The AI Revolution Call: The Compute Shortage

get. Um, I looked a little bit like a cross between myself and Gordon Ramsay.

It's, it's hard to get it right on point.

Yeah, that was, that was, uh, hilarious once again though. Welcome to the AI Revolution call. I'm Cody Willard. That's Greg Delon. We work at 10,000 Days. We invest in tech, revolutionary tech like AI. John Dio, my old friend from college, um, at UNM and has been working in the AI slash machine learning slash ontological software world for 30, 28 years. 28 years. Is that about right?

Yeah, sure, I think it's about right.

No, I think it's even longer than that. 30 years.

I go from '98, so—

oh yeah, all right, then 28 according to my AI math. Um, you know, it is just— we've been doing this call now off and on most Mondays for almost 3 years, I think. And, um, yeah, are we officially going to be, uh, distributed through Apple Podcasts? Are— can we find this on now Spotify?

You could—

you can search for it on Apple Podcasts. You can't search for it on Spotify just yet.

Why are we not on Spotify yet, Craig?

Because Spotify is a different animal. It's a It's a different beast.

I think my daughter's going to need to find us on Spotify. She's not going to Apple.

Okay.

All right. So eventually you guys will find us on Spotify, but we are now officially podcasted. This one, this will be the first official podcasted AI Revolution call from this group. Um, anyway, very exciting. Uh, we used a little bit of AI to get there, did we not, Craig?

Yeah, just to figure out how to do it. Um, it was something I've never done before, but there I got a few different leads on different companies that facilitate that. So Apple is very picky, Spotify is even more picky on how to get everything set up, and it won't unless your picture is perfect, the correct amount of pixels, um, and You have to make sure there's no copyright stuff anywhere and all that. You have to like, I don't know, it's a lot. But Podbean was really cool and it's very easy to use.

So, all right, we got to get you a better mic. You sound like you're in a cave.

Okay, let me see if I can fix that.

I sound like you're in a large room with a piano in the back, which fits, you know.

Yeah, actually, I mean, that it, yeah, that's The background works with the sound effect, purposeful or not. You know, I read something today. You know, there's a couple billion people online and we're already over a billion monthly users of AI. Read something today, it's something like 600 million daily users now. 1.3 billion daily users of the internet and 623 million something like that, users of AI on a daily basis. And, you know, I certainly— what I was going to say was over the last 3 years, you know, we've gone from when we first started doing this, it was new, you know, ChatGPT moment had just recently happened within the last few months, I would bet. And John, of course, had been doing this stuff for 20 or 30 years, so it wasn't necessarily new to him. But, you know, there was a true inflection point there. And it changed everything. And certainly over the last 3 years, we've gone from it being a little bit of a novelty to being integrated into our lives. Like, I notice my 50-year-old wife who is constantly on Gemini. I mean, just asking, how do we keep my sweater from shrinking? How do we— what is this stuff in the, in the new fish tank. It's just remarkable how it is slowly but surely being integrated into our lives. And I think the real point I'm trying to make is that, you know, we're probably still in the first inning, maybe halfway, maybe a little more than halfway through the first inning of AI, of the AI revolution going mainstream. And it still hardly any corporations are really integrating it. I'm still amazed when I play Madden Football, which I haven't played in many weeks now, I should get back on it, that, you know, it still like has this stock footage and stock clip voiceover for the announcers and it's not real time, you know, like this stuff is still not even barely being integrated at all yet. Anthropic is so desperate for compute that, you know, and not to mention OpenAI and everyone else, there's trillions of dollars of backlog on all of it and we're not even using it yet. John?

Yeah, first of all, Madden never updates things. They don't even bring them up. They make me so mad. So, um, you're a Madden football player? I used to, but every, every— I expect, you know, a couple years go by, I get another new version, and I'm like, this still haven't changed anything, right?

But come on, you and I, we should have a, uh, a tournament. That sounds like a lot of fun. We got to do that. Please continue beyond Madden.

Um, but in terms of like, I, I think of Anthropic. Anthropic's the topic of the day, uh, for what they ended up doing because Yeah, they're so short on compute that they went and made a deal with SpaceX, xAI for Colossus and take the whole thing for them.

