Big Tech Selloff: Dan Niles on What Happens Next

Yeah, that it would be horrific. I mean, at that point, you'd have to really think about whether we have seen the top back in June, because to me, out of OpenAI and Anthropic, Anthropic is the better company, and so they are in the best position to raise money. You're going to have a tough road if that deal can't get done. I think OpenAI has got a huge problem because they're stuck between Anthropic and enterprise and Google and consumer. And so I think they're going to get squeezed.

Tech is awesome in that the winners change all the time. The pioneer is not necessarily the one that's going to end up being the ultimate winner. Where's AOL today? Where's Netscape, right? Where's Nokia, which had 40% market share in smartphones? You know, where's IBM in PCs? You can go through all of the different verticals, and the guy that started things off aren't necessarily the ones that are going to dominate on the other side of it.

I've been flagging the whole data center risk for a while, and it was Texas that really got me going, 'Oh boy, we could have an issue.' I got obviously more negative around the AI stuff because of it. Because Texas is a red state, right? Texas has been very promotional for data centers. And so when you have the governor of Texas saying, 'Hey, you know, we're going to slow some of this stuff down,' you go, 'Uh-oh, if Texas is willing to slow this down, I potentially have a problem.' And we're all terrified of what happens on the other side of midterms.

It's great to see you too, Wilfred.

Sure. So if you think about it very big picture, if you believe that AI is one of the biggest transformational technologies ever, and judging by the fact it's about 3.5% of GDP at this point, which to put that in context, if you take all the big innovations in the 1900s, so highways, electrification, and internet, all of those three combined don't equal what we're investing in the AI trade right now. So, it's obviously huge, which then means that you're eventually going to have a big bust, because when people realize you have some kind of revolutionary technology, they're going to overinvest, because they know if they're the last guy standing, if they're the Amazon or Google or Meta in their respective areas, they're going to dominate. And so, I think that's what you see going on.

And so what everybody's trying to figure out right now is, have we seen the top in June, or is it a top like what you saw in 1997 and '98? And I think it's more like '97-'98 with the internet, where you had a currency crisis in '97, stocks got slammed mid-year but finished up strongly. 1998, you had the Long-Term Capital Management issue off the Russian bond default. Stocks got hit, finished the year up strongly, and then continued to obviously surge into '99. And I think that's where we are, because agentic AI — and I'm sure you've talked about Muse on some of your podcasts from Meta that just came out, or OpenClaw which came out in January — agentic AI requires 10 to 100 times more tokens, and we're at the beginning phases of that, and enterprises adopting it. But these intermediate sell-offs can be obviously very nasty if you get caught in them. And so that's why I think being nimble around this and focusing on, you know, the different data points that come out, and more importantly how stocks react to them, that's really important.

I mean, and this goes back to my experience with the internet. I remember in early 2000, if you remember, there was a lot of the same stuff. It was called vendor financing back then, which is called circular financing now. And there were deals that weren't getting done. And if you are an AI bear and you think we've seen the top, you go, 'Well, we're seeing that today, right?' So, what have you seen so far? Well, Fermi, which was backed by Nvidia, which is a NeoCloud, that deal got shoved out. NScale, another company backed by Nvidia, that deal has gotten shoved out. AirTrunk, which is backed by, I believe, Blackstone Canada pension fund, that deal got shoved out. You're also seeing all the debt that's already been issued.

So, take CoreWeave as the big example, the poster child arguably for NeoClouds. Their debt that they issued is now trading at 15%. And credit default swaps have surged, including for names like Meta or Nvidia, etc. So you're seeing the cost of money go up tremendously, and the more marginal players are having a real tough time getting it. So that's the first piece of it. The second piece of it is growth. And so if you look at the two poster childs for native AI, which is OpenAI and Anthropic, from the beginning of the year, if you look at something like the Ramp Index, the token spend for those two through early August was up about 9x. From early August to today, it's actually flat.

