The Zitron Bear Case: What's Right, What's Wrong?

0:07Aaron: Good morning, good evening, wherever you are, and welcome back to the Enterprise AI Show. This is your host, Aaron, and we are continuing the Models and Markets series, but we're bringing back Brandon for a few episodes. Today, Brandon and I tackle the provocative critiques of AI by Ed Zitron, a vocal opponent in the tech industry. We delve into the bear case against AI, exploring both the merits and flaws of Zitron's views. And all of that is coming up right after this quick break.

0:40Today's show is sponsored by Nasuni. There's a growing gap in AI right now between what's possible in theory and what successfully works at scale inside an enterprise. The difference comes down to unstructured file data. Many AI initiatives struggle because the file data they depend on is scattered, unstructured, and disconnected from where and how work actually happens. Nasuni changes that. It brings your unstructured file data into a single secure foundation so AI, both generative and agentic, can access it with the context, governance, and performance it needs in production. Bring AI to where your unstructured data lives.

1:15See what it takes to activate your data for AI and request a demo at nasuni.com slash AI.

1:22Aaron: And we're back. and we have a guest host, guest star. I don't know what you want to call it anymore, but regular guest at this point. That's right. I'm going to take over the Reasoning, not the Reasoning show, the Enterprise AI show. I'm taking it over one episode at a time. That's right. So here's what's going on. We're going to continue our, what we're calling models and markets. We keep saying, oh, we're going to name it, we're going to name it, and then we keep calling it models and markets. So we're just going to go ahead and call it models and markets now. But we're going to do a slightly different approach with Brandon and I doing this one today.

2:00Aaron: What we're going to do is maybe get into a little bit more on the technical side of some of these and also try and figure out some of the bear and bull sides of all of that. And the one we're going to start with today is everyone's favorite critic in the industry. It's Zitron and the bear case for AI. Like, we're going to kind of introduce this as like, we're just going to say what's right, what's wrong about AI. And as we have in a lot of these shows, we're going to do kind of a what's now, we're going to do the past, we're going to do the present, we're going to do the future. And so, Brendan, you originally actually came up with this idea.

2:41Aaron: So I'm going to do the why now. So I'm going to kick it over to you. Why now and why Ed Zitron? Well, I mean, I think Ed Zitron, as you already mentioned, he's made his name as the biggest critic, I think, of AI, at least the most vocal. And I think the question to ask him is to kind of go, I think he lays it out this way. He basically says AI, for the most part, doesn't really do anything productive, and therefore people are over-investing. And he would say something along the lines of Silicon Valley is like out of ideas, right? That's kind of like making his very long case much shorter. And I think, you know, if we pick that apart, I guess I'll start with this.

3:19I always think his clear mistake in my mind is he can't separate valuation from kind of the benefits, right? Like you, and I'd like your take because I know you're in this world very, you know, day to day. It's like two things can be true. And this is like, this will be unsatisfying to everyone, right? It is possible that the AI companies that we know of today are overvalued, right? Simultaneously, it's very possible that AI provides a lot of value to people in lots of different ways, right? And as with any company or any type of new technology, this is like the boring part or the ugly part of capitalism.

4:01This is what we're trying to figure out. How much is this worth? What does it do? And it kind of goes back and forth. And over time, right, industries usually settle out, right? If you think about the railroad or shipping, it's really well known. This is what it costs to ship a container. It's been like that for a long time. It's very predictable. And when I say a container, I meant like a physical, you know, shipping container, not like a technology container. And then I think that's where I think Zitron is just constantly conflating these things. Because when I hear him talk, I do understand where the valuation argument comes from.

4:32And it's very well grounded. And like, we could all probably make a similar case. simultaneously, I always think like he's not taking the time to like use some of this stuff in his personal life. Right. And like, he's a writer and you know, anyone that writes a long newsletter, I mean, I'm going to go out there and say, it can always help you edit a newsletter. It's no wrong, no reason AI can't give you ideas about what did you miss? Is it a hearing to my writing style typos? I mean, this is just general editing stuff. So at the absolute minimum, right, that would help him and help us all newsletter writers.

5:03And I don't think that's cheating by any means, just like it's not cheating. Sometimes I send you guys ideas in the Slack and you guys say, this is good and this is bad. Well, you're not always available. You're busy. You have real jobs. So sometimes I need someone else to do that. So that's kind of, I guess, the why now. Why are we talking about this now is I think Zitron is just conflating those two issues. And I think it's our chance to kind of pull them apart. What are your thoughts?

