The AI Revolution Call: Day 1 of NVDA GTC! The word of the day: Inference!

Hello and welcome to the AI Revolution call. I'm Cody Willard. It's Bryce Smith, Craig DeLonn. They work with me at 10,000 Days. We invest in things like the AI revolution. John Dio, dear friend of mine for 30 years. He's been working in the AI revolution longer than he's known me. No, actually, I think I met you before you officially entered the AI foray.

Yeah.

Craig, I think every week we got to improve The intro video, it's AI-generated. It's a little too long. I think it should be about 15 seconds, not 24 seconds. And I think every week people should tune in to see the next iteration of our endlessly not revolutionizing but evolutionizing AI intro.

I got one more change for you, uh, let the final part of the music run, like after the thing goes away and we're still here. Oh, couple—

I might be asking a lot, John. I'm not sure that this is a video intro. It's not like tied into StreamYard, which is what we use.

Well, I might, I might be able to overlap the music in a different way to go transparent on there.

All right, it's a good, it's a good request. I just, I think it's a high bar.

All right, we'll see.

Um, one thing, I mean, if you're trying to build a brand, you want to build something that's recognizable and consistent. So there should be some consistency across.

Should there be? Are we trying to build a brand, Bryce?

Always marketing, always like— yes.

I told Bryce and Craig just probably 4 times last week, I think I might be the worst marketer of all time.

Up there.

We're not— we're not like— the intro video is not a piece of creative art.

It should be.

That's my point.

I want it to be a piece of creative art.

It can be both. It could be both.

And Craig is an artist, Bryce.

If the Fox News logo was different every time, that wouldn't be very bad marketing.

So here's the Fox News—

nobody's confusing us with Fox News.

You guys, you guys notice that, uh, you know, AI Revolution is AIR. So like, take the place of like Nike Air or something, and you got like an air— no, that's—

I literally— it's an homage to Michael Jordan why I named it AIRs. I could have gone like SARE. Space AI Robotics, or I could have gone Rays, but I went with Airs as an homage to the one, the only, Air Jordan.

So you could have been like dunking on robots or something.

Yeah, that would be awesome. I think, Craig, I would like to see myself dunking on robots in that intro video, or have a robot do the dunking.

Kind of, it's coming up.

No, hell no, I want to dunk on the robot. I don't want the robot dunking on me. See this pinky? That's a result of someone catching— someone threw an alley-oop pass and I went up to grab it in front of the rim. And then another guy came behind me and grabbed the ball and reverse dunked my hand and the ball into the rim and to the stick. So no, I don't want any robots dunking on me. I'm the dunker, not the donkey.

I can go so many directions with that, but at least—

oh man, that's true. But at least in the AI revolution in which my own mind exists.

Yeah, yeah, yeah.

Let's—

speaking of the AI revolution, let's, uh, 3 minutes into the call, actually get started on some AI revolutionizing.

That would be the soonest ever.

Yeah, no, I don't really think so. I think we've done better. This was a long drawn-out, sort of like Craig's 24-second intro video. This was drawn out a little bit. Poor Craig, he had no idea I was going to complain about the intro video today.

The first 3 minutes, there's everybody ripping on it, on my work.

Mostly just me. I think the other guys are actually Don't care. They're defending you. All right. Anyway, the AI revolution. You guys hear about this guy Jensen Huang? He's the CEO of like this semiconductor company.

Yeah, something, something.

It's a startup, I think. Some VC— NVIDIA. Jensen Huang gave a speech today. Bryce, did you watch the whole thing? I watched about 5 minutes, moved on. I had some stuff. I'm trying to get through my 137 unread emails.

Yeah, I watched it.

Tell us, give us the 5 most important bullet points.

Yeah, I think, um, it's like not a ton of surprises per se, but, um, they— he just gave the roadmap for Reuben and then Reuben Ultra, which is the next iteration, um, and some new chips that are coming along with that. You're assuming people know what the next generation of GPUs and graphics processing units, which are an advanced form of chip which powers AI. AI is a semiconductor built out of glass. I don't know how far the backwards needs to go.

No, that was good. I just— you were like the next generation.

And I just thought people should know the AI chips that power everything from ChatGPT to Google's different things, Gemini. I mean, they have NVIDIA everywhere. They, so they announced the company, or they acquired a company called Groq, not to be confused with the AI service from Elon. And that was one of the highly anticipated things, how they were going to integrate the Groq chip. And basically it's going to run like in parallel with the main GPU, and they're going to offload some of the high inference tasks to that. So they're going to be working together just like a CPU works with a GPU. They call this an LPU. I don't remember what that stands for right now, but logic—

logic processing unit, perhaps.

