The AI Bubble Question

Hello and welcome everybody to The Attention Mechanism. My name is Justin Robert Young, joined as always by Andrew Mayne. How you doing, buddy? Hey, Justin. Great to see you. Now, this podcast is not always... show up in your feed. But when it does, it is here with purpose and it is here with clarity. There are a lot of conversations that we've had off mic because we work together a lot on a bunch of different stuff.

The biggest topic that I think from the general public is probably just chat GPT. It being three years old. That's how they interact with it. But a lot of the business press, especially last week.

was talking about the bubble. The AI bubble. Now, some of that seems to have been quelled because NVIDIA had strong earnings, which I guess shows you how much the hype cycle really matters. But we've had a lot of conversations about what that actually means. And when people compare the... current AI industry to, let's say, the 90s dot-com crash or the Great Recession with Lehman Brothers and the subprime mortgages. How-

You think about that how we think about that. So I'll direct it to you. Are we in a bubble? I... Every situation is different than the previous one. And also when we say – we have to say what is the what and what is the bubble, okay? When you said the –

You know, the subprime mortgage crisis, you'd say, OK, what was the bubble? Well, we had propped up prices on houses really, really high and and found out that, you know, it wasn't able to sustain people's ability to pay mortgages or rent them. Right. And so you'd say there was there was. Yeah.

pricing, you know, you'd say was there was was not like an oversupply of houses, but it was like, hey, we've said these things are worth way more than people have incomes to support them, what have you. And utility of a second home maybe is not people what they thought. In the dot-com bubble, you had several things. You had, on one end, you had companies that were... Not even profitable, like not even profitable that were pre-even really any kind of revenue that were being traded at.

you know, huge amounts. And there's been the criticisms of the defense of like the pets.com, like, well, you know, Amazon prove pets.com. Right. I'm like, well, funny. There was a company called Amazon at the time as pets.com that actually said, no, we have a better. They wound up doing good.

Yeah, they're like, hey, logistics and also make sure you make a profit on the things you ship. And I would say that was the argument against test.com was that like literally shipping you dog food when it's really expensive, ship it. And I'd say that was the idea that people.

Pet owners are going to spend so much more. And it's like, nah, you could – people are like, well, it turned out – I'm like, well, no. Amazon was already there doing its thing and also said, man, that seems stupid. We're not going to do that.

And I think that you get the other side, you have like Cisco, which is building out all this dark fiber, all the fiber, all the fiber optic. Right. Which was, you know, the big story was that, you know, who's going to use it? And this is in a pre Netflix world, a pre YouTube world. And the funny thing about.

Any type of internet, you know, and that rolled after 6 p.m. Not super useful. And even during the day, it wasn't that useful with the that one of the big drivers for usage of that what ended up being, you know, streaming video YouTube. the iPhone, et cetera, what have you, okay? Now we're in a world where you have companies like OpenAI and Meta, XAI and Anthropic who are like,

Yeah, we need compute. We really, really need compute. And we're seeing record revenues between OpenAI and Anthropic and what they're doing as far as the API usage and whatnot. they're compute restrained. And people are like, might be a bubble. I'm like, well.

I can tell you from the compute demand part, it's not. It is literally there's not enough capacity. And at OpenAI, you know, with GPT-5, you know, the GPT-5 model, they went for reasoning. They probably would have loved to have, like, you know, done a bigger trend.

They've done more on this, whatever. But they also found out for the consumer market, people were happy to 4.0. There's this weird place where what consumers want versus what coders want when it comes to high-end models are different. And they said, let's kind of do a consumer focus on this.

and hopefully more compute will come along. And, you know, they build out the compute they're building for. So I look at it, the people I know, and I'm friends with the people who are like, you know, running Stargate and like trying to build compute. And they're looking at going like, there's...

We don't know where the processes are. The energy is going to come from because there's just not enough because they look at the demand curve. They look at every time there's an improvement in adoption rate. There's a limit there. And it wasn't like I'm like point to me to the person in 2001 who was like, I need more fiber. I need more fiber. I need more fiber. You know, we could see at some point.

that there would be an advantage when we got to streaming video and stuff and do that. But their demand wasn't as present as it is now. And also... So if we want to talk about data center, let's separate all this up. If we want to talk about pure data centers or data center bubble, I absolutely do not think so because I know the people who – we do the podcast, The Opening Eye. What's the one thing that keeps coming up? Compute restriction, compute restriction.

I also know the revenue goes up. Anthropics is a similar story. So I don't think there's a data center bubble at all. Do I think that there might be some players that might be building or build out data centers and don't know what to do with them? Maybe, but. I can run an agentic system inside of a data set. You give me a gigawatt of compute. I can make use of this. Yeah. And that is not going to, at least if you look at the demand right now.

That is unlikely to change. There seems to be at least a market demand for more and more of this stuff. And, you know, in the stuff that we've even been in the room for with Sam Altman and other people at OpenAI. They say, look, we have two gigawatts now. We serve everything on two. Everything you know about OpenAI is served on two gigawatts. If we went to 10.

