Transcription
What's up everybody? It's LG Ducet here and welcome to Milk Road AI, the daily AI show that bought the dip, and the dip before, and even the dips before that, and now we are finally getting the chips to go with them.
Today is July 22nd, 2026. We're recording a day early on July 21st. Believe it or not, there's another bottleneck forming. It's one that we've known about for a long time. Energy. And it's one that's becoming more and more unique to the US models, whereas China has excelled at matching their energy needs for their data centers. So, can the US catch up? And if they don't, what happens to the market?
This is our weekly rollup episode where we sit down with all three of our AI analysts to discuss the market, what they're watching, and what matters most to their Milkroad Pro portfolios, which you can check out for just a dollar at the link below. A reminder that our podcast today is free and it wouldn't be possible without our partners at Secure Ties, the Regulated Rails for tokenization, and BitGet stocks 2.0 with real liquidity and real dividends. Keep paneer out later in the show for a message about them.
All right, boys. Market is looking really good today. Hey, lots to talk about. Melvin, how are you feeling, man? Give me your 30-second spiel. You feel good. Your your low DCAs are up. How you doing?
>> You can exhale.
I feel [laughter] I feel amazing. Um, Nebius is up 15%, Micron's up 13%. Um, we had a huge selloff last week, which we called in the show, which is basically a FUD. So all our analysts basically bought into these companies like Nebius and uh Micron and they're basically up like 20, 30% up from previous lows. Um, so yeah, I continue I continue for this to run up to happen into the earnings. Micron has earnings in the next month or two. Nebius has earnings next month and I think we're just going to continue to go higher. So I remain extremely bullish on the buildout and extremely bullish on compute and I I think everything's just going up.
I do want to talk about earnings later in the show, but Melvin, I just want to know right away, what is what to you is the most important earnings call of this season?
>> Um, I really want Oracle uh to go up this earning season. So, I [laughter] they've been they've been really down. I think they're they're around 125. Um, so I would really like the market to recognize that Oracle is not is one of the key beneficiaries in in the AI buildout and that'll further push up bunch of the stocks that Oracle is connected with as well. So, it would be really nice for the market to validate that. So, I'm really looking forward to Oracle earnings.
Okay, we'll come back around to it. Kyle, how are you feeling, man?
>> Yeah, I feel great. I mean, I think we all said we think this pullback was one healthy, but isn't going to sustain. Um, we also talked about I don't think that it's going to be a V-shaped recovery. At least a lot of people have said that. We'll see what happens. This was obviously a huge bounce. You know, up some things are up 12, 13, 14%. So, that's been nice to see. Um, some of us got bids at the very low, some of us got bids kind of halfway down. So, um, I think everyone did pretty well, at least from the Milk Road Pro portfolio side of things. U, we'll see if this continues. Um, if you look at the downtrend that Micron had, I flagged this a couple days ago and on the live call yesterday, we did smash through it today. So, that looks pretty good. Um, and, uh, I think Google's earnings tomorrow will tell us, you know, kind of what's the hyperscale. I mean, they'll speak for Google only, but I think that's going to be um they're going to tell us what the hyperscalers doing. Are they going to guide higher for capex, which they've done for however many quarters straight or not? And I'm pretty sure they will. And so, I think that'll send us flying once again.
Nice. Vincent, how about you, man? Same sentiment, I assume.
>> Yeah, definitely. I mean, the Micron dip I bought is already up more or less 20%. That's pretty That's pretty good. Of course, I think there's some like this was a really busy week because there's happening so much. We had first earnings, we had earnings of the banks, JP Morgan, Goldman Sachs, great commentary around AI. We had TCMC, ASML as well. Both of them raising capacity on crazy good earnings as well. We had the uh memory data from today uh from from South Korea. It's chart looks like this up 527% year-over-year and another 22% month over month. So that was really bullish of course for for today's runup in the memory stocks. And we had the models there were actually three different models coming out over the last round about 10 days uh or actually seven days. Uh Kim K3, Inkling, and Bonsai, all different models, open-source models.
>> And Gemini today.
>> And Gemini today. Yeah. It's it's it's it's crazy. So to me, models are really commoditizing here. Um, so a lot's to cover today.
Just going to pause there for a second to point out that the market is showing signs of something kind of different happening. And our analysts at Milkro Pro are all over it. They spent the last couple weeks making a lot of trades, getting out of some positions, and then getting into a lot of new ones, getting ready for the next wave of robotics space, or even kind of picking some different AI winners. If you want to see what they have in their portfolios, what position is their opening, it's just a dollar in Milkroad Pro at the link below.
What Vincent, can you tell me this Kimmy K3 thing? It's like almost Tuesday afternoon. I've been swamped. What What is it? What is it? I just saw the name a couple times. You guys are going to nod your heads like you already know, but what? Tell me what it is.