Yeah, that may be right. I don't remember. And the idea is that he showed this chart which I thought was really interesting with the different sort of tiers of token generation and how companies that are producing tokens have to decide how much, you they have a fixed power budget. Power is the ultimate constraint, right? And so Jensen's like, look, you want to get as much revenue per watt as you can out of your AI factory. That's your constraint. So here's how—

literally, people watching this know that we, like, months have been saying it's all based on, like, processing per unit of energy.

Correct. Um, and so the, the, um, Obviously you can produce a lot of like low-quality free tokens. And then to get to really the complex high-intelligence stuff, you're going to have to use a lot more tokens to do that. So the idea with Groq is you stretch out sort of this long tail further than you could with GPUs alone. So he's talking about— and they measure it in tokens per second. So he's like to get up to 1,000 tokens per second, which you really need for these heavy, complex reasoning tasks, agentic tasks. They're going to use Groq and Groq allows them to do that. But you're selling, so you're producing more tokens, but like more tokens per prompt, I guess you would say, right? So if you ask AI to do something really complex, it's going to sit there and spin its wheels for a long time and then you charge a lot of money for that. And so the value of that long tail is really high. And so Groq even allows them to push that frontier further. So that you can do complex research and true agentic things that work, you know, the tokens or the chips are running for hours or days at a time. And so it's really efficient inference, and it's really high-value stuff. So that was— I thought that was cool. But I mean, in general, NVIDIA, they're, they're the best. It's funny, all of the I mean, it's lonely at the top is the old phrase. And I think everyone loves to punch up at NVIDIA, but they are the best in this industry. People are like, oh, NVIDIA is good for training but not for inference. I mean, that's like a 3-year-old story at this point, but it keeps coming up. And Jensen's like, look, inference is actually more difficult than training, and we're the best at that too, by a mile. And he gave lots of reasons why.

Well, Bryce, I've been saying on these calls and elsewhere for a very long time that the last year's best training chip is this year's most efficient slash best inference chip. Is that still true?

No, I wouldn't say it's the best. It's that last year's training chip, it is still a good inference chip, very good inference chip. But what they're doing now is like the frontier chips, Vera Rubin Ultra with Groq is gonna be the cutting edge GPU and it's gonna be the best at inference. So it's not like—

Oh, it's literally like first, its first purpose for existing is inference, not it's the best training chip that then eventually becomes an inference chip, right?

Yeah, it's purpose-built for inference. Um, and NVIDIA— I mean, Jensen said, you know, this is with— especially with OpenCL, the OpenCLaw, Claude code moment, um, we're at the inflection point of inference, and it's skyrocketing token generation, skyrocketing token demand, and the demand for compute. Like, you know, there, there is no— like, I think he's— I don't know if he said it exactly this way, but he's like, basically there is no limit on the demand for compute. If we can produce more chips, we would sell more chips.

And that actually follows along not just for their piece of it, which by the way, it's language processing. Language processing, okay. Yeah. So, and that like language, like large language model. So, but I'm saying like continue along from there, you know, this, the NVIDIA, the supplier, they could make more and it would sell, go where what it's used for, which is like selling tokens, you know, OpenAI, Anthropic, et cetera. And it's the same for them. Like the reason that those chips, you make more of them and they'll sell is because you make more tokens available and those will sell, that those sell out as well.

Yeah.

John, literally we're using Manus from which Meta bought in like November. And Bryce has gotten all of us addicted to using Manus as our agentic AI application. And I had him go through today and help me figure out how much— what's the cost per token or per— what, Bryce, what do they call it?

Not token, credit.

Yeah, the price per credit. And like, I had 1,500 credits based on Bryce's referral code to me to sign up for Manus when I signed up. Bryce, what was the project I assigned, or you assigned to it today, that cost?

Yeah, I had it just generate an invoice for our management company and send it via email to the person that needs to pay us.

Right, and so we did that, and you normally would have spent, you said, 15 minutes on it, and today you spent 15 seconds on it. So it was like— like an order of magnitude times 6 more effective for you. And then I said, what does that actually cost? How much per unit does that, per credit, did we spend for that 15 seconds? And you figured out it was, what did you say, 0.4 cents per credit? Yeah.

At least the current—

How many credits?

Zero that I'm on.

We spent 300 credits at 0.4, so it was 12 cents. Is that right, Bryce?

I thought it came out to like $2. I thought when I did the math, it was like 4 tenths of a penny and it used like 500.

Okay, which is $8 an hour.