We believe we'd be capped out within a year or so. Yeah. Because the demand is so much higher. Yeah. Like they would saturate it very, very quickly. And they are planning. I mean, because they're building all this. They are planning on saturating that. very, very quickly because not only of the demand, but also what they're going to offer, which they feel will be even more compelling to people that will pay even more of a premium for. And the thing that I'm looking for.

So one of the ways that I try to find truth is find what's the consistent story. If this is true, what else is true? Are these things true? OpenAI, the other AI companies, you know, the people we talk to says, hey, we need more compute. We need more compute. We know when they get more compute.

They do more stuff. Things come out. I don't think they're making that up. I don't think that they're making up that they want more compute and also that they feel resource limited by the amount of cue when it comes to inference. So I believe that story. So then there's the narrative of, but then you get the other side of it, which is people like, ah, it's hype.

You know, like, you know, oh, MIT released a study. And I have a friend that has a startup that just used the study because it reinforces the value of her company. But it's a BS study about like 95% of all AI projects fail at companies, whatever. But you get that, meh, AI is not working out. Yeah, okay. then why am I seeing it's harder for entry-level computer programmers and copywriters to get jobs? If you're saying it's hype.

Yeah. Why are we seeing? We are seeing disruptions. And, you know, you and I are the biggest optimist in the long term about AI employment. But we do know there's disruptions. Why are we seeing this disruptions? Why are we seeing that?

You know, I just saw a report that reportedly McKinsey's frozen like new job offer, what they're offering, you know, new people coming in because they're facing competition from all sorts of things, including companies using their own AI tools, et cetera. And so either you're telling me that's not true.

Nobody's ever – no sectors are being faced to have to transition because of AI, or it is. And if it is true, then that means that we're having to begin to address the demand. So I would say that if you think it's a bubble – But also – Yeah, but also all of these companies are just passing a promise to pay each other later. And so all of these press releases that come out between NVIDIA and OpenAI and XAI, they are all just a circular.

a value chain where no money is actually being exchanged. They are just saying that money will eventually be exchanged. Well, I mean, yeah. And it's funny because part of what's happened is that. A great example of this was years ago.

So Nolan Bushnell, who I think is an amazing visionary, this is the guy that started Atari and Chuck E. Cheese, visited a friend who went with some computer lab and they showed them like the first, you know, computer video game called Space Wars. And he looked at Space Wars. This is really cool. is it cost? He's like $20,000, like 20K in 1960s dollars. He's like, huh, let me know when the price falls to a certain point.

And then one day somebody says, hey, it's cheaper. We found this new processor. We can build this cheaper. And he's like, cool. And you get Atari. And he created Atari, right? And so he's building Atari cabinets and hiring every young person he can, including a barefoot. you know, fruitarian guy named Steve Jobs. They had to put on the late shift because he smelled so bad because he refused to bathe. Smell too bad. Yeah, you know, that was the height, but Steve, but no.

Nolan Bushnell realized, he said, you know, as he's watching these arcade cabinets get shipped out of the warehouse and off to taverns and bars and places like malls and kids were putting, you know, thousands of dollars, millions of dollars and quarters into it. He didn't get any of that revenue.

He sold the cabinet once and he didn't get any of that revenue. Right. And every 25 cents that went into a cabinet he didn't get, didn't matter how popular the game was, he could sell more cabinets. He said, you know. I need to be at both parts of that market. Yeah. And then he sold the company to Time Warner, to Warner Communications. But, you know, it's a different story. But the point was he said, oh.

I really want to be in both parts. I just don't want that to be the go out the door. And now you get companies in video like, oh, so you want compute. All right, we'll sell you the compute. And we'll loan you the money for the compute.

But we're going to want a piece of every quarter that goes into that compute because we think that's where the market is. And to me, it's like, oh, let's circle. Like, no, it's very smart economics if you really believe that's where the future is, because NVIDIA has a very healthy business. NVIDIA is making. money and nvidia is not trying to trade future business to prop up current revenues nvidia is literally saying we think there's a future that may accelerate very quickly and also like

Who knows? Maybe at some point you're going to see wafer scale compute or some other thing that can be competitive in video. So they're trying to diversify their speed running what other companies like Microsoft have done.

which is to try to get into stuff. You know, one of the things that Apple did, you know, everybody knows what Apple is. Very few people know what Blackburn is. And Blackburn was the investment group they first formed in like Nevada, which was to take when they were getting, you know, billions of dollars of cash flow to reinvest and whatnot.

And it's, you know, you look at Apple and you go like, well, there's the Apple. There's when they sell you the iPhone. There's also what all their investments are doing around the world and God knows where. And so I'd say that if you say that, hey, I believe in the sectors a lot, then you're going to make these kinds of investments. And people want to say like, well, it's Enron, but not quite Enron. And like NVIDIA had to come out and say, no, we're not having Enron.

You know, like, like, like, come on, guys, let's be. We're selling a popular product that we can't keep on the show. Yeah. And we are. And it's it's so in demand that people are willing to make. deals where if it speeds them getting this product that they are willing to give up something on the back end. Yeah. And yeah, it's and they yeah, they want they want to participate in the upside. You know, the idea that if they own a piece of the upside, it means that when the chips leave the.

or they still continue to make money from it, which I think is a smart move. So on the chip side and the data center side, like, yeah, I think there might be areas where people may have overpaid or misallocated or whatever. But we're talking about the actual demand for this stuff. Like, no, the data, there are things like even, you know, Michael.

breweries, you know, you know, his, he, one, he pulled back his investment from NVIDIA as I understand it, but then two are his hedge against NVIDIA rather. This, this, this is, this is the, the Christian Bale character in the big short. The guy that was, that was in.

