SPEAKER_00 0:00

You are listening to the AI Investor Podcast from 24-7 Wall Street. On today's episode, we talk about personalized agents going mainstream with the launch of a previously behind Meta. We ask the question Is AI suddenly paying for itself? We dig into the data and an update on my favorite sector, robotics. All that and more is next. Eric, very good to be back filming again. I enjoyed seeing you earlier this week. I came down to DC and we hung out for a little bit. We were not spending time with all of our senators and lawmakers. They were not looking for our uh insights on AI regulation, although maybe maybe we'll save that for the next trip.

SPEAKER_01 0:37

But good to see you. Hey, not yet. That's all I have to say.

SPEAKER_00 0:40

But yeah, that's the next trip.

SPEAKER_01 0:42

It was great seeing you, Austin. And um we've we've once again had a very action-packed week in AI. We're gonna spend a lot of time on today's episode going through what's happening in personalized AI. Uh, Muse has had a very strong launch. It's led to strong returns for Meta stock. We've talked quite a bit about on this podcast. So it'll be fun to introduce there. As you mentioned, we've got that high-level data. And then we're gonna do a little preview about what's coming across the next month because I think we've got a really fun slate. And, you know, for long-term listeners, you're gonna like hear what we have planned for the month ahead. So diving into the market, Austin, we had an extremely robust rally on Monday. It kind of appeared out of nowhere. You had a basically, if you had a tech portfolio, it was probably up three, even four percent on the day. But the big news continues to be as of Thursday, we are filming at 1010. I'm gonna look up rank now. We're having a little bit of a retreat in the markets, uh, largely driven by interest rates which have reached a 19-year high. And importantly, as well, interest in AI continues to falter as an investing topic. You were mentioning before we had filmed, you were on the homepage of the Wall Street Journal. And what was it, 17 mentions of AI on the homepage?

SPEAKER_00 2:00

17 mentions of AI on the homepage. And this more so than other environments has felt to me sort of like a zigzag environment where I'm having a hard time uh bringing data that seems to be seemingly uh opposed to each other. So, on one hand, you've got interest rates above 5.5%, and you know, companies like Oracle are definitely getting you know Delta Heavy Blow there because of their big financing build outs. But yet then there's also at the same time, there's ongoing data also in the journal this morning that the AI build is projected to be larger as a percentage of GDP than any of the prior major build-outs at telecom, railroads. Uh, we are seeing incredible traction from the Meta's new Agentich release, which I know we're going to talk about in a little bit. So it's hard to figure out what is up right now, right? On one hand, we have all this negative pressure from high interest rates, persistent inflation, um, which threatens the financing build out. And then on the other hand, we continue to see increasing capex, we continue to see models outperform, and we continue to see models that are more capable in ways that you and I have discussed, which is agentic. So that's really where you get like the multiplication of memory and increasing chip demand. So it's there, there's a lot of seemingly disparate factors going on right now. It's hard to distill it down.

SPEAKER_01 3:17

Yeah, precisely. And it it leads to we we see our own data. I think new subscribers, um, especially in an area like Apple Podcasts, uh, I think it's at its lowest figure since April 2025. And it's because we saw that furious rally up until June. We've had stocks largely trading sideways, and as we've explained, a major part of that is these geopolitical issues, the fact that interest rates are rising, things, things outside the AI space. But again, this past week, Austin, anthropic. They announced cloud-led discovery of a potential undiscovered uh gene editing mechanism. Uh, this would be something akin to CRISPRing. There's a lot of work still to come in front of it, but a pretty fantastic result. We've talked about whether or not AI can solve novel problems. And it's happening in the same week that OpenAI announced it's resolved more than a hundred long-standing mathematics problems. So we are at a time that looks like we are seeing a major breakthrough in AI's capabilities and moving from something that tends to be a bit more theoretical at spaces like math, to something like biology that could have an incredible impact on humanity. And then we've got continuing model releases. We'll talk about Muse, but Opus uh 5.5 came out in the past week. It's a new model for Manthropic. It's gonna, it's it's on this cost reduction curve we keep seeing, but it's a it's very capable, especially in um visual mediums. I'll put in the show notes, but one of the coolest demos I saw across the past week was someone wanted to illustrate how camera focus works. So they built an interactive lab. And the creator of this said it took about an hour and 26 minutes at an API cost of 2566. And it's just completely mind-blowing. And it is once again a good example of how capabilities in AI are going to increase the more we have these longer horizon agentic tasks where you're able to get AI to execute for much longer periods with much less feedback, reaching an end point. So that's going to be very transformative across the next year. And we just continue to get kind of small previews of it along the way.