Yeah, exactly. So we were trying to— that's literally what the next project for Matt is to do, is for it to like graphically and analytically explain to us at this current moment, snapshot in time, how much cheaper is it to use an AI to do a lot of your work than it is to use the, you know, whether it's a $50,000 or $100,000 or a $300,000 employee per year, what this kind of breakdown is.

And so we've been playing with that.

And it's really interesting to see exactly, John, that it's all about the per unit of energy, of credit that gets spent, and how it applies to opportunity cost of a human being.

See, I prefer a flat kind of, I'll pay this much per month, and then depending on how efficient I am with it, I can bring that cost down a lot. Just, I'll know what the maximum cost is.

Well, John, I like how you just did that because for 30 years, 25 years in my mind, I've been angry at like telecom and internet internet companies for shifting from per unit use to monthly cost. And it's because I feel like they, like, they should charge more to the people who use more and less to the people who use less.

But you're right that it actually—

it's the old show me the incentives, I'll show you the outcome. If the incentive— your cost is set. You want to get as much use out of it as possible, which, by the way, is terrible for the telecom companies providing that access to you. Now that I think about it.

Jensen said, I thought this was interesting, that in Silicon Valley, you're— every software engineer is given a token budget. Now, this is at least starting to happen. Like when you sign your compensation package, it's like, okay, I'm going to pay you $250,000 per year. And you get $100,000 per year in tokens to use on Cursor and everything. And it's like part of the perks of your job is like how many— like, does your job give you enough of a token budget to where you think you can be fully productive as a software engineer? And I had the thought using Manus somewhere along those lines, but it's like, I think employers will soon, as agents start to— agents start to like proliferate, proliferate. I think employers will soon be like, if you're caught doing something that an agent would do more efficiently, like, that's a fireable offense. It'd be like, like, like you ask your employee like to like do some numbers and you see them like with a, like, like on a piece of paper writing, like, what the hell are you doing? Like, use Excel, right? Like, it's just inefficient. And I think employee— pretty like agents are going to be much more efficient at so many things. Employers are like, why did you spend a week doing this?

Why didn't And this goes back to why if you are a good agent user, you will not lose your job because you will be so incredibly productive. It is the people that don't use agents well. And by the way, the only thing you need to do to use an agent well is communicate effectively, write a good prompt, or continually write good prompts is really the only barrier to entry. If you've got a knowledge base, I was arguing on my father's birthday dinner on Friday, March 13th, Friday the 13th, every 7 years, I guess. He was— everybody was lamenting. They were mocking people who get like philosophy degrees and like art history degrees because you can't ever make any money and you got a bunch of student loan. Like they had some anecdotes about it. My family did. And I was like, wait a minute. That's actually why I keep arguing those people are going to have jobs. The art history major who learns how to use AI effectively might actually contribute to society. They might actually provide a service and an application and some sort of like productive thing that actually helps society instead of just being able to like pontificate over a cocktail on a Friday night and not act and then go work as a, you know, manager of a Starbucks or something, which God bless managers of Starbucks. I don't mean to make— but if you spend $150,000 getting your art history degree and you've got $100,000 of student loan debt, it would be nice for you to actually make hundreds of thousands of dollars using that degree. And I think AI actually enables that. Over the next 5 to 10 years.

Yeah, because all those people use tech.

They all use tech. They all— it's in everyone's day lives. But imagine they're able to actually create using tech because they don't have the barriers to entry. You know, they can just go online and start coding stuff. And some of this stuff is niche.

It's not just about coding. You're going to create more writing, more everything about art history, about philosophy. Like, that's what I'm always saying is like, I don't know all the ideas that people in art history are going to have, but I do guarantee that AI will enable them to actually monetize those ideas.

I think two things from what you're saying. One, I agree that the knowing how to communicate with, you know, to prompt, to ask for things is a table stakes. Okay, so it's— you need that. It doesn't get you as like, oh, since you can do that, you're gonna go make millions. It's like, no, that's the starting point. There's a lot more you can know about how to be effective with, uh, AI and agents and things like that, even without programming. There's a lot more. So it's— that's a starting point. Um, so I, I agree with that. The other is, um, I— we've, I think, talked about this before, but, um, there's— there are those people who are the entrepreneur types, the creator types, the whatever, who come up with, uh, the, you know, new ways to, to do this or that thing. And then there's the rest of everybody who— they're not busy, uh, inventing from scratch. They're going to go work for somebody. So somebody else is going to do, uh, that kind of break new ground thinking, and then they can use their communication then to go fit in to get things done.

