The AI Trade Just Changed... Here's Where the Smart Money Is Going

stocks are delivering blockbuster earnings, but many of the biggest names are still selling off. Today, we have a special guest joining [music] us to break down the latest moves in AI and technology stocks, and you won't want to miss his insights. Welcome back to Naval Market Buzz. Joining us today is Tieran Ray, publisher of the technology letter. Tieran, it is so good to have you. Welcome to Navler Market Buzz. >> It's my pleasure, Crystal. Thank you for having Yeah, it's really exciting to have Tier in here because we need to make sense of this crazy oscillating market in tech stocks. >> I don't know if we can make sense of it, but we can try. >> Good. >> So, let's start with what has everyone scratching their heads. ASML and TSMC both beat earnings expectations and raised guidance. Yet, investors sold the stock anyways. So, what message is the market sending? You know, it seems to me that we're getting less of the information people used to depend on and more of the kind they're not used to. ASML used to talk about product orders and they sell product, you know, in the dozens of units a year. It's not a high volume. They stop this year disclosing order numbers and they speak vaguely and in the meantime they talk about sort of future possibilities way out into 2028. So I actually think the business is quite healthy but people are getting a lot of information now from management from the market about something's going to happen in 2027 2028. They saw the same thing from TSM yesterday right we're building 265 billion worth of manufacturing in the US based on expectations for big chip demand and again it's projected out into the future. So you have companies that, you know, it started with Apple like 8 years ago not disclosing iPhone units anymore. And now we've got companies such as ASML that for years were very careful about telling you what they're going to sell. And now they're saying, you know, we'll give you a vague idea or we'll report once a year. We're getting information that is far in the future that is speculative and we're losing some of the information from semicmpies for example like ASML that used to be rock solid because it was just about current demand trends. >> Right. >> Tieran Bpoke has documented that AI infrastructure stocks are performing better than pure AI stocks. Does that boot well for GE Venova's earnings announcement on Wednesday? >> Sure. Absolutely. The GE Vernova stuff as you know Louie is backup generation and there's companies having to reach for anything they can find because the utilities are not moving at AI speed right to get a utility to provide generation for a multiund megawatt data center is slow and that's if everyone agrees New York City New York state I'm sorry this week you know governor Kathy Holchel put a moratorium on new data centers uh there was something in Florida where there was a tango data center from one of the projects down there, another big AI data center that local officials blocked. So, this is a trend to not even approving permits. There's time, right? There's permission. Then there's actual energy generation capacity and all these things are causing people to reach out to things like Bloom Energy and to GE Vernova, which as I understand, Lou, you might know them better. My sense is it's a better gas powered backup generator because it comes from rocket engines, jet engines, and so it's it's more sophisticated than traditional kind of GE generators. That's my sense. >> Tierin, Elon Musk just bought APR Energy. Is this a start of a bigger trend of AI companies buying power assets? Now, Crystal, it feels to me more like an Elon move. Meaning, what we seem to know is that he has a big facility for X AI, you know, which does X/ Twitter and Grock in Tennessee. And apparently, they needed 150 megawws to power it. They could only get 50 megawws from the utility. And so, lo and behold, Elon decides to buy a company that produces these power generation like mobile generators. I'll just buy my energy. But I think that it's flashy and it's interesting. But I think that most companies that have to contract for hundreds of megawws or possibly gigawatts would uh want to either go to a reliable provider. Let's say it's Bloom Energy which has fuel cells that they can put on site or they'll deal with the utilities and they'll just go through that process as opposed to sort of this outside outside move to capture something which feels to me like it's a rush to say, "Oh, I need it now. Uh, I'm going to, you know, I'm going to just spend a billion dollars and buy this thing. So, I don't know. I mean, maybe I'm wrong, but it seems to me like an Elon move where it's I had some money lying around today. I need it. Who knows about >> That's a great point. Let's go back to this data center ban that basically started with New York. It's really not a ban. It's just a one-year moratorum. >> Yeah. Right. >> So, because you know New York is perfect for data centers. You got hydroelectric. You got a direct current grid from Quebec hydro data centers like green energies. Obviously you have that in upstate New York. You got lots of land. You got fast internet. It seems like this fuss is you know the people in Buffalo got upset that their electricity went up 20% in the last few years because of the data centers. But I would think up upstate New York is perfect for data centers. >> Well, you're saying that politicians should not cater to constituents. >> Well, no. Well, you can build your own energy, you know, like you can get a natural gas line or you can negotiate a a deal where everybody gets free electricity, you know. >> I I mean, I think that the urgency, you know, what you're pointing out, Louise, the urgency is not at all clear to citizens, meaning that the industry is saying we need this yesterday because we want to sell X amount of AI. No pun on X AI. and the citizens are saying we failed to see why chat GPT is not functioning