SPEAKER_00 5:33

And yet, interest remains at a perpetual low according to Google search trend volume, which is just it's again, it's baffling because we're now starting to hit model capability where people are experiencing it in the real world in a way. You know, we have talked about how most people's interaction with AI historically has been a free model through chat, which means you're probably one or two model generations behind and you're not getting full agentic capabilities. And yet, so I could understand if many people were like, well, when is AI really going to get capable? If most people are still in that experience with it and they're not actually building with cloud code. And yet now we're seeing meta launch Muse. They had a great day on, I think the stock was up about 12%. So a lot of well-deserved recognition there. But I you would think that something like this would be a tailwind for the entire AI trade as people really on their day-to-day level can can appreciate what AI can do for them. And yet here we are. So talk to me about what happened with the launch of Muse, and maybe we can put it in the context of all of these sort of confusing signals between like breakthrough models and high interest rates or geopolitical tensions of the Trump Geeche summit and ongoing CapEx build out.

SPEAKER_01 6:40

Yeah. And so Muse, what it is, it's a personal agent. And as you noted, it led to some fantastic returns for Meta. Uh, they were up 12% on Monday. Importantly, this is part of a continuing rally where Meta had dropped, I don't know, it was probably down to the low 500s, and now we're trading back in the mid-700s. It's back to near all-time highs. So the narrative on Meta is shifting extremely rapidly, in large part thanks to Muse. So what we have is a new product. And delivering consumer agentic AI is been very difficult in part because agents are very resource intensive, as we've talked about a lot on this show. And what Meta is delivering is they're going to deliver, or they are currently, an experience where each person has um a dedicated secure computer via virtual machine that's going to have an agent within it and it's going to keep your personal data hived off. Um, you know, this is something with its own browser. So when you look at what this is driving excitement about, it's not just it's not just Muse. It's also, I think the CPU sucks. Austin, was it Intel up like 12% on Monday as well as Yeah.

SPEAKER_00 7:56

Back over 110, I think they were ripping.

SPEAKER_01 7:59

Yeah, and it's just because people see that if if we're going to have personal agents for each consumer, what that's going to mean as you you have this demand for CPUs for these agents running all these uh tools and resources. So the big thing is that the news has been very optimistic about Muse. Uh, it's it's number one on most app stores, I believe Apple's App Store, and third-party data from companies like Sensor Tower has been amazing. The sensor towers uh showing more than 200,000 daily downloads. And most importantly, people are continuing to use Muse. Its daily active users hit 400 and almost 50,000 10 days after launch. So it's doing what took most other consumer applications months or years to do in a matter of days. So this is what's really positive. And the other thing, a lot of the media attention on it that generally is a bit more negative, has been very positive. The New York Times wrote an article. I gave my life over to Mega's AI agents and was blowing away, which that that's not as good as you're gonna get in terms of media coverage for personal finance products, right?

SPEAKER_00 9:08

Yeah. And and and from a and from an organization that probably, you know, of the major pubs, you know, would be more AI skeptical, right? I think the New York Times still has a lawsuit against open AI. Um, so it to get that sort of commentary uh out of that organization, it just speaks to how remarkable the tool must be.

SPEAKER_01 9:25

Yeah. And if you haven't used Muse yet, I've been using it a little bit. I'll talk about that in just a little bit. But some of the use cases from the author of that profile, things like trapping spending, getting refunds, it includes a call feature, which I think is gonna be rolled out to more users. But they cite an example of being able to actually contact the customer service department for their health care, which I think is something we would all love to have happen in our daily life.

SPEAKER_00 9:52

That's it. Three three X levered AI on that alone. On that alone, if you are able to actually get somebody from your health insurance on the phone, there is nothing AI can't do. We are here, we're AGI, ASI, whatever it is, RSI, we made it.

SPEAKER_01 10:07

I mean, I think that is close to the killer use case, but it it also did things like generate a personalized podcast. One of the first things I had done was hook Muse's access up to my Gmail and started thinking of ways beyond just sorting email of understanding who I was and being able to build out personalized features. Now, of course, one of the biggest problems across Agentic AI at consumer level has been it requires high levels of trust. If you are using AI in your business, you know, you have different trust mechanisms. You have that you're gonna need to have an API key that might get out there, and you might have security concerns. But for a personal level, it is a big ask that to use these things at their most efficient level, you're going to need to give things like credit card logins. And this is an area that Meta spent a lot of time on. Um, you know, it's things that most consumers won't care, but I don't even think Muse will share any personal information with their advertising engine, which is surprising for a company like Meta, but shows they really want to nail just kind of engagement and user experience first. They're they're not even thinking about monetizing in the sense that it's down the road. They will figure something out first. They just want to get people using this. One thing for my wife, I talked with um an old coworker, the old host of the show, uh, co-host David Hansen. Um, he he was talking about we have a uh calendar that sits on our wall called, I believe it's searchlight, and giving that access, and then you can upload documents like we still get something like a PDF with food uh options for our school for the next month. Right now it's it's a paper calendar that sits on our fridge and something that doesn't integrate. Well, you give that PDF to Muse, and all of a sudden it's going to be able to build that into Searchlight. It's just something where it shows it it takes a little creativity at the beginning for people. Um, but once you figure that out, um, how how kind of impactful it can be. Um, so let's now that we've described what Muse is, uh, Austin, I I do think last week we didn't really talk about stocks at all. So maybe, maybe, maybe we'll get back. We're we're a little out of practice, not having as much stock talk on the show.

