SPEAKER_00 0:00

You are listening to the AI Investor Podcast from 24-7 Wall Street. On today's episodes, we look at momentum stocks bouncing back hard. The next deep seek moment arrives at the heels of Kimi K3's release. Investors panicked about a new Chinese model. We're going to see what that means for leading U.S. companies. We're also going to discuss whether or not natural gas is actually the next AI bottleneck. And we cover the recent news from AMD signing a massive deal with Anthropic and all the latest from the earnings season. All that and more is next. Eric, uh big market news to talk about. We saw a massive bounce back for the momentum trade last Tuesday. And this was after stocks had sold off hard after the release of Kimi K3, which I know we're going to talk about. Many stocks in the portfolio had seen a lot of volatility after the Kimi K3 release, and then a huge rebound on Tuesday. Air Test Systems, which is one of the, you've been pretty clear, one of the most speculative positions in the portfolio, but also one that has the highest upside and is perfectly positioned for where chip manufacturing is going. Um, had a 25% day in the green. So what was going on here? A very volatile week. And I know we're still in it, and we're gonna talk about that. Some of the hyperscalers selling off hard again today, but tell me about this momentum rebound you saw last Tuesday.

SPEAKER_01 1:21

Yeah, volatile's the word. You know, Austin, sometimes you feel like a million bucks, sometimes you feel like a momentum stock in July. I I kind of feel like a momentum stock in July right now. I've been sick all week. We've we're gonna, our editor asked us, don't say you're actually gonna keep it to 30 minutes because you never will. I'm saying we're gonna do it because we do have a hard stop. So there's been some really interesting news that impacted the market in a big way. There's been some macro issues. We'll kind of get to them all here. We're gonna try and do it in a pretty tight 30 minutes today. But Austin, the the big picture is what you said Tuesday. I've talked about this indiscriminate selling happening across the market, that it's a rotation, it's a factor rotation, it's moving out of risk, it's moving out of momentum stocks, and it's kind of doing it uh indiscriminately. So you're seeing these uh large sell-offs on a lot of positions, sometimes, sometimes regardless of what we see as the underlying quality. Um, so on Tuesday, though, we did see that bounce back. A lot of the stocks that were down the most commensurately were up the most on Tuesday. But today, Austin, we are now seeing kind of a reversion to this. The Nasdaq is down 2%. What I like that I'm seeing, Austin, today is that it does seem a little bit more logical in selling. You aren't seeing just this um broad sell-off across AI stocks. You are seeing some stocks that will have had positive news, such as the impacts from Google earnings, relatively strengthened today. Well, a lot of other stocks are seeing kind of um more exaggerated sell-offs. What I do like about what I'm seeing today, Austin, is um a lot of the selling, it seems less like this large rotation of momentum in AI infrastructure and a little bit more um, for lack of a better word, logical. You know, a lot of companies that will have benefited from news flow are actually up today. Um, I think my portfolio is only moderately down despite a 2% sell-off, which stands in contrast to the kind of selling we've seen in recent weeks. So that that's kind of a positive, you know, when when you see this kind of factor sell-off, stop being the predominant reason stocks are selling, that that often um speaks relatively positively. But, you know, we do have the shadows that um we we've talked about probably the worst thing that could happen to the AI trade at this moment would be soaring interest rates because it's going to affect this next phase of needing new financing for a lot of the build-out. And it's going to generally hit a lot of stocks with this momentum profile. So definitely probably, probably a little bit more ominous these past couple of weeks. But in terms of actual news flows from companies, as we'll cover, it's been a very good past two weeks.