And let's explain. Colossus is the giant AI factory. Like it is the giant data center that xAI built for Groq. But they were willing, and the terms were so good from Anthropic, which has at this moment unlimited access to capital basically. That they— that Grok, SpaceX AI, now was like, yeah, we'll rent you the whole thing for a few hundred billion dollars for the next few years.

So it starts with, how do— how often is a company's revenue like almost 100% gated by the supply side of things? Like, how much compute can they get? They're not held back by demand.

John, on that point, even moreover, when was the last time a software-based, cloud-based business was constrained by compute? Like, that's actually interesting. Certainly, you know, semiconductors off and on for decades have had cycles most of the time where there's more demand than there is supply being able to be cranked out. But in hardware in general, you know, good physical goods are different because there's a constraint of factories. But you're right that it's actually— that's an interesting point. It's not just that it's this huge number that's not being constrained by the law of large numbers and still growing at incredibly fast levels, but it's a software-based business that is completely constrained by compute. By the amount of processing power it can get its hands on.

And if they can get their hands on it, then they were 10x last year and they could be 10x this year. And if they— they want one more of those in a company of that type, um, basically it's kind of game over if, if that happens. But I don't know if it will be. You know, Google has a shot still. OpenAI maybe It seems like less so as time goes by. They're all still in it, but with that deal Anthropic made, they really shifted into the pole position.

The deal getting that much compute from XAI.

Yes. Right. And then from an XAI standpoint, I think like SpaceX— what are they called now? SpaceX AI? They talk about data centers in space and how they have to do that because the capacity is going to run out. Um, but really I think it's a great hedge for them that, well, people could say, yeah, you know, there's, there's risk associated with like tons of things, all the variables of is this thing that's never happened before going to work. And you know, you could fall back and go, look, it's Elon, things end up working. But, you know, when and so forth, people waiting for that to, to take hold. And while they wait, now they're off making $3 or $4 billion from Anthropic.

John, I'll take it to another— just taking that to its logical endpoint there. It's— this helps finance the data centers in space and it offloads the risk of needing to grow revenue inside of Groq. Which, by the way, now in the Tesla, having access to Grok has another just incredibly small improvement, but it is so revolutionary to be driving around and, hey, we're going to a Neil Diamond tribute concert at the Spencer Theater. Let's ask it about the Spencer Theater. And it's incredible. It's just, you know, what you learn while you're driving along or having it give you a tour of a, of a city. It's, it's fun. Anyway, your navigation point, that is not— it's a distribution of Groq, but it's again not monetized. And Groq has not been really monetizing its compute and, you know, improving its ads and things like that, but certainly not at the scale that like Meta has been doing. Um, but this actually offloads the risk of needing to try to monetize Groq and get data centers into space. It helps finance it. Yeah, directly.

Yeah, yeah. I mean, let's face it, Groq is a little behind. Uh, it, you know, it's just, it's just not the leader. It, uh, but it's also newer, so it needs some time to catch up. But, um, these things can go on and happen, and he doesn't have to worry about that. He can be like you know, it can, can, can handle the sort of the competition or whatever and say, yeah, I'll get you your compute, but it's going to be my compute later.

You know, it's a bit of a pivot for xAI slash SpaceX AI into becoming a neoscaler, not a hyperscaler like Microsoft, Google, and Amazon, which have giant $100 billion existing cloud businesses and are now hyperscaling the AI side of that. But it's more like a CoreWeave or a Nebbius that is building AI compute that can be rented out profitably at scale. And of course, you got Elon with his first principles and endless refining and risk-taking approach to it all. So, you know, it's truly exciting times. Anthropic is the topic of the day, John, and by that you meant the deal with xAI, but in general, just the lead of Anthropic ahead of ChatGPT. Everyone, I'd say, sans Gemini, Google's Gemini, it's still just growing leaps and bounds in large part because they have distribution of every Google search, not to mention every Android device and soon every Apple device. But the underlying core of what you were sort of hinting at there, John, would you elaborate a bit on Well, I think it almost feels like maybe Anthropic's lead is increasing.