SPEAKER_00 12:23

Let's do it. Let's do it. I did get some great feedback on last week's episode, um, but I think people are really here for the stocks and the companies. As I hear you talk about what Meta's Muse does well, which I have not personally experimented with yet. So I'm excited to hear about your use cases. I'm looking forward to trying it myself later this week. I think that Gemini has to be right around the corner with the killer agentic app. And because one of the things that makes these so valuable, and you and I talked about this maybe a year or so ago, maybe Cloudflare was like a play here, was the login and integration for agentic apps, right? You're you're for your agents to be truly useful, they they need to have access to your personal information. They need to be able to log into things and act on your behalf. And when I think about the organization that does that the best, it's Google, right? They have the most login, uh, the the smoothest login experience, login with Google. They they've integrated with all of these different partners. So I would not at all be surprised if in the next three months we see Gemini's killer version of this right around the corner.

SPEAKER_01 13:20

That's a fantastic point. The first thing I think most users of Muse will say is why hasn't Google done this? Right. And and and it is the low best.

SPEAKER_00 13:30

Here we are again. Here we are again.

SPEAKER_01 13:32

It's a condemnation of um the company's capabilities, which they they have been struggling across recent months, but it is also trying to fit their business model into what this is, right? So I I think for Google, this is potentially one of the most disruptive areas. If you're a meta, getting people integrating their life around this more, it doesn't stop them from seeing a feed of entertainment. For Google, it could lessen the amount of times you are doing active search and giving them revenue. But they do have the advantages that you talked about. When I needed to start with um with Muse itself, I need to immediately give a third party access to Gmail or other areas. And as you noted, well, Google's gonna have Gmail, it's gonna have your search history, it's gonna have often the operating system that you use in Android. So they are playing off third base, and it's a little stunning that that they are playing from behind in this space. But again, it is pretty early. Um, some some individual stocks are worth talking about. Uh, Inadata, not gonna be a household name. Ticker, I N-O-D, Hunterbrook. We've talked about them before. They're a company that does deep research on companies. They recently highlighted as a muse winner, they're selling specialty data sets for our agentic use cases. You could think of creating these simulated environments or practice versions for digital assistance. Because what you're gonna need to come to is, you know, if if you're booking and there's a flight that's $200 cheaper, but it has a six-hour layover or it's an airline someone doesn't use, it's it's gonna need to use judgment, right? And we've talked about the gap for truly usable knowledge work and agentic AI being this judgment that humans do well and computers have struggled to catch on. So they're selling kind of the data that provides judgment to these models. And 58% of their revenue comes from Meta. And they're up 24% across the past week. I don't know if this is a stock I'm gonna be recommending, Austin. We've talked in the past about some of the data plays around AI. I worry with some slight architectural shifts, how much these companies they feel more like a trade than something I want to own for the long term. Absolutely.

SPEAKER_00 15:51

Absolutely. Interesting story, but yeah, I would not touch this one. And and this probably has echoes of your um behind the meter power build out for SpaceX, where I think it was SEI, right? Where you're saying, well, they have a special relationship, there's there's gonna be a big build out, so therefore they're gonna be in a good position to power it. That's a position I liked. And this would seem similar to that, but this is largely like for inno data to be up 24%, it implies that Meta's Muse is going to be the runaway agentic model and default of the future. And yet, if one thing has proven standard in this industry since we've been doing this, it's that the model leaderboards keep changing, right? And like it was OpenAI for a long time, Gemini had its huge comp from behind, then Claude uh with Claude Code, now MetaMuse with their agentic model. So the the first place horse here always changes. So this feels like it's just a point-in-time beneficiary of this model going great. So like good for inadata, it's it's a great one to watch, but it's it's not gonna add to my portfolio because this just feels like a point-in-time convenience because of what Meta's model is doing.

