So you have extra capital. What are you going to put it into? You'll put it into like a humanoid robot that'll kind of do the dishes for you. It'll fold most of the laundry, but at first it won't be able to fold all of it, but it'll still be better from a convenience perspective. It'll walk your dog. You no longer have to pay for dog sitting. Like it'll fill in kind of some of these gaps, but imperfectly, and you'll be satisfied with it because it'll save you, you know, half an hour or an hour of housework a day, which is great ROI from, you know, taking that capital and devoting it to that thing. And once it's in household, it'll get better over time, it thinks. And so I think that's where, you know, it really, this cycle relative to other cycles is very different in that way where you, you know, even iPhones today, it's like they upgrade the operating system, your phone gets worse. Cathie: [1:03:09] Yeah. Brett: [1:03:10] You know, and so it's like the here, the the software upgrade can backward apply into the hardware for much longer and deliver, you know, big performance advantages rather than kind of like re-skinning of previously available capabilities. Cathie: [1:03:25] We should size both of these markets. So the robo-taxi market, globally, the entire ecosystem, we expect within the next five to 10 years to be an $8 to $10 trillion opportunity with half of that roughly going to the network providers like Tesla or Waymo. We really do think Tesla is in the pole position. And then the humanoid robot we size as a $26 trillion opportunity in the next 5 to 15 years. And to put that in perspective for people, the global GDP today, so just one year's output of goods and services, is about $130 billion. Speaker0: [1:04:14] Trillion. Cathie: [1:04:15] Trillions. Trillions. Oh, I'm sorry. Trillions, trillions, of course. Trillions. Yes, $130 trillion. So these are needle moving. They're needle moving. That's good. Ryan: [1:04:27] That's making me feel a lot better about the deficit situation we got ourselves into. Maybe we can grow our way out, right? Thankfully. Cathie: [1:04:33] This is a argument we have all the time. And you'll notice that the bond market has been rallying. Yeah. The inflation is lower. Even though the deficit and government spending is coming down, but the deficit as a share of GDP, people are saying this is unsustainable. If real growth, because I lived through the Reaganomics years as well, that was the secret. Growth, productivity, and growth accelerated, and we ended up with a surplus in the 90s. So we absolutely are not freaking out over the deficit. Ryan: [1:05:10] Well, let's talk about this then. So we're just talking about humanoid robots, and that's already starting to sound a little like Jetsons, you know, like sci-fi era. But I want to talk about something that's even more sci-fi, which is like brain chips. So we've got Neuralink computer interfaces. Prior to this episode, Josh was just showing me someone playing Call of Duty with their brain using Neuralink technology. And according to Josh, he's got a clip from, I believe it was Elon Musk, right, Josh? Yeah. Josh: [1:05:38] It was pretty amazing. Elon was just kind of describing. So in the natural extension of this conversation, Elon was describing for disabled people that Neuralink could actually allow you to embody a physical humanoid robot through the digital connection of the brain machine interface. And that was something that totally blew my mind. That's crazy. Brian and I were snooping around the website. We were like, of course, Josh: [1:05:57] ARK has a position in Neuralink, just invest in the Series E. So clearly you have some thoughts about this. What does the world look like? Are we going to be able to telepathically embody these humanoid robots that we're now building? Is that a thing? Cathie: [1:06:07] You know, because we had the opportunity, I will say it was a privilege to, you know, listen to the entire team talking. I mean, they took us, they took, well, you probably, Brett, had already been there, but they took me to places that was like, oh my gosh, I'm going to do this. I'm going to do this before the end of my life. I think I would love to do it. So maybe, maybe Brett, you want to launch into this. Brett: [1:06:37] Yeah. I mean, I think, you know, Neuralink is a really One, it's a profoundly human-positive company, as in, you know, you were taking, there are, I think it's upward of a million people that have profound kind of paralysis and communications issues that just don't have an easy mechanism by which to communicate today. Right now, they're installing chips that go in the kind of hand motor cortex. So you can imagine the experience of the user is to, they can imagine handwriting, like, or not, imagine's the wrong word. They can try to handwrite, but their hand won't move, but then it will move for them. Or they can try to point with their hand, their hand won't move, but they'll be able to move the cursor