Yeah. Yeah. A bear on everything. Like, well, he said this like, yes. And he said that about the last six rallies, too, which if you had if you did what he said. And I think he got hurt. I stand corrected if I'm wrong, but I think he got hurt very bad by Tesla, whatever.

He did the housing crisis, absolutely figured that out, but a lot of other times not so much. But his analysis, I found like, no, I think these things last long in this, but I don't know. It's not my job to do this. I just know where the demand is. Now, if you ask me. Do I think every AI company is going to be successful? Do I think every AI startup?

The answer is no. I clearly know. I've seen really brilliant people with companies I think had great products that have been acqui-hired. You know, we've seen this happen. And there have been some very high value acquisitions and there have been some other stuff where the company just couldn't be. revenue.

couldn't do their their series B or series C and just ended up, you know, folding. And so that is going to happen. And we're going to get to get high. I think we're going to get some very large scale, not open air anthropic because those guys are making like tremendous revenues.

a solid customer basis, but some of the upstart competitors, things like this, I think you're going to see that they're going to find it harder to raise unless they're able to show that they have a significant amount of revenue coming in. Um, I think that'll happen. I, but I also think that there's also going to be, it's funny cause you know, being in venture capital now, it's funny cause like, I'm like, yeah, like I think that, I think a lot of the bigger dinosaurs are going to have problems.

I think that a lot of the nimbler ones are going to come in and have a huge opportunity. And I get this thing like, what if we're in a bubble? Like, well, if you're in a bubble, I know from somebody who runs a consulting firm that does automation and gentic stuff and efficiencies.

That's when I get the most calls. This is the paradox where people say that we're facing. There may be an AI bubble. There's an AI bubble. These companies are going to be in trouble. They're going to have to use something to solve this problem. They may have to buy a lot of AI tools to... Yeah. It's all for this. It's like, well, okay. So, but yeah, I am-

I think that when it comes to the data center situation, I don't think we're building anywhere near the amount of commute that we can. It's not – it's not – You know, it's different. Again, economics are always the same. You never want to go. I remember in the late in the 1990s, you know, when we saw these insane valuations, you know, there was these justifications like it's the new economy.

new. We just trade on perpetual value in the future. It's like, well, eventually that comes due. Eventually you got to ship something out the door. Eventually you have to show something works or else you're just going to want to put your money elsewhere. But, you know, I think, like I said, the data setting is fine. I think that as far as valuations, you know, I ask people this, like.

You know, I have a document here as pulling this through because you ask people like, OK, what was the P&E ratio, you know, of, you know, of. You know, one of the three – 3Com was one of the top companies. Like a huge – you know, this P&E ratio looked nothing like NVIDIA's. You know? Yeah. It was absurdly high.

There's one or two companies right now that are publicly traded that I think have very, very ambitious profit-to-earnings ratios that I'm not confident that they're going to be able to – some that I think are actually just – slightly more sophisticated consulting firms, masquerading as technology firms, but we'll see.

Well, then the other big thing is that I think that you made a very good point that we've talked about, which is that if you look at the big players here, the ones that are investing and moving this stuff forward, their profit to earnings ratio are not. Crazy. And also they have other businesses that they are diversifying into this.

And aside from Microsoft, which would be the publicly traded proxy of open AI, the one big exception that people would say is open AI. And open AI is burning so much money. They have a gigantic. a hole that they are investing a lot of money in. And while they are bringing in a lot of money on API, they're bringing in a lot of money on partnerships. rumors over the weekend that ads may be coming to free chat GPT, which we'll see what that does to their bottom line. The reality of...

Them investing a lot in their future is there, and the critics' argument is they are burning money so fast, eventually they are going to hit a wall. They are competing with forces that they cannot. Hope to match. Google can outspend them and they will eventually Icarus and fall to Earth. Obviously, as you know, Justin is a producer and I'm the host of the opening podcast, the official opening podcast.

We are going to come from a certain point of view on this, you know. Yes. Please factor in that into your shill of meter as we go on. But I would also say that, you know, as somebody whose fortunes are tied to many different things. now um and were before even when I was there uh um I I am convinced that

Last week was the third year anniversary of ChatGPT. The third year anniversary of ChatGPT, right? The 30th year anniversary. The 30 years that – third. You're saying third. Yeah, it feels like been around for three years. There is. I know feels like 30. Yeah, it does. Yeah. So it was three years. Not a billion active users. Three years. I was three years ago. I'm interested. I am at the the Mexican restaurant in Burbank that Rick and the.

But Rick and Morty is like they based the Mexican restaurant that they go to there on. Talking to some friends because my wife and I are getting ready. The next day we're moving to the Bay Area because I've been working at opening. I wanted to be closer to it. I remember telling people, hey.

