Dan Niles: AI Won’t Kill Us—But Some AI Stocks Won’t Survive

I'm an optimist. I believe humans can figure it out.

A lot of investors have gotten used to, oh, stocks don't go down and stay down.

Yeah, you should have been pulling back a few weeks ago.

Oh, yeah. Oh, definitely.

My pleasure, Amber.

I mean, it's like if you went back a few years ago and you said, 'Well, we'll never have a global pandemic. That's impossible.' Anything is possible. So, do I think it's likely to happen? No. Could it? Sure. Could you see in some case something bad happening? Some bad actor. I mean, we've already seen the Federal Reserve get hacked, right? You've seen the power infrastructure grids of the US get hacked. So, could you have something get a lot worse? Yeah, it's possible, but I don't think it's likely. And so, that's the same reason we get in airplanes and we drive on the road, even though you could get killed, right?

You know, I think you have to remember Dario also predicted that there would be mass unemployment. And I think we've seen the employment picture hanging in pretty well, right? When did ChatGPT come out? At the end of 2022. It's late 2026. Have you seen mass unemployment? No. And have you seen a lot of jobs actually created from this? And so we'll just take that as an example.

Are you having a slowdown in hiring at the massive tech companies? Absolutely. Are you also, though, seeing non-tech companies now able to get those workers that they couldn't get before? Yes. And you have people being added to help non-tech companies, you know, learn how to use AI. So, to me, there's, you know, I'm a big believer. I'm an optimist. I believe humans can figure it out. All the big technological advances in the past that were supposed to cause mass unemployment and unrest — there's a, you know, the Luddite movement obviously is the one everybody remembers — we've managed to figure it out and get through it. So, I do believe that there are also other competing interests. If you're in the lead in AI, like Anthropic or OpenAI, what you'd love to do is shut down any ability for things like open source or open weight to cause a problem to your business. And so having more regulation to choke things off if you're in the lead is really good. And there's a reason that they're wanting the government to protect them from any charges of antitrust if they talk to OpenAI as well. So, I'm not saying that there isn't a chance of something really bad happening, but there can be multiple forces at play at once.

No, absolutely not. I mean, if you can give yourself a monopoly, wouldn't you? And have it be government protected? Like, it's a great idea. You can pull it off.

Well, I mean, it's obviously bad. I mean, I've been worried about this ever since a few weeks ago when you had, you know, states out there which you would have never expected, like Texas saying, 'Hey, let's pump the brakes on new data centers and power to these data centers.' And that's something I never expected. Most people didn't expect that because Texas has been deeply supportive. It's the second biggest, behind Virginia, state in the US that supports data centers. And so when you have that happening, you should have already been worried about a slowdown because you can't on the one hand say, well, access to power is a limiting factor, and then in the same breath say, well, the fact that a lot of these states — you know, Pennsylvania, Texas, etc. — are saying, hey, let's put new connections on hold for a while and data center expansions on hold for a while, you can't simultaneously say, 'Well, that doesn't matter.' And so, this is something I've been concerned about ever since, you know, a few weeks ago when that happened. And the midterms aren't until November 3rd. So, you've got another couple of months for that. Plus, now you've got these comments from OpenAI, Anthropic, Elon Musk is supportive of that too, and, you know, of slowing the pace of improvement in AI, which means slowing the spending if you're not going out and training new models.

Yeah, you should have been pulling back a few weeks ago.

I think it just tempers it. I mean, you have a lot of people out. So, let's just back up and set a very big picture. Any great technological revolution in history — and that's whether you're talking canals in the late 1700s, or railroads in the early 1800s, or radio, electricity, or most recently the internet — if you have a great technological revolution, you're going to have overinvestment by human nature because everybody knows, wow, this is something that's going to change the world. And if I'm one of the early companies that can participate in this, I'm going to make a lot of money. And companies have even said that — Microsoft has said the biggest risk is underinvestment, not overinvestment. You know, Larry Page has said, I'd rather go bankrupt than lose this race. So, you know, there's overinvestment now. So, then the question is, when does the bubble burst?