on the screen. And so then for the Call of Duty one, it's basically move thumb for joystick and wrist for the other joystick is what they have to cast themselves into doing. And then the chip that's installed can successfully interpret those signals to provide the output that they intend. and the technology is improving over time. And so they first installed this technology on a profoundly paralyzed patient. And in, I think it was the first day he broke the brain-computer interface word output, you know, or bit output rate that had been set across all the previous kind of generations of trying to do this. Brett: [1:07:58] And they're going to go, you know, right now it's just in the hand cortex, but you can imagine they could get into the vocal kind of like cortex so that instead of imagining handwriting, Brett: [1:08:09] You can, and imagine is the wrong word again, but instead of trying to handwrite and being unable to, you can try to speak. And it could take kind of the signals coming off of trying to speak and then actually produce the speech for you on screen. And so people who are basically limited in their ability to communicate with the world because of kind of input-output limitations will get or should get unlocked again by this technology. And then the next direction they're moving is to and to be able to not just read output but to write to the brain and so within you know cochlear implants are a common operation for people that are deaf where you can provide kind of like basically direct signal into the auditory kind of part of the brain to to to mimic what would happen with an eardrum well with blind people there's no comparable technology well they're going to try to write to people's kind of like optic center of the brain and and so you're But imagine the person they can't see then can potentially have the experience of seeing. Not only that, they could see it in whatever spectrum they want. They could see in infrared if they want. They can, you know, attach a camera and do anything. And so you can, in some cases, for instance, on the input-output basis, you could have superhuman type capability where you actually can communicate at the speed of thought with AI agent. Whereas, you know, us mortals are restricted to our ability to type or our ability to speak. And so that's their design is towards making this Brett: [1:09:38] As kind of economical and easy to undergo as like a LASIK type procedure, where in the future, people might opt in and say, hey, listen, listen, I'm going to go to the Neuralink Center, and I'm going to, you know, upgrade my memory. Josh: [1:09:52] And Brett, to your earlier point, right? Like, we were talking about how, like, we were basically trying to increase the throughput of input and output. And it very much feels like Neuralink is kind of that final frontier, right? Where it's, we are merging with the AI, we are embodying the AI. And I'd love to hang on this, but there is one headline that I really wanted to get a chance to talk to you about, which was your SpaceX valuation. And I read the headline, my friends and I love SpaceX, and it said, by the year 2030, about $2.5 trillion in enterprise value. So that's a huge number. That puts SpaceX at one of the most valuable companies Josh: [1:10:25] in the world in only five years time. And when people look at SpaceX, the Starship program is blowing up. They're quickly iterating, but it hasn't quite worked yet. So I'd love for you to just walk me through that valuation. That is a huge number and why you think SpaceX is going to be so successful. Cathie: [1:10:38] So, you know, just so your viewers know, we open sourced it. You can go find it and see exactly the variables that are making this forecast work. Josh: [1:10:51] It's a great report. We'll have a link in the description for everyone to check out. Yeah. Cathie: [1:10:55] And Brett worked with Sam and Daniel and Mach 33, which was our partner in doing the model. Brett: [1:11:04] With SpaceX, you know, Starlink is such a unique asset. So yes, the Starship is important to that valuation, as in they are, you know, upgrading the size of their rocket so that they can loft stuff into orbit for, you know, on the orders of hundreds or maybe even as low as $100 per kilogram, perhaps even lower than that, as opposed to thousands of dollars per kilogram. So that just means you can send up more satellites and then you can afford to charge kind of a lower cost for access to those satellites or send more bandwidth against those satellites for the same cost, which is the direction we think things are going to go. And, you know, the net result is you have a global connectivity player, whereas prior, you didn't have a global. Just like Netflix went from, hey, we're streaming over the top and people are like, oh, that's a funny duck. You know, actually broadcast is tied to the terrestrial cable nets that deliver the content. Well, you know, Brett: [1:12:01] Think about internet ISP