Don't want to say too much, but there's a really cool thing coming out tomorrow. Really cool thing coming out tomorrow, you know, and nobody cared. Nobody cared. Nobody cared. You know, kind of day of nobody cared. It's like, ah, that's cool. And then, you know, I having been having been somebody who's part of my job was to try to figure out how do we help people adopt this? How do we people learn how to use this? How do we make this happen?

The mistake that I made and everybody else made in thinking was it's the most sophisticated tool in the world. How do we adopt? How do we figure this out? Well, it turns out the tool talks to you. The tool tells you what to do. The tool, you know, can you do this? I can try. Sometimes it works. Sometimes it doesn't. And you go, oh. And that was the mistake I think everybody made was worried about how would the individual adapt. Turned out individuals adapted quite well.

They took to it second nature. They understand these things have limitations in this, whatnot. There have certainly been some outside problems, some outside things, but overall, close to a billion people using this tool. Amazing. The cost of medical information, and by the way,

Chat GPT, when compared to doctors, gives higher quality information now than the average doctor. OK, you may have a great doctor, but the average doctor better than that. First time in history I'm aware of the cost of medical knowledge went down year over year.

Right. So now you're at a point where you have a question. You ask chat GPT, you go into a hospital. Guess what? Your doctor's asking chat GPT. Everybody's doing this sort of stuff. Right. So three years in adoption accelerated way beyond. I thought it would be industries having troubles, which we'll talk about in a second.

industries have a trouble because companies are doing things over several years that have a lot of time to adjust been hard and i would say that's why you get layoffs and stuff like this but you know you're seeing a lot of this so now The rumors have it that OpenAI may be doing ads. They may be looking at ad revenue and whatnot. I think smart. They have a free version of it.

Pay for the free version, you know, do a wall, make a really clear wall, make sure the model's not being manipulated and that's going to be clear. They're going to have to figure out a way to navigate that to maintain respect. But touch upon a thing and then like, you know, are they going to crash out?

I think that when you can make your product continuously better and you find a way to do it, it's a great way to do it. If you can do it through ads without compromising the product, that's great. Google had a really good run. before it started to suck as far as a search product, okay?

Still a useful search product, not as useful. And also, you know, the Google, the problem. Frustrating. Yeah, it is. It is. It is a degraded experience. And you can't pay for a better experience. That's the problem is that I am. I am very much a believer in. Ads are fine, but also give me a way to put a dollar value on it.

and not have to deal with ads and to know that I'm paying for money to make my product better and not to pay your advertising team to do it. So I think that they add ads. By the way, let me also say, because we often beat up on Google on this, Amazon too. Dude, the amount of time that I have to scroll or dodge the sponsored thing for stuff on Amazon has just become...

Like, give me the things that people are paying for. That's fine. I understand. Put them in line with the stuff that I'm looking at. But just for the love of all that is holy on this, the holy season, can we please just let me.

see what the the projector to make it look like it snows on my house is without 90 000 different sponsored versions yeah we'll touch about that and that's probably gonna run to but there was amazon found out that they could make a huge amount from advertising they looked at like when you saw And that was the, you know, it's funny because I remember we talked to, I don't know, somebody who worked with Tech TV or years ago or The Week in Tech, not somebody there, but one of their pundits.

And he was arguing how Amazon would have to open up stores. I got an argument. I'm like, they're not. I mean, they may open like token stores, but Amazon's plan is not to open up large brick and mortar stuff.

They have stuff, they've done experiments in retail and they've been terrible. And he's like, well, people have to know. And I'm like, it's advertising. And now Amazon's like, yes, not only will we buy advertising, we will sell advertising and we will use every square inch we can. And so I'd say that's sort of. pathway there. But anyhow, the point is, I said that OpenAI has...

you know, opened up these different revenue streams, you know, from different subscription tiers for the models. They opened up when they made ChatGPT available in India. They came up with a discounted version of it to sell in India. There is a large number of consumers, though, that are reticent to pay for anything because of just how they view it, what have you. It is the reason that... You know, Amazon sells has prime. You have that in places like India. They have Amazon.

Uh, they have their own free, free streaming platform there and free content there. You know, we talked about fast channels are, are huge. Yeah. And not just here. Yeah. There, but yeah, Amazon has a fast channel and I think they're going to bring one here and that's, we.

forget here that how much they have one. Yeah. They have free V or whatever they're calling it. Yeah. And that's now being folded into prime, but that was originally this weird standalone thing. And so now we're getting this ruled of like, there's a lot of consumers. So I think that makes sense because there's certain consumers like you can.

You could make it 50 cents and they still won't be bothered to pay for it. And it's like, cool, sell advertising to them. As far as OpenAI, what do I think about OpenAI? I think that-

We've seen open eyes revised upwards their profit projections within the last year. You know, they start at the beginning of the year thinking, hey, we think we're going to make this much profit. We think that we're going to have this much written on profit revenue. And those revenue numbers have climbed considerably. They came out with finally.

very highly competitive code product with Codex. And this summer, just this summer, and we've talked to people, worked on it, and talked about the numbers they're seeing through there are incredible. Now, that's great and all. But, you know, if you're saying you're going to make, you know, 100 billion a year in two years time or whatever the point at which they do it, but you're spending, you know, 500 billion dollars on compute. Yeah. It sounds like, you know, well, there are.