SPEAKER_01 16:59

Well, that changes in how you build the models, changes, right? So it could be a point that two years from now the data that they sell is even more valuable than it is, but it could be we've had some architectural changes that makes their offering significantly less valuable as well. So it's it's an extremely hard space to predict. And sometimes you need some humbleness that I don't feel like I have a good enough understanding. I don't know if anyone does, maybe some people out there, but it's hard to predict. Another area you had brought up was content delivery networks and security. That pretty soon, if we have massive adoption of consumer agentic AI, the majority, the vast majority of traffic on the internet is going to be agents. And there's going to be a whole layer of companies benefiting from this. The biggest thing is Cloudflare. That's obviously an active recommendation. We we probably should have re-recommended it, but we talk about constantly in terms of it's expensive.

SPEAKER_00 17:54

That's always been the issue there. But it's a front, but it's a lead, it's a top dog, though, right?

SPEAKER_01 17:58

So it's it is the most well-positioned company to be the agentic toll booth for this next era. Um, I we we've also talked about Fastly. Um, you know, it's been a reader suggestion in the past. I haven't recommended it because I would rather have the top dog in the industry, but it is worth noting that there's been some research into Muse's request. I found that Fastly handles 29.3% of them. It's most of the browsing the agents are doing. Now, I would know Cloudflare has a higher percentage, and they're handling what I would consider more of the value addition tasks, such as the security features. But but it is worth talking about. You could see Fastly as a company in much the same way that um AMD has actually been a better performer um than Nvidia recently. Sometimes these number two companies are able to really outperform when a market itself is red hot. Uh finally, just kind of like what's facing challenges. Broadly, you know, I would talk about Austin, you talked about health insurance earlier. The big thing is if Agentic AI takes off, there's a lot of business models that rely on customer friction. Are you gonna cancel the subscription? Are you gonna actually call in to challenge something? Areas like that face a lot of headwinds if we see large adoption. Health insurance. I mean, we can we were talking about, well, you are in town, our health insurer, who who I won't name by name, but it seems like they're trying to kind of Costanza. If anyone's seen Seinfeld the episode where Costanza's working and they don't want to fire him, but they're just gonna make him climb through a vent to get out of his office. It feels like they're doing that to us. They're increasingly not covering things they used to cover. It's hard to fight with them and you run out of inertia. But if I can have an agent that not only understands the game here of how to talk about codes not being covered, but will initiate the calls and get to the point of talking to a person, well, that might fundamentally change this industry. Insurance, right now, a lot of people still buy insurance through human agents. If you can go and do a comparison of insurance that will tell you you can save 50% with no activity aside from your agent booking it, that's a potential game changer. And then we have what were the aggregators of the original internet and what are the aggregators of what would be this Internet 2.0 with agentic AI. You look at a booking, this is a company, it's actually joined news. Um, it's a company that people thought were gonna be killed in the 2010s by Google, but survive that. But there's the other side of it that, you know, their value proposition when you're just talking to an agent, they're now maybe another company that can go over the top of them. And and we should finally note a company like Amazon, which we've recommended, it blocked access to Muse, which isn't surprising. Um, they've been blocking access to agents across the past year. Go ahead, Austin.

SPEAKER_00 21:01

Well, they're obviously also gonna roll out their own shopping version, right? They don't want somebody else shopping on their site without their behalf. So they're gonna obviously do their Alexa, whatever version.

SPEAKER_01 21:10

Correct. Yeah, they're they are definitely this is in terms of existential threats to Amazon, this is up there. So they're gonna try and be their own platform as well. You know, we're especially challenging. Go ahead.

SPEAKER_00 21:22

Let's let's pause on that because I want people to understand what you're saying. It's when you say existential threat to Amazon, you mean to their ad business, right? Because if people are placing orders on Amazon, it's great for their retail business, but that is notoriously low margin. The business that actually makes Amazon profitable is AWS and increasingly ads. So it's probably not it's probably not an existential threat to the company. AWS looks like it's gonna be the kit the ongoing grand slam of the last two decades in terms of business build outs. But one of the big growth drivers for Amazon has been ads, and that thesis could be totally tanked.

SPEAKER_01 21:55

Correct. And and we still own Amazon. So yes, we need to frame this challenge up. I believe, aren't they? I think they're making like $80 billion a year in ads.

SPEAKER_00 22:04

It's something shocking. I mean, it's throw out a big number and I'll agree to it right now because it is something it is something absolutely hard to imagine.