provision. It's really limited to like these local domestic cellular providers in every geography. And Starlink is actually a global credible competitor to that. And so if you look at the overall spend on communication through cellular and broadband, I think it's on the order of $1.3 trillion. It's definitely more than a trillion dollars. And so we think that Starlink can get to the $200 and going towards $300 billion in revenue because it's providing basically like this competitive set of communication equipment to people. Now, that feeds into them being able to launch more satellites, more satellites, and then they take all the excess cash flow and they throw it at Mars. So as an investor, you have to be comfortable with the fact that you're not investing for the cash to be returned to you in a distribution that you're going to get your distribution check and take it off to the bank and be like, I'm going to cash this. No, you're getting a share of a future Mars colony that they're going to build. And so- And humanoid robots. Brett: [1:13:04] Yeah. Yeah. The first thing they're going to ship is cyber trucks and humanoid robots. And so the way we model it is the incremental cash flow goes into building out the constellation and continuing to build it as long as it's sufficient to do so, and then beginning to throw more and more basically up mass to Mars and beginning to build infrastructure on Mars. And so that doesn't meaningfully impact your valuation consideration in 2030. It begins to in 2040, like what is the value of Mars? But it is the big caveat that you should not be investing in SpaceX if you're wanting the cash back. You need to want to have a piece of what's going on in Mars for it to work because that's the mission of the company. And we think it's like a profound and exciting and interesting mission. And the reason SpaceX can be so successful on Earth is because they are designing for Mars. So their technology is like literally extraterrestrial and they're coming back and they're competing with terrestrial players. And so they have this huge performance advantage. Like you would never design the Starship rocket as big as you would, except that you need to get it to Mars. You would never land the rockets on their butt, except you have to need to land on Mars. But as it turns out, that engineering badassery is really important to kind of winning on Earth, And which is why they're dominating the space on earth right now. Cathie: [1:14:28] Yeah. 90% plus of all of the satellites. And Elon often says, and just if you look back to the early days of venture, Defense really created Silicon Valley, right? Now that world is flipped. It's private, public, but we see SpaceX, as Brett said, in going to Mars is going to bring back so many enhancements. We can't even imagine them, whether it's material science or whatever, back to Earth, that really that's where we get our dividends, maybe in 2050 or 2060, whatever. Brett: [1:15:14] Yes. No, I mean, in the future, it should be cheaper to service the Earth satellite market from Mars than from Earth, because you spend less energy to get back to Earth from Mars than you do to get from the surface of Earth to low Earth orbit. And so the Mars base actually gives them a competitive advantage back on Earth. And then over time, you know, if there's going to be asteroid mining and anything that we're like extracting resources from the inner solar system, kind of Mars should become the industrial base for kind of that sort of transformation. And so there is option value in that Mars colony. And like from the perspective of Tesla, they're going to ship robots up. But you'll learn a lot about how your robots perform when they're trying to perform on Mars. And you'll make them really robust to kind of like doing Earth moving if you're testing them out on Mars, which then can be useful back on Earth as well. Ryan: [1:16:07] Yeah, you got to wonder what the money the robots are using on Mars, which brings us maybe to the last topic, which is cryptocurrency. And Kathy, this has been another massive change. I think the last time we brought you on the podcast was December 2023, and we're on the cusp of releasing Bitcoin ETFs. We are in a completely different world right now in 2025, the summer of 2025. And there's many things you could comment on, I'm sure. You guys have a Bitcoin price prediction for 2030, $1.5 million per Bitcoin. Ryan: [1:16:41] Bull case. Yeah, bull case. We've also seen massive IPO market with Circle, their IPO, and this is on the back of stablecoins. I think it's been like a 6 to 8x since IPO, which is absolutely staggering for Circle. We have Coinbase being inducted in the S&P. We have a completely different regulatory regime. We have just this morning Robinhood announcing a Ethereum layer 2 and a tokenization strategy, bringing stocks on chain. All of these things, I know you guys have a full fintech, digital assets, cryptocurrency type of innovation strategy. But