SPEAKER_00 4:07

Now, if people want a good visual or data point of what we're talking about here, you know, being the maybe a macro driven moment right now, look no further than Alphabet. It was down 7% this morning, it's still down about 6.5% today. This is after earnings, and they announced that they're going to be raising CapEx. If that sounds familiar, right? That's the story of the last three years for all hyperscalers, up above 200 billion. But as we've seen, CapEx is now starting to enter the finance moment. So they're, you know, Google, Alphabet down 6.5%, Microsoft and Meta down a similar amount. Um, so this is one of the things we're talking about. Now there's others, right? And high energy costs also do inhibit and to, in some way, the data center build-out. Now, behind the meters of solution there. But there's a lot of factors going on, high interest rates, uh, high energy costs that, um, and you know, maybe a rotation out of this trade. But glad to hear that you're still seeing some good news behind the scenes, and I want to get to that. Let's go to a very specific news item, and this is uh the release of Kimi K3. So the model may be Kimi K3, but you may as well call it DeepSync2, because basically the same thing happened. So K3 is an open source Chinese model. Uh largely seems to be derived and distilled from US models, including Fable, but it's open source. Uh, it's very affordable, I believe. And as far as I have seen, it's it's performant and comparative with uh Anthropic's current Fable model. One thing worth noting, though, is the current version of Fable that we have is not the mythos level model that was previously, you know, gave birth to Fable. What we have is sort of a nerfed version of Mythos. So what we actually don't know is how far is Kimi K3 from the Frontier version of Mythos that got sort of nerfed before it was released to the public. But what we can see is it's at least very close to Fable. But with that context, with this Deep Seek 2.0 moment, what is Kimi K3? Is it better than US models? What happened here?

SPEAKER_01 6:03

Yeah, I think we'll attack a few different paths here. You know, first of all, uh, like you said, what is it? Second, is it better? Is it slightly worse than US models? Third, how did they do it? Which you kind of got to with distillation and that tee up. And fourth, you know, who does this ultimately benefit or what does this news mean? So um, Kimi K3, as I mentioned earlier in the podcast, it was released right after we filmed our last podcast, um, which we're kind of almost missing the big news of the week. But um, the the kind of panic, Austin, comes from the fact that upon its release, it it's released and it shows benchmarks and it was competitive um with many of the leading US models on the market today. The US's goal has been to maintain something in the range of like a six to nine month lead above Chinese models. And they're doing that with trade tactics, you know. This is no leading edge NVIDIA chips into China, no EUV machines into China. So if China is able to catch up with the US with less of this compute, well, you know, there's a number number of quandaries that would exist, you know. Is all this money spent being wasted? Um if you're invested in the AI infrastructure trade, uh is this compute actually not necessary for the best models, which was the original deep seat question, right? That that really brought down a lot of these companies. So let's unpack a few aspects of this. Is it actually better than the US models? As I noted earlier, the benchmarks of products um when it was released were very strong, uh, especially in areas like coding. More importantly, though, it was a significant leap in capability from prior Chinese models. And it was released just seven days after GPT 5.6. We talked about the growth rate from GPT 5.6 for OpenAI and 37 days after Fable 5. So, you know, does this show that the leads have been cut? What's what's important to understand here is how models are evaluated. It's kind of like humans, right? Austin, if you were talking about how smart someone is, you might have a friend that um, you know, in college, they get grades across different subjects with different strengths and different weaknesses. And then you have an overall IQ test to assess someone's general intelligence. And Kimi K3, it seems to be very good in a couple of areas, um, specifically agentic and vision. Those happen to be almost all the benchmarks that the company released. But when you look beyond that, it's less broadly intelligent for models from open AI and also anthropic. So there's a there's a third party. I've I've talked about them before on this uh podcast, it's uh Epoch AI, and they do an IQ for models. And where they put Kimi K3 is generally kind of on a trend line of actually what you'd expect. And where they would put is about there's multiple ways of looking at, but somewhere four to eight months behind the leading US models. And Austin, I think another misconception about this is a lot of people think open source inherently means um kind of cheaper, more efficient models. But when you look at these benchmarks, they actually don't consider sometimes what the efficiency of the model is. So Kimmy is a 2.8 trillion parameter uh monster. It's a very big model. And and when I talk about parameters, I'm talking about the amount of kind of tuning knobs in an AI model. Because there's no database of knowledge that they're relying on, like traditional computer systems. The database is the learned knowledge of the weights. So this is a very large model. And where that matters is if you're comparing it to Opus 4.8 for an anthropic, well, that's estimated at 1.5 trillion. That's about half. So, so Austin, when it's when it's twice as large, when it's got twice as many parameters, that requires twice the memory footprint, which means fewer copies deployed, smaller batch sizes, more of that interconnect overhead for all the networking, and ultimately higher costs per token. So, you know, if I had to do just a quick summary of that, well, there was a lot of headlines saying it had caught up on those initial benchmarks. The true story is it's probably still four to eight months behind in many ways. And the true story is as far as uh there's there's always trade-offs in engineering, and it it's kind of brute forced its way to a lot of its capabilities. That means it's a little bit less efficient. So next question, how they do it? Well, you you mentioned earlier distillation is certainly a part of this. Uh, there's there's been some research that shows that there is heavy distillation, especially of fable. Um, but Austin, you know, the other side too is that we need to keep watching. Uh in in China, they they have constraints that don't apply to the US labs, and they they have probably equally smart people working there. And and smart people learn to work around constraints. Um ultimately, you know, I think the big question, Austin, as to um who does this ultimately benefit? The market's initial reaction, right, would be to have a sell-off across AI stocks, because if anthropic and open AI are losing their lead, you know, the the next step in a lot of people's minds would be the the AI trade is over, right? Oracle is about to implode. Um, you know, you see all these deals being struck with anthropic and other companies, they're gonna go away. There, there certainly would be short-term pain if these, you know, couple companies with this kind of oligopoly at the moment of kind of the lead in models were to fall behind. But there's probably a long-term gain. And that is if you have more model choice. It's what we talked about last week, Austin. If you have more model choice, if you have more open source catching up, that's going to mean increasingly that value is going to accrue to the infrastructure layer as we have more choice and more competition. So I would say, you know, the market tends to get these things backwards. It got deep sea, deeply backwards. Um, and if Kimmy was as powerful as people initially thought, which, you know, I described why it's not, that would be a pause for AI. The last thing I want to mention on this, Austin, is that one one kind of interesting side note from this is that in the past couple of days, we've seen a lot of discussion from the Chinese ecosystem about how they're getting away from NVIDIA. Because as we've talked about on this podcast before, NVIDIA's greatest moat is CUDA, which is their computing platform for GPUs. These companies are increasingly saying they've essentially used AI to be able to build software to work their way out of NVIDIA. So it is, it is something we've talked about, this kind of idea of what holds value in an AI age, what is the value of software? Um, this is something we will continue to hold NVIDIA, and I like NVIDIA's position, but it is interesting how you give these companies these constraints that you block NVIDIA from exporting their chips to the country and how they work around that impediment and what they're doing is kind of um shoveling out the moat of NVIDIA, which is something I think is a little bit of an undercovered storyline.