And exactly my point that, you know, like there'll be some incredible 25-year-old entrepreneur, AI-entrenched art historian who hires 5 other art historians and they use some sort of LLM or something else, or they build some application or they tour the country teaching something. It's— that's where the human creativity and human element, I think, really gets discounted by the people who are so sure that AI kills all the jobs. There's, there's an element of understanding and there's an element of life principles and just being a part of being alive that humans bring to the table, that especially creativity-wise, is very valuable.

And maybe less— sorry, Bryce— less creative, but one other thing is that people are going to be needed to be the check and balance for a long time. I think it was maybe Amazon just recently had a big outage, and it was an AI-based outage.

It was a war-based AI outage. The Iranians bombed a data center in the UAE owned by—

I'm thinking of a different one where they themselves pushed some code where it was their automated AI said, oh, here's a good thing to do, and basically downed production for a while. And it wasn't a great thing to do. And it was shortly after some layoffs. And it's like, you got to understand where people fit into the gate process of reviewing things, you know, getting done. And there's a lot of need for people on that and will be for a while. But on that end of it, Bryce, go ahead.

Yeah, 2 points. Um, first, just on the demand and how you can't picture like everything that will be done in the future with technology. I saw an advertisement for Febreze air fresheners that have a built-in microchip that like monitors the room and timing, and I, you know, optimizes. And I was just like, imagine if you told like Gordon Moore when he wrote Moore's Law '67 or whenever that was, like, transistors will be so cheap that you can put them in like a $2 air freshener that you throw away after a month or 2 months, you know, like no way he could.

And it literally is like a good— like for the supply, for the company, Febreze makes more money. Was it Procter Gamble probably owns Febreze? Yeah, somebody's making— shareholders make more money on an— on the margins and on the cash flow because they bought a 22-cent chip that they sold as part of a $3 product.

Yeah. And then the, the second point is Travis Kalanick has a startup that came out of stealth, and he wrote this paper, I forwarded it to the team, but I thought it was really good. But he actually believes like we do, unlike most tech people I've seen, that he's like, AI and physical AI robots is going to create like, so more jobs. Then it will destroy, if it does destroy some jobs. And he gave like a few examples, but he was just talking about like Waymo, how many people work at Waymo per car. And he's like, if you extrapolate that out for millions of robo-taxis, like there's still going to be like millions of people working in robo-taxi companies doing things to help facilitate how robo-taxis work and stuff. And It just like, he's like, you just don't know.

Where did you send that?

Let me just show people actually a couple of the quotes.

I look, I read it after you emailed.

Was it? Yeah. Email. Email.

It was email.

Yeah, the company is called Atomz. And it's gonna, the mission is like bringing AI to the physical world. And he's got a few ways he's going to do that.

You know what is annoying? And this is what AI solves.

Having to remember the channel in which you received something— was it email, Slack, or app? No, uh, my AI agent should go find that shit that I'm looking for. That's all I need.

All right, so now I gotta figure out where we're being—

and I guess I should say for the record that Travis Kalanick is the founder of Uber, if people don't know who he is, and he was forced out by the board.

And as you said, I'll share your email. Like, Travis is not the caliber of operator that Elon is, but he does think like Elon. And we should all read the paper. Bryce is right here, right now quoting this article. But Moore's Law squared is our friend, not Moore's Law, but the acceleration of Moore's Law.

That's a good point.

The cost per unit of intelligence is going down in price not by 50% every 2 years, but by 90% per year. Total capabilities in general intelligence have increased nearly 1,000x over the last 3 years. Less than what? What is that? 1,100 days, about 1,000 days created 1,000x increase in performance. Do you guys hear that? That is 1x per day. Like we're doubling per day like this. I can't— this is the exponential acceleration stuff. That we always talk about as one of our fundamental principles, but that we can never comprehend because we're human beings.

Not doubling per day on the, on the previous day.

Yeah. Fix that for me. I knew I messed up the—

that's one that they talk about where, you know, you pull up the chessboard and you're in infinity pretty quickly.

Um, right.

Yeah.

So it's not, it's not doubling per day. I knew even that. That's why I sort of tried to discount what I was saying, but it is, we, you know, we could figure out the CAGR growth that is 1,000x increase in performance over a 1,000-day period. It's incredible. It's mind-blowing. It's why I truly believe in my entire career of investing in revolutions and all of the hits I've had.

This is the greatest pitch.