adequately and you're telling us that we need this and so there's a real disconnect and maybe you could say if you want to be cynical you could just say it's a marketing fault on the part of Alphabet Amazon Microsoft Meta Open AAI Oracle SpaceX that they have not adequately communicated to people why other than you know we're going to bring jobs we're going to bring local commerce we're actually maybe contribute let's say they we're going to contribute to local infrastructure You know, there's a lot of they talk about benefits in the traditional way that, you know, if you're going to build a cement factory, you talk about benefits, but they are not communicating to, I think, a population that feels baffled by what's the big rush and and they see that it is in some way caught up with equity markets, etc. And so, you know, if you want to one point of view is you could say it if you believe it's an ultimate good to have the data centers and you believe that it will outlast the current boom for decades possibly, which is quite a possibility. Maybe that has not been adequately communicated to people are seeing it come in a rush. >> Yeah. Yeah. Well, you know, Larry Ellison bought Lion Country Safari down in Florida and I don't know what happened to the animals, but uh well, he lives down the street from me, so I know there's I haven't seen any elephants or giraffes running around, but that's anyway. Uh he does have a big estate, but what I would say is there's local opposition building in Florida to putting a data center uh where Lion Country Safari was. So we don't know if Ellison's behind the data center. We assume he is because he bought bought the park and but yeah there's local opposition everywhere in >> and has it been explained to people we really need this here because some public benefit let's say public good or if not you know to have that debate we don't need to justify this to you this it's private property etc. Yeah, this is uh not the most affluent part of Florida. It's on the way to Bellglades, so I know Florida is growing, but you know, I really don't question the location, but yeah, I'm not part of the local community and so we'll see what they what happens and whether it ever gets built. Even in Reno where there's lots of data centers, we get occasional grumbling, but I haven't seen any projects be stopped. Let's move on to a harder question, and I think you're more qualified to answer this than any of our other guests. Can you please explain uh to our viewers what AI tokens are and what happens when they run out? Is this going to bode poorly for OpenAI in the future or even Anthropic? When you type into chat GPT, help me with my resume or help me with my pitch deck. Everything you type is getting sent to the server on the other side. And everything that comes back that says Lou, I would be happy to help you. Shall we start with an outline is all that data sent to you that appears in in the response in the prompt. All that's being counted both outgoing and incoming. And the AI providers such as OpenAI decided that's how we'll price for this thing. That's how we'll monetize. You have a free version of chat GPT and then you have a pro version. And then for corporate types, they pay for every chunk of data. So all they did is they arbitrarily chopped it up and they said for such and such amount of words that you type phrases, we'll charge you this much. And so they charge ranging from pennies, literally pennies, couple pennies per let's say character or per word that gets sent out and comes back on up to tens or hundreds of dollars. And so those are called tokens. The token is just a fancy word for this is the you know the micro payment for this amount of data. And it's exploding. So one estimate from this company open router Louie this is 50 trillion tokens of data a week right. So you're talking about hundreds and hundreds of trillions of tokens on up to quadrillions each year of people as they use more because it just is expanding. More people are starting to take up the tools in corporate enterprises uh to use this stuff. And so the whole hope for SpaceX's XAI unit for open AI for anthropic. The reason they have trillions in valuation is we're going to make hundreds of billions selling these tokens because more and more people going to use it. It's like an expanding gas. And you've seen probably Louisie this week a number of discussions in sources such as Bloomberg about the Chinese AI models. The Chinese surprise price lower. So my sense of anthropic for example might be like a $100 per token at the high end of what they offer like the premium offering for enterprises. These companies such as Moonshot which makes is a Chinese company startup makes something called Kimmy which is an increasingly prominent so-called frontier AI model. They charge pennies I think it may be 10 cents or lower per token. So you got 10 cents in per token from Chinese models say 100 plus from the topend anthropic model similar for open AI and SpaceX. So it's a east versus west discussion of token deflation and I think it's real. I think that the stuff from everyone else is going to be just good enough to use. I don't think you'll need to use you know anthropics claude fable to do a lot of stuff. I think that the gains have plateaued and so I I've thought for a long time Louie that it will commoditize and I actually thought Amazon would be a big beneficiary that they would provide you know the same way they did with books. They just bring down the cost of using this stuff. You buy enough compute and you can make it cheap. And so when that happens, you have a tremendous deflation in what was expected to be the money you'd make for every bit of data, every word, every character, every phrase, however, it's chopped up into tokens. So token deflation is a is a real issue for all the aspirations for the revenue that's supposed to come to SpaceX's XAI, to anthropic, to open AAI, to Google Gemini services, etc. That's my sense. You know what's interesting