SPEAKER_01 22:11

2.8 trillion from ads right now. Um per day. But yes, it relies on a human actually going and shopping. And and the other thing as well in tests that have been done with agentic AI, um, the basket size is generally smaller because you're not browsing and going, oh, I need this, I need this, I need this. It's just a dedicated kind of tool to get one thing at the cheapest price possible. Now, Amazon still has plenty of advantages. Like you said, they're also benefiting AWS, which is a very high margin business on its own, uh, could could soon be a trillion dollar revenue business the longer we take this out. For example, its logistics business is still unchallenged, and it intentionally built that out to give it this physical end to its digital dominance. But you know, this does present openings for companies like Shopify and Walmart. You you do need to accept that. And finally, I do think one question that's really interesting from kind of this birth of agentic AI to consumers. We've we've had it to enterprises for a year, and now we are seeing it in applications that are being widely adopted by consumers would be what does this mean for anthropic and open AI? Because we talked about them last week. They're they're doing these warnings about slowing down. They also have trillion dollar IPOs coming up. So I'd say in the short term, it probably impacts them less than people might think. It's more competitive with OpenAI, which has always had a large consumer base, but you have to remember almost all the recent growth has been enterprise APIs. Um, OpenAI recently has been catching up with Anthropic in this space, thanks to Codec. Their new models like Astra are aimed at being really competitive on selling to companies. But OpenAI, they also announced a device for agents that's supposed to be a little button you wear and it allows you to interact with your agent. What did Meta announce last night? Well, their own version of this, Charn, along with some VR glasses that put Apple to shame. But that's that's maybe a topic for another time. So, Austin, I I think the big question here is the LLM behind Muse isn't state of the art. It's not competitive with the best models. That's Muse Spark 1.3. But, you know, what is the pressing idea for consumer adoption, I should say? That's do you have distribution, which Meta does. Do consumers trust your privacy while we talk about how they're going to virtual machines to do that in the world.

SPEAKER_00 24:45

Meta might struggle with that one. Meta's gonna struggle with that one. And the the own goal of the glasses is definitely not doing them any favors right now. And I actually saw, I think Zuckerberg's on stage today rolling out Muse with the glasses. And I'm gonna call that a PR fumble because I think you want Muse to be totally as far away from the glasses as possible, which uncharitably got labeled as pervert glasses. Um, so so I'm gonna call I'm gonna call that a PR. I'm gonna call that a PR miss.

SPEAKER_01 25:13

Well, and let's that this is an opening for Google. It's an opening for Apple, right? That you are going to need to trust for a for agents to be most capable, they need to make decisions. So the bar for trust is so high. And the last thing I want to say is just what harnesses that you've built to work with other tools? What's your infrastructure in terms of having the available capacity to create these virtual instances for, you know, basically mini computers for agents to work on? Very few companies can do this at scale. It is essentially Meta. It's essentially Open AI as a challenger, and it's essentially Google. I wish I could say Apple there, but you know, one problem, Austin, they're not gonna have the compute capacity as a problem.

SPEAKER_00 26:01

They're not gonna have the compute capacity. And also they um they've signaled, and I Apple is a company that you know it moves slowly and thoughtfully and comes out with the best version. And they almost were probably too early with their Vision Pro. And they've signaled everything is saying that that's gonna be you know, dustbined. And I think it would be a little bit of egg on phase for them to go back right to to wearable. And you know, another one that you would want to mention here, but they sort of pulled the plug on it was Microsoft. They had their HoloLens um experiment too, and they sort of they they've backed away from that. So it's either like issue a bit of a Mea culpa, hit light speed on development and try and be competitive here, or just try and go ahead to the next thing, which is I think what they're going to do.

SPEAKER_01 26:44

Well, or partner, right? They're gonna probably need to rely on someone like a Google who will have I mean, Google has, I think, 12 gigawatts under construction. They could be a willing partner to Apple, who gets relatively uh great terms uh as Apple might need to race to create an injectic offering.

SPEAKER_00 27:03

Aaron Powell And there's obviously been partnerships between Apple and Google as well, like being the default search. I was talking strictly about manufacturing glasses, but from it when it comes to when it comes to renting compute, yes, you're absolutely right. They're gonna be compute constrained, and they're now gonna find themselves in the very rare position of being a price taker instead of a price maker, and that's just what it is. Um I want to I I want to continue to move down this road of like there's seemingly uh diametrically opposed data points, so you can draw whatever conclusion you want about AI right now. I'm seeing you know that AI is the bubble, interest rates are above five and a half, five point one percent, so you know it's it's gonna stop this deadness tracks. I'm also seeing one of the most important numbers in investing that you and I talk about a lot, which is ROIC, so return on invested capital. This is also something that Warren Buffett has always often discussed as like one of the key most important metrics that he looks for in investment. So yeah, so just for our listeners who I'm sure know this, but ROIC is a way of measuring how efficiently you can use the money that you have to generate more money, right? It's the return that you get on your existing capital. So it's a way of saying, you know, if we invested all this money for the last two years, what's the return that we get back on it? That matters so much because the big criticism of AI has been, sure, you're spending all this money, when are we going to see the return on it? So let's take a look at some of the ROIC figures because this now is like one of the most bullish signals that you could be seeing. And this should silence all of the haters who, for the last two years, have been able to say, like, you know, oh, well, like where's the return on your investment, right? Look here, right? So, like, what where and we've talked about this in the context of Meta, but I want to hear this in the context of other companies. Meta was predictably first to show high ROIC because they were able to deploy AI models internally for their ad business, which is just a perfect fit. But what what are you seeing for return on invested capital across this landscape today?