5:24Aaron: Yeah, no, I actually completely agree with you. And it's so interesting to me. also i'm gonna do the why now in the past all in one how much of this is and not just ed but the industry feeds on critics and there is always going to be a market for the critic side of everything everything we've ever done in this market there's always a bubble there's always something coming there's you know every good side has a bad side there's and especially with technology there's lots of ideas of like okay hey this is just a fad and there's the idea of like okay you can pick some of these things over the years like all the different waves you know this this podcast has written over the years every single one of them has a downside and some of it might be right some of it might not and a lot of it is just normal product cycles i won't even call it hype cycles i'll just call it product cycles right and is ed and others just predicting this because it goes up so fast it has to come down that fast as well and so i guess like that's where I'm like, okay, you know, if I take the past and I kind of apply it to where we are right now, and this is where I'd love to, to, you know, get your take on, on the present.

7:01Aaron: Cause I, I haven't studied it as much as you have. Like he has very specific claims and, and to your point of like valuations and very specific things, like tell me about what the present looks like to you with something. Yeah. Well, I think his big thing is, you know, something along the lines of, you know, OpenAI and Anthropic, right? There's, they're ways overspending on inference, right? It's not profitable, right? Which I think is like, I think it's reasonable fact. Like I don't, I would say, and this is kind of the classic, you know, all startups have to invest before they can make money.

7:32And I think his point is like, well, it'll never make money. And what we're going to end up with is all these companies doing this weird financial transactions and borrowing money they don't have to build things that will eventually be worthless. And there'll be some type of financial, you know, pullback, chaos, whatever you want to call it, the bubble will burst. So again, this is back to what I was saying before. Well, let's just pull those apart. It's just like, just last week, Amazon's recent earnings, I think Andy Jassy kind of laid out the case for what they're doing. And I think it's pretty simple when you hear him say, he's like, well, we're borrowing, we're going to spend$220 billion next year.

8:08Well, we got to build out these data centers that take about two years. They're really expensive. But when you have a data center, it's going to be making money for like 30 years. Then as we get close to them, actually, in that two-year cycle, when we get close to actually utilizing those data centers, we'll have to go buy all the servers. And that takes about six months. And those are going to last. We're going to depreciate them for three years. They're going to run for five or six. So in year four, so after you do all this, just classic investing, it's like we get the data center for three years and then we're going to refresh that equipment as necessary.

8:38But we're always going to get two or three years of like essentially pure margin from those things. those things. And so this would be what Zitron's like saying like, well, no one's, you know, Anthropik and OpenAI and others, they're not going to be here. They're not going to use it. And okay, let's just grant him. Let's say he's right. But at this point, you know, the utilization of these, you know, GPUs is going to be used by someone. It could be a new model, could be an open model. You know, there's plenty of models you know about, right? And that could absolutely happen. It could be the case that Anthropik and OpenAI lead the way forever.

9:11And it's always their frontier models. And they're the ones that are always the biggest customers. It's possible when these Chinese models becomes, or an open source, something along those lines. So I think he's just always like this idea that like, it won't be utilized. Now I think his case, I mean, his real bear case is just like, no, there is no one using these models. They just like, that's all just fake, right? Like no one's doing anything with it. And I just like, to me, it's like, it just doesn't pass the sniff test. Like I use it all the time. All of software development's fundamentally changed.

9:38And I think, you know, this is where maybe he could say, I don't get it, But I'm just like, it's slowly going to, if you will, make its way into like, quote unquote, normal people jobs. And they're going to use it more. And so even if like, you know, the example that people always like to say is like, well, in the dot com boom, it's like the dark fiber, you know, laid in the ground. And then eventually it was utilized. And I just kind of see the same thing here. It's like these data centers and this technology, I think it's going to get used barring something just unforeseeable. Maybe tomorrow, maybe you do it.

10:08Maybe someone else, hopefully you do it and you'll share the secrets. you're like, I have something better than the transformer. I don't need to do this complicated mathematics anymore. I can do it all on a laptop. Well, okay. Well then, then I just would be like, yeah, that money's wasted, but we're probably are even in a better place, right? Something even better is about to happen.

10:25Aaron: Yes. Well, and I'll say this. I think that the dark fiber layout is the best technology we have of like, okay, yes, absolutely. There's no doubt our industry is burning through money at unprecedented levels get it but i also firmly believe it is exactly that of it's laying train tracks it's laying dark fiber it is setting up infrastructure for a whole next generation of things that is absolutely kind of this you know ai forward world if we will and and yes to your point i mean there's so many ways in which it optimizes things i do you know both personal and professional i don't think the genie's going back in the bottle there now to maybe lay out a little bit about this though like what i actually want to talk about is like when you were talking about okay we do these build outs and you know these are only good for so you know we're gonna have like okay gb 300s and gb 200s and you know you keep going back models and models and models and as these things kind of depreciate like let's all put on our hardware and data center hats for a second.