Yeah, I think that if their growth rate is above Gemini as well, and Google would be the ones that could turn this around if they turn it around. But Anthropic, if they're able to maintain the rate they've got, which has been barn burning, they, they, they could be, uh, what, a $3 trillion, uh, revenue company within 2 years or something. It's, it's something crazy like that. It's been 10x times 10x is, is what, right?

And it's something, you know, tens of billions already. So yeah, yeah, let's call it $30 billion, 10x, $300 billion, 10x, $3 trillion revenue. I don't know that they're planning there, that they're actually going to be able to do that, that quickly. But, you know, we talk a lot about what I do as a revolution investor and trying to invest in the most revolutionary technologies and the companies/platforms that drive them is in many ways inevitable. Like these, these outcomes are inevitable. The question is one of timing. So someone's going to be doing $3 trillion of AI LLM revenue within the next few years. The question is, is it Anthropic and is it within 2 years? You know, it's— that segues me to another point I'm struggling with internally. It's an eternal struggle for me as a revolution investor. Maybe verbalizing it out here will help me process it to a new point for once, that the biggest mistake I make is ever doubting how big these revolutions get. Because I talk about them creating— it used to be trillions of dollars of economies. Now I look for technologies, revolutions that create tens of trillions of dollars of economic activity, and how quickly the market can discount that stuff versus how quickly it actually inflects and becomes reality is this ongoing battle. And I strive as the way I trade and invest, be buying fear, buying when markets are down, like I was doing 6, 8 weeks ago when everyone was in panic about the Iran war and we were getting some great pitches and literally went out for a money raise for the first time since I launched my fund because I was so excited about the setup. Here we are 6, 8 weeks later and many stocks have doubled or tripled. Many of them have, you know, rallied 30, 40, 50%, and the markets themselves have been on fire and we're at new all-time highs. And it's this, John, you know, like you just talking about, like the potential and even maybe probability of Anthropic hitting $3 trillion in 2 or 3 years is why these markets and these stocks, these AI infrastructure stocks are doing what they're doing. It's trying to start, continue discounting this incredible shift in economy and in society.

So, and what's holding it, uh, potentially holding things back is, um, energy. We've talked about it, and, um, there's getting to be some real pushback, um, just from diff— Like, how many is it? 30 out of 50 states are fighting against having any of this power stations happen in their state.

New data centers, you mean?

Yeah, I think Maine. I'm in power.

I think they go together maybe because if it's the data centers, the data centers—

I don't think anyone's against, John. I think maybe it's semantics, but nobody is against more power stations.

No, they think the power—

They might not want them in their backyard, but everyone wants—

They want them for the data centers. They're going in as—

Right. It's because it's power centers for data centers.

Yes.

It's the data centers' demand on power that is the reason there's a political backlash against data centers. Craig, you were about to contribute there. What were you going to say?

Yeah, Maine is the first state to ban data centers in their state. So they don't want any states in there. And any—

It's like it's a disease.

Well, but look, I mean, you, it certainly, you can understand politically how that works. You know, there's always backlash against trillion-dollar businesses that are, and those are the companies that are most aggressively spending to build these data centers. There's always backlash against higher energy prices and more demand on the energy grid from data centers would make energy prices and probably has made them go up at least a little bit thus far. The fact is that there's this other side of the argument, sort of like there always was about Bitcoin and crypto mining technology, and everybody's like, hey, it's such a demand on the grid. But yes, also they're building these Bitcoin mining and data centers with as much of their own independent energy generation sources as possible. And so when, if and when there are times that the grid needs more energy than it currently has, say in a heat wave when there's lots of air conditioners running, that excess capacity from those data centers could and probably will be at least partly allocated towards the energy grid to contribute and keep the grid flowing and to keep prices relatively low. So, but yes, there's major backlash against it. And, you know, it is what it is.