Now, the tokens underneath it have continued to grow really strongly. They're up like about 80% in that August time period, but the actual spend is roughly flattish. So, what's going on? Well, that goes to the second part, which is: is AI revenue growth slowing? And why is that happening? Well, you've got open-source or open-weight models coming out of China, but also in the US. You've got things like Nemotron from Nvidia. You've also got Meta introducing their own open-weight models. And these are pretty competitive. And if you saw what Google just did this last week, they also are now back to the frontier in terms of model performance with very low cost. And so you also have this going on in the background. So you have both cost of money going up a lot and revenue growth looking like it's slowing down.

Anthropic, which is getting ready to go public, you know, you look at the data and you go, 'Wow, Q2 sequential revenue growth was up 145% sequentially. Q3 was only up 45.' And as I said earlier, like, we'll have to see how Q4 turns out. I'll be actually curious to see if that deal gets done, because you've also got a ton of companies trying to get out in front of them and raise money, right? Meta talked about raising, I think, $95 billion this morning. You're seeing Oracle out with a bond and debt offering as well. You saw OpenAI is also trying to raise money in front of Anthropic's deal. So all of these guys are competing for the same pool of money, and it's the same group of investors, and they're draining a lot of that capital, and investors are getting a little nervous about having all their eggs in one basket, which is the AI trade.

Yeah, that it would be horrific. I mean, at that point you'd have to really think about whether we have seen the top back in June, because to me, out of OpenAI and Anthropic — and I'm sure I've said this on your prior podcasts — Anthropic is the better company. And the reason I say that is OpenAI, when they started, they focused on consumer. Well, consumers don't pay, right? We've been trained that, oh, we get to use Instagram or Twitter or, you know, Google search for free. And so only 5% roughly of their customers actually pay for the paid service. Anthropic started off focusing on enterprise, and that's why you saw Anthropic go from being much smaller than OpenAI at the beginning of the year to now they're bigger, and they focused on enterprise, so they're profitable on an adjusted operating profit basis as well. And so they are in the best position to raise money.

If for some reason — and I think the deal will happen. I think it might happen at a much lower valuation. They might be able to raise less money than what people thought. And you've already heard talks of that following their investor day. You know, that's going to be another thing where you have to consider. And so I think whether or not you think the top was on June 22nd, I think selectivity certainly kicked in on June 22nd when, you know, the semi-index peaked. And so for me, I love this environment because it means you just can't throw money at everything and everything's going to work. There's going to be names that blow up. There are going to be names that do well. And if you do a good job, you can hopefully, you know, hedge the risks that you run into when you're trying to figure out: is this 2001 and 2002 post the internet breakdown, or is this more like '97-'98 where you had horrific drawdowns, but they were just temporary, and then you ripped higher in '99?

Yeah, I would think you're going to have a tough road if that deal can't get done. I think it will get done, but I do think they might end up having to raise less money at a much lower valuation than what was initially anticipated just 90 days ago or 30 days ago.

Yeah. And I think this is why I don't think it's the top. Enterprises are just starting to adopt AI. And so if you think about it, Microsoft has 450 million Microsoft 365 licenses. And the number of people taking Copilot licenses is, you know, less than 10% of that. So you have a long way to go for this being adopted. And then underneath that, you've moved from just chat-based AI, where, you know, you ask ChatGPT a question, it gives you an answer, you move to more agentic AI, which is you give it some task. So let's take you for example, right? Theoretically, before you did this trip to DC, you could have said to whatever you're using, 'Hey, you know what? Book me a trip to Washington DC. Have it leave at these different times. Book me a hotel. Get me a car. Oh, set up a schedule for me, because I got this interview with Dan Niles. Get me some notes on — give me a summary of what Dan's been saying over the last 6 months. And then send an email to my wife telling her my schedule, etc.' That requires 10 to 100 times more tokens than if you just sort of Googled — or not Googled, but used a chat-based AI and said, 'Hey, give me just the summaries of Dan's tweets.'