SPEAKER_00 13:27

Lots of really great points in there. And there's a lot of questions that this moment raises for me that we sort of just have to wait and see play out. But one is it it seems pretty obvious that a lot of these performant Chinese models are just distilled off of US models, and a lot of data suggests that. So if China's advanced and people are looking at the gap between Chinese models and US models narrowing, and one of the implications there, of course, is that as the gap narrows, eventually China will exceed the US and have more powerful models. But if the way that these models are closing in that gap is just that they're getting better at distilling our current frontier models in the US, that that doesn't suggest um some sort of overtaking. It just means maybe, you know, fast follow, nipping on your heels. One of the other things that's also interesting right now is we, I would argue we have not really seen the impact of limiting our most powerful chips to China yet. Because while we have been limiting chip sales, we have not seen the first big wave of blackwell-produced models at scale from all of our uh hyperscalers. Even Fable and Mythos were largely trained on TPU uh and training. So now that we have this new Blackwell cycle coming, we need to see more US models developed on that. And we could see a performance jump domestically that China doesn't have access to short of distillation. So there's just a lot of questions here that this raises more questions than answers for me. I don't think it's inevitable that because the gap is closing, China will eventually overtake US models if the way they're closing that gap is distillation. But more questions than answers. Um, I personally still I it still seems that the US hyperscaler infrastructure AI trade is where I want to be. So that's where I'm gonna be putting some more money. But uh bottleneck after bottleneck, let's go forward to you know, the strategy that has made the portfolio perform as well as it has has been you largely identifying up until this point the early bottlenecks in the equipment build out. One of the next bottlenecks that might be the you know a limiting factor here is natural gas, right? There's optics and interconnects, there's chips, there's memory, and maybe natural gas next. So what's going on with this thesis here? And is there a role for uh or strategy for us to play here to take advantage of it?