So roughly one of three scenarios that happens. One is usage levels off, revenues level off, and there's this huge frigging bill to be paid. Because guess what? It turned out they overestimated the demand for compute and AI products. I don't think we've underestimated the demand for AI. I kind of think that seems crazy to think like, I don't think there's much AI in the future. So I'm skeptical of that. You know, another outcome is, you know, they –

Get by the other – the current projections meet the world at which we're going into, which is just a slightly more bigger economy are now, whatever, and is the national growth, whatever, and they're fine and whatever. The other outcome is that we see a fast –

We see the scenario where we continue to use AI in more areas. And I think to see that more companies are either having to use it or being out competed by companies using it. And I think that there's two of those scenarios are fine for open AI. You know, the one that the one that doesn't work out well means means that we have to be heavily heading towards this plateau. Here's the funny thing, too. When I talked about their compute plans, one of the things they mentioned was.

The cloud, the idea of making a cloud available, because this is the situation. People go, oh, who's going to win? You know, open air Google. Like I think everybody wins. But also that the public statements they have made is, you know, yeah, we're going to build a lot of compute. Maybe other customers might want to use it too.

And let's say for some reason, you know, OpenAI says we're just going to focus on consumer and chat GPT. We're going to let the frontier model, the AGI thing, let Google or Anthropic worry about it and be like, oh, my God, what happens then? Oh, by the way, we have all these data centers we have access to. Yeah, we can now get into the AWS, Azure, Google Cloud market.

which last I heard was pretty profitable. Yeah, if you're doing, yeah, particularly that. And I think the opening I hedge, which nobody's really called it, that seems to be kind of like, oh, well, and I think they're betting on the idea that they are going to be, you know, continue to see the increased demand, but it's also this kind of.

of thing you'd be like what if what if they don't have the best model then they lease their compute to other people like it makes sense yeah i also think that we are moving and this is a larger conversation that

You know, models matter in a lot of really, really different ways. And it is fascinating to track all these things that have that have come out. Google has a huge release over the last two weeks. We saw Deep Seek today as. Great model. Gemini 3. Great model. Yeah. Gemini 3. Phenomenal. Opus. I've played with it. Yeah. Anthropics model is great. New Opus. You've had a great runway. Released a new video gen model. It looks phenomenal. I haven't played with it. Yeah.

I don't know how much that matters compared to product right now. I think that we are in an era where these models can be cool. What am I doing with it? How does it affect my life right now? And the one thing that I think OpenAI has been smart to focus on has been product. Let's put these things in people's hands. Atlas is a browser. to me is maybe the most impressive thing that OpenAI has done from a product perspective. It has now totally replaced my usage of its desktop app.

It is the number one way that I use ChatGPT and I use it in more ways than I ever have because it has put it exactly in the place where I spend most of my time online on a browser. Does that necessitate? the best model in the world? Does that necessitate AGI? No, it just was exactly the tool that I need right now, the best tool that can be available to me on that platform, which right now is ChatGPT5. That's great.

It knows my email. It knows my browser history. I can ask it to go find this browser that I was in before, and it finds it by way of natural language. They have, under the radar, built an extraordinarily compelling search tool. So it's like... Where are we in the world of email? Or can you spin Canvas out into more of a docs competitor?

These all seem like things that could happen because they care about product and they care about how people are using these tools in a native AI way that I do believe is the future. I think that we are. There's no going back. Everybody is going to be looking at technology and demanding from their products the things that are now available by way of the latest technology. And so I would say it to me is a reasonable.

projection to not say that ai is just a feature that will be layered on top of the way that we understand products right now but we might want to more think of this leap as like the analog world to the online world where now we demanded, no, I want to interact with my bank account. right now. I never want to balance a checkbook again. I want to download a PDF. Like those were things that became just market expectations.

And if that is the future with AI, and I believe that it is, I don't think that there's a lot of evidence to suggest that it's not, then the compute demand is... Yes. And I think that that's why everybody who's invested in this is acting like it's exponential. Yeah, I think that the the bubble.

Like I said, you know, there's going to be winners and losers. There are some winners. There's going to be some companies with bad strategies, whatnot. And that's why I try to tell people we have to separate out. Do we mean a private capital market bubble, a data center bubble, a public company bubble?

Yeah. Do I think that, you know, Google's at a 30 P&E right now? Like, historically, not like the craziest range it's been in, you know, like, you know, more traditional companies or but I mean, when you were talking bubble territory. 25 years ago, these were 300 P&E, 1,500, like insane P&E, you know, or something just had no profits whatsoever. So when it talks to the public stuff, it's like, do I think Google and Anthrop, do I think Google and, you know.