maybe I'll throw this one to you first, Kathy. What do you see when you look at crypto right now? What's the most exciting areas for you? Oh, my goodness. Cathie: [1:17:22] Well, you know, we're very proud that we were the cornerstone investor in Circle. And I heard Tom Lee describe the Circle IPO as the chat GPT moment. And I understand it because now, whereas before we were saying to institutional investors in particular, you know, you may not like this thing called crypto or digital assets, but it's a new asset class. And you have to have a point of view. If you don't like it, you need to have reasons why you are not including this new asset class in your asset allocation. And so we were kind of spitting in the wind, to be honest, not until the Bitcoin ETFs came out. We got a little more of an audience around that, but you still had the SEC kicking and screaming even around the ETFs. And then you do get a new administration, a complete change from a regulatory point of view. Cathie: [1:18:29] And yes, circle going public. And I think if you ask, when people ask me, what's the biggest surprise, I thought, and you know from probably our last podcast, I really thought that Bitcoin would serve the role that stable coins are going to serve right now. From the point of view of emerging markets in particular, their tether is more the stablecoin backed in large part by treasuries. So I think that's the biggest surprise to me. But I think because of that IPO, I don't know if it was the safety that, you know, being backed 100% by Treasury securities conferred upon it. But finally, finally, we have institutions studying this and studying it hard because you cannot miss, you know, the equivalent of AI and AI and crypto together as we go towards agentic. Also with DeFi threatening all of them as well. They better get with the program. So you're right. Light switch. It's almost like a light switch has gone on here. And it's thrilling. And I think... Robinhood launching its tokenized strategy, Cathie: [1:19:53] Launching its own layer two using the Arbitron, launching perpetual futures in Europe, in the UK, Cathie: [1:20:04] And rolling out staking or re-rolling out staking. I mean, it's going for it. We think this could become one of the most important financial services institutions in the world. And, you know, when we first put Robinhood in our portfolio, we were accused of investing in meme stocks. And you'll notice we did not invest in any of those other meme stocks because they weren't worthy. There's nothing about technologically enabled innovation about them, but there was with Robinhood. And I will say, in our early days with Robinhood, we were a little impatient that they weren't going hard into crypto. I think what held them back was all of the regulatory trouble they got into, and they had to play it safe. But they have taken off the gloves now and it's going to be extremely exciting. I think it's going to accelerate the crypto movement in more of the traditional financial world much faster now. I know all the banks are trying to figure out, you know, how to participate, lower their costs, whatever, however, or, you know, how to cannibalize themselves so they're not disintermediated. So it's pretty exciting. Ryan: [1:21:23] Brett, how do you think, putting your analyst hat on, do you have a theory for how stablecoins are going to scale? Let's say a post-Genius Act. Let's say that moves forward, that gets signed by the president. Then what happens? Brett: [1:21:35] Well, I think it'll be helpful to catch this in how we model the DeFi space as a whole. So this is how, one, I think there are two financial revolutions happening simultaneously, maybe three. But one is kind of cryptocurrency, which is direct displacement of kind of like traditional currencies with digital assets in some way. With Bitcoin as the, you know, maybe it's the digital gold, like really, maybe over time it also becomes really more than stored value. It becomes medium of exchange and unit of account. But stablecoins as another kind of like thing that's accelerating the cryptocurrency phenomenon. And then there's, you know, smart contracting and smart contracting protocol calls, which is really going right up against the traditional functions that financial institutions provide to people and doing so in less expensive and more efficient and quicker way. And so the, you know, there's like a functional theory of finance where it's like you don't pay attention to like what the regulated institutions are allowed to do. You pay attention to what people need. People need a way to like store their money. People need a way to transform money through time and space, like borrow it from the future or like put it in place to lend it to other people. People need to like kind of socialize risk through insurance contracting spaces. And DeFi is going to like Brett: [1:22:57] Applications are going to be composed upon DeFi to enable all of that stuff, right? And