SPEAKER_01 15:47

Yeah, I just wanted to quickly touch on this. It's it's definitely interesting that shows we've talked about on this podcast before. You start out with something in AI, and you it's really hard to know what directions it's it's gonna go in and and and what industries you're you're never gonna be able to come and expert in everything, but you're going to need um kind of a level of knowledge and some things you never expected. You know, we'll talk later about jet engines as part of the AI trade, which who have guessed that, you know, five years ago. But Matt Smith, I believe he's the LP of Chronometer partners, I could have that wrong. He was on the podcast Invest Like the Best, and he gave basically a thesis on the fact that the U.S. was going to run into a significant uh shortage of natural gas in the coming years. His thesis essentially uh believes that by 2028, we're going to be in a situation where demand is going to exceed supply. And by 2030, uh we are going to be in an area where we have exhausted all storage. And in his quote, a true crisis begins. The downstream impacts of this in his reading is that natural gas is essentially about to have a memory moment. You know, this being Austin, when we started the podcast, memory uh was still a single digit percent of the spend of an AI build-out from hyperscalers, and it's quickly moved on to, you know, estimates range from 30 to 50 percent, depending on where you look. He believes natural gas could soon be go from a similar rate, about 10% of costs, uh, to 20 to 40 percent the cost of hyperscalers. And that's largely just due to the increasing cost of natural gas. So, you know, Austin, the interesting thing here, a few things. Number one, it's it's not so much that this is going to be a cost increase driven by AI itself. AI is a contributing factor, but the largest factor is actually um what is on the docket for exporting natural gas and the fact that we're going to be exporting more of it. But second, you know, there's a company perspective and there's a what does this say about the future of AI itself? And natural gas is responsible for I think like 41% of utility scale energy in the country. And if the price of it moves up by 100 to 200%, that's definitively going to be felt by customers, and it's definitively going to become a political boondoggle. So I think number one, this is worth following from every person who's investing in AI stocks because this could have spillover across the entire AI space. The second aspect of this is it's it's going to have some real losers beyond consumers. Uh, Matt Smith, he highlighted hyperscalers. He said the turbine and engine companies, because a lot of people are going to uh they're gonna look for alternatives, uh, energy and construction firms, also fuel cells, something like a Bloom Energy. Some winners could probably be amongst the solar supply chain. He sees one of the clearest ways outside the situation as some innovation among battery technology, which, you know, is already an interesting area, but could get even more interesting with this. So, Austin, the bottom line is this is an extremely tight demand situation. These markets are extremely well balanced. But I do see his point that you know, you can have companies that are doing all the analysis around what hyperscalers should be buying in terms of their energy. And they're largely saying natural gas, and there is analysis from natural gas companies and flowing all this analysis through, there could be a blind spot. I I've read some excellent analysis before on this, it hasn't come to play. But what's interesting about Smith is um kind of giving some of these specific dates and data from tracking all the projects. So I think if if there's anyone out there and you're listening to this, uh, number one, you could listen to the podcast, which was uh the latest invest like the best episode. Or also there was a uh essentially a 20-page paper, which we can put in the show notes as well. I think this is definitely something to track. It's something I'm gonna be continue tracking. And uh, you know, maybe maybe it could impact some of the investments we have going forward, uh, you know, especially if it makes some areas like solar more attractive that we haven't made investments in. But solar has incredible hooks into um the growth of AI itself.

SPEAKER_00 20:15

I'm hearing some potential echoes there as well from uh a pattern that you have uh made core to the portfolio and some of your trends, which is the transition from EVs to AI and these companies that were set up for EV success uh having you know being sold off as EVs have had a weak um you know moment. And but then they have a lot of potential as their technology is now adopted for AI. I'm hearing the same same potential maybe with battery storage, right? And I'm thinking about Ford's new energy business that they've announced. I think GM is also doing one as well, looking to become large-scale um LFP battery manufacturers and re uh get some return on their EV spend that they've done the last um few decades here. I think it was actually one of our listeners also had pitch Ford's energy exposure as an opportunity. Now, that that's not an endorsement, and we're not saying we're gonna put it in the portfolio, but just hearing some echoes there from a pattern that has worked well. Um so definitely an interesting angle to watch there on the battery storage. Let's take it from the macro and get down to the specific. I want to talk about some specific earnings news in the AI space. AMD had a major deal with Anthropic. They've signed two gigs of capacity for a $5 billion investment. As an aside, it seems like we've been at the I don't know if you have the same impression, but the last few months of 2026 have had fewer of the of these big gigawatt deals announced. It felt like the end of 25, every week there was some new one gigawatt, two gigawatt, five gigawatt facility being announced. Haven't heard as many of those in the last few months. And I wonder if that's just because they're all committed from those deals that they signed in 25. But talk to me about this AMD anthropic capacity deal. This is the first one I've heard in a couple months, it feels like.