I mean, it's got to— that's what you're describing has to become less of a hypothetical and more of a this is how it works. So there's some talk of in a given community where a data center comes in, you're paying for people's electricity. You know, you basically, you're going to subsidize everybody in the community. Yeah, and there's something—

by you, you mean the company that's building the data center, not the—

yeah, but I'm saying subsidized sounds a little like, oh, you know, like untrustworthy, like we're gonna knock 10% off your bill. Oh sure you are. Versus no, we're like going to pay it, you don't pay for electricity anymore kind of thing, because the level at which they need to come in with power makes the power of the community they come into be like De minimis, like nothing.

Yeah, de minimis relative to how much the actual data center will be generating and consuming.

So there's that. Yeah, right.

But John, you're exactly— I like how you phrased that. It does need to become less hypothetical and much more grounded in reality in actual terms of agreement. And we are still in the hypothetical phase of this incredible, which is itself sort of incredible. It underscores the fact that the government cannot stay up with the pace of change that is happening in technology and AI, which is probably a good thing. But it is— they are like— we're like— it's literally $1 trillion or almost $1 trillion of CapEx in the US alone on data centers this year, up from $100 billion. And the government is still like, yeah, hypothetically, such and such.

What's the annual GDP right now?

38 for the US. Craig, will you ask?

What's the debt? Isn't that the debt?

38?

Okay, it's about 7.

More than that.

It's about 7.

Our nominal GDP is 31 trillion.

Wait, if it's 31 Uh, and the debt is— okay, oh, I was thinking of— no, I was thinking of these, right? The deficit, the, the whole— the U.S. budget, the government budget is what, $7 trillion or some number like that? Okay, the, the, the debt is like $40 trillion. So the debt— so GDP is 30-something. So you, you get, um, $3 trillion from this one and the data centers, and then the other ones and adds up, you got like a couple years from now, $6 trillion coming from this side of things. It starts to be a, like, a sizable piece of, of GDP.

No, not starts, it's already. Like, this year, the spend on AI data centers alone is equivalent to almost 2.5% of US GDP. That does not include the magic That happens when you spend $800 billion in one year. And that includes what happens with pipefitters, plumbers, electricians, contractors, construction workers, Caterpillars, dirt work, like, and then the restaurants in that area that has all those even just temporary 2-year workers building the data center, buying food, having drinks, and then the waitresses buying daycare. Like, that's— they used to call it trickle-down, and I'm not going to use that term because it's politically loaded, but that is what I call the percolative aspect of the economy. Just percolates and it goes everywhere and it just builds upon itself and starts bubbling and there's steam that rises and it all happens and There's trillions of dollars, maybe 5 or 10% of impact on the economy this year because of data center spend.

And we're starting to see it in the labor numbers, in the new jobs. And that doesn't— I mean, I say starting because you'll have these like 155,000, you know, some number related to that building data centers. So, but now once those things are built, then you have another kind of jobs that—

Well, and John, it's like, that's the thing about technology is that, you know, stuff like this can have 10, 20-year cycles of, even if it's not doubling and tripling in its growth every year, it's still a trillion dollars per year for 10 years is the low end of what you should expect now. It will probably be more than that. Some years it'll $2 or $3 trillion of spend. Some years eventually it'll go back down a little bit. But it's not like it just happens once. It like the amount of compute that's going to be continually demanded to do all of the AI generative stuff, all of the AI, the flywheel of AI creating AI that uses AI that creates more AI that needs more AI that uses more AI. Is self-reinforcing for many years to come, at least. And so it's not like it goes away. It's not like all of a sudden, hey, well, we built $1 trillion this year. We can go back to $100 billion and spend next year. It doesn't work like that. It's, it's, it's, it's not cumulative, but it is like it provides a base and especially If we are right, which is the data is, is proving this out, that the return on all that investment invested in the data center stuff is probably 30 to 40% ROI.

And then so, and then what's it all getting used for? You know, I, on a daily basis, I'm working with this stuff for developing software. Today, I was working on a bug and working with Claude, Anthropics system there, and it's going through and looking for my— what might be causing the problem and coming up with this and that and going through. It went through a system that has a million lines of code to go and come up with what was actually happening that was causing this particular issue, and then what came back in English with here's what happened and why. And it was this last piece of code that got put in that was supposed to do this. But instead, in this situation, does this. And, you know, it's like incredible.