>> Yeah, so it's an uh open-weight model from China. It's a 2.8 trillion per parameter model. So, it's a it's it's a pretty big model, pretty good one. The I think the most important part here is that a when they launched it within 48 hours the GPUs they had were pushed to the edge. So they actually had to restrict access or actually pause new uh consumer subscriptions because the model was that good and people just ran at it. Then you had Trump coming out uh based on that saying we're thinking about restricting access to Chinese open-source labs which in my from my perspective would be a really bad idea for the US. Uh good for the labs short-term but long-term really bearish for kind of AI adoption. Um, so yeah, it's an open-weight model that is crazy good, that is really cheap and uh, yeah, that's basically the news.
>> And it's their new model, right? The K3. I mean, it implies it has two before. I didn't even know about this Kimmy ones. I was just I just know like Deepseek and the other kind of more high mind share ones, but this is one of them now.
>> Exactly. Yeah, it's
>> And what does open weight mean? What does open weight mean? Just just explain that.
>> Yeah. Uh so so you can basically as a company or or also a private human being you can download the model and then you can set the parameters you can fine-tune the model. You're not giving away the the data to kind of a centralized uh company or institution. Uh that's kind of the the different and it's much cheaper than than than the others.
Now Melvin, you said this is bullish for for the market. You tweeted it this morning. Why?
>> Yes. So there's two sides of the trade I want to look at here. Um first this is really good for you know Nebius and the AI data infrastructure buildout. Um the whole just to explain what the open weights is one more time. So like closed models, right? You have GPT5, Claude, um those they only run on on the makers, the own servers, you can't just download Claude and go into your hardware. But open-weight models like Kim K3, DeepSeek, Llama, you name it. You can actually go and download these and run on them on any GPU cloud you want. So every time a a huge like a competitive model opens, it creates like a fresh new demand for GPU capacity. Uh specifically for clouds that specialize in serving those models cheaply and fast. And like Winston said, Kimmy K3 tweeted that out that they're turning away customers because they're they don't have the because the demand is so high. Now why does Nebius benefit from all this? Because their model, their business model is renting out GPUs. Um, and the inference plumbing so other models like Kim K3 can run fast and cheap on top of their infrastructure. So, a bigger, better open model isn't the threat to Nebius at all. It's it's actually a fuel for them. This is why Jensen wants an open world model as well because it's going to create a fresh new wave of demand and it's going to have him sell more GPUs. So he benefits from that.
>> It diversifies the customer base, right? Instead of only being open AI and Anthropic, you now have however many people that are going to use this. So it's it diversifies.
>> And and this also ties into something Nebius already has, which is called Token Factory. Um if you're not basically familiar with that, it's basically a a menu of ready-to-use um AI models that businesses can plug straight into their apps uh through one simple API. So instead of going to going and building their AI infrastructure from scratch, think of it like a vending machine for AI. You pick a model like Llama, Deep Seek. Now Kimmy um Nebius already has Kimmy there too, their previous models. So um they can they the Kimmy K3 will likely be on there as well. So this is this is a huge huge win for Neoclouds because they will just run this.
>> Yeah, let me ask you guys a question. So this I think we realize is bullish for for semis. I think we you know semis have pulled back like crazy. We're like guys buy semis. It makes no sense. Whatever you want. Buy Micron, buy you know DRAM ETF, buy Nvidia. Like it's all going to go up because there's still so much demand for semis here. Um, but there was a uh and neoclouds and everything. There was a chart that went out this morning. I don't if you guys saw this. I'm just going to try and share screen real quick. um from semi analysis and it was looking at data center capacity that is being uh data center power that is uh contracted and uh uncontracted and so you see they have contracted a bunch in 2027, 2028 and then it kind of falls off. They predict 140 gigawatts of power that still needs that is uncontracted. So they still need that power uh versus 36 that that is contracted. So you can and Vincent, you've talked about this a bunch and I think this is probably a very bullish chart for for Bloom Energy. Um, but so we need these chips, we need GPUs, and all these models are going to continue to push that demand. Um, but is the is power going to be and already is I guess becoming the main bottleneck? Is that going to be an issue for these data centers, do you think? So they're not going to connect to the grid obviously. They're going to have to have on-site power, which is why Bloom Energy this is kind of bullish for or things like Galaxy where they have data centers already connected to to power is is what they're going to have to use. But are we even going to be able to get this power? Because I think what we all understand is we need way more data centers, way more GPUs. Where the heck are we going to get the power in the US? China's fine. They have so much power it's not even funny. Uh the US though struggles and is this an issue of just they haven't signed the contracts yet or is it they can't actually figure out where to get it? What do you guys think?
>> It's really bullish Bloom Energy actually. That's the only thing [laughter] really.
No, I saw this chart and and Bloom Energy actually came out with a report two weeks ago or so where they were showing the terawatt demand and the outlook also from semi analyst and some banks. I think the biggest issue so a I think it's pretty clear that the data centers moving forward will not be able to connect with the grid just because of societal pushback. uh and this is also part of the Bloom Energy report where they show one of the largest constraints actually is pushback from society because of high energy costs etc. So this leads to the the solutions like off the grid right behind the meter solutions. [snorts] I mean we saw Bloom Energy also news from this week that a data center project with Oracle was delayed uh because the gas pipeline like Bloom Energy needs natural gas to run their fuel cells um was not permitted by by by the state where it was built right? So do I think it's one of the biggest bottlenecks in and and hindering the data centers? Yes, absolutely. I think this will slow down the the data build data center buildout a lot. That's by my and and and that's you could you could argue that this is actually bearish for GPUs, memory, etc. because there is no point in having all your your chips on inventory.