SPEAKER_01 21:57

Yeah, I think partially open AI is. They were doing for a while where they were just doing like a deal a week, and it was just like deal here, deal here. And uh they they almost broke the market. It was getting to be so much. The second thing is the deals often have gone larger. You know, we talked about uh Broadcom forming its unit that's going to do up to uh 20 gigawatts, and it's such a fantastic number. You just kind of breeze through it, but that kind of announcement might be worth 10 to 50 uh regular deals you might have heard about a year ago. So some of them have gotten larger in size. But also, I think there's two key pieces of news here, um, just around Anthropic and OpenAI as kind of the leading model of companies. Number one is this AMD deal with Anthropic. Uh, I think one thing that's really in focus is uh AMD, they're releasing Helios, which is gonna be their competitor to the large Vera Rubin systems. They announced not only this deal with Anthropic, but the deal with Microsoft as well. So we're just increasingly seeing what you would expect from companies having their own diversification strategy where they're gonna look for a custom strategy, they're gonna look for AMD, and they're gonna look for NVIDIA. You know, in prior generations and early AI, AMD wasn't able to get the traction that a lot of investors hope for, but it's now priced for a lot of traction this next round. Second, open AI, I talked about this kind of good news happening in the background. Um, they're raising their capex estimates from 600 billion through 2030 up to 750 billion. So, in terms of news from these leaning AI model companies, well, the market was initially fretting about this Kimmy threat. In terms of their own planning, in the recent week, there's been nothing but basically they're planning to go even bigger.

SPEAKER_00 23:44

Uh that is that is the one story that has been perpetually true since what, 2023? You know, whenever whenever we think we fit a ceiling, people are like, nope, we're going even bigger. Um, so you uh it is your favorite time of year, right? Which comes four times a year, and that's earnings season. But with earning season comes opportunity, and it also comes volatility. So talk to me about what you saw this earnings season. We're gonna talk, I'm sure, about Alphabet's 200 billion capex move that puts them up above 200 billion increasing their capex. I think they're one of only two hyperscalers that are at that level now, uh, with Amazon being the second. I think Microsoft's about 185. So I want to get your thoughts there. I'd also like to hear about GE Vernova um and STM. STM had a pretty rough day down double digits. So talk to me what you saw in some of the early days of earning season here.