Like, it would take people the minutia, the, like, the detail to which it can go down and find that stuff and the pace at which it does it. And then, as always, the whole point of what we're talking about is like, this is the worst it's ever going to be. It's just going to get better and better as more and more spend and more and more build and more and more compute happens.

It would have taken, uh, a team of people a week to find what, what it found and to, for it to, to work it out.

Used to be like a competent, hard-working team at that.

Yeah, yeah, right. Yeah. And we used to be like, oh, okay, we'll let it participate, we might get something from it. And now it's like it lets us participate. You know, it's doing most of it.

That's, that's just great. And that's— you laid off half your team?

Nobody, because we're making—

well, that's so strange. I thought it was going to—

no layoffs. Hey, hey, knock on wood, right? But it's so far, you know, what we're doing is so sophisticated that maybe we didn't grow the team we would have had to. And maybe if we tried to grow the team, we would have run into the complexity of the too many people trying to, you know, teach each other about what's up with different things. So we're probably getting more done with the same team than, you know, we otherwise could have.

I just want to repeat, this is where like the magic things of the economy start feeding upon themselves, is because it's not like just because there were fewer people hired at John's company because they're able to do more AI stuff. But if John's company is able to do that much more sophisticated and that much more productive stuff internally, there's more profits inside of their company, there's more growth inside of their company. And then thereby other companies, suppliers, customers, etc., all benefit. And the whole thing, their employees benefit.

Yeah.

All of it feeds upon itself.

The piece I'm not sure I understand is I can see how this technology, at least on the side of it helping write code, which is an area where it does really well, it's one of its best areas so far. I can see how much value it provides. What I don't maybe get is how much demand there is for, and we talked about this a little bit, but how much demand there is for new software out there in general. And I think that there was that— I forget the name of the law, like the one of these things, it's, um, uh, you know, it's ironic or something, but it's that the, the— uh, Jevons Law. Yeah, that like the more, um, uh, the, the, the, the cheaper it gets, the more demand that's needed anyway. So it's, you're, you know, like if you're saying, oh, revenues are going to go down because the, the you can't charge as much for it. It's like, no, you, you don't charge as much for it. And it—

but if you, but if you do 10x as much and you drop prices in half, you still make a lot of growth.

Yeah, but why does the 10x come from? And it's the fact that there's— there is, um, the demand, like, that was held back because it was expensive, but there's also a demand— well, Yeah, and there's a demand that never could be fulfilled because it would have involved trying to hire more programmers than exist in the world.

And even just complexity of me— of managing any of it at all is more than most people would ever want to do. John, the thing is, I would argue that, look, low talent, low value-add employees who do coding. Okay. Like, yes, their jobs are threatened and I'm not sure those are ever coming back. I'm sure there will be other things to do because AI, the whole point of AI is its ease of use and all you have to do is talk to it. So if they have any internal drive at all, there will be some opportunities and there will be other jobs and other things you will be able to do. Beyond that, the amount of growth of just random software apps that can do random quick things for you because AI is so effective. You just tell it, hey, I need to, you know, create an app that tracks my entire itinerary, make a visualization of it, show me on the maps where I'm going. You know, I just did a whirlwind 10-day trip with— I was in 16 different airports on 8 different flights, including connections, 2 different boats, a train, 2 countries. And it would have been amazing to actually just have a one-off app that tracked all of that and, I mean, you know, helped me manage it. And that is, you know, at some point in the next few months, couple of years, Just a random example. What I always tell people is I'm smart enough to realize that we need all of this compute because there's so much demand, but I'm not smart enough to tell you all of the millions of ideas and services and applications and random stuff that people are going to do with it.

Yeah, I feel like I am not as smart as you're describing, and I, I guess I had this naive sense that for some, any given company, that's like, just because they were struggling to get like a scheduling app and an app that helped them do financial analysis or something, and you know, that that's basically, if they had those things, that was all they were out trying to get anyway. So they'd be done basically because they've been spending 2 years with teams of people and projects and outside contractors trying to get those 2 big things done. But it turns out the only reason those were the two things they were trying to get done was because they didn't know— there was no way for them to get 15 things done. And now there is. And really, their whole business, which is around kind of getting themselves to a new level of service, of support, of the products they can put out, of how fast they can put those products out, of how much goes into getting the supply chain, It's like everything they do has the opportunity to be optimized. And, you know, before I didn't think of it as that, that, that there was that demand there. But now that I see what's possible, I can see how all this demand could be like, that's it.