And there's some people who think it's already burst. I'm in the camp of — agentic AI showed up arguably on January 31st when this thing called OpenClaw got formalized. So, that's just earlier this year. You haven't seen the full impact of that in this short a period of time. And agentic AI means that you're producing 10 to 100 times more tokens than you were when you just had a chat-based AI, right? You're asking a question, it gives you a simple answer back. And so I think you have a lot more room for that to continue to run, which is why I think this is, you know — and I used this analogy back in late June when I warned of a speed bump back then, and I said, 'Look, people are focused on token minimization,' and I think you could have a problem with tech stocks. And then we saw what happened between late June and late July where they got absolutely crushed. They bounced. I think you've got a different problem right now because you've got regulatory pushbacks obviously with, you know, data center pushouts, what you talked about, and then over the weekend what's just come out from the big frontier labs where they want to slow the pace of innovation. And so you need to pay attention to that. There is a possibility obviously that you hit the peak, but if you look at the data —

With open source coming out, the cost per token has gone down about 50% from the end of May through now. That's a pretty big drop. The good news is the number of tokens being produced are almost up 4x during that same period of time. So that obviously makes up for the fact that what you can charge per token is dropping. But if you are sitting at the frontier like an OpenAI or Anthropic are, you really don't like to see that because that shows that open-source, open-weight models are taking a lot of share. And, you know, how does that play out ultimately?

No, I think — what I always say is you don't need a Ferrari to go to the corner store to get milk, right? And what you saw earlier this year is people were using Ferraris to go to the corner store to get milk. They were using Anthropic models or OpenAI models to summarize emails. You don't need those models to do that. You can use a very cheap open-weight model to do that. You need the highest-end models to solve things like mathematical proofs that have never been done before that can improve businesses. Right? That's the stuff you want to be using those models for. So over time, I think what you're going to see is 90% of the requests that people put in, they're going to be done by open-weight, open-source models. There's going to be like 10% of those use cases that are done by the leading models. Now there's a question within that of, well, how many leading models are going to ultimately be out there? Well, if you look at the history of technology after the internet bubble broke, how many leading e-commerce companies do you have? Well, you really only have one. You have Amazon. Well, what about search? Well, you have really one. That's Google. What about social? Well, you kind of have one. That's Facebook. What about streaming? Well, you really only have one. That's Netflix. So, do you think you're going to have five plus models that everybody uses?

I don't think so. I think you're going to have some losers within that group. And for me, I've said this now for a long period of time, I think OpenAI to me is the one that potentially has the most issues because from the beginning, they were focused on consumers, right? ChatGPT. Well, only 5% of those consumers actually pay. Why? Because we're all used to getting things for free from Google. Google is integrating their consumer-based AI into the normal search product that everybody uses. So, they're in a great position on consumer. And Anthropic from day one was really focused on corporate. Why? Because corporations will pay. So, I feel like OpenAI in particular is stuck between those two guys: Anthropic on enterprise, Google on consumer, and that potential issue going forward.

Yeah. I mean, and I think you need to watch — I mean, obviously early on OpenAI was much bigger than Anthropic. The latest data we have is that Anthropic now is much bigger than OpenAI. That it's about — you know, the last set of numbers we have, we don't know how they're counting revenues exactly, but is that Anthropic's at about 65 billion annualized run rate and OpenAI is right around 40. Right? You go back a couple of years ago, the ratios were reversed. So it's just something you want to watch. Obviously, the shakeout from this when you do ultimately have the peak and people really starting to cut back, it's going to be pretty ugly. The good news is valuations for the most part are not where they were in 2000. Right? In 2000, Cisco, which was the leading — you know, they were the Nvidia of their day, right? They were the most valuable company in the world. It was trading at over a 100 times PE. Today, Nvidia is trading in the mid-teens.

So, when you do get a shakeout, it shouldn't be as bad. Having said that, I think everybody remembers the semiconductor index went down 29% from, you know, the end of June roughly to the bottom when Situational Awareness got taken out. And that was a pretty ugly drop in the space of 30 days or so, right? In the space of a month. So, you know, rallies always overshoot to the upside like we saw with the meme stocks, you know, way back when, and they always overshoot to the underside as well. And what happened between, you know, June 22nd and July 29th in semis should be a warning shot because there wasn't even a real — you know, there was a problem in the sense that people were focusing on token minimization, but it wasn't as bad as today where you've got data centers being pushed back and you have the leading model makers saying, hey, we may need to slow down improvements.