These are the greatest pitches. This is the greatest moment in time. For an investor to be buying Revolution Investing stocks because we have this type of accelerating fundamental performance and we have stocks trading at like Meta trading below the S&P 500 multiple. I don't— we're not going to get that pitch in 3 years. We're not going to have that pitch in 3 months. This is your moment in time to do it. Bryce, I literally wrote down Kalanick robots quote as one of my notes here when we were all talking. I'm glad you could— yeah, got to that one. Another one I've got here that I think we should— you sort of hit on, or what, maybe, uh, Craig, you might have alluded to it, whatever it was a minute ago. Look, what's really— oh, it was John, that Nvidia— I don't remember who it was. It was Nvidia. If they can make more chips They would sell more chips if they could make more chips. It would be all more efficient. There would be like enough supply of inference if they can make more chips. Who is building chip factories? Well, Taiwan Semiconductor and Intel. And as of Saturday, we had actually heard this from Elon 3 months ago on an earnings report. But as of Saturday, He said they're breaking ground. What was— what did he say, like breaking ground in the next month on the Tesla TerraFab is what they're calling it?

He mentioned the 21st, but I don't know, it sounds—

all right, so there you go. Like, yeah, they're gonna break ground and start— like, Craig, you mentioned in our morning note that you send out to the team every day about notes that You're like, hey, I hope that Tesla can do similar to what Facebook and Amazon, Apple, Google have done, design their own chips. And as I wrote you back, it's actually so much bigger than that because they're not sending the chip designs, proprietary chip designs to Taiwan Semiconductor to build and maybe someday Intel to build. Elon's actually planning for Tesla to own the actual factory that makes the Tesla proprietary chips instead of having to get them from Taiwan Semiconductor or Intel.

What's the goal there?

They— he is pretty darn sure that over the next 2 to 3 years and even right now, like we're talking about, the biggest constraint in the AI revolution is chip manufacturing. If we can make more chips, there'd be more compute.

So he doesn't want to kind of go hat in hand to anybody.

It always— Elon is picturing the number one vertical integration of whatever business he's in, down to like Tesla mining lithium so that they can put it in the batteries that they put in their cars. True vertical integration in that way. But specifically in this case, It's about the— it's addressing what is and will be the biggest constraint. Where is the bottleneck to make robotics go to a billion units of Optimus being sold?

Have you heard of the chips?

Yeah. Have you heard of the story of the first McDonald's in Soviet Union or something like that?

Sure.

Because it was like they went to do it and they could not find good enough potatoes to go make their, like, french fries. So they had to literally set up their own farms to be vertically integrated on that.

So great analogy, John. Yes, exactly. This is a potato factory.

It's a potato farm. Yeah, that farm of a— yeah, something like that.

Yeah, so I mean, and so that's where, you know, Elon's the only one who dreams that big. He and Jensen Huang is, you know, arguably as good a visionary and executor as Elon is. But he doesn't— like, the one thing nobody's going to argue is Elon's got the biggest balls on wall in business. Like, the guy is scared of nothing, and he's got the huevos to go out and be like, I'm building my own fabs, MFs. You guys are all scared. You all are addicted to Taiwan Semiconductor and you're praying Intel figures this out. And I will tell you, if there were anyone, if there were anyone outside of Taiwan who could figure it out, Bryce, who would it be?

I put my money on Elon. You know, on that note, Jensen said that They're actually selling a lot of CPUs on a standalone basis, and that hit Intel. And I haven't checked AMD stock, but, um, Jensen laughed. He's like, we never thought we'd be in the CPU business, but we're selling a hell of a lot of CPUs, and I think this is going to be like a multi-billion dollar business for us.

Basically, they're selling as many CPUs as they can build, just like they are selling every GPU they can build.

Yeah. But Nvidia, like, I wrote, I don't know, 2 years ago, I did, it's a platform company, not a product. And as time goes on, I think that just becomes ever more evident that that's the case. And, you know, even Jensen argued today, he's like, hey, we're vertically integrated. And they don't build their own chips, they design chips. But they build so much software, they help their customers design the data centers and server racks. And like, even today, he unveiled the system where they— all of their— all of the supply chain that goes into building and operating a data center, that's all going to be like on NVIDIA software too. So like, designing the, designing the equipment, integrating it, and then like operating the data— like you're— you build a digital twin of your data center, which is just a virtual representation of the thing. and your contractors can use that to— this is all changing. I don't know why this is— all this is changing, but, um, like, they do— I guess the point is just that NVIDIA just does a lot more than like, hey, here's a chip, hope it works for you, you know, we think it's really good, hope it works. But like, they're even designing the software to help people optimize the electricity usage so that they can maximize the revenue from their AI data center that they build. They design software that helps them build the data center, to hook it up to the electricity, to hire subcontractors, like everything in the stack other than the fab itself Nvidia is doing. And he's— I think— I don't think Nvidia gets credit for being that sort of an AI platform.

