Dan Niles: We’re in an AI Bubble — But It’s Not Over Yet

We're in a bubble. But you were in a bubble in the late 1700s with canals. You were in a bubble in the early 1800s with railroads. You were in a bubble when radio first came out. You were in a bubble when the internet infrastructure came out. You're in a bubble with fracking. Any great technology that changes the world, people look at that and they go, "Okay, if I'm the survivor, I'm going to make a ton of money." Make no mistake, the people that are saying we're in a bubble is right, but it doesn't mean you can't keep going up further.

My pleasure, Zach.

I mean, I think we have at least a year to go. I put out a post on X on March 31st saying history may not repeat itself, but it often does rhyme. And I compared where we were in 2026, so we're in year four of the AI infrastructure buildout, to what happened in '97 and '98 when we were in years three and four of the internet infrastructure buildout. And if you go back to '97-'98, we had macro scares then as well. '97 you had the Asian currency crisis. '98 you had the Russian bond default. The S&P was down 11% intrayear in '97. It was down 19% intrayear in 1998. But underneath that, you had this internet infrastructure buildout going on. And the full years finished up 31% in 1997 and up 27% in 1998. So obviously macro scare with Iran. This time around market sells off 9% through March 30th. And then obviously you've seen the rebound since then with the S&P through a couple of days ago up like 19% and the semi-index up 80. So very similar pattern. The one thing that makes this even potentially better than back in '97-'98 is you didn't really have a new version of the internet come out. What you saw in January of this year is you had Open Claw come out and be formalized and that kicked off Agentic AI. So instead of Zach Guzman asking a question to ChatGPT and then you get an answer back, Zach Guzman may say, "Hey, you know, go find my latest interviews. Go find data on Bloomberg. Go get my latest Twitter posts, then go create an Excel spreadsheet and put all of this in there and then, you know, create a PowerPoint deck out of it." When Zach Guzman does that, all of a sudden you're using 10 to 100 times more tokens to have Agentic AI. And you can see that in the data where token production in the couple of months prior to January, it had gone up about 20% in two months. In the two months after the formalization of Open Claw, it went up over 100%. And that's why you've seen, I think, capex expectations which started the year at about 30% now are sitting at about 70% for the five big hyperscalers. And so I think you're going to keep this going at least until you anniversary that agentic AI push that started just a few months ago.

Yeah, I mean, it's a great point, Zach, because as you know, there's only two things that determine stock prices, right? It's earnings and the multiple you put on those earnings. And on the earnings front, S&P earnings this year are expected to be up 25%. So if you just did the simplistic calculation of, well, if multiples didn't change, the S&P should be up 25% this year. And likewise, if you look into next year, and I don't like to use forward numbers that much because they obviously can change, but for right now at least people are thinking, hey, we get 15% earnings growth next year in addition to that. Obviously, and for good reason, people are concerned about inflation and oil prices seeping through to inflation. And we've seen some hot inflation prints recently. But don't forget there's a big offset in that you have Kevin Walsh that started as a new chairman of the Federal Reserve and he wants to cut rates. That's why he was appointed by President Trump and he's focusing on things like trimmed mean PCE which has been going down. It's actually at 2.4% versus core PCE at 3.2 or headline PCE at 3.5. And he also believes AI is deflationary in nature. And so that's an offset to the bond market which is pricing in right now one rate hike by the end of this year. And so that should keep the easy money party going, which we really had since the end of 2022. As well as you've got Agentic AI driving that. And to your point, you've got 25% earnings growth overall in the S&P 500, which is faster than anything we saw in the 90s during the internet buildout. And that's a pretty good recipe for the stock market.

Yeah, I mean, you brought up some good points. I mean, one of the big gating items is obviously access to energy. And so at that point you go, all right, well, what's going to make that better? And so there's obviously lots of different ways to play that. I mean, for me, semis is a good way to get to that. You can look at some of the analog semiconductor names within that. Also with some of the changes being brought about by Nvidia's new platform, Vera Rubin, that's going to be coming out in Q3, you're going to need a lot more high bandwidth memory that's coming for that. You're also going to add a lot more optical backbone to this architecture going forward because if you can transfer data at the speed of light, your energy costs go way down. And so some of the optical names, and then obviously as we talked about before, if you go ahead and you have, you know, your agent do a bunch of stuff, that's going to require a CPU. And so CPUs, and I wrote about this back in March, talking about how Agentic AI, you go from chat-based AI, you maybe needed eight GPUs for one CPU, you go to Agentic and you can get that potentially closer to one-to-one. Some people even think it flips. And so, you know, the traditional microprocessor vendors also benefit. So there's a lot of different ways to play this. And I think they all work in different ways, but the main thing that you go back to is very simply Agentic AI requires 10 to 100 times more token generation. So all of that ecosystem is going to work well. And then you can get into the architectural changes and, you know, what that's likely to drive. Agentic requires 10 times more memory. You're going to have more optical to try to transfer data faster, etc. And then obviously need more energy. So there's multiple different ways to play it. And yeah, you know, they'll all work to some degree.