SPEAKER_01 24:32

Yeah, Austin, I've avoided saying this is gonna be a shorter episode because I don't want to jinx us, but I'm doing it now that we are gonna go fairly rapid fire through these because we do want to keep this episode a little bit shorter. So let's just go through these one at a time. ST Microelectronics, it rose significantly after we had recommended in early March. Um, like a lot of things in this power trend, it's taking a breather and it's down about 17% today. Austin, the bottom line is that this company has a huge data center opportunity, but still a small piece of its business, only about above $1 billion this year. They're gonna be more reliant on a rebound in autos and industrials in the near term, and and that's where they struggled with this quarter. So investors are saying maybe maybe the prices got a little bit ahead of us, but I'm I'm actually fine with this as I am with mini drops because if if the valuation continued trending down on this company, we wanted to add it as a safer play on a large trend around power management. So I would I would welcome any drops and and consider adding more to the portfolio. Uh, you had mentioned G or Vernova, they dropped 9%, uh, largely because of an EPS missed last quarter. Um, but they raised their full year guidance. Uh, and their backlog is now awesome. Their backlog is four years revenue. Uh, we we haven't added G Vernova to the portfolio, um, but it's definitely an interesting category. Another company in the space is FT Aviation, ticker FT AI. Uh, they've been converting jet engines to power turbines, and this week they won a big contract with their JV. 1.47 billion, that's about half their entire revenue from last year. Uh, this is a company that's trading at about 12 times expected profits in 2029, about half the level of GV. They're more speculative because they need to grow this industry. But Austin, one thing I think people ignore about these stocks is specifically they're going to have a massive services and maintenance tailwind from all of this build-out that even if they were to hit peak deliveries in the future, um, it's gonna be buffered by their ability to continue kind of servicing a lot of this equipment out in the field. Uh, I'll keep going. Alphabet, you mentioned earlier shares down 6%, but you got to look and say, is this alphabet or is this just kind of a broad sell-off against kind of comparable companies? And Microsoft's down big today, Amazon's down big. The market clearly doesn't want them to say they're gonna keep spending on data centers. But here's the contrasting factor 82% cloud growth with massive acceleration continue. Their operating leverage continues improving in their cloud division, revenue is up 82%, operating profits up more than 200% year over year. Their backlog's now at $514 billion, more than half of which they said is gonna be recognized in the next two years. And, you know, their CFO on the call, Austin. They were just emphatic that they said of anything we've gotten more bullish recently. And they said the dynamics look healthier than they are a year ago, and they see continuing positive ROI, and they said they're going to spend much bigger next year. So, you know, a sell-off in Google, but I think it's less about anything Google announced, the very strong quarter. And it's more about kind of the market wanting some of these hyperscalers to take their foot off the pedal. Well, the hyperscalers are saying we're not gonna take our foot off the pedal because we are seeing such fantastic ROI. Um, last company I want to highlight that report last night, ServiceNow. I brought them up last week as one software stock we are considering adding. Austin, the reason we haven't had these software stocks is because a question of when could the sentiment possibly hit a bottom? And we see with ServiceNow excellent earnings. They had a 7% drop yesterday as everyone was getting out of the stock ahead of earnings. It was up 7% after hours. Now it's down today, which kind of just shows where's the sentiment? It could not be more in the gutter. Like what's below the gutter, Austin? It's like in the core of the earth right now. That's where sentiment is for these SaaS companies. But you know, as far as what they announced, they raised full-year guidance. They said their agentic deployments are up 9x in the last nine months. They said their uh AI products grew 5.5x year over year. Austin, the bottom line is right now, there's this battle of how quickly the AI business can get material for a company like this. Uh, you know, their goal is relatively modest in the near term, but by 2030, they want basically 30% of their contract value to be from AI. Um, so you know, a good quarter. Uh, they they noted as well. They they had a controversial cybersecurity acquisition recently, but they have a $1 billion cybersecurity business and growing, which the markets generally like cybersecurity while the rest of software is sold off. I I continue as I research service now and think about would this be a software company we want to add despite sentiment, to like what's happening. And um, as I will often say in situations like this, I will welcome lower prices. Um, and and I think it is one that we would strongly consider in the future. And I am not shocked by the market's reaction at all to what seemed to be excellent earnings. So, Austin, that's that's my earnings recap. And I think that is kind of all I had to discuss this week as well. In about, is that 30 minutes?

SPEAKER_00 30:04

Is that I'm gonna I'll give it to you this time. This is the one time you said we're gonna have a short episode and we did it. I credit to you. I'm proud of you. I'm happy to be here on our first ever journey. This is like uh, this is like our Kimmy model, right? This is like our lightweight model, right? This is our this is our 30-minute episode. So well done.

SPEAKER_01 30:19

Um I go land on aircraft carrier with a big mission accomplished banner unfurling behind me.

SPEAKER_00 30:24

I think we did it. I think we did it. Perfect, perfect. Uh, we did it. Eric, thank you for all of those updates. I look forward to hearing your thoughts on ServiceNow and other positions in the in the SaaS Pocalypse trade. You know I can't resist a deal, so I I've been doing some shopping in the SaaS space. Um have not bought ServiceNow, but it's been on the list. Um, I did pick up some block, I've mentioned some other names that I bought in the last few weeks that um I would love to see in the portfolio as well with your seal of approval. Although I will like you, I will always welcome lower prices. So until then, we'll see you next time. The AI Investor Podcast is for educational purposes only and should not be considered investment advice.