>> They can't be powered.
>> Exactly. Exactly. So that's why Bloom is one of my big or is actually the biggest position in my portfolio. Um however there are many if you listen to semi analysis and to Dylan Patel there are many options that you have to get power on site it's not only fuel cells you can use any sorts of engine diesel generators everything you just need the input which is gas uh and I think this is the core core bottleneck that we're having now.
>> Yeah, Dylan actually said this in a podcast recently that um in about 2 years uh solar plus battery will be cheaper than gas. Um so I like company if you look at companies like EOS right I think Vincent had a um in a position in it. Um I expect these like these companies like this to um build these out and it'll it'll be massive massive catalyst for all the buildout as well as also I think algorithmic changes that will happen over time with the way we use energy and how we can make it more efficient. Uh but I do see this as a like a sort of a bearish news for the buildout. I think this this would be a this would be a really bad for all the all the stocks that we were in because if you can't power your GPUs, how are you supposed to generate um you know money for Anthropic? There was actually a report that came out that got deleted by um one of the Cutrini re um analysts yesterday. They were saying that Anthropic is on pace or currently by the end of this month they're on pace to generate about $77 billion in um in ARR. So they were at about 50 40 to 50 like like a month ago now we're at 70. So if if somebody like Anthropic can't power their you know all their GPUs or they can't train they won't be able to grow and that will just have a cascading effect on the entire market. So
>> I think everything behind the meter will be interesting. Everything. Fuel cells, batteries, solar is yes, I'm bullish as well, but it's connected to the grid, right? You cannot build the solar farm next to your data center. It's not possible because it's way too big and you need permits, etc. Uh but yeah, it will be definitely a really interesting trade from here.
>> What's the solution? What's the actual easiest solution? Like just stronger policy like just ignoring public?
>> No policy. We need policy. Yeah, just turn into a collective because even Bloom has Bloom Energy has been having these problems too of like permits and and policies around these data centers. So they can't even um provide the power even if they want to um gas for them.
>> Because they cannot get the like it's not of the fuel cells that they want to build on site. It's because they cannot get the the pipelines which has the gas towards the the data centers. That's the issue.
>> And every state's a bit different in how they allow it. So like we we really need to deregulate this and just let them get their own power and let them figure it out by themselves. It's hard to do though. It's why like Elon had I forget what factory it is that he has. He put it on like the border of two states because like the one state they had good regulation around uh around the power. So he put the power on the other side of the state like on the other state and then the factories on this side. [laughter] This is insane that he was able to do that. But like these are the kind of things that they have to like jump through.
>> The one thing is is like these businesses figure it out. You know what I mean? Like they find ways to get it done. They bribe whoever they need to bribe to either change the laws or to make it happen. Like I I feel like these are very powerful organizations and they they find a way to make it happen. This is why like Galaxy, who we talk about a bunch. I don't know if anyone saw but they sponsored the Texas Tech uh stadium last week and everyone's like why are they doing this? And it's because they're trying to build a bunch of data centers. They need to pony up to the you know to the um congressman and and the uh the mayor etc. like the people that that make decisions in Texas. Uh and so they do that by putting their name on their stadium of their favorite football team. So these businesses find a way to get around it hopefully. But at some point this this could be problematic.
>> But one thing is pretty clear. Look at China, what they did over the last couple of years. They are by far the leader when it comes to solar buildout and adoption.
>> That's crazy.
>> Um and they started a couple of years ago. Yes. it was all kind of state funded and without the state kind of money there would be no kind of viable business uh in in solar there but now they anticipated it started to build a couple of years ago and now obviously with with the whole AI craze they are benefiting massively and that's their kind of core strategic advantage in this geopolitical race between China and the US next to the rare earth that they have.
>> Yeah, China basically took the you know the different approach or the different bet than US. US were all focused on the frontier labs, the models and everything. But China realized that the bottleneck is power. So they, you know, built it up. Now what are they doing now? They're building their in-house chips now. Like you have Huawei and Alibaba, DeepLike all of them are building their custom chips. Now their bottleneck is not going to be power. Now once once those chips start to you know catch up to Nvidia the newest generation they can just go ham and we can't because we're limited by our society%.
>> So.
>> It's all about getting energy and converting it into intelligence and China has so much more energy than the US does and so that's where the US needs to figure their out and if the Democrats win then they're even more [laughter]
>> But I mean we're talking about this now for five minutes. What's really clear and I wrote about this on X two or three days ago when when Bloom Energy was down 40%. I mean buy the dip on this name if you have no position in it yet, right? It's the same like memory for Micron, right? It's also down 35%. The demand for those things will just increase and and and the chart Kyle you you showed earlier just it's just another data point that proves this. So, um
>> And their pricing power is going to go up if there's if it's this scarce to get power. Do you know what I mean? Like that's going to be huge for for Bloom. Um, all right.