if you look at on a global basis at what does the traditional financial services sector provide to the economy, it provided some amorphous set of risk transformations and store value functions and all of these things and, you know, transaction facilitation. But it charges us. It charges us on relative to the number of financial assets in the economy, we think 3.4%. Brett: [1:23:27] Well, that sounds like a small number, except by 2030, there's going to be a quadrillion dollars of financial assets. We're paying a huge amount for bank real estate and bank lawyers, right? And so the promise of DeFi as you tokenize assets is we think that the payment out into the underlying protocols and applications will be around 1%. And you can argue, you know, full blended stack. And so you go from playing 3.3% to rent Morgan Stanley's balance sheet and ask their structured note guy to write the transaction contract for you so that you can hedge some risk to, Brett: [1:24:02] Listen, you can pick a contract off the shelf. You know it's going to execute because you have a blockchain insurance execution on the back end. So you don't need Morgan Stanley's balance sheet and it net costs you one, right? And so then that means kind of, and you can do it at any scale. So you don't have to be a hundred million dollar business to catch the attention. You can just like lift it off the shelf and do it yourself. And so it provides basically better financial services to everybody in the world because of that. Okay, so then how do we model it? We say, well, what's the adoption rate? And the adoption rate is mimicked off of like early internet adoption during dial-up days. So less than 5% penetration by 2030. You do all the math on the yield that accrues to the underlying tokens. You end up with a $5 trillion in market cap estimate for kind of like the DeFi stack. Now, I think the introduction of stablecoins and the acceptance of stablecoins here could actually be the equivalent of the dial-up to broadband transition in internet. Brett: [1:25:01] Bitcoin is like dial-up entry point into kind of like DeFi space. It's hard for people to understand. It's actually on a per-transaction basis. It's expensive, but you can move large volumes of money. It's kind of like, it's generally clunky. And DeFi still is generally clunky. I think average person approaches this stuff and is like, oh my gosh, I'm going to accidentally lose all my money and I don't even know what I'm doing. And so there's clearly lots of kind of like development needs to happen on top of it. Now you have extraordinarily well-capitalized enterprises that are incentivized to get people onto digital assets and to get them in an asset they already hold a lot of. They already hold dollars. You're just going to have a dollar that you can instantly transfer and that you can do all of these things with. And so I think it'll serve as an accelerant across the DeFi stack where people are going to realize, why am I accepting this lower yield on my savings when I could get a better yield? Why am I accepting this exorbitant rate for this bank, for this personal loan, when I can actually kind of like borrow in this way more efficiently against the assets that I already have? And so I think that the, it's, you know, Brett: [1:26:14] I think the Genius Act and all the traditional financial institutions that are going to try to partner with Circle and try to roll their own. And, you know, it'll provide this onboarding for that average, like, you know, normal person set of users into a bunch of these, like the starting point to actually exploiting these applications. And so that, like, if anything, my bias would be, eh, we need to accelerate the adoption curve some here, given that we've crossed this threshold, that there is this chat GPT moment in kind of stable coins, that people are like, one, they're looking at Circle's stock price and like, whoa, maybe I should hold some, yeah, yeah. And then two, it gives Circle like instant credibility to sign on with a partner. Like they can go to the banks and be like, hey, we can white label this thing for you. And the banks are like, oh yes, look at their stock price. They're doing well. we want some of that juice, you know? Ryan: [1:27:03] So I think it'll be- Yeah, well said. And back, Brett, to the convergence point that we opened this podcast, what do we think the AI agents and the robots, do we think that they're going to want the future of programmability in their money? I think so. And that's gotta be digital programmability. Kathy, maybe we'll end the episode with this. And this has been a fantastic through all of kind of the innovation S-curves that you guys are seeing. It's been a fantastic overview of all of these things. The one monkey wrench that could gum up the works with 7% GDP. Maybe there are several, but one of them is certainly macro setup, right? So we've got tariffs. We've got the potential of a new Fed share. There's just generally some