I mean, every guy who runs any business, any woman or man who runs a business has, knows all of, like Craig used to help manage restaurants. Like any manager of any restaurant knows all of the little things that they possibly be using. I have a medically fragile daughter and we have a log that tracks every time she pees, poops, gets medicine, every, you know, whatever's going on. There's a little diary at the end of it to make notes. And we've been doing it that way for 10 years. And I tried maybe a year and a half ago to have AI help me create an app to do all of that. And it wasn't ready for primetime yet. We're going to try it now again. Probably this weekend and rebuild it. And there's just like personally ideas, like personal things I need software coding for that I never would have considered trying to actually build were it not for that being as easy of use, right? Craig, I can tell you wanted to say something.

Yeah, well, I mean, if you listen to, uh, Meta's earnings call, I mean, every— what we're talking about here is basically what Zuckerberg was describing, because you have OpenAI and Anthropic really focusing on enterprise and coding, and Codex is great and 5.5 is great. And then you have Opus over here and, you know, these two titans over here and Meta's in the corner and people aren't really sure what Meta is trying to do. But what Zuckerberg said is he, he's not going to give Codex or Claude Code or any of these really I would say hard to use tools for like the regular person to his mom. And what he wants to do is make something like we've used Manus, but something even easier than Manus that's integrated that just does stuff. You know, like this is, it was like, oh hey, I know that you post a lot about, like Lori posts a lot about your daughter. So, you know, on Facebook. So it's possible Meta could be like, hey, you know, did you wanna do an app or something like that? Or this, that, and the other. There's all kinds of things that Meta could assist you with some— with a really good harness on a top-tier model, which I'm not going to say MuseSpark is better than 5.5 or better than Opus 4.7, but MuseSpark is a, a great model. They're already training the next one, and as long as you have a solid harness around it, what you're describing is not far off. I think it's like the end of this year, uh, to be super solid for whatever you want to do.

That's awesome input, Craig. Thank you. It is the TV anchor background in me that makes me want to give everybody a little, like, explain a little bit of what you were saying there. Simply, Meta's latest AI model is called NuSpark, and Meta is trying to use that internally in the, like, the 5.5 and the Opus 3.7 or other AI LLM models competing against Meta's NuSpark. Which is—

Muse. M-U-S-E.

Yeah, sorry, MuseSpark.

Yeah. Yeah.

I think I was thinking of Muse.

MuseSpark.

You know, on all of this note, I had in London a few weeks ago, might have been Manchester, wherever I was at this for this lunch, I had a really insightful businessman talking about AI and whatnot. And He made an analogy that, uh, if you're trying to put a hole in a wall and you've been using a drill and a drill bit forever, you might think you got to just build a better drill bit. You got to build a better drill. And then someone else comes along and they're like, hey, we could blast a laser through the wall and it's free and instant once you buy the laser. And you don't work down that. And it's like, you have to remember, John, sort of what you were talking about. Like the goal was never a calendar. The goal was to simply facilitate doing better, quicker, more productive business. And that's always what software's only purpose truly is. It became an industry unto itself, obviously software to improve using software and software built on other software. It's endless loop, but In the end, every application, every service is simply someone trying to put a hole in a wall and something that helps you do it. The drill and the drill bit are an application themselves.

Yep.

Yep. You can transcend it all. Robotics. I'll simply say— What's that?

That was my next thing.

All right, well, you want to segue it or you want me to?

I just want to— I want us to be evidence-based about how we pick our winners. As opposed to kind of, uh, reputation-based. And I think that if you look at Optimus and you say, well, it's, you know, it's Elon, so of course it's gonna— it's gonna be number one. Um, I'm not seeing it yet from Optimus 3, but we're not being shown Optimus 3 yet. Uh, and I think we need to see something there in order to, like, be objective about what's, you know, what's coming.