SPEAKER_01 28:52

Yeah, that was a great tee up that there is this idea, okay, you're spending $700 billion this year, what whatever the number is going to end at. Um, but we are seeing a lower amount of revenue. Well, the the problem is that that spend, it is over a useful life. And there is a certain amount of analysis that needs to go back to say, are we actually seeing cost-effective spend right now? And what does that say about the ability of AI to continue? And and this is a little bit more in the weeds, but but it is just so centrally important. And again, if you are watching on video, which I think Apple has finally adopted videos and podcasts. Thank you, Apple, for finally catching up.

SPEAKER_00 29:36

Um go ahead and where did they get that idea? I feel like this is a little bit like dump on Apple hour after they just rolled out their first foldable phone. But at the same time, I don't know, Apple, like I've been I'm deep in your ecosystem and like the the Vision uh Pro was not interesting. You're like 10 years late on the on a foldable phone. AirPods were the last great product they released. So come on, Apple, get your act together.

SPEAKER_01 30:01

Yeah, and uh their their podcast app, it definitely seems like it's uh they started the podcast medium.

SPEAKER_00 30:07

They started it. They should have had video. All right, whatever. Let's stay on task. We're getting aside. Let's stay on task. Video podcast Apple. Yes, way to go, but let's let's stay on task. Your return on invested capital, what are we seeing?

SPEAKER_01 30:18

Yeah, and and again, we will if if you're watching this on video, we'll we'll have some graphics. This comes from Exponential View. It's a service I follow. Uh they they publish something called investmentbrief.ai. We can we can link to the sample here, but what they have is some of their high-level data for how AI spending is looking. And and the big takeaway is that the investment, the AI economy, I guess they would call it, it's up about 3.5x in the past year. It's at $229 billion annualized. And by their analysis of the annualized AI revenue over the cost of capital, including uh a 15% ROIC hurdle rate, they're seeing AI delivering 160%, which puts it into a positive, a very attractive return area. And most importantly, if you're looking at the chart that we will have on screen, this has increased throughout 2026. So it is a little counterintuitive that as concerns of the AI build-out have skyrocketed across 2026 because it is increasing in scale, where you actually had the worst ROIC was in times like 2023, where you were essentially getting none of it. In 2024, you're still getting less than a dollar back per dollar spent. You were you were less than 100%.

SPEAKER_00 31:43

And you're ramping doll, and you're ramping dollars at that point. So whatever dollars you're getting back, your return relatively looks worse because you're hitting this incredible capex boom, right? So even if you are getting a return, it's it's muted because of how much you're spending. Sorry, please keep going.

SPEAKER_01 31:56

Yeah, and and it's just we've talked about if if if something like DASNAR spending is increasing 100% year over year, that looks fantastic. Like it it seems inconceivable. But if the amount of dollars flowing into the AI economy are up 3.5x, well, that is going to mean better ROIC. It's going to mean that the amount of revenue from these AI applications is exceeding kinds of spend rate when you when you amortize it across its useful life. So, you know, we'll we'll again have that chart up, but a few other charts. Where is some concern right now? Well, if they identify it, it's the funding quality discount. They have a metric that they look at where is funding coming from? If it's coming from cash flow, that's great funding. It's very easy to continue growing from it. Where they see as negative uh kind of funding sources are things like are you using these related party transactions? Things like NVIDIA providing funding to the neo clouds, that that's just a little bit less high quality. And we've talked about some areas like debt as well. That is a bit more in the caution zone. So that is one area that they've identified as something to look out for. Another area to highlight is right now, still a very small percent of revenue from this AI economy is coming from the actual apps themselves. We're gonna look for some growth across that in the next year. And one more area of concern would be that this data comes from Ramp, but they conclude that 80% of OpenAI anthropics enterprise revenue on ramp. And keep in mind, this is just one company, so they don't see everything, comes from 1% of businesses. So again, this is Austin, we've talked about the barbell, where some businesses are using AI to an incredible amount, but the vast majority are using very lightly, it's still mostly some chat interfaces for employees. And the more that we see this adoption where it's part of your infrastructure, where you're adopting Agentic, is where you continue seeing these incredibly high growth rates. But I did want to highlight this just because, again, as we've taken off this year and there's a lot of talk of how much this capital could be chasing very poor returns. Here's a third party with relatively rigorous methodology saying we've actually seen a massive acceleration in ROIC across AI in 2026, which is very positive news for the continuum build-out.