uncertainty. It has felt like a shaky year from a macro perspective. We've got wars, lots going on here. What's your outlook for macro as we sit in July 2025? Cathie: [1:27:53] Well, the market has been climbing this wall of worry, which should tell us something. And I think, and this has been our assumption, happy to see that it's starting to play out this way. We believe that President Trump, and believe me, I understand how chaotic and how crazy this environment has seemed to everyone. I've been just throughout Europe and Asia, and they're like, what is going on? I said, okay, yeah, we agree. It seems chaotic. I think that the methodology here is that President Trump decided to take all the hard medicine and do the dirty work. Very early on in his administration, he will define himself as, and remember, he wants to be considered the greatest president who ever lived. Well, tariffs and wars and a bad economy don't get you there. Taking all the bad medicine, I actually think, and from a timing point of view and given our point of view on innovation... Cathie: [1:29:04] What will happen here is we're going to move as we go through the second half of the year into next year into a massive productivity driven recovery and then expansion, which will be accompanied by much lower inflation than anyone expected. And it's not going to come just from all of these deflationary technologies, but home prices are starting to fall. There's a huge inventory glut that builders are facing, and they got to clear that. So I think inflation will come down. We know oil prices probably going to come down as well. So if this sounds a little bit like Goldilocks, it is. But why did he do the dirty work now? It's because of the midterms next year. And that campaign season is probably going to start sooner than normal this fall into winter. And he needs to extend the Republican lead in both houses, right? And in both the House and the Senate. Otherwise he'll be a complete lame duck. And he has a lot more to do in terms of lowering regulations, Cathie: [1:30:19] Getting the government out of our hair and really allowing innovation to take us to a place that will benefit everybody. And I might sound like an evangelist on this, and I am not making this political, but this is because, believe me, I was scratching my head saying, what the heck are you doing? Because tariffs, tax increases, that's what they are to me. And I don't like those. Cathie: [1:30:46] But I think at the end of the day, we're going to have much fairer trade at the end of the day across the board, because the worst part of trade for us was the non-tariff trade barriers that we faced. And a lot of those are coming down. I can't wait to see a paper written about that. And so to be honest, I'm very, well, not be honest to, as I've just said, I'm very optimistic about what's going to happen. If you ask me what's my biggest concern, it is the commoditization coming out of China and how that is going to make its way through the system. Of course, it'll mean much lower inflation, but for many companies, it could mean much lower profits too. So really trying to understand that dynamic if I'm being realistic about the risks out there as well. But I think we'll look back and this chaos will have obfuscated a lot in terms Cathie: [1:31:48] of the progress that we're going to see. And I'll give you one illustration of what happened in his first administration. Cathie: [1:31:54] How did he cut the corporate tax rate from 35% to 21% given the very, he had a narrower margin in his first administration. How'd that happen? It's because he was distracting everybody with all of these crazy ideas and policymakers just said like, okay, we have to get something done here. And this seems the least threatening of what he wants. He's going to lower the corporate tax rate again, but not explicitly. It will be through the Cathie: [1:32:27] 100 expensing of capital spending in year one, that will get us to the equivalent of a 14, 15% corporate tax rate, which will make the U.S. Cathie: [1:32:39] The most competitive nation, I think, in the world combined with the deregulation. And I think you hear many people, they're very bearish. They're saying, oh, the dollar is going to go down. The U.S. is, you know, investors are moving away from the U.S. Into these other areas. And some of that has been going on. But I believe the return on invested capital in the U.S. Will increase relative to that anywhere else in the world, and that we will attract capital, attract more investors, and the dollar will go up in the years ahead. Ryan: [1:33:18] There you go. Let's end on that optimistic note. Kathy and Brett, thank you so much for joining us today. The message, it sounds like, is it's a good time to be an innovation investor, and you guys are certainly there on the frontier of that. So we appreciate all your insights today. Cathie: [1:33:33] Thank you, Ryan and Josh. It's been fantastic. Music: [1:33:34] Music