Well, so I just want to be clear about this. Between now and midterms, I'm very negative on the market overall, which means I don't want to be involved in anything. Now, I have long positions, I have short positions, and so you can make money even in those environments hopefully if things are set up properly. But if you're an individual investor, you know, there's this whole mantra that's sort of been popular of buy the dip and the market will always go up and it'll never stay down for more than a short period of time. That is going to be proven incredibly wrong at some point because you have a lot of macro things going on at the same time. And, you know, so if you just step back and you look at history, the first one is September is seasonally the worst calendar month of any year. It is up less than 50% of the time and on average you suffer losses. Second thing is going into midterm elections, between the end of July and November 9th, you have a median drawdown of 10%. Since 1990, if you look at non-midterm years, the drawdown is about half as much. So that tells you you're in a seasonally bad time to own stocks. We just talked about data centers getting pushed out. We talked about what the leading AI vendors are going to do. And then let's focus on the really big alternative to stocks. That's bonds. You haven't been able to get these types of returns on bonds. In some cases, if you're Japan since the 1990s — in the US, you're going back to the 10-year today is pushing, you know, on 5% or so. And so that's a great return for something that's considered riskless. And you haven't been able to get these returns literally in decades in some cases. And it's not just here, the UK, Europe, like it doesn't matter. They're all viable alternatives when there's some people that think obviously that, you know, this is a circular bubble just like we saw in 2000 and you're going to get absolutely crushed. And let's not forget NASDAQ went down 78% from peak to trough over two and a half years. The Dow almost got cut in half over that same period of time.

Well, you have to remember, Amber, anytime you're negative, if you go out long far enough, you're going to be wrong, right? So to say, well, you were negative in May — the question is, well, when did you get positive again? Right? So that's why, you know, we were negative. You can go back and your viewers can see what I wrote on June 22nd. We were negative. We thought there would be a problem in tech stocks, specifically around the internet. Then we got bullish on July 29th. That worked out pretty darn well. And so if you just believe — if you're a young investor, if you're in your 20s, you should just buy a low-cost index fund and compounding will do all the work for you, right? Albert Einstein said compounding is the eighth wonder of the world. And so if you've got decades ahead of you, of course that's going to work. If you're sitting in your 70s and you're retired and you don't have decades to make your money back — and I have, you know, some relatives that unfortunately couldn't convince them to get negative in 2000 and get rid of a lot of the stuff that they own, and they got absolutely wiped, and they didn't have decades to live to make that back.

So I think — and the other thing is it's foolish to think you know how much things are going to go down. When I got negative in 2000, when I was on the sell side and thought, 'Yeah, I think the internet bubble's breaking,' I would have never guessed the NASDAQ would go down 78% over two and a half years. So what my view is always the same: if I'm worried about a big sell-off, I try to get out before that because it could be the top and not just a temporary speed bump. And so for me, just like in late June when I was getting negative and I said, 'Hey, you know what? I think we potentially have a problem,' my goal is until I get to November 3rd and I get to the midterms because there's a lot of things that can change. And I'm glad you brought up the Fed. The Fed is probably going to hike on Wednesday. But they're probably in a hiking cycle. And there's a reason that phrase, 'don't fight the Fed,' exists, right? It's because right now the bond market is discounting two Fed raises this year and three and a half in total by the time you get to mid-next year. That's not a positive either, right? That's why the phrase 'don't fight the Fed' exists. So for me, I don't think this is the top, but I certainly am not going to be stupid and say, 'Hey, don't worry about it, and you'll be fine,' because you might not be fine. Because if you end up with all the leading labs saying, 'Hey, we're going to stop — you know, really slow down training for a while,' well, it's not like the internet stopped growing in 2001 and 2002. It just grew slower.

Are you going to see the market get cut in half? I don't know. But if I feel like I've got a potential problem, just like we had between late June and late July, I don't want to pretend that I'm a lot smarter than the market and I know exactly how much these things are going to go down. Because I think you saw in some semiconductor names they got cut in half in a month, especially in memory. So that's just something investors need to keep in the back of your mind, especially when a lot of the tricks that central banks and governments were using — running up huge debt piles. There's a reason that bond markets are getting hit hard. And it's not just what's going on with oil. They're getting hit hard because you've got deficits that are ballooning all across the globe. You've got debt-to-GDP ratios that we've never seen at these levels. And you've got deficits that are the highest ever outside of a major war. And so all of those things combined should make you kind of nervous to say, 'Oh yeah, it's not a problem.' Yeah, it's not a problem if you have decades, but it can be a pretty big problem over the short term if, you know, I'm being too optimistic and this in fact is the top and we're about to see a slowdown like we saw in 2001 and 2002, and, you know, these things are going to get hit hard.