>> Hold on, Vincent. Wait, just one one last thing. Vincent, can you just remind people in like 30 seconds what exactly Bloom Energy is and what they do?
>> Yeah, so it's a it's a behind the meter behind the meter energy company that put fuel cells, so fuel cell boxes on site, so next to the data center. And once you put gas through those uh bloom boxes, you get energy and that's an energy source for AI data centers without the need to connect with the grid which is the biggest issue.
Got it. Great. Thank you. Go ahead, Colin.
>> Vincent, at the beginning of the show, you mentioned the banks had their earnings, which by the way, their earnings were insane. Uh which is a great thing for the economy. Uh so that's one good news, but you mentioned they talked a bunch about AI. I I didn't hear this. So what was that? I'd love to hear what what you learned.
>> Yeah, so two quotes here. One from Goldman Sachs, they said or the CEO said AI is an accelerating technology super cycle and the infrastructure financing cycle remains in its early innings and Morgan Stanley came out saying uh the capex buildout is only 10 to 15% through uh basically from from the money we spent today up until the 1.5 trillion in 2028 that they're seeing today. So for them it's just a a massive business. That's why their earnings were so good because they're financing all those those projects, right? For for instance, Bloom Energy, uh Morgan Stanley was involved and and and some other banks in providing the money for the comp for the customer of Bloom Energy to buy those boxes, right? And and they're saying saying this is only 10 to 15% there. That's really bullish for the entire AI AI buildup. But then also from an AI adoption side of things, there were so many data points in their earnings report where they're saying, "Hey, 90% of our AI of our employees are using AI. We have massive efficiency gains." Um, that's just another proof point on what I was saying in the Robin Hood um podcast. Finance is one of the primed industries when it comes to AI adoption because they have so many standard processes, they have so much data um that the AI can use and I was just really bullish and and and that's also a core part on why the why the broader market is is is is now green and not that much.
>> Yeah, so Jamie Diamond actually said some interesting thing about the AI capex as well. Like uh he was saying 400 billion last year to 700 billion this year. He sees that crossing in one trillion in 2027 against a total US corporate capex of roughly $4 trillion. Um so he basically he was saying AI could soon be eating a quarter of all US corporate capital spending. But Jamie also confirmed that or um you know confirmed that AI has slashed headcount by 30 to 40% um targets inside JP Morgan. So they're they're seeing massive gains and they're actually seeing their margins go up which is so which is what you want to see from uh from all this token spend right and one of the interesting thing that you know they said was JP Morgan's generative token spend is quote tribal right now but a significant acceleration the second um half of this year and token cost was named a budget line in the brand new cost category of Fortune 500 uh balance sheet And um you know JP Morgan's already you know routing workloads to cheaper models like Kimmy K3 or any other models that come around to manage that.
>> JP Morgan is not routing to Chinese models. Come on. You think so?
>> Cheaper models. It may not be Yeah. It may not be Chinese models but there are other you know open-source models.
>> They're open. Wait. So actually this is a really good point because a Coinbase came out today saying they slashed token cost by 50%. But the the tokens that they produce are all-time high.
>> Yeah.
>> Yeah. That's them routing between the cheaper and more expensive models, right? Um, but then the second point I want to make here is the So we talked about Kimmy K in the beginning, but that's actually only one model that came out, right? There was uh the a model called Inkling from the former OpenAI CTO um which is 80 to 90% cheaper than Frontier Labs and it's an uh it's an open-weight model as well downloadable finetunable what we were talking about earlier and there was Bonsai 20 27B uh it's also a US-based model for on-prem so it's a smaller model that does not run in the cloud but actually on phones on laptops uh etc. etc. And this is just speaking towards we're moving into a world where a models are commodity and b the token costs are getting much cheaper and obviously this is really bullish for AI adoption.
Can I make a case for memory here? Um why?
>> Go for it.
>> Kim K3 is absolutely bullish for memory. So Kim K3 or any other open-source model I'll specifically talk about Kim K3. It has a 1 million uh token window context. Um and it's it's runs on something called reasoning mode, meaning it constantly has to uh thinking think through steps uh before it answers before you know it spits out a quick answer. Um so the way I think about it is picture it like this, right? A closed model like GPT or Claude is basically one giant uh restaurant kitchen serving the entire world out of a single building. Millions of people. [laughter] Yeah, millions of people are ordering it once, but they're all basically cooking in the same kitchen, sharing the same ovens and same stuff, but like open-source. Kim K3 is the complete opposite of that since anybody can download it. You don't just get shared one kitchen anymore. You get thousands of different companies, separate company building their own kitchen from scratch just to serve their own customers. Same, you know, it's not the same recipe. It's the same recipe, but different equipments are needed. Um, so this is why I remain bullish on memory because you all need different memories for these open-source models. It's not just it's it's not one it's not one shared building like you know what OpenAI and Claude is doing. So
>> So two questions. One question, how did you come up with this story and this analogy?
