Dan Niles: Be Nimble - 30-50% AI Crash By 2027

It's obviously Google. I mean, as I said earlier, they have the full stack and they're the ones you should bet on. And unless something drastically changes, they will continue to be the winner going forward because they've got everything and they've got the massive cash flow to fund it. So, in the short term, are semis at the most overbought they've been since 2000 or 1995 before they had horrific corrections? Absolutely. But if you're taking a longer-term view and you're trying to pick individual names, Intel's just gone back to where it was in 2000 this past year, and it's still undervalued. To me, it doesn't make sense that the stock market's at an all-time record high and oil prices are up 60% this year. And the bond market, we saw last week, the 30-year yields and the 10-year yields hit the highest levels for the year. Like, one of those is wrong.

100%.

My pleasure, Wilfried.

Yeah, I mean, if you look back and compare where we are today, you're about three years into this buildout for the AI infrastructure. ChatGPT came out at the end of 2022, so you're, you know, entered year four. If you go back to that time period, '97-'98, and you use Netscape Navigator's launch as kind of the starting point of this new thing called the internet, then '97-'98 you were in years three and four of the internet infrastructure buildout. But back then in '97, as you may remember, you had the Thailand currency crisis. And so then the S&P dropped 11% intrayear in 1997. And then in 1998, you had the Russian bond default and the failure of Long-Term Capital Management, and the S&P went down 19% intrayear. But if you look at the full years, '97, after that 11% intrayear drop, finished the year up 31%. In '98, the S&P finished after that 19% intrayear drop up 27%. And why was that? It's because you had this backdrop of this huge underlying buildout of this new thing called the internet. And so that's kind of where we are today, in that yes, you have this macro issue around oil and Iran, but in some ways it's a lot easier to deal with, right? Because that was a, you know, a man-made event, the war, whereas those currency crises, Russian bond default, you could say was more structural, several years in the making. And so that's why I thought this would be the bottom, we'd get through the war obviously, and then the markets would, you know, continue to bounce off those levels and continue to run up because of this AI infrastructure buildout and one specific thing being agentic AI really kicking off in the beginning of the year.

Sort of yes and no. You did get a reset because of the Iran war, but the big thing that happened was this thing called agentic AI. And by that what I mean is your viewers may be familiar with this thing called Clawbot that came out at the end of November, and it morphed, and by January 30th it was called OpenClaw. And so what that enabled you to do was before you would ask your ChatGPT or Gemini a question, it would give you an answer. So that's chat-based AI. With agentic, you might go ahead and say, 'Hey, this is Wilfried, and go get me this data from the BBC website. Go get me this data from the Bloomberg website. Go to CNBC, get me this other stuff, and then go ahead and put this all in an Excel spreadsheet.' Well, that requires 10 to 100 times more tokens to be generated. And you can see that in the token generation data, where in the two months prior to January 30th and the formalization of OpenClaw, token growth was about 20% over the prior two months. In the two months following, the token growth was over 120%. And that token growth is continuing to grow because corporations are just really starting to adopt this because this is kind of what the promise of AI was, which is you replace people with this agent that can then run, you know, 24 hours, 48 hours and get whatever project you want done versus just a chat-based AI query.

You need to have it reopen. There's no question. I mean, you might by some miracle get out of this, but 10 out of the last 12 recessions have been preceded by a spike in oil prices, a sustained spike in oil prices. So, if oil stays up at these levels in the mid-90s, and remember, we were sort of in the 60s before this started, you're probably going to be flirting with a recession by the time you get towards the end of this year. I mean, McDonald's, which reported recently, you know, surprisingly came out and said, 'Hey, we're seeing some stress in our low-end consumers. Their comp store sales missed.' You know, obviously they're all over the world. And so, when you see things like that start to pop up, you have to worry. And oil prices, they just seep into everything, but it takes time. So a spike and then a drop doesn't really matter. If it stays up at these levels for one to two quarters, you're going to have a huge issue because that'll seep into inflation. It'll go ahead and sap consumers' earnings power, and what we heard from McDonald's, you're going to hear from a lot of other companies going forward, and this is going to turn to a huge problem, and at that point you're going to get a massive correction in the stock market. But the thing we know for sure, right, is we have an election in the United States coming up, midterms in November. The incumbent party, when you have a spike in oil prices leading into that, just gets absolutely crushed. And so President Trump's going to want to get oil down, and that means coming to some kind of resolution. And from what you've seen, right, that's where he's going. He's calling these things love taps, these, you know, exchange of military actions by Iran and the US, which tells you he really wants this thing to end.

Sure. Well, as I said earlier, with agentic, you're doing multiple actions, multiple calls to APIs, you know, opening up applications like an Excel or a Bloomberg or, you know, a CNBC website or SEC website, etc., and doing all of these things. And so, you need an orchestration engine. You need something that goes ahead and organizes all of that. That's what CPUs do very well. And if you go back to training, you're just doing the same thing over and over and over again when you were training these AIs. That's what a GPU does incredibly well. And that's why you saw like an Nvidia absolutely explode higher because CPUs weren't really valuable in training. You move to chat-based AI, again, that's pretty decent for a GPU. But now when you move to agentic, that's great for a CPU. So you went originally from about, let's say, eight GPUs for one CPU. When you move to agentic, that ratio gets closer to one-to-one. And so that just again shifts what kind of hardware you need to run this. And you know, Intel has been given up for dead for decades, right? The stock has just gotten back to its peak in 2000 this year. And but between this move to agentic plus the fact in Intel's case that they're probably going to end up with some pretty big foundry customers by the end of the year to help with supply being secure, being produced in domestic fabs run by US companies in the United States, and in this case Intel, that's really good for them. Obviously benefits names like AMD as well, which I also mentioned back then.