Google and Microsoft are going to go away? No. Do I think people are going to stop using Anthropix products? No. No. So I think that- You know, will there be corrections and stuff? Yeah, there's every day. You know, will there be major ones? Could be. Get a major event. You know, we saw this with DeepSeek. You know, the first – People, when Wall Street heard about DeepSeek, which was well in several years into the, you know, the models of being released.

for the first time understood what it meant to distill or make, or excuse me, like find some efficiencies there and go like, well, they made this thing 10X cheaper. It's like, well, sort of, maybe kind of, but also OpenEye did that. Google had done that. Anthropic had done that. Like every one of those companies, like every eight or nine months, some company got their big gain. Nobody – it was too technical. Nobody cared. And that happened to people like, oh, look at this. And they were like –

Yeah, it was a good gain. Also, like, you know, they're using tons of distilled data, but it was like, yeah, there was also, it was an update to people like, well, if we cut off their H100s, you know, their NVIDIA chips, they won't be able to do it, which was silly. Like, oh, they'll never outsmart this. A billion really smart, heavily capitalized people won't outsmart this. Well, they can. Yeah.

Think and will and are. Yeah, we've talked about this before. But, you know, we watched the market just like – we watched like NVIDIA drop like in half of its value for right now. I was like, this is just crazy. Well, that was what was fascinating about this stuff last week. was that there was this drumbeat, this very weird narrative drumbeat, and Wall Street is an insular world. It is several dozen people that can set a narrative inside that entire community.

And so this drumbeat built of a bubble, bubble, bubble, bubble. Now all anybody wants to talk about is a bubble. And for whatever reason, NVIDIA's earnings was like the proving. Is this a bubble? Will NVIDIA have good earnings? Guess what? The earnings were phenomenal. And so now everybody's like, oh, I guess there's not a bubble, which means that your initial thought was unserious because like, why would you even pin it on?

NVIDIA. If you are seriously concerned about this, why is that the linchpin for it? And so NVIDIA wins. The narrative goes away. Everybody eats Thanksgiving. It was just the strangest thing. Yeah. I again, it's a lot of people who spent decades in Wall Street and people writing about that. making guesses about a field that they didn't see where it came from, didn't see how it got here, had no idea. A year ago, couldn't predict where we are now.

And then now it's hard. It's hard. Yeah, it is. In fairness, it's hard to imagine that things keep scaling, getting bigger. And, you know, I'm a super techno-optism and I'm like, there's, you know, but, you know, the biggest, you know, the biggest limitation I hear now is energy. This energy, like we're going to, and it's just the fact, the amount of, you know, when, when Sam's talking about wanting to put one gigawatt of compute out per week, per week. And, and, but.

That's a huge scale up of energy generation. Here's the thing that when we do the Tension Mechanism podcast 10 years from now, which we have one podcast going on 15 years. Oh, yeah. Oh, we don't quit. We're not. We have one going on almost 20 years now. Here's the thing that's going to be the story. And we're going to be like, yeah, we tried to tell everybody nobody believed us is when.

You have all of industry is now looking at AI, looking at data centers, looking at compute. And we're looking at building up this data centers. And we look at we found this point at which we found the formula to turn. energy into intelligence. We found this formula to take energy and actually directly convert it into intelligence. It's an amazing thing that we don't talk enough about the fact that literally I can tell you what I can get for one watt.

I can tell you how much intelligence I can get from this, and we will get better and more efficient, and we'll probably look at the way we do it now as grossly inefficient, but it works really well right now, okay? You will now have Microsoft invested in Fusion.

And fusion reactors. And you have other people invest in fusion. There's probably a dozen fusion companies. I think maybe 30% of them might actually have an idea of what the hell they're doing. Some may be BS. I think the same thing with some of the small, medium-sized nuclear reactors. I think some of the nuclear companies.

Some have an idea, some maybe not. You have more capital being going into right now into clean energy, by the way, when it comes to fusion and atomic energy and whatnot than we've ever had before. The chance that we've had a breakthrough, we'll have a breakthrough is considerably high because more capital is going into this now than ever. Right. And a lot of different you need different bets. You need you need smart people not to agree. You need smart people to disagree. And that this is why.

There's people like, what if all the AI companies got together? awful. Yeah. Like what do you want them all racing? We want, we, this has to be a, it's a mad, mad, mad, mad world where everybody's trying to get the cheapest energy possible. Literally part of what open eye success.

was a lot of people have deep mind and send Google and said, I'm tired of it. They're on the wrong track. We need to try something different. You need to make multiple bets. You don't all, you cannot predict what's going to be.

With any certainty of what's the best outcome, in many cases, you need different people trying different things, right? So now we're making a lot of different bets when it comes to energy. We have a lot of different bets when it comes down to companies trying to do things. And the result is that... Just like the mobile phone, largely mobile phones and laptops made the modern electric car possible. Because when you are trying to.

build, when you're trying to build, when you're trying to invest in battery technology, you needed to figure out one, what's the best new formula? Two, you need to invest the money in bringing that from the lab into the factory. And that is where it's the highest risk. It's easy to say, I have a thousand labs, try to think.

It is high risk to try to bring this thing to mark. But once we did that and we started learning a lot more about chemistries, we understood a lot more about lithium ion, et cetera. Now, all of a sudden, if you look up, if you take apart a Tesla battery, what do you find? A bunch of smaller, tiny cells, you know, and you realize like, oh, this got us here. Learning how to build batteries is how we got to electric cars. We got drones and a lot of other cool stuff.

And that came from the cell phone. The iPhone came from the other earlier cell phones in the 90s, actually. So now we're at this point where the data centers. have a very high probability of getting us to really cheap and abundant energy.