Oh, it's going to end and it's going to end badly. But I mean, the thing that people kind of get confused on is when you say you're in a bubble. And make no mistake, we're in a bubble. But you were in a bubble in the late 1700s with canals. You were in a bubble in the early 1800s with railroads. You were in a bubble when radio first came out. You were in a bubble when the internet infrastructure came out. You're in a bubble with fracking. So any great technology that changes the world, people look at that and they go, "Okay, if I'm the survivor, I'm going to make a ton of money." And so you have more companies. You've got, I always like to joke, you have a hundred companies that each think they're going to get 50% market share, right? Like that's not going to work. And so when you get to the other side of this in a year, and you're already seeing it in the market, right, with a lot of OpenAI had these grandiose plans of spending, you know, 1.4 trillion in capex. You don't hear them spouting that number anymore over eight years, right? They've backed off that. They said, "Hey, we're going to focus on a few of these other projects, etc." And you see Anthropic now growing faster than them. And for the longest time, I've said OpenAI, I think, is going to be the one that has the problem because Google is the dominant person in consumer-based search and they've got the cash flow to fund it and they've got 13 different products with over a billion users each. And then on the corporate side, Anthropic started off saying, "Hey, consumers aren't going to pay for AI, right? Because we're all trained to be on Google and we get answers for free." Anthropic focused on coding and corporations because corporations pay for stuff. So I think when we get to the end of this year, and actually you already started seeing it last year, people start to question, well, how's OpenAI going to pay for all this stuff? They have memorandums of understanding and etc., but they don't have the cash flow, right? They're not expected to turn cash flow positive for several years. And Anthropic, to their part, you probably saw that in Q2 of this year, it looks like they're going to be profitable because they've grown revenues so fast. And originally that was supposed to be in 2028. Now playing devil's advocate on the other side, you can say, well, yeah, but that's because you had all these stupid things going on like token maxing, which was blowing through engineering budgets because if you tell a bunch of engineers, hey, I'm going to judge you on the amount of tokens you produce, they're going to write very inefficient things to just produce a lot of tokens. And so you're having to see companies back off of that. So Uber said they blew through their entire budget in four months, right? Uber doesn't want their stock to go down. So they're going to have to go in and figure out where am I going to cut so I can afford to run my Claude code. Microsoft is backing off of letting their engineers use Claude. Now having said that, you can say, "Yeah, but they're telling them to use Copilot and then that can access Claude underneath it." And you know, there's ways around that. But the point being is that you had a lot of companies now going, "Wait a minute. I'm blowing through my AI budget. I got to cut." I think the issue is going to be where they cut. And I think you're going to see software continue to have issues. You're going to see IT services, the consulting companies continue to have issues. And then you're going to have headcount at these companies. You're going to have one really great engineer that's going to get paid five times as much as he used to because he can do 10 times the work because he's using all these agents to help him. So we're, I fully expect sometime next year we're going to be, you know, I'm going to be on your pod and talking about why certain sectors like semiconductors are down 30 to 50%.

Yeah. I mean, I think again, there's a time element to this. All right. It's like if you get sick, right, it's going to take you a certain amount of time to get over it. And so in this case, you've got Agentic AI. We're all just trying to figure out, every corporation is trying to figure out how to use it, how not to blow through their entire budget in four months, what to cut, what to keep, where to invest, where not to invest, etc. And so there's a time element to this. And so that's why I say I think it's going to be at least a year from when Open Claw got formalized at the end of January until you figure that out. Now, as we talked about earlier, you can say, well, the fact that Microsoft is getting their engineers off of, you know, Claude by June, the fact that Uber is dealing with these issues, and there's a whole bunch of others, right? You can go through like Shopify had problems around this, etc. And then you've had some software companies having issues, and then other ones, like you saw today with Snowflake doing well. Like, there's going to be winners and losers. And that really started, you could argue, back in like October of last year when OpenAI people started to really question, hey, how are they going to spend all this money with Oracle, right? And so as these companies also come public, we're going to really be able to see, okay, what are people willing to pay for this, right? SpaceX is going to be first, obviously, and then, you know, that's obviously a little different, but that'll be another real interesting test because right now it looks like it's going to go public at 100 times revenues, which is pretty insane, but it is Elon Musk, right? So we're going to have a lot of milestones along the way. I'm a big believer and have strong opinions loosely held, and so for me, I strongly believe we're probably, you know, until next year at some point, this bubble keeps inflating just like it did in the 1990s, late 1990s. But hey, if OpenAI all of a sudden has a problem because Anthropic's really killing them, then that timeline could change.