Right, so that's why — so you move forward to Microsoft, enterprises will pay for AI. Consumers, it's hard to get them to pay for anything because you get search for free, right? So I think you're still very, very early on in that, which is why I find it hard to believe it's the top, because if you think about the internet, there weren't different stages of the internet. There are three different stages of AI. You had training first, right? You got to train these models up so they're smart. Then you had the chat-based back and forth. And now you've got agents that are running at night organizing Wilfred Frost's schedule for the Washington DC trip while he's sleeping. That's why I think we've still got more to go.

And Microsoft is giving you a very safe environment if you're an enterprise, because what are the big AI companies telling you? Some of them at least, which I think is stupid, but they're going, 'Oh, we're going to kill 10% of humanity.' You know, that's not going to make people feel comfortable just going off and buying your AI. They're going to want a safe environment to run it, where if something goes wrong, they can say, 'Well, you know what? I was using Microsoft. Like, how could I have known it would have gone wrong?' So, that's why I think Microsoft is in the catbird seat on enterprise. And they'll route Wilfred Frost's query to whatever agent or LLM it thinks is best. Doesn't have to be OpenAI or Anthropic. The easy stuff, summarizing my, you know, tweets or whatever, it may go to a very cheap open-source model. Something a little bit more complicated, it might send to — you know, and none of what I just said was that complicated, but let's say you wanted to, you know, build some web portal for your Master Investor podcast and set up payments, etc. That might go to, you know, an Anthropic model or one of the high-end models. And so, that's how I'm kind of thinking about it. And I think Microsoft for enterprises is a safe space to ramp this.

Well, they're not going to get it for free, though, because remember Microsoft's going to charge you for Copilot. What it will do for Sky is Sky can say, 'Well, do we need this many people?' Like, I don't know what your staff is there, but I'll give you my personal experience. I thought I was going to hire a research analyst this year. I'm not, because what I'm doing instead is I'm going ahead and using AI to go ahead and get all the stuff I would tell a junior analyst to go do.

So if you think about the example I used with you, you know, do you need as many travel agents? Do you need as many assistants getting you prepared for your interview with me, or crypto — you said you had an interview later? Like, you can go ahead and tell an AI while you're sleeping, 'Hey, go do this.' And the AI doesn't mind if you come up with an idea at 1:00 a.m. in the morning and you give it something else to do, right? Because some brilliant idea came to you. So, you're going to pay for this. Sky will pay for this. Now, Wilfred Frost may go, 'Hey, I'm used to doing stuff for free, so I'm going to use the free version of ChatGPT' — or, I mean, I pay for the ultra version of Gemini. And, you know, so it depends on where you are, but enterprises are going to pay for this stuff, because if Sky gets hacked and all that information gets leaked, you guys are in big trouble, right? If you did something yourself, if you're inside the Microsoft environment, you can say, 'Well, hey, I'm in this environment. I did everything the way it was supposed to be done. I'm protected.' So, we'll see what happens. But this is why I think we have a long way to go, to your point, based on, you know, where you work.

Yeah. I mean, I think the big thing — so I'm going to bring up Google as well, because I'm sure you'll get there, but if you think of the two of them. So with Google, you've got search, and there's this question of, okay, if — to use you as the example again, if you're having your agent do all of this while you're sleeping, the agent's not looking at an ad, right? Which that was the big issue with Google. Their search revenues came in below plan for the second quarter in a row, and their YouTube revenues also missed. And so people are going, 'Huh?' And if you remember back a couple of years ago, you had the same question with Google where people were like, 'Well, wait a minute. Is search getting disintermediated?' And then it was like, 'Oh, Google's starting to monetize. It's all good.' That question's come up again.

If you look at Meta, well, what do people go to Meta's site for? Well, they go there to be entertained, right? You watch — I like watching dog videos, you know? You watch whatever it is, keep up with your family, your relatives posting their Thanksgiving plans, Christmas plans, etc. That's more of an entertainment site. So, the agentic world is not necessarily going to disintermediate Meta as much, or there isn't that threat in my mind relative to Google. And so I think Meta, with the fact that they got out what I call an agent for the rest of us, like OpenClaw, which was formalized on January 30th, right? You had to buy a Mac Mini, you had to set it all up. Meta makes it easy with Muse. And I think you're going to see every company come out with that. And, you know, Meta is in a pretty good spot to potentially monetize that. And you're seeing it in the efficiency of their own business. The negative, of course, is negative free cash flow, the amount of money they need to raise. But I think from a 'how protected is their model, their business model, their base business model from AI disintermediation' — because you go there for entertainment — I think it's got a pretty big moat, relatively speaking, because of that.