You know, very, very cheap and abundant energy. And I think that's the thing. Ten years now, we're going to look back and say like, oh, yeah, we had this breakthroughs. We're doing nuclear. We're doing this. We see we've now figured out how we can keep energy can get cheaper year over year, whatnot. I think that'll be an amazing moment. And we're going to kind of forget like, oh, I remember when we thought building up data centers and stuff was a bad idea. You know.

I obviously agree with you. And, you know, there is another podcast that we will have to do down the road about some of the conversations that are being had now on the political side of the aisle, which would make that exact argument. That building out these data centers are a bad idea and that this will lead to energy restrictions and higher costs.

And, and, and, and, and. Yeah, and we know there's some genuine questions there and some very bad faith arguments we know too. That's the frustrating part. There's been some stuff about water usage. It was just flat out not true. There's some players out there saying things. There was a very high profile book. We've never talked about it.

about because a person who wrote it I never thought was a fair actor who got called to task for some things that were completely not accurate like well I'll fix this and I'm like yeah maybe everything is broken about that but I'm like, it's weird because you're like, what do we need? We need people to spend billions of dollars on new energy systems. What do we have? People spending billions of dollars on new energy systems because they're incentivized for a reason.

Sounds great. They are desperate to. Sounds great. They are buying old decommissioned reactors. They are trying to build their own. They are investing in new technology. Like right now there is. I don't know. I wish we had a capable scientific and tech press because like that's the stuff I want to read. I want I want the like several, you know, the 20,000 word.

uh, article about like the race for power right now. Like that's that, that I would, I would love to, to have that. And it's just, it's, it's such a sad state that like we're, we're in, in such a. depressing place when it comes to tech media that like no one's writing that kind of stuff. Well, and you know, we see too, like, like I think that.

Whenever media feels directly impacted by a thing, we get a lot of stories about how terrible a thing is. We saw that with social media. Once you started realizing that social networks were impacting it, then we got the stories about it. We saw that earlier, the Internet and now, you know, New York Times, you know.

you know, has their time for another New York Times hates opening eye story, you know, and trying to go twice, twice monthly. Let's rehash the same three stories. But yeah, you know, they're put a new headline on it. Yeah.

Two things I want to talk about, though, I think that directionally what's interesting where things are headed, though, is that we mentioned this before. Like, you know, as tools like open a shopping research tool become more prevalent, you mentioned we mentioned Amazon and Amazon AdWords.

Now when I have an AI and I tell the AI, go out and find the thing. We have to rethink a lot about shopping. We're going to have to rethink a lot about how ad tech works because guess what? I just want to – I want my – computer to have access to a spreadsheet and maybe some legitimate reviews from people about products. Going to be a very, very interesting space.

It is going to be a very interesting space. We kind of glazed over the ad story with free chat GPT. We know nothing about, by the way. We don't have any insight. We have no idea. No, this is a report. A report that came out over the weekend was that they are prepping this as a product. So take the word of the Internet for what it is. But it is something that.

Sam has talked about openly as something that they have thought about in the past. We have no idea it's reality right now. But as a concept, I think people are underselling what this would mean. to the online world because we're not just talking about another big platform coming online, which I think right now, ChadGBT, you would look at it as a much more potent platform than something like TikTok if you look at its user base and stuff like that.

And how often people come to this. It is far more like Google or Facebook in terms of the kind of regular usage from free users. But beyond that, if it were just that, it'd be a big story. But this has the capacity to revolutionize how we buy things. And if it revolutionizes how we buy things, then that means it has an even more potent ad space. The reason why Amazon can sell their stuff for the price that they sell it and they want to jam as many ads into your.

search history or your search results as possible is because they know you're about to spend money. So those are very, they were very valuable ads and open eyes as a capacity to possibly build that real estate for themselves. And it's not that far away. And if that's the case, then they are. They are in a huge catbird seat for a massive free user base. Yeah, there's this point where there's a lot of things we would like.

You know, there were a lot of great ideas in the arts, a lot of great ideas from like using AI tools to help you manage your budget and whatnot. But often the incentives were completely wrong in the idea that you're like, oh, well, the person paying for this really isn't out to help me. You know, they're out to do this thing. That's the thing we always have to analyze. Are our incentives aligned or not aligned with the, you know.

going through the agent stuff, who knows, we might end up with a protocol, which is like agents get 2% of any transaction if you want to allow it. Then you just sort of solve a lot of problems like this. I know that a lot of companies are going to have to think. Very, very clearly about their value proposition, because I mentioned this before, like I use ChatGPT Atlas with my Amazon AWS bill and like cut like $80 or something off per month.

Cause I couldn't track down where it was going through. And I told Atlas to go through it. I will let the agentic system go through and open up screens and go through and track down where the money was going. When you start to think about this, this is tangential, but I wonder if this is a very important thought was, um,

In the Caltrans, there was a guy that got arrested and charged because it turned out that he had built, he worked for Caltrans. He built himself his own little secret little apartment in one of the train stations. OK. Really? And he used Caltrans budget to do this. Right. And they found out another guy that two guys, two guys have pulled the same scam. Right. Built themselves private little apartments to go live in whatever on this thing, using the budget to go pay for the stuff. Right. And.