Well, do you remember what happened in 1999? Oil prices went up a lot, doubled actually, and the Fed was hiking rates. What did the stock market do? Stock market absolutely ripped, as did semiconductors, as did NASDAQ. And then in the first three months and change in 2000, I remember NASDAQ was up 86% in 1999 and it was up 24% year to date through March 10th of 2000. Yeah, and now by the way, you're in year five and then year six of the internet infrastructure buildout because if you use when Netscape Navigator got launched at the end of 1994 as the start of that, right, then you have, that's the ChatGPT moment arguably of the internet infrastructure buildout. You have 95, 96, 97, 98, 99, so you had five years. And in the fifth year, you had NASDAQ up 86% while the Fed's hiking rates, while oil's going up. Does that sound familiar? It's kind of what people are worried about now. Yeah. But you have such strong underlying demand from this thing called Agentic, which is brand new. Like we didn't have this last year. I mean, arguably we had Claude out in November, but like it was still evolving into Open Claw, right? So, but that is what I think makes this a little bit different. But make no mistake, the people that are saying we're in a bubble is right, but it doesn't mean you can't keep going up further. I'm always reminded of Alan Greenspan. You probably remember this. He made his famous irrational exuberance speech at the end of 1996 in December. The S&P then went up another over three years by 90%. I think NASDAQ was up close to 300%. And the semi-index was up another 450% through the peak in 2000. And then what was funny about it is if you go back and you look at his speeches, he had a speech that said structural change in the new economy. And that speech was in July of 2000 where he was like, well, you know, the internet really changed all my theories on the market and business cycle could be attenuated. And of course, that's when you had one of the worst recessions in forever, right? And the market got killed and the NASDAQ goes down 78% and I think it was semis that were down 86% from peak to trough. But so like, yeah, you're going to get people getting sucked into this and then it's going to end very poorly. I mean, you think about that. You go in 1996, he's worried about irrational exuberance and in 2000 he's talking about the new economy, right? So we'll see what happens this time, but I mean, Agentic is real, right? It really makes things very different for the world really. And so we'll see how that plays out.

Yeah, but here's... Well, I guess it comes down to who you are. I'm an optimist by nature and I believe like human beings will figure this out, right? It's like the Luddite movement going, "Oh, we got all this mechanized farming equipment. It's going to put us all out of work." You know, the US, I think, when it was founded was 90% agricultural. Now it's less than 10%. I think we could agree the US has done just fine, right? And so now if you talk to some corporations, some tech companies, they go, "Hey, yeah, we've gotten rid of all of these jobs, but then we've created all these other jobs." And you haven't seen it in the data, right? We're in year four now of AI, right? With ChatGPT coming out in late 2022, unemployment's at some of the lowest levels we've seen in decades. So I know that people can dig into the data and they can look for little nuggets that show that, oh my god, we have all these unemployment issues, but there's jobs being created elsewhere. Okay, so let's say you do have engineers, obviously you're seeing layoffs at some of the big tech companies for sure. But what about the non-tech companies? They can use Agentic AI to really help their businesses. So you fire a bunch of engineers from Microsoft, but if you have Boeing going, "Hey, maybe we don't want to have, you know, planes taking, you know, years to design," and they actually talked about that on their call, and you can use AI to help with that, then they'll get a bunch of engineers, and Boeing and Microsoft are both in Washington, right? And so you've got, I think, the benefits of this moving to other companies outside of tech. And I know we all spent tons of time talking about tech because that's why you have these trillion dollar companies, etc. But then this is really going to benefit those other firms that aren't tech-centric that can still use this because Agentic touches really everything. You know, whether you're designing a plane, doing accounting, doing law, you know, my son's dealing with this who's in law school, and so it touches all these other areas that the internet potentially didn't transform as much as Agentic is going to. And so I'm pretty excited about human beings being adaptable and other areas outside of tech taking up a lot of those jobs.

You're 100% right. Like my son had to make a change in law school because he's looking at this. And I have another son. I have twin boys. My other son is, you know, in computer science. He was focused on AI before any of this came out. And so, you know, he's got 10 agents that he spins up to help with his startup that he's doing. My son Warren, my other son, you know, he's the one in law school. And he came to me and he goes, he wanted to focus in transactional law and he's in second year of law school. And he goes, there are going to be no jobs when I graduate. So he switched to litigation because until you have a robot walking into a courtroom, right, and talking in front of the judge, you're going to need human beings to do that job. So you're right. I feel really bad for some of the new grads in coding and, you know, software engineers because it is tough when you have corporations saying, hey, I've got these guys who've been at my company for two to three years and the top half of that or the top 10%, if I arm them with AI agents, they're going to be five to 10 times more productive. It creates a really tough spot. And I know some, you know, I'm an engineer by training, right? I have a master's from Stanford in electrical engineering. And, you know, so I love to keep up with the latest and greatest technology and have friends with kids who are graduating now. And yeah, it's a rough job market. So I really do feel for them.