>> Uh to a Claude. [laughter]
>> I use Claude.
>> I love it. No, it's it's it's it's so true and and I and I completely agree with you, Melvin. I think this whole war between the models is bullish for the entire AI infra buildout because what it is it it is more people building models. So, we need more GPUs. So, we need more memory, optics, fiber and all that. Uh and also when we're talking about on-prem models, so smaller models that you can run on your phone and laptops, we can start thinking about Apple here because they are the kind of distribution. They they have the phones, they have the laptops, right, that those models will run on. And obviously that's really bullish for them like for an upgrade cycle where people actually start to buy new iPhones. Real world assets like funds, treasuries, and private credit are still running on rails built decades ago. Gated, paperwork heavy, slow to settle. Everyone's talking about tokenizing them, but far fewer can actually do it and do it without cutting regulatory corners. Securitize can. It's the SEC regulated infrastructure bringing real world assets on chain. 9 years in native tokenization not wrapped, backed by BlackRock, Morgan Stanley, and Kathy Wood's Arc Invest and chosen by the New York Stock Exchange, Van, BNY, and Apollo to do it at scale. It's the regulated bridge between traditional finance and crypto. Tokenize the world at milkroad.com/securitize.
Everyone's tokenizing stocks these days, but almost nobody's doing it right. Thin liquidity, prices that drift from the real thing, dividends that just vanish. Bit stocks 2.0 is different. Real NASDAQ and New York Stock Exchange depth through licensed brokers, prices mapped one to one, dividends paid to your account in real time. Plus, you get the lowest fees in the market at just 0.04%. And you can trade them like any other crypto as margin in earn in grid trading. Tokenized stocks finally done right. Head to milkroad.com/bitgget to get started.
Yeah, both Apple and Google I think are positioned really well for that. But um I think for this earning season coming up or that we're currently in, I guess a big thing you want to keep an eye on if you're listening to any of these is the ROI of using AI, right? So like the first kind of instance of this was like Meta came out I think two quarters or three quarters ago and was like yeah for for every you know piece of intelligence we put into our ad model we earn more right? If we can spend more on GPUs, we can earn more dollars uh from from our ads. Um and so we've seen that on the ad side. You just mentioned Vincent that the the financials so the banks are starting to see kind of ROI from from using AI. We want to see more examples of that because there's there's two things that can really kill this um AI And for buildout, it's one, energy. If we can't get enough energy to power these GPUs, then the GPUs are all useless. It's two, it's there's not ROI, right? And this is the question people keep asking, well, is there actual ROI there? And now we're starting to see there's more use cases available, especially as like the agentic economy comes out and the models are getting cheaper, right? Gemini just came up with their model this morning, and it's much cheaper and faster um than, you know, Claude or or ChatGPT Grok two weeks ago. Much cheaper. Obviously, some of these open-source models much cheaper. And what you mentioned was some of these companies are starting to to sort of like be an aggregator and use different models for different reasons. And so, if companies can succeed at doing that um and get ROI on their business, they're only going to use more and more of these tokens. And that's what we need to see. If they can't figure that out and they're not using more and more tokens, that's like the other bare case. I don't see that happening. I mean, us in our company, I I max out freaking every single day. Uh, and I know most of our people are that are using it. Maybe not Melvin because he only uses Claude to get his little analogies about buildings and memory, but [laughter] but the other power users. And so I think it's this earning season is how many companies are going to come out and say because I'm using X model, I'm getting this type of ROI. And I think if we can see that across many different products and services, that's going to be the most bullish thing that we can have for the market this earning season. Yeah.
>> Plus capex uh expansion from the hyperscalers. Yeah, I I fully agree because LG, you were asking in the beginning Melvin, right? What what are you looking for? Me personally, I'm really looking for ROI. So, AI adoption. What Kyle was talking about like names like Eli Lily, right? Hopefully, they are coming out saying, "Hey, we did this and that with with a we seeing some first results in ter I'm not sure if this going to happen, but like this is so important and it's also tying.
>> It needs to happen.
>> Yeah, it needs to at some point in time, right? Their names like the banks they were talking about the the insurance companies are starting to talk about it as well. Robin Hood um also a name in that space why they're launching agentic credit cards or agentic trading and just to tie it back to the models what I was saying uh earlier because Donald Trump is thinking about restricting the access of of US companies to Chinese labs I think that would be really bearish for AI adoption in the US because what it does. It Yes, it protects demand for the US labs, but it hinders competition.
>> More expensive.