about that is I personally have experienced kind of that cap limit. I use a bunch of different AI platforms, but like say for example, chat GBT, there's a certain amount that you can use it free before it stops giving you good responses and then you have to start paying. Now, I actually have a little bit of a cheat code because I have a bunch of different Google emails. So once one kind of taps out, then I'll move to the other. But obviously >> Wow. Okay. >> Yeah. [laughter] Louis and I as newsletter publishers hate that because somebody comes back to get the starter discount [laughter] with a new email, right Lou? >> Yeah. Does Google have a police force? >> Hey, listen. There's nothing wrong with it, but obviously I do have to pay now. So, you know, I am personally experiencing kind of that difficulty. So, I'm wondering, do you expect more of a push back when it comes to these tokens? And do you think that this could hurt companies say like Meta or maybe other stocks that could suffer from this? >> Yeah, I think in aside from deflation, Crystal, aside from we're just going to go to the Chinese stuff because it's cheaper or we're going to run it locally on our data centers, meaning the enterprise will run it. So, Proctor and Gamble will run it or JP Morgan will run it locally. They won't have to pay the the tax, the fee of tokens. Aside from those two things, chasing the cheaper offering or putting running it locally and avoiding the cost altogether, I think there's going to have to be a reckoning about what are you getting for the tokens because if you're typing for free, help me with my resume chat GPT or help me talk about my relationship with my kids and you're getting stuff. It's like, okay, who cares? You're not running up thousands. It's it's a $30 a month or whatever. But when you talk to corporations and they struggle to, as I understand they are, they struggle to find out is this stuff going to like replace our HR department or replace our finance department, they start to scrutinize what is coming from all these tokens. And it's not just the cost. It's the endless activity that doesn't necessarily generate a clear output. And so I think at some point you have to replace I would think if you want to be anthropic and have a trillion dollar valuation you have to replace we're going to charge you per token with something that's more like you know a price for value based pricing. We're going to give you a deliverable. We're going to package a result if possible. This is going to really double the output of I don't know your finance department's general ledger. something that's meaningful that you're paying for instead of just it's a bucket and you pay for it in, you know, three different, you know, fast, medium, and really slow, which is what they're doing. >> Yeah. I'd love to ask, are there any stocks right now in terms of technology that are up your alley or you're looking at right now? I still like the semiconductors Crystal because if you think of all this token discussion we've been having as the froth meaning we can't figure out how much is really going to be sold how fast it will commoditize and the price will go down. We don't know exact aggregate data center demand because to your discussion you know bu about nova there's promise in what's being built out but there's also an uncertain horizon I think of the semiconductor stocks as doing something very predictable over time which is they get better at the incremental cost of making a chip. They build factories and at periods there is over supply but basically what they're doing is they're getting better and better at driving down the cost of making units of chips. It's not clear to me that the world's going to stop using computer chips. So probably in aggregate there'll be more chips and things smartphones and servers and AI data centers and the progress economically of companies such as Micron Technology, SanDisk, Broadcom, ASML as the company that serves them with tools. Taiwan Semi is the company that actually manufactures runs the machines to manufacture. All these companies get better economics over time in aggregate over the long haul because they get as as they as they serve demand it allows them to retool factories they build more efficiently. And I think we don't see it crystal in a two-year time frame or quarterly time frame. But I think it operates over decades for them. To me, it's just simple. That's all I can handle. I can't handle what is the gas powered turbine trajectory for Rena because I don't know their business well enough and I don't understand the industrial economics although I think it's very valuable and I don't mean to discredit them I can understand you make this wafer and you chop it up into chips it's like brownies and you get better at doing it over time and that makes the asset value >> wonderful so how do people find out about your technology letter >> yeah thanks Louis you know the technology letter which is a website and an email that goes along with it www w.thetechnologyletter.com is the main page of it is like the landing page is like a public portal. So you can without subscribing you can just browse it in your web browser on your phone on your iPad and see chunks of what are the day stories every day of the week and then if you want to go deeper on it if you want to read it click on them and you can subscribe it's 30 bucks a month and it's currently a dollar a week for the first four weeks so it's an incredible value for $4. And if you decide that after $4 you're not completely satisfied, you can cancel me. >> Great. Well, as always, we will have your links in the description. But Tunan, thank you so much for joining us. >> Thank you for having me. It was delightful. Thanks, folks. >> And thank you all so much for watching. If you enjoyed today's discussion, please like and subscribe and also turn on your notifications so you never miss another episode of Navier Market Buzz. But we'll see you this Wednesday for a midweek update.