Yeah. And I think that's another reason why the bond market is having issues because now it's not just a government issuing debt. You've got the biggest companies in the world also issuing debt and equity, which, you know, I never thought I would see Google having to issue equity or debt. And so because they're such a cash flow machine. But when you have that going on, and, you know, one of the first things I look at every morning is, hey, what are the credit default swaps on the debt for these companies? You know, what are they doing? Because I'm trying — because bond investors, you can get a lot of great data from. Because if you think of what a bond investor does: if Google stock goes from 100 to 1,000, they just make whatever the rate was they lent them the debt at, right? They don't make anything extra, but theoretically they could lose all their money if the company goes bankrupt. So they have to do a lot more work than an equity investor that says, 'Yeah, I don't care if one of my 10 stocks goes bankrupt if the other nine go ahead and double.' The bond investor doesn't want one of the 10 that he has to go to zero. So if you look at what the credit default swaps are, which is the insurance for those bonds, you can see that they've all been going up for some period of time. And that should make you nervous because you're being told — and by the way, the cost to insure the debt for some of these big giant tech companies is actually higher than the cost to insure the debt for the North American investment-grade companies in general, which is kind of surprising, right? You would think, oh, well, Google should have — you know, or pick your favorite company — and should be lower. And actually in a lot of cases with these companies it's actually higher. So those are all things that I watch to go, hm, you know, because I view investing as what's the risk you're taking on for the reward you're potentially getting. And you always want to balance those two things.

Oh yeah. No, definitely.

Well, I mean, if you've been looking at some of the stuff I've been writing, yeah, you would have seen it. I try to put out a note every Sunday and I'll talk about the big picture and then I'll bring up, you know, individual companies. But Google's interesting because it's very similar to Meta today because if you go back then they were having issues and then if you remember the whole antitrust trial got decided and it was basically a slap on the wrist. It didn't matter, and then they came out with models that, hey, were really good, and so things started to improve. If you look at Meta, it's in a similar situation right now where a few years ago they had the leading open-source models before we ever heard of this thing called DeepSeek. And then they got into huge amounts of trouble and fell behind. And then Mark Zuckerberg, who I think is probably the best in the world at this because he says, 'I don't care what I need to pay to catch up, I'm going to do it,' and he went and got Alexander Wang. They went ahead and overhauled their models. And if you look at what came out two weeks ago, they introduced Meta Muse Spark 1.3 and it did really well. And so all of a sudden they were back in that conversation of having a leading model in the middle. And by the way, they're going to release open-weight versions of that.

No, I mean, I like Google as well. As I said earlier, I think Google and Anthropic are going to be two of the big survivors. And, you know, you've kind of gone through this thing again right now where people are like, 'Oh my god, Google's falling behind. They've got all these issues.' I think that's true, but if you think about AI models, what makes them smart? Obviously it's the algorithms, but it's also the data you have to train them. And Google, I forget what the number is up to now, but they've got over 10 products with over a billion users each because we're getting all of this stuff for free. What do we do? Well, we give Google the data to train on. And so I think they're always going to be in the hunt because they do have a tremendous amount of user data to train on. And by the way, that's also what makes the Chinese models so good. Because they've got 1.4 billion people they can train on. And, you know, there's other advantages those models have as well. But yes, I like Google. But right now, I think the play is Meta has gone from this company that's, you know, doubling capex roughly year-over-year and doesn't have another way to monetize that spend, right? Where Microsoft, Amazon, Google all have a platform to sell that and to monetize that separately. Well, guess what? Meta —

Actually is starting to build that now as well, introducing an API two weeks ago. Then they introduced their agent product last week, and it wouldn't surprise me at some point in the future if they have a public cloud as well. And the multiple, again, back to risk versus reward, the multiple is well below in the mid-teens, or had been in the mid-teens. I think it's moved up a couple of points relative to some of their competitors that are now trading in the low 20 times based on calendar 27 PE. So, you know, valuation is another thing that I think about as well. And so, I like Google right now. I like Meta more because of the new things that they have being introduced and different ways to monetize that big spend. But yeah, I think they're both very interesting companies.