SPEAKER_00 34:28

Well, let's let's shift to another build out. And I'm gonna bring together a couple elements here. Um, and I want to talk about the robotics build out because this has felt to me like it's you know going to be the next big cycle, but it's definitely one where hype and excitement could precede actual returns, just because robotics is going to be an inherently more complicated build out because it's you have longer supply chains, you get to physical world AI, you have to have you know real atoms in bits supply chains, you have to think about things like batteries, and it's not clear that we have like world-class, real world AI models in the same way we have world-class coding models. So I really want to hear about what's going on with robotics and what people can expect. And I also am curious, right now it's very timely because we have a Trump G meeting as well. And in many ways, the US has been trying to hold China over a barrel with access to chips to slow down their AI ambitions. But it would seem that if we really want to hit in a robotics AI inflection, we're going to need China's help. Their manufacturing capabilities are way, way larger in terms of total scale. In many ways, they have more capability and they certainly have lower costs to make robotics uh achievable. So I'm curious how this, like the US being dominant in chips and China being dominant in many of the manufacturing inputs required for robotics could play out. I know that's extremely difficult to talk about and to project, and we're not a geopolitics show, but it's top of mind for me right now because I do believe robotics and AI is the future. And I'm just curious what we can expect to happen given that meeting right now and the fact that the US has already tried to slow down China's ambitions where we have leverage, and now we're entering a place where they probably have leverage.

SPEAKER_01 36:16

That is a big question, Austin. And allow me to completely disappoint you. Uh, I'm I'm not going to discuss too much today because what I did want to do was more just tee up the fundamental HP4 robotics to preview what we're going to have coming in October.

SPEAKER_00 36:30

Um Yeah, sorry, sorry to throw a curveball there. It's just it's it's top of mind because the summit is happening right now, but but I am interested to hear just about the robotics industry overall. And maybe in the back of our mind, we should keep in mind that this is going to that there's a there's a risk here that this has the same sort of like geopolitical pawn dynamics that we have talked about many times on the show. It's like an ASML or an NVIDIA, right? They sort of get to get stuck in between nation-state powers and the entire robotic sector could experience that as well.

SPEAKER_01 36:59

And and that is going to be a key area we're gonna have to look at as we're thinking about investing. Because right now, the fundamental conflict, and we've covered this a lot with robotics, is that when you look at the researchers who are, you know, people at universities, and you follow them across various social media networks and their papers, they are far more measured. A lot of them have much longer time frames. But when you see the leaders at commercial companies, many of the engineers there, they're much more bullish about what's coming and uh about this potential Chat GPT moment where robotics is able to take off and get far more generalized. So of course, you need to have you need to have some balance here, right? That, well, if you raised money for robotics, you have one incentive, and that's to say it's gonna be incredible faster than people think, right? But you also have to consider that there was a bit of a similar dynamic that played out with AI itself, where again, a lot of the researchers, especially in academia, as LLMs took off, were far more conservative than what ended up happening um out in the space. Um, you know, you you do they do sometimes have incentive themselves to put some brakes on things and be more cautious, right?

SPEAKER_00 38:20

Um I I would also say there's there's a principle there that you actually have taught me over the years, which is the human mind is really bad at projecting exponentials. And yes, and even if you're an expert, so you know, academia, which which is largely, you know, extremely brilliant experts in their class, even they will struggle to understand how exponentials compound in technology. I mean it's hard enough to understand in finance. When you take it to technology, then capability starts growing, that's even harder. So the the naysayers, you know, I guess they could be forgiven because that is a hard, it is a hard thing to project where this capability goes.