Well, I mean, here's the thing. There's obviously the short term and then there's the long term. If you think about it, we still haven't had Nvidia or Apple or, you know, any of the people that have talked about getting a relationship or Tesla with Intel starting to produce yet. If you look at it from a valuation basis and you say, 'Okay, the one thing you know about semiconductor stocks is the earnings estimates are always wrong, right? They're either way too high or they're way too low.' If you look at valuation and you say, well, I'm going to just look at it from an enterprise-to-sales basis, and you look at where an AMD is trading or TSMC, you can say, wow, you know, Intel is still undervalued relative to those companies on an enterprise value-to-sales basis. And so the stock should be able to continue to rise. And from a terminal value standpoint, like what is this stock worth two to three years from now, the composition of the business should look a lot different, right? They'll probably have some combination of Nvidia, Apple, Tesla, others producing chips on their foundry business, and that will make the company a lot more valuable than what it was in the past. So, in the short term, are semis at the most overbought they've been since 2000 or 1995 before they had horrific corrections? Absolutely. But if you're taking a longer-term view and you're trying to pick individual names, Intel's just gone back to where it was in 2000 this past year. And so, and it's still undervalued relative to what it can probably ultimately earn. And so, would it surprise me to see the stock go down 15 to 20% along with the entire semiconductor index over a short period of time? Absolutely. It wouldn't surprise me at all. But do I believe that by the end of the year they'll be higher than where they are today? Yes.

No, because the other side of it, Wilfried, is that agentic requires 10 to 100 times more tokens than chat-based AI. And so that's the difference. You're not talking about a small change. You're talking about an absolutely enormous change. I think when you look at where capex growth is right now, you can see that the numbers have moved up pretty dramatically. So you started the year thinking hyperscaler capex up maybe around 30% for the year. Then Q1 guidance was given when companies reported Q4 results in, you know, January, February, and then that number moved from 30% capex growth for 2026 up to about 60%. Then, you know, a couple of weeks ago the big hyperscalers all reported again, and that number moved from 60% capex growth to 70%. And so then this ties back to the fact that you're not talking about a small change in token generation required with agentic. You're talking about 10 to 100 times more tokens needed when you do an agentic query relative to a chat-based query.

Well, on the margin, it'll change how the stock performs. That one obviously, Broadcom's another one relative to some of the other players, right? Because agentic requires more memory. It requires, with some of the architectural shifts underneath, it requires more optical. And so, you only have a limited budget. You don't have infinite money out there. And so what you're going to spend that money on is going to shift. And if the ratio was eight GPUs to one CPU, and that ratio is going to move closer to one-to-one, then obviously the stocks that are tied to that, those will be the ones you want to take a look at. And so that's why for me the microprocessor vendors are a lot more interesting right now than the GPU vendors or other parts of the food chain.

I mean, absolutely. I mean, you're seeing it in the numbers, right? Where if you look at Google Cloud for the fourth quarter, the growth rate year-over-year was 48%. When they reported the March quarter, that growth rate went from 48 to 63. So that's a 15% acceleration. You look at AWS, that growth rate went from 24% to 28%. So 4%, given that they're the biggest of the hyperscalers, that's not bad. You look at Microsoft, however, and you go, that growth rate went from 38 to 39. And so, you know, they're clearly the one that's struggling, but it's still accelerating slightly. And that's why you can look at kind of these are the numbers versus what these companies are all saying and see who's actually executing well and picking up a lot of share.

Well, the thing is Google has the full stack, right? If you look at them, they've got a leading large language model in Gemini, right? Microsoft relies on OpenAI, so they don't have one of their own. Amazon, who's heard of Rufus? So, they've got that. They've obviously got their own chips. They've been producing their own internal chips for over a decade, which is the longest of any of the three. You've also got enormous cash flows funded by their ads business. And then you have end devices that they can distribute this technology through with their Pixel phones, which are really the model for the entire Android ecosystem, which is over 75% of all smartphones sold worldwide are Android phones. So they've really got that full stack from top to bottom, which no other company has.

I mean, I'm not sure exactly what his motivation is, but I would say this. If you look at what Tim Cook did well, he was really good about the supply chain continuing on the legacy of Steve Jobs when he took over. Don't forget the iPad had just been launched. And so the year he took over, Apple's revenues were growing 60%. And the next year, I think it grew like 40%. And then the growth obviously has slowed down since then. And so what innovative products have really come out? There hasn't been a lot. If you think about the last one, it was this Vision Pro ski goggles, as I like to call them, and nobody bought those. And so I think now putting a hardware guy in charge that hopefully can give you some innovative products, I think that's going to be interesting. And this year you should get an AI-enabled Siri, right? And they're a couple of years late on this product, and finally get a foldable iPhone probably late this year, early next year. And that should drive a pretty big upgrade cycle because when you've seen big upgrade cycles, typically they're driven by changes in the form factor. And you saw that with the change to the iPhone 6, where you went from a four-inch screen size to a five-and-a-half-inch screen size, and the growth accelerated from 7% growth on the top line to 28% growth on the top line in their next fiscal year. So, I think that's what I'm kind of looking at and excited about. Obviously, the valuation's really high, but what you've seen is that people sort of do seem to treat their iPhones like consumer staples products. I mean, it's shocking to me, and Apple was one of my top picks coming into this year, but it's still surprising, even with oil prices as high as they are, that you've had the fastest two quarters of year-over-year growth rate since 2021, when you had COVID and everybody had to figure out how to work from home and were buying, you know, iPhones and everything else to make that happen.