I would argue that I'm being objective when I think Brett Adcock at Figure is not going to ever produce a viable commercial robot, and that I think Elon's history of delivering credible products eventually would mean that he probably will. But I give it a trustable limitation. But right now, presently, it's not like they're like, hey, look at the latest Optimus stuff. I think they're keeping it a bit under wraps for several reasons. But there's actually a bigger discussion to have here. The reason they're keeping it under wraps, they do want to leapfrog ahead and not just show some incremental improvements. The other main reason I think is Elon's even verbalized that people constantly, as soon as they show something, the Chinese and Brad Adcocks of the world, competitors of the world, immediately go out and start trying to copy it. And so I think they're trying to do several leaps forward and then show something that's, you know, 3 to 5 years ahead of anyone else before they show anything again. I have a question on that same sort of thought process though, is Elon's approach to doing this, let's call it Tesla's approach to doing this, is, um, to, you know, try to make it basically as holistic and turnkey from the beginning before they start selling it. And I think what we see with a lot of like the China, the robots from China and a lot of just the even other American startup robot companies and like the marathon, half marathon gimmick that they ran with the robots a few weeks ago in China, um, is performative and probably not actually all that helpful. Um, and I think what Elon has learned from Full Self-Driving and how long and hard it's been to go from idea 12, 13 years ago, or probably 15 years ago, the idea to implementation and eventual actual execution and pulling it off and having a fully autonomous product is that, you know, you've got to to get it at least to a point that blows people's minds and it does have value in everyday use case scenarios that it can do from the minute you deliver it to people. And I wonder what that, you know, I'm sure there's some curve we could probably use AI to help us figure out what that curve looks like, but there's a minimal level of competence that that robot needs to have before Elon, A, shows it again and certainly eventually sells it as a deliverable product.

How can he make a call about how, like, well, 3 is the one, you know, that's the one we're going to be able to go mass produce? It's like, if it hasn't already performed those use cases. It seems a little early to go say that.

Elon Musk, ladies and gentlemen. In other words, you know, it's partly to drive his teams to do the impossible. Um, you, you make impossible timelines and then you try to get them filled in. Um, that's a large reason why he does that, but I don't think 3 is going to be ready for prime time, and I think it'll be 4, and it'll probably be next year before they'll even really start rolling them out en masse at any scale. I'm even talking hundreds, um, inside of the Tesla factories. It's okay. I mean, you know, this is one of the most impactful revolutions of all time, and it's not going to happen as quickly as everyone wants, but it's gonna happen more quickly than most anybody expects.

So to my original, uh, kind of comment, it sounds like we can't do what I was describing. We can't be evidence-based because we're not gonna get served the evidence until it's already a convincing case.

Um, I think that is, that would be the, a better, more succinct answer than the one I just droned on about.

No, I got it. I got it from what you said. So, all right. So I can't really— well, then I mean, nobody can really make the claims. You can do the— you do what you've been doing, which is— and it's kind of like you've been doing for decades of, you know, identifying things before other people because you kind of know what to look for and so forth.

Funny you say that, John. That's right. I mean, that is what I'm paid to do. That's what I strive to do. And I work very hard to try to continually do it. And yeah, intuitively and logically both. My instinct, like, I'm all— I think it's the Optimus robot's going to be the killer, not killer, not killer robot, but hopefully not of robots. The most amazing. It'll be the category dominator.

Yeah, yeah.

And there will be millions, hundreds of millions of other robots sold, billions of other robots sold collectively. That was another thing I was going to say when I was going to the Robotics Revolution is that, you know, anecdotally, I'm hearing constantly about robots being used in industries they have heretofore, heretofore not been used in from from farmers in rural New Mexico using robots with lasers to shoot the weeds in their crops to robots tying rebar and hanging sheetrocks in high-rise constructions in Abu Dhabi. And these are not humanoid form factor robots, obviously. That's the point. Like, there's robots everywhere. Drones are a form of robots. Like, robots are already here. The robotics revolution is happening, and hardly anybody realizes that. Everybody's focused on the AI revolution, but the robotics revolution is inflecting right here, right now. We're having the ChatGPT moment of robots over a drawn-out moment, call it a year, two years span of time. The moment is a year or two. But it is happening aggressively and acceleratedly right now, isn't it?