You're like, I can't be the one guy did it. And then they found out after that guy, somebody else did it. Like, you know, I bet there's a lot. of other kinds of scams going on there. And some people, $10,000 here, $20,000 here. I bet you, if you could actually afford to do an audit, like a real audit. Okay.

We might find there's a lot of money to do other things we'd like to do with, like high-speed rail or whatever like this. And that's the thing I started thinking about is I think looking into – and the venture fund I'm going to start looking in for.

Companies that are looking at doing really, really – because audits are expensive, right? But AI can do it. And also, in AI, it's like, well, who's going to get it wrong? Great. Then I have a human go look at that and double check it, right? That's the beautiful thing. It can flag things.

Think about a world where you're running AI audits on. Imagine doing it on the public sector and the private sector. And I just think an extension of the shopping thing. All the efficiencies that can be found that are too often you just deal with.

But the inefficiencies, the corruption, the theft, they just say, well, it's just a cost of doing business. But the problem is over time, that cost increases. It goes up, up, up, up, right? But if you can bring in AI and you can bring that stuff effectively zero, it's insane. And so I was thinking about that the other day.

the day. I was like, because one is using AI to go in to figure out how do I lower my costs on everything? Like right now you can give ChatGPT your monthly bills and say, can you help me figure out a better strategy? It probably will find stuff for you. But when you start to think about larger companies, say, hey, where?

Is is where I noticed that we're paying a lot of a lot of a lot of money to, you know, such and such, you know, restaurant, which is right next to the suspicious tavern, you know, and like you can start to find all kinds of stuff. And I think that's I think. If AI was only deployed to figure out inefficiencies and theft, probably pay for all the data centers. Yeah. I mean, that's the other thing right now is if the clock on AI metaphorically would last 24 hours.

I think that we're three minutes into the story. Yeah. I think that we are, we are at the very, very, very, very, very beginning of, of where we are going and the adoption of this tech. is so widespread already. The value is so apparent that really, and this is a recurring thing that we hear on the OpenAI podcast, is that the bottleneck now is...

human comprehension. It's just discovering, oh, this works for me right now. And that's a crazy place to be because that's a place that the internet was at back in the, you know. Late 90s, early aughts. Well, let's talk for a moment just real quickly. OpenAI's partnership with Thrive Capital. Yes. So that was the news. Let's talk about the partnership with Thrive Capital.

OpenAI announced that they're taking an equity stake in Thrive Capital. And Thrive Capital focuses on basically they go in and they invest in companies and basically try to figure out better efficiencies for them using AI tools, automation, what have you. They're primarily involved in IT, which is, you know, everything, and accounting. And OpenAI says, hey, we're going to take an investment. We're going to send OpenAI engineers.

to go help you figure out how to deploy and how to build on top of this and to make it improving. And it's a very interesting bet. You know, often the, the, you know, You'll hear this like, if NVIDIA believed in AGI, why would they sell any of their computers? It's like, well...

You need people to figure out how to deploy this stuff and put it into use. It's not like you say, I'm going to hold on. We're going to just put them in warehouses and wait. You get people who have a very bad idea of how business works telling you how businesses should run. Yeah, because that's, you know, it turns out turning out chips at that.

volume and capacity is its own business. It's something that you really got to keep your eye on and succeed in. Yeah, I'm like, well, and also like if the NVIDIA really believed in AGI, then I don't know, they'd be making investments in opening IANTHROPIC and other stuff.

and continue to produce these systems and not, you know, GameCon. I don't know. I think they're behaving like that. And when people said about OpenAI, like, oh, if they believed in it, would they deploy it? Like, well, they learned a lot by deploying. ChatGPT accelerated things in a way that...

Nobody foresaw. And now I think this is really putting money in your money where you're putting your compute where your mouth is. It's like if your tools are really good and we've seen from our podcast talking to people at OpenAI, they've got some really great.

practical experience now of how to employ these systems and how they work really well. And now they're like, yeah, maybe we should just partner with companies that are like actually deployed out there and just literally, like I said, make something off of every arcade cabinet. Yeah, it turns out it's a venture firm that is there to identify companies that can get efficiency from AI. I think...

OpenAI is invested in the idea of being a solution to create those efficiencies. That is a huge part of their business model for which they would like to continue to grow into the American and international economies. to being the thing that does the efficiencies. Yeah. Yeah. All right. Well, we will be back probably next week. I think next week. You never know. We're meant about...

Well, we'll see. We've got a fang we've got to record on. Oh, never mind. We won't be back next week because I'll be on a flight back to Texas. So a week after that. Oh, who's to say? Andrew, where can people find me? Well, you can check us out on our sister podcast called The Open. Yeah, there is a new episode that's coming out. Um, uh, probably when you hear this, uh, all about model behavior, a fascinating conversation with, uh,

You know, people who are literally in the trenches determining the way that these models speak to you, specifically 5.1, the latest update. So go ahead and check that out. Otherwise, you are at AndrewMaine. On X, on Twitter. Yep, and andremaine.com. You can check out my blog. And I'm Justin R. Young everywhere that you can find a Justin R. Young. Until next time. See you later. Diamond Club hopes you have enjoyed this broker. Dog and Pony Show Audio