So I saw a documentary about the making of a fab, and I think it was about the one happening in Arizona, North Texas, uh, and, uh, Samsung or something. And they're talking about how it, like, a train passing 20 miles away has effect on the vibration of stuff. Then so basically they made the whole building like float on, um, like shock absorbers, like thousands of them that are all individual pylons going in and, and controlled and everything. And just like the, this, the person putting this thing out said, it is so incredibly difficult to get right. And I mean, yeah, you put your money on Elon, just know that what he's on there is like really hard thing to do.

It'd be like trying to imagine someone blasting a rocket ship the size of the Statue of Liberty and landing it in chopsticks.

Yeah, it'd be like that. Like, nobody's ever going to do that, so never happened.

Yeah, um, Bryce, the other—

the only—

the last note I've got on my 6 notes here is my question is, did Elon— I mean, did Jensen and Nvidia re— like, punctuate the idea that their lead is getting bigger? Do you believe that is true?

He said this before, but he's like, if our competitors' products were free, they would still be a subpar option for an operator.

That's like shit talking right there.

Yeah, he's like— and he— because his point is that, look, it's all about revenue per watt, which we've— we mentioned earlier. And he's like, our token production is so much higher, um, and tokens translate directly to revenue, that you couldn't— on total cost of ownership basis, doesn't matter if the initial chip was free, it would not be, um, profit, you know, on a comparison, comparative basis, would not be a better choice than Nvidia.

That's mind-blowing.

Yeah. And like, I guess, I, you know, you think about like, we do— my family does construction, and it's like If you're, you know, moving dirt, right, you get paid per yard of dirt you move. Like, if you have a tractor, like, what, Caterpillar can move 10,000 yards of dirt a day. Like, it doesn't matter how much the competitors can charge for their piece of equipment if it can only move 100 yards of dirt per day. They're just— there is no—

you'd have to have 1,000 of those units to compete against one of the Caterpillars, right? Yeah, it's mind-blowing.

So you got like these, uh, units of measure that are becoming really the thing. And at the data center level, it's now measured in terms of electricity. It's like energy. That's a 100-gigawatt, uh, data center. And then for the AI folks and everything, instead of it being, you know, AI calls or users, whatever, it's, it's this token thing, you know. It's how many tokens, you know, it's crazy.

Is we proudly do not invest in commodity companies. We do not invest in commodities. We avoid gold, oil, electricity, anything that is purely commoditized. And in some sense, on a very fundamental level, AI inference is a pure commodity. But the thing is that There's only one platform, one company that enables you to generate any meaningful return on that commodity of inference token generation at all. And that company is NVIDIA.

Yeah.

He said, you know, he predicts that NVIDIA will have at least $1 trillion in revenue over the next 2 years, which is huge. But he's like, If you're going to spend $1 trillion on our stuff, you've got to be pretty confident that you're going to generate a return. So he's like, the rest of this presentation is me showing you how you will generate a return on a trillion. Wow.

You know, in the end, it truly is always about the economics.

I mean, this is—

we've been talking a lot for the last few months at our 10,000 Days Capital Management company about the idea that we, like Elon, want to inspire to be by Elon. At least we want to operate on first principles of economics and investing. And that's what— that's what Jensen is getting at there, is that on a very fundamental principles level, the first principles themselves being electricity generation on a per unit of intelligence is so much better on theirs, probably 50 to 100x better performance than the free guy AMD. Which literally AMD is free because they go to ChatGPT and they're like, hey, we will— if you will buy $100 billion of our chips over the next 5 years, we will give you an equal amount of our equity.

Right.

That's how bad the AMD chip is compared to the Nvidia chip. Nvidia might give you cash to take some of your equity But NVIDIA is not giving away their equity to anybody under any circumstance.

So you're saying they're Webvan?

Back up the truck. It's time to buy some more Webvan.

Remember Webvan? It's like where they would pay you to use them. And so that is just thought like—

Craig and Bryce, you guys do not know this reference, but in the year 1999, 2000, And I'm not even sure John knows this part of the Webvan reference. Webvan was a home delivery company that, like what Amazon does now with first-day delivery and groceries, but they literally were like, you know, like losing an extra 50 cents for every dollar of revenue they generated. And it just was an endless pit of losses. And the point that I remember is I actually remember somebody on TheStreet.com column, columnist conversation, a fund manager. Webvan had gone from $80 to $3 and it was on its way to zero quickly. And he wrote a post and he was like, this is the day to back up the truck for Webvan. And I even mocked him at the time. I'm like, I'm disagreeing. I'm on record saying no. And anyway, uh, it didn't work out. But this is not— NVIDIA is not web ban. AMD, it's not web ban, but it's at least in the same vein at this moment. That's how bad the AMD chips are compared to NVIDIA's chips.