>> Yeah, it's way more expensive and and and and you will not be competitive if you have no chance to access the cheapest models. So, it would be a really stupid move, but hey, there's some smart people in the White House that can protect the US. Completely agree with that. Um I think competition is just we need competition. Like there's no way if if there's no Kimmy K3 these models these Frontier Labs is not going to push out better and bigger models or better models, you know, you need competition uh to further this buildout um and also um the whole drama around Fable 5 now that's really going to come into question like in the next like weeks or so because we have China beating us because they beat us in coding with Kimmy K3 they're better than or top three um at at one of those things. So, is that going to be like how are we gonna deal with that? Like are we gonna just push out models without any checks or how are we dealing with China? What do you guys think on that? I hope they're not dealing with China like they did with the chips like because what they were doing is restricting uh Chinese companies basically access to the most like up-to-date Nvidia chips, right? What happened? They designed their own which are working really well. And you can take that kind of template and look back the last 30 years. China was always innovating everything that the US were not giving them access to. So I hope that they're learning from this and not doing that. But you I mean no clue on how they how they will be handling it. It it's just bearish for adoption. There is a bit of a bare case for Anthropic and OpenAI out of all this, especially if they let everyone use open-source models um because obviously they're getting undercut like crazy and you know you just said Coinbase, other companies I'm sure are doing it as well which are kind of routing to other models at some point this is going to have to be problematic for the valuations of Anthropic and OpenAI at 1 trillion or 2 trillion whatever they're at, I don't know what they're at today but like I don't think and Vincent you've you've talked about this a lot but the the frontier models are becoming a commodity and now we're seeing frontier models that are not American-made models, but are Chinese models and they're open-source and they're still at the frontier, which was not the story a quarter ago, even just a few weeks ago, right? Um, and so how does does that have a knock-on effect to the market because obviously there's a bunch of companies that are investing in these and they're raising a ton of capital and they're a big part of this um buildout. Now, I know it diversifies, right? Because now you have all these other models that need to buy the GPUs as well. Um, but you know if this starts to hurt their revenues because right now Anthropic and and OpenAI are actually just incentivizing and subsidizing a lot of their token usage. They've been keep extending Fable week over week of the last few weeks. I don't see them stopping that anytime soon. So is that going to be is that a an issue to the market do you think?
>> So let me make a quick point real quick. Um, so interesting fact, Kimmy is Kimmy K3, the company is worth 48% less than Anthropic, yet it's it's insane. It's only valued at $20 billion. And if you look at
>> Yeah, but but that's cuz you can no one like at least Anthropic can satisfy, you know, I don't know how many users they have, but let's say 100 million users. Do you know what users they have? Like Kimmy can only satisfy, you know, a few whatever it is, a million users or whatever. And all of a sudden, they don't have the capabilities, the GPUs, right? Right? Whereas Anthropic does and OpenAI can power a billion users, right? So like just because their model's good, it doesn't really matter if you can't power that model to a certain amount of users and you don't have a good UX because that's the other issue with Kimmy is like Anthropic. You go in and you can build agents and it does it all for you and it's super easy. You have projects and skills and connectors and all these things. There's no UX for Kim K3. It's terrible, right? So like most people can't actually use it and then they don't even have the capacity. So like I think they should be valued way way way way less than an Anthropic or a Chatb because now like the mode is becoming I don't want to call it a mode but like UX is part of it, you know.
>> Like a 48x though, you think that?
>> Yeah, I don't know about that much. I don't know about that much but like because they can service their users whereas Kim K2 can't like that's why there should be a big step difference in the valuation so I agree Kyle while we're in a world where you and I are using AI the most right once we're moving into a world of agentic AI and companies adopting AI agents. I think this will be no issue. I think the UX is no issue over there, then it's really about costs and what is the most efficient model and there I don't think that's a barrier anymore.
>> Yeah, but so if UX is not the thing, which like I don't know, I think it is for many and even like switching costs are are real difficult if you're a power user, right? Um, so like for me even like I don't want to switch off what I have because I have all these agents set up and these workflows and like it'd be so hard to move that. Um, but then there's just the for a while I've always called these like frontier model businesses like capital allocation businesses. They need to know when to buy compute uh and when to raise capital etc. And you know OpenAI has been the best at that for a while. Anthropic has now started to be really good at that. That's where I think Kim K2 I don't I don't know but I'm assuming they're far behind. So, like, can they even get the GPUs now to go and service all the people that want to use it? How long is it going to take them to to to catch up on that? I don't know. But I imagine it's going to take a while.
>> Yeah, exactly. But but that's exactly the point on why I'm I'm bullish, right? Because
>> We need more AI infrastructure and the more open-source open-weight uh kind of models we're we're having out there, the better for me as an investor because we need more AI infrastructure, right? Um so I'm more on the bullish side of things also when it comes to [snorts] Chevan's paradox right? So the argument there is for instance for on-prem model they found a way to to to use much less memory right? So if you need less memory then probably you you you're adopting more right and in the end you have more demand for memory with those on-prem models so I keep leaning on the on on on the bullish a lot of things here.