Great question. This is the thing I'm looking at that I think means it should be okay longer term. They said their query growth was the highest in five years. So think about it this way, and to me this is the analogy I've drawn in my head, which is: I remember when Meta came public and the stock collapsed on its IPO, and I was short the stock from the day it came public, because I'm like, well, everybody's accessing Meta through this thing called a smartphone, but they don't have any ads on smartphones, so how are they going to make money? It's the same thing with Google, which is they have all these, you know, the Gemini-related AI overviews that you see, but they're not showing ads on some of them. And then the number of ads that they're showing on the ones that they do isn't that much. You've got the queries clearly. So, people are using this. So, now it's just a matter of them showing ads on more of those results, and then, you know, the number of ads. So I think Google longer term will be well positioned.

But let's get to the end state. The end state of this, I think, will ultimately be Wilfred Frost accessing his AI through his smartphone. Now, if Wilfred Frost uses Apple, he'll just go, 'Hey Siri, go do this and this for me.' Siri will then access whatever agent it's associated with. Now, as you probably remember, Apple has an agreement with Google. They're actually paying Google for AI, whereas with search, Google is actually paying Apple. So, it shows you where some of that balance of power is. Now, obviously, the other 75% of the world's phones run on Android. Well, that's Google's operating system. And so my belief is the consumer, it will access AI through whatever smartphone it is, because they have that on their bodies, right? It's in your pocket 24/7. Most people sleep with their smartphones, I think, right next to the nightstand. And so that's how you're going to get to it. And Google and Apple are both very well positioned, given they dominate on smartphones ecosystem, and Microsoft dominates in enterprise. And so that's how I see a few years from now how the winners end up evolving between all of that.

And that's why I think Google, given they're on 75% of the smartphones in the world — obviously the lower-end stuff, Apple's got the higher-end quarter. That's how I see this all playing out, and why I think Google is fine. But you know what? This is one of those environments where, you know, another great Brit's quote, right? John Maynard Keynes: 'As the facts change, I change.' What do you do, sir? You have to be very nimble. And that's why for me, when I previewed all these results coming in, I was very nervous about a selloff. It happened. But that gives you a chance to re-engage at much better prices and say, 'Hey, you know what? This is how I see things shaking out.' And, you know, if you run a hedged strategy, then you can say, 'Okay, well, I hate the NeoClouds. I can be short those, and I can be long a Google or whatever — pick your favorite.' And then hopefully that'll work out on the margin going forward. We are at a more selective phase.

The short answer to your question is I don't know. It depends on how investors take this, because that's where the opportunities are, right? Is when investor perception and where the reality is really get out of whack. So for me, I always like to let the dust settle to some degree. But I think next year, if I think about it, because obviously, you know, we got a couple more months left to go in the year, things will get sorted out. I think next year Google will have a great year. So for me, my goal is to get involved with the stock when I think the stock has itself bottomed. And, you know, we'll see where that is, right? That's a valuation question more than anything else.

I mean, think Facebook, right? If you think Facebook back in 2022, the stock at one point, I think, was down 75%. I think it finished the year down 60-plus percent. And then I think the next year it, I want to say it at least doubled, potentially tripled. And so you don't want to get stuck in the 75% drawdown, but if you can get it properly, which thankfully, you know, I had some CNBC interviews and I have actually done one with you back then, but we bought it. I mean, couldn't have gotten much better, and one of the names we got right. Right. And the stock's done really well, but you don't want to get caught in that big drawdown either. I don't think Google's in that position because of that query growth being the best in five years. And I think investors, as they think through this, will get there, but you don't know how long that process is going to take, given the experience with Meta back in 2022 when Zuckerberg went down the whole metaverse path and then, you know, reversed that into 2023.