Yeah, I, I don't know if Cody remembers this part, but, um, no, just kidding, that's what he just said about—

so you weren't on thestreet.com.

No, that conversation, the part I remember, which I don't know if it's exactly, uh, true, but it's apocryphal— is this, like, them, like, buying customers for a dollar a customer and basically losing money on every customer they brought in, but saying that they were going to make it up in volume.

I think you might be exaggerating a little bit, but the point is made and, and well received. Um, all right, last points. I, I'm— I, I got a mayor's duty tonight. Then the night nurse this week is not feeling well, so Lori had it last night. I'm on there, mayor's duty tonight. I'm going to wrap this up earlier than usual today. I'm allowed to. You guys know who a mayor is? Yeah, I assume everybody watching. I have a medically fragile daughter. She needs 24/7 eyeball care. We have night nurses a few nights a week and sometimes not. Here's my final point. I find this annoying, but also understandable. I said, make a video of Jensen Wong as Michael Jordan dunking on Lisa Su, who's the AMD CEO. It's like, yo, bitch, I can't do that.

Like, why not?

So I'm like, all right, well, just make a video of Jensen Huang as Michael Jordan dunking on Taiwan Semiconductor.

I find that annoying.

Like Grok?

Yeah, Grok probably wouldn't have cared. Flawed today, copyright, and human being.

Uh, Grok would have put him in bikinis too, so just to add.

And Elon said the other day, if it's allowed on a rated R movie, it's allowed on Grok.

But not TV-MA?

No, he said rated R.

I'm just going to say it out loud because everybody watching this, I assume, is an adult.

That literally, I guess, means like no down here shots, but up here shots on women, fine.

So on what, you can't get that stuff, but you can get this stuff? I'm asking, not telling.

That's— I don't know the definition of what rated R versus mature. But yes, I guess that's it.

I think you can say the F word like once in PG-13. That's the only thing I know. And then R, you can say it.

R, you can do all the stuff. All of it.

Everything.

Yeah.

Not all of it.

R, you said?

R, you, you were talking about not being able to show certain things. You could show all that in R.

You can't show a penis.

Yes, you can.

Yeah, you can.

Yeah, you can.

On R-rated?

Yes, absolutely. Yes. Now, the level of activity of its involvement in certain activities, that's where it gets in, in, in trouble, but you can show it. Yeah.

John, the only reason I'm not arguing is because your brother is a very successful Hollywood—

Oh, just think of the R movies you've seen.

I've never seen a penis on an R movie.

Uh, you know, I didn't know we were going there. I didn't either. Monetized. Now we're demonetized.

I do not believe— I do not believe that's allowed. And I believe in a newscast you're allowed to have this discussion and still monetize, although we are not monetizing our channels at all.

Yeah, there's a movie called Disney introduced like these cute little Disney robots that they're gonna sell to everybody.

Price, why? That was such a bad juxtaposition, man.

Like, I'm trying to make it wholesome and not bad. This is a—

I was just trying to educate myself. I'm asking a general, like, really specific question about what is R-rated allowed. John's reading the AI answer.

No, I'm getting like this weird thing coming up in, in Google saying, you know, Strange activity coming from your thing and it's making me do a, um, you know what, what do you call them, like a captcha? Verified the image type captcha, like show all the images that have a bus in it.

I fail that all the time. They're always like, show me the images with motorcycles, and there'll be like a wheel from a motorcycle on one of the images, and they're like, no, that didn't count.

By the way, the answer was yes, you can. So no, I still don't believe it.

Yeah, yeah, there's a movie called The Piano from 1993, Harvey Keitel. You can go see his—

there's lots of movies.

I do not believe this. I'm going with it. You guys can't make me, uh— you just said if we went and checked into it and, uh, it was AI hallucinations.

No, I've seen the movie.

All right, can we change topics? All of us are dying here.

You're the one who started it.

I didn't mean to.

Oh, okay then.

All right, last words.

I'll go. So I use Manus today to— I created the website for our fund in Replit. But I needed to port that over and have it in something hosted on Manus, but through GitHub. And that would have been like an all-day project for me. And I just sick Manus on it. And it was done within like 45 minutes.

And while you were doing other stuff, Of course.