SPEAKER_01 38:53

Correct. And that's that's where it is so hard because you use the word compounding. It's not just that the human mind struggles to understand what exponential growth is, the the growth rates and how they stack on top of each other, but how exponential growth fosters additional breakthroughs. So if if you are working in a lab in 2013 or 2014 and you're just seeing the limited capabilities from compute, especially the GPUs, what you don't see is how some excitement about some of the initial exponential progress is going to lead to a lot more investment in compute, which is again, it's it's going to start becoming very material very quickly because it's got this second-order impact. And I think the goal is for robotics is not to invest so far ahead of time that you know you you are going to have to go through multiple hype cycles for years before there's anything real. But it's also not to wait so long that you're on the other side of this exponential compounding. So we've we've seen a lot of kind of excitement around Generalist AI recently. Um company Generalist AI, they released the Gen 1.5 model. Um, and it really is pushing towards robots performing tasks without task-specific training. This is the unlock that we've seen in self-driving cars. When I used to, when I would visit Google in 2015, they would have entire training courses in the desert trying to get every single niche that they could, every single scenario. And it was it was a way of approaching the problem that could never fully solve it. What it what you needed was generalist self-driving car algorithms. And a lot of the progress we're seeing from Waymo and um uh Tesla this past year, and a lot of advancement from their self-driving cars is because they've moved on to having truly generalist uh models that are making self-driving cars so much better and um don't have this natural endpoint, continue improving to the point we will see these commercialized. So last week we saw Figure, which is a company we've talked a lot about. They've got a pre uh pre-IPO, they're at 40 billion. They put robots in 30 homes that they've never seen before, and they just released videos on them successfully kind of doing everything you'd want from a robot on your home, making beds, folding towels, you know, finding novel objects that you don't know what is on the floor and knowing how to pick it up, right? So, what I want to do in this next month, I I want to talk to our listeners about a few different things we're gonna do. So, first of all, I want to do a broad state of robotics video. This is gonna be a deep dive, relatively similar to what we had for a presentation on AI, but I want to unpack this problem of where is robotics today? What is still holding us back? What is a reasonable time frame? And then go through the stocks that are likely to benefit. And Austin, I think in there we're gonna really unpack what you've been asking about, which is okay, how how do geopolitical situations impact this as well? How are we going to have we we had AI and data centers, which basically we had the supply chains for the US and China kind of decouple from each other? Now we have something where they're deeply interwoven. And what's that gonna mean for how you're investing? Second, we're gonna try getting a robotics interview um with a specialist in the space within the next month as well. So that's that's something you have been facilitating, but I think we're gonna work really hard to get that um into our October programming. Third, in October, what we have on our agenda is an AI and energy deep dive. We've previewed this, but we have one of the best experts in the space. They're lined up. You know, things can always happen that doesn't it doesn't come to pass, but we we feel pretty good about this. And we want to dive really deep. As I previewed it in an AI presentation behind the meter, I think people are underestimating how much this is gonna take off across the next one, two years, and how essential is owning some actual energy assets going to be during this time. Um, so I think we've got all this planned for October. So at the end of the day, what this really provides, more stock ideas in new spaces, what we've been talking about, some new novel things. But I will say too, I think we'll probably begin pushing. Last week we talked about what this podcast looks like in the future. We'll probably push for more things that are more natively YouTube first. Um, you know, one of the areas is just with the podcast companies, um, I they just have almost no incentive to push new discovery, new programs. Um, and again, we're we're just not seeing much growth there, which isn't anyone listening's problem. And we appreciate all the comments that we have and feedback, but we'll increasingly build things that um are gonna kind of be YouTube first and we'll also put them within our feeds. Spotify has video, Apple has video now, but I think we're gonna have a lot more of these explainers, like robotics. We talked about where we are still working on optics one. We're we're gonna have these interviews. So I think what we'll see is probably a little bit more of a cadence moving forward where we'll have one week that's more of a recent news, and another week that's either explainer or an interview. And we'll probably move a little bit forward in that fashion. But I'm extremely excited about what we have this next month between uh a deeper dive into robotics, uh, interview with robotics, and also the fact that we're gonna be able to start tackling new subjects like energy. So for anyone out there, um, you know, we appreciate you continuing to listen. As I said, the interest in AI stocks is off 85%. We we realize that for a lot of investors, if if you were someone who just came in because you're watching those profits across April or May, you're out of the space right now. What we have is a lot of the people who are kind of long-term investors and willing to invest that it's not as peak frothiness, but as we've talked about on the show Austin, time and time again, generally investing at those times is when you do the best. So we will be sharpening our pencils, looking for even wider opportunities. And uh, I think we're gonna be able to really express that well throughout uh October.

SPEAKER_00 45:28

Like you, I'm very excited about the episodes across the next month. And uh a shout out to anybody who's listening in San Francisco. I've been very curious to hear from anybody who's experienced Tau Robotics. This is the humanoid in-home cleaning service. I think it's like 30 bucks an hour. They bring some robots into your home. And it your figure demo reminded me of this. And I've really wanted to hear about anybody who has experienced that or knows anyone who has. So if one of our listeners is in San Francisco and has used that service, please reach out to us, let me know, give us a comment because I want to hear about that first person. Uh, Eric, also just a shout out to you in recognition for the deep dives you put out. Um, I strongly encourage all of our listeners to make their way to YouTube and look at some of your most recent videos. They are exceptional. And if those are the format going forward, um, I know they are going to be happy listeners, soon to be viewers because uh we're gonna they're gonna be moving to a video format. But with that, I think that's a wrap on this week's episode. So, listeners, thank you so much for your time. We do appreciate it. Please leave us a comment on YouTube, Spotify, or wherever you kept this podcast. Um, and until next week, take care. The AI Investor Podcast is for educational purposes only and should not be considered investment advice.