So, where do you guys think we go from here? We had pullback 20, 30, 40, 50% uh and I think it was Nibius was down 50% was it or close to anyway? Uh and
>> 40%. Yeah. By the way, it's up 17 and 12% today. So, it's thing is just having an absolute [laughter]
>> Having an absolutely day. Where do we go from here? Do we chop for a bit because it's summer? Do we rip back up to all-time highs and they get repriced, you know, another leg higher? Or what do you guys think happens? Over which time frame?
>> Walk me through your thoughts for the rest of the year.
>> Next week. [laughter]
>> Q3, Q4, where do you think it goes?
>> Until end of year? Definitely higher.
>> Yeah.
>> I think I'm I'm I'm still really bullish on this whole AI infra builder as we just outlined. I'm also really bullish on consuming agents, right? I was talking about it in the Robin Hood pot. We're moving into this world where agents start to consume, start to spend money, not only uh uh do coding stuff, right? Code coding work. So I'm bullish on on on the adoption layer as well as digital assets. Melvin [laughter] um so much higher. Um do we reach all-time highs in names like Micron, etc. Don't know, but I think we will at least be close to to where they were. And and that's why I think the the buying the dip here is still even though Nebius is up 17% right it's down a lot from all-time highs and same for Micron same for blue manager is is really attractive.
>> Yeah, I pretty much have the same view. I think we're going higher. I think this earning season will be a key uh key push towards that. I think the big question like we talked about is going to be like the ROI. Is the ROI going to start to come in? Um and also the I believe every every one of these hyperscalers or almost all of these hyperscalers likely going to guide for you know higher capex. Google's probably going to do that this week uh because they they can't even use their internal employees can't even use their models and Gemini was partly delayed. I mean it got released today but like it was delayed because they don't have capacity right to to do that and then they signed a deal with SpaceX. Um so the question will be um the ROI and I think ROI will start to come in maybe start to come in this quarter and into the next quarter and I think we're just going higher from here. So remain extremely bullish. a macro side of things that could be a bit choppy right if if especially Iran. I mean oil is now up what 10 or 15% from from the recent lows again because there were some news obviously that can have some impact on the inflation side of things that could be something uh that may hinder the slowdown from a macro side of things but I don't view that as a really big issue because midterms are coming up and Trump has to make sure he's he's winning and he's winning that by low inflation and and high stock market. So I'm
>> I mean inflation surprised and I say that in quotations because I was saying it's not going it's not a surprise it's going to go down but it surprised the downside a lot last week and so
>> Yeah, but that was last week that was before the kind of renewed tensions down there. So but I think you know that's just wishy-washy. It goes up and down up and down but down but we know that that war is not going to continue. Uh I'm sure oil will come back down. So I'm assuming the Fed doesn't raise hikes the end of this month or doesn't raise hikes doesn't hike rates. Uh at the end of this month. But uh yeah, let's see. Otherwise, I don't think honestly.
What was your view, Kyle?
>> Yeah.
>> Uh yeah, I'm super bullish. I I I I do think we're gonna I think we're gonna The fact that you said Micron, you're not sure if it'll be at all-time highs by the end of the year. I think that's crazy. Um I do think we'll be uh at all-time highs because I think hyperscalers are going to guide higher um coming for the next few weeks here. I think we're going to start to see ROI in companies uh and we're going to get that this earnings and next earnings. Um, and ultimately I think that's just going to be a bullish scenario. And then if you look at
The macro, I think the macro looks great. Uh, I think, you know, the fact that you're seeing banks do extremely well, uh, is a really good sign for the economy. Um, unemployment remains low, inflation remains low as long as the war doesn't continue, which I don't think it will. So, it's like you got macro that looks really good. You got capex that I think is going to continue to to increase.
Actually, a friend of mine I found out last weekend that their partner is a friend of mine's brother, their partner is um dating someone that works at Google and he has an unlimited budget. So, he buys uh land to build data centers for Google. That's his job. I was like, "Holy that's his job. Can we chat?" And I was asking a bunch of questions. First of all, he has to go into these um places with a different name. Pretend he doesn't work for Google until like the very end of the contract, but he has an unlimited budget. It does not matter what it costs. If he can find land that will allow them to build the data center and get energy, then he has just free will to buy at will. So, there's your like kind of insight on what these hyperscalers are are planning. And I think it's just buy more, buy more, buy more. And I think that's just going to be bullish for the market overall. He >> he could just buy my house if he wants to. [laughter] Dude, get this guy on the pod, man. I want to know what he's buying. I want to see that. I want a live call. That's your next live stream is a live stream surveying land, man. That's what I want. >> Yeah. No way Google would allow that. But yeah, I think we're just going to continue to to go up here. So, I I'm very bullish.