Yeah.

So I was doing other stuff, sending emails, getting all my other stuff ready.

So now I'm sorry, who owns MANAS?

Meta.

Meta, right?

I am more excited about the AI revolution than ever. And I was very excited about it to begin with. But, um, I just— I think it's becoming more real every day. Companies integrating it across the board. Um, I just— I get a little tired of the, like, so much of the bubble talk and the short sellers and all that kind of stuff. It's just like, I don't think they actually use the product. If they did, if they actually followed how companies are using it, I think they would be far less skeptical. They would be more objective. Like, The lack of using the product just makes them tainted and not objective. Right.

I totally agree. You know, Michael Burry, the short seller from what's the movie?

The Big Short.

The Big Short.

And, you know, like, he's been proudly bearish and calling AI and basically the entire economy a bubble for months now. And I guess he tweeted for a while and then didn't. And then 6 months ago, he closed his fund and went like retail. Marketing for a newsletter product thing, which, you know, more power to him, whatever. But I quit following him this weekend. He had just one too many mindless trolling, like just droning about what— like, I literally— I got about— it was a long-ass paragraph too, and I got about a third of the way through, and I just like, you know what, I don't want this in my head anymore. I, I like to read Jim Chanos. I like to read people who analyze. And, you know, I don't even care if you're a permabear, but I really do feel like a lot of the, the mindlessness about the AI is a bubble and stupid and isn't going to help, or it's going to destroy jobs also while it doesn't help, or whatever it all is.

I just— I'm done with it.

It's here, it's working, it's, it's real. Johnny, last words?

So last week I showed an example of, um, some comics I made to show what's possible with generating images for AI. And I don't know if this shows up here. Can you make that show?

There you go.

There's an illustration I made, and it's like, find the typo, because there's like one. Or no, there might—

whatever.

It's very, very little. So this is what's possible now. I described in what I wanted to communicate about AI and AI-native businesses. And I got this very detailed— takes a bit of time to talk through what's going on here— type illustration totally made by—

and John, 6 months ago this would have been bad, and 2 years ago it wouldn't have been— like, it would just have like foreign, like made-up alien symbols for words and stuff, right?

This is where we've gone in that in that short amount of time.

That's why Travis Kalanick is saying we've got a 1,000x improvement in 1,000 days, because this exists now.

It's incredible. It is very good.

All right, I guess my last word—

let me share this again.

Um, stop screen.

All right, now I'm sharing my screen.

I took, uh, some pictures that we made, and I was like, hey, um, turn the picture— this picture into a horror movie.

See how, like, this was at my land where it got burned in a forest fire, and I thought this looked really creepy, like a devil or something. I thought it'd be scary in a movie. And I made this video, which actually I guess that's just a picture, which is all right. I wanted it to like make this the mouth. It couldn't quite do it. See, and it's not that good.

Um, but I did think this was even—

this is cool.

We took another image.

Oh, this stuff isn't going to load right now. What is going on?

Do you have to upload it first or anything to make it work in this?

Yeah, I don't know.

Environment.

There we go. A yucca plant I found on my property picture and told Tamika.

Oh, crazy.

Isn't that scary looking?

That's intense.

Yeah.

Here's another one. Hey, let me stop that one. Let's see if I can make it look— there you go, better, more, more visualization. This is the one it made before. It's also scary.

Like, isn't that creepy? It's awesome.

Here's one, it's got music. I don't know, I just— I think this stuff, like, I can't imagine what we're about to unleash with the world's creativity and just random people being able to make incredible movies. And it's not there yet. I mean, that stuff still— like, I had to, you know, I burned all my tokens. I— Bryce, what am I on, like a $200 a month Gemini subscription? Yeah, like, I'm paying up for that thing. I made 6 or 7 of those videos in about 10, 15 minutes. And it was like, you're done till 3 o'clock this afternoon. And my point is, is it has been many times in this call. My last word is until you can make an hour-long video like that, that I just showed you off of a single prompt, there's not enough inference. And we are probably 100 or 1,000x away from that being reality.

Or next month.

Let's go 1,000 more days, John. 3 years.

So it's the other thing to say with AI is that was super cool and that's the worst it'll ever be.

Yeah, that's correct. All right, thanks guys. Do you guys do last words when I'm not here, or am I the only guy who makes this?

Yeah, yeah, yeah, yeah. Okay, you're doing it.

I like it. I think it's a stylistic thing, Bryce, trying to keep consistency for the marketing purposes.

There you go, see?

Yeah, yeah. All right, here we go. Watch the last time you're going to see this particular outro to our video.