Wouldn't Tesla— wouldn't, uh, Elon Musk say that, uh, a Tesla with Full Self-Driving is a robot?

100%. Yeah, and it is. I mean, I just, uh, you know, we got the latest 14-point whatever update in the Tesla Cybertruck in my wife's Model S, uh, X the other day, and They're just— again, it just is better than ever. It just feels better, safer.

It's amazing.

It's amazing.

We are regularly now, and it's like before she wouldn't use it, and now— because my wife's the one who has the Tesla, and before she really wouldn't use it, and now, um, we got the new Y with the, you know, it's got the Hardware 4 and the latest, uh, upgrade there, and it's like a person driving.

It's funny, my, you know, I've had a Tesla now for 7 years, several different Teslas over the last 7 years, and my parents just finally broke down and got one after they saw my Cybertruck drive me from a hospital appointment home. Uh, my dad was in the driver's seat but didn't actually have to touch anything the entire way, and so he was like, oh my God, actually maybe I need a Tesla. So they got a Model Y and maybe 6 to 8 weeks ago. And just last night, we're over at my parents' house for Mother's Day. My dad's grilling burgers, and he turns to me, and unprompted, he's like, "Boy, I love that Tesla." It's a pretty great vehicle. All right, last words.

Oh shoot, I just used up my last word. No, um, let's see what else is, uh, happening, uh, AI-wise. Are you, are you guys still using, uh, Manus or anything like that?

Yes, on a weekly basis, not a daily basis.

Craig?

Um, so interesting, that was going to be my last word. So ChatGPT, OpenAI has been doing this thing to get people back on their platform after they had a a little mini exodus after their snafu with consumers and working with the Pentagon and all that stuff. So they've been giving people like a month of free Pro. So I got back on there just to try it out. And 5.5 is really freaking good. I think it's the best model out there for everything that I do. I had to do a I had it make a— so I was working on Palantir today and doing some modeling, and I had it break down their commercial side of the business, and it thought and worked just like Manus did. It was pretty impressive. Now, I don't think it's— it doesn't have the harness to be able to go and deploy different things and do all the really cool things that Manus can do as far as workflows, but as far as creating spreadsheets, getting data really thinking through processes, 5.5 is very impressive. So the people over at OpenAI, you know, can't stand Sam Altman, but I mean, there's some really smart engineers. They're doing some really cool things. So I know, and I wanted to bring up the fact too, Anthropic, you know, they're struggling to find compute. They just did that deal with SpaceX.

Don't steal my item. Don't go and throw another item on and then it turns out to be my item.

Hopefully it's not, but I just want to say—

Anthropic up doubling their, uh, no, no, it's not. No, no.

Well, I, I'm just saying there is one company that, that has no shortage of compute deals, and that's OpenAI. So we can't count them out just yet. So that's all I'll have to say about that.

Okay, so speaking of compute and availability, Anthropic, they went and made this deal with SpaceX AI. Um, and the same day they announced it, they announced that they, um, doubled the limits, uh, basically like doubling the tokens for everybody on the Pro plan and the Plus, the Plus, the Pro, and the whatever on up, so that your, um, Max, I guess, so that, you know, people who felt like they were kind of— they kept running out before it would hit the the— is it a 5? I think like every 5 hours it resets, you know. They— and they kept having problems with getting enough, um, kind of tokens to work with with them. And they're like, bets are off, have what you need, here it is. They doubled it for everything right when they did that.

My final word will be If we used to use a hammer and nail, then we went to a drill and drill bit, and now we're at the laser, what comes after the laser?

Not needing to put a hole in the wall.

We're living in the simulation, I guess, would be the answer at that point. All right, thanks everybody. Peace and love. See you guys soon. Find us on Apple Podcasts and YouTube and, yeah, Twitter, X.

Yeah, all that.

We're on— we're social.