11:35Aaron: What I see happening is it actually will allow cheaper and cheaper, lower rate inference. Like, you know, basically not everyone's going to need the Ferrari every time. And so it's going to allow more economical low end. And you build out the top end, you just keep building out the top end and then everything keeps funneling down over time. And I think that's actually plays really well into the AWSs of the world. I mean, that's where they know how to, you know, absolutely maximize something like this over, say, an open AI or some of these others. You know, a lot of folks write the hyperscalers off, but I think, you know, one thing they know how to do is build for the future and then optimize that for, you know, ops over time.

12:21Absolutely, and I think, you know, maybe we won't get into it too much today, but maybe we'll talk about it in a future episode. It's like, you know, how all this could play out in multiple ways. One example is that, just what you said, the hyperscalers love the fact, more models and they would love to be the infrastructure for all the major model companies. But coming, you know, at some point is like, you know, just ripping off some Ben Thompson, you know, discussion here at some point, you know, let's assume for a second, Anthropic and OpenEye get so big that they're the dominant. They have the value, right?

12:47People want to buy the tokens from their models. At some point, they'll do the math and say, we should build the data centers. We don't want to, if you will, be renting this out. And then they would start to be in conflict with primarily Amazon as well as Azure. Now Google's like in this other weird category. They both make a model and they also rent models. So they're like in this, I don't even know what you call them. So, I mean, it could happen that way. So you could say like, and I think this is why Azure and AWS, they're investing so aggressively. Like they don't, they've kind of lost out.

13:16I know people don't like, sometimes people don't like when we say this, but like they're not really making frontier models. I don't think that's news to anyone. So the best business for them is to be like, we want to host all the inference and we want to make sure that we do that in a way that we have all the models people want and we don't allow one of these companies to usurp us and become the only thing. But who knows? And back to Zitron's case, it's like, it's possible that happens, that like Anthropic goes away and we just pick, we go there. And I always say this in like the harness, like it'd be fine with me.

13:44Like I use Claude a lot when I just do slash models and I just switch. It'd be great if they were like, fine, put a hundred in there. I don't care because I do kind of move around. I try to like, if you will, manage my own little home budget, right? Right. And that may happen. That's a good way to think about it.

13:57Aaron: Yeah. Right. And the same thing with like, you know, or it's possible it goes the other way. It goes that like Anthropic does become the behemoth and they end up buying all these data centers from AWS and Azure because like we just don't need that from you anymore. In fact, we're going to buy that and we're going to keep it. We're going to build it out. And we've seen this play out like millions of times. It's a little bit of the whole private public cloud, you know, discussion we've been having. You guys have been having it for like, I don't know how long this podcast has been going. 15, 20 years.

14:19So it's the same thing. And again, it's like it's maybe unsatisfying to everyone. And it's again, like two things can be, you know, always two things can be true, right? There's a lot of promise in AI and the valuations can be out of whack, but it's okay. We're going to, it's going to work its way out. Not saying, I'm not going to say it's going to be very windy. We're going to do many podcasts about it, but like it'll work its way out.

14:39Aaron: And I'll add this and I'll kind of close on this as well, like a potential future kind of thing. I think what could really happen here is it depends on who you believe. And the reason why I say that is, you know, as we're recording this, a bunch of earnings calls are starting to come out. And again, all you keep hearing from earnings calls is like, hey, we are hardware limited or we are GPU limited. Like if we could go build data centers or if we could get our hands in more GPUs, we could hit demand. Now, I think they're always a little hand-wavy when it comes to what is that demand? Is it customer demand?

15:12Aaron: Is it like their own demand to make their own models? Like I feel like they kind of blend all of that together and just go, hey, look, we have so much demand. It is amazing. But if you really peel that back to your point, maybe there's something there, maybe there's not. But I think there's enough evidence to say, okay, it passes a sniff test. And I think what you and I are both agreeing on is Ed's classically curmudgeonly, the sky is falling, the sky is falling thing. Because I mean, honestly, I just think that's how he makes a name for himself. And that's just his thing. He's the the classic bear guy, no matter what the markets are.

15:47Aaron: And he's just continuing that trend. And that makes him money. He makes his money. Like everyone, everyone makes their money how they can. And he's figured out a way to make money by going, Oh, nope, that's crap. Absolutely. Totally agree. All right, man, we're going to end it there. Thank you very much, Brandon. I appreciate you taking the time today and for everyone out there. Thank you very much for listening. And of course, as always, we would love your feedback on the podcast. And of course, reach out to us on all the socials as well. And we will talk to everyone next week. Thanks for listening.

16:23Check us out at theenterpriseaishow.com for past shows, newsletters, and all things enterprise AI.