>> Vincent, I think you had you also had a chart you wanted us to look at today. >> Yeah, exactly. Right now, >> it's it's it's just another data point that supports the kind of bullish teases that we all three share. it's uh the the export numbers on or the export prices that South Korean companies charge for for DRM. from memory basically that came out uh today and I mean if you look at the chart it's just one vertical thing uh going up uh and it just shows the the demand resulting in pricing power for those mar and to me this is a really good chart showing how healthy this buildout is how much demand there actually actually is um and uh yeah just supports the teases that we're Here. >> There we go. >> Yeah. Ju just to second >> Yeah, just to give you some context on that chart. That that chart rocketed from 10,000 per kilogram for most of the past years to 100,000 per kilogram for DM. >> A 10x or even more than a 10x in a couple of months. >> And people think and people are staying bearish on memory. You're dumb. [laughter]
>> And here's why. Here's why this so the the bears will say this is cyclical, right? This always happens. it goes up and then it's going to come back down. And I think the thing that they're missing is all the things that we've talked about that we're bullish on. Like we haven't even got to the part yet where a bunch of companies are saying ROI is here from using just basic AI, right? So like that is to play for and then there is agents and then there is robotics, right? And like these are the infinite demand for these things because the ROI on these things is going to be absolutely insane. And these are all coming back to back, right? So it's not a cyclical thing because you have all these ramping up at the same time, right? Or or one after another. Um, and sure like so you have human robots or robo taxis you need memory like edge like inside the devices but then also to train those you need it you need even more data centers which need even more memory right so like you need it from both sides. Um, and as more hardware devices get put out that need AI, whether that be phones, which we talked about to run consumer agents or even like trading agents or it's, you know, cars for robo taxis or it's humanoid robots, they all need more memory and they need more training for that. So like we have so much more demand to go.
So, so two points on that. This chart is basically showing you the pricing power of a company like memory. They can increase prices. They're selling the same volume. So the cost basis stays the same. It directly flows into EBITS and free cash flow. And because of that, UBS and I have it on my notes here. That's why I'm looking down. UBS has um a forecast that Micron can generate more than 400 billion in cumulative free cash flow by uh 2028. And at today's prices could reby 40% 40% of their stock. That's pretty >> insane. They won't do that. They could do that. >> Yeah. And obviously prices will increase so the number will go down. Uh but it's crazy. And and and and and the second point you need to know is so so you may be asking yourself when is this chart coming down right it's basically a a supply versus demand question and I did some analysis on that and a lot of demand will like ratio to what total demand we have today in markets will come online by 2028. That's kind of where the pricing power could turn over. Again, it depends on how much demand we have from robot taxes, humanoids, agents, etc. Uh, but up until that, I think it remains a good buy.
>> I I do think that a lot of the buildout will probably most likely get pushed into 2020 as well, like the micron uh fab that they're building in New York is delayed because of, you know, all the issues with it. And this is going to be a continuous thing for the whole buildout. So, this could even go into the shortage could go well into 2028 and into 2029 if I'm being honest. So, >> yeah. >> What about when we move data centers up into space? How much memory do those use? [snorts] [laughter] >> More. >> More higher. SpaceX earnings August 4th. They just announced that. So, we'll get our first SpaceX uh earnings. That'll be fun, too. >> Yeah. Chart is down a lot. By the way, LG, you were calling it. You were calling it a couple of weeks ago. We were all wrong. We said chart was were was higher short term, but no, >> I should have gone lower. I should have I should have said a dollar >> and we said we said where would it be at in a week and it was at like 215 and you guys all said like something in the high 200s. I said 199 >> and now it's what is it today? 1157. Okay. It bounced up a little bit. >> It bounced. Yeah, I'm buying it lower. That's what I'm waiting for. >> I'm not buying it. Anyways, [snorts] uh guys, [laughter] if you guys do want to see what these guys are buying, uh and when Kyle does that, go Milk Road Pro. Uh you guys have been busy in the last couple weeks and and like you guys have said, um you guys have been been walking the talk. Uh you guys bought up a lot of that dip in the last couple weeks and and obviously today is a great day to to kind of for that vindication. So, um great work guys and uh thank you for the thoughts. This is a long one and you guys you guys love to talk about this stuff and I love listening and I think a lot of people do as well. So, if you guys want to see what they're doing, Milk Road Pro, it's just a dollar to try it out. So, check that out at the link below. Otherwise, guys, uh it's earning season, so I feel like we're going to have a lot of really great pods coming up, both with individually Melvin and Vincent. We have a great one that came out uh today when we're recording um about Phutonics. And Vincent, we got another great one with you this week. Uh and Kyle, if you guys want to check out something a little bit different, Kyle's doing a few live streams. I don't know if you want to plug that, Kyle. He did one. He did one. We'll see if he can do more, but he did one uh and it and people people really liked it. So, uh lots going on right now. No. Come on. >> [laughter] >> Melvin was too shy to join. Melvin, you can't even talk, man. You weren't even there. Anyways, uh, lots of great stuff going on at Milk Road AI, so make sure you check it out. Make sure you check out Milk Road Pro. Gentlemen, thank you for a great show. >> Thanks, guys. >> Thanks, LG. [music]
Want to stay ahead of the biggest technological shift in history? Subscribe now to get insights straight from the sharpest minds in tech [music] and finance. Quickly, you'll note this show is for educational purposes only. Nothing here is financial advice. Investing always carries risk. Never invest more than you can afford to lose. [music] Thanks for tuning in.