AI's Next Big Crisis Is Already Here

If you really want to know something's becoming a bottleneck, just look at what Nvidia is investing in. >> What's up everybody? It's LGAT here and welcome to Milk Road AI, the daily AI show that is buying yet another dip and will continue to buy the dips so long as the market is dishing them out. Today's August 18th, 2026. There is a major power crisis on the AI side in the US. We've spent the last couple years worrying about whether there are enough GPUs, memory, packaging, all that. But we're approaching a point where the US could have more chips available than it has electricity available to actually run them. Today, we'll dive into how our leading analysts view this issue and whether it puts a damper on the entire buildout trade and how the guys, despite these power issues, are still very bullish on sectors like memory. There's a lot packed into this hour today. And if you want to see how they are expressing all these views, that's in Milk Road Pro, which costs just a dollar for a 7-day trial at the link below. In there, you'll get all their trades, including the handful they've made today during the dip. And a reminder that our podcast today is free and that it wouldn't be possible without our partners at Saver. Money, the stablecoin payments platform built for Asia. Keep an ear out for more information about them later in the show. All right, boys. We're back. Markets in the red. We are going to talk a little bit about that, but we have other things we want to discuss, but we always start with how you're feeling right now. Melvin, what's going on? >> Uh, it's good to be back. Uh we had a bit of a outage last week and in our neighborhood with tornadoes and stuff, but luckily everything's all good. Uh but uh today there was a bit of a market selloff. And uh and this is not so about like earnings or AI bubble story. This is just a macro selloff because the biggest issue right now is the bond market. Um because today the 10-year-old treasury yield is around 4.7 I believe and the third year pushed past like 5.3% which is like the highest level since uh 2007. So basically when like risk free yields like yields get that high you know um stocks you know get less attractive because you can actually earn five plus you know in bonds without taking equity risk right so that's like the the big big uh reason behind the sell-off and another reason is um Trump came out today saying that um there has we're not making a deal with Iran or he said something along those lines which also caused a uh pullback as well. Um, other than that, AI story still remains intact. I'm super bullish on the market. This is a buy the dip moment. Um, and I'm going to buy some definitely going to buy some today. So, >> what happened last week? Did a tree actually fall in your house? Cuz that's we kind of made fun of you for it. But then we But then simultaneously later on in the day, we were kind of worried. So, you the house is fine though. You're sitting in the same house. >> Yes, I'm sitting in the same house. The problem was I couldn't leave my house or leave my street because there was like 10 trees blocking me off everywhere on the roads. So I was stuck in my house for like two days. >> Too tall. >> He couldn't climb. He couldn't climb over the tree. >> Yeah. >> Melvin's like 4 foot Melvin's like 4 foot one. So he couldn't get over the trees. >> What am I supposed to do? Walk to the stores like 2 miles away. >> My god, man. This generation Jed Z can't use their can't use their legs. Can't use their brains. Just need everything to be just served to them. >> You should you should walk to the store in your dad's pajamas in a blizzard up a mountain. No, I'm just kidding. >> All right. Vincent, uh, no no natural disasters in your neighborhood this week naturally, but how are you feeling about the, uh, the market? >> Yeah, except of crazy heat wave in Europe. It's all good, I guess. No, it's been it's it's it's been a completely normal week in the island. We had four new models. Uh, we had Nvidia coming out saying that the 800vt DC infrastructure will start actually uh in second half of 2026. So earlier than what we expected. And then we had some massive geopolitical swings across US, China, Taiwan, Korea. Um so yeah, pretty normal week in AI I guess. Uh no I think actually there are a lot of investment implications from that especially from the 800 volt DC kind of um uh part of the story has a lot to do with power with cooling. I think we'll definitely cover this uh in today's pod. remaining bullish. Bought the dip today as well on on on some of the memory names and uh yeah, nice to be back. >> Great. Kyle, how about you, man? >> Well, we just just before this LG and I recorded a podcast uh I don't know what the title is going to be, but something about the everything bull market, so I remain very bullish on the market. You'll get that tomorrow, by the way. Uh we had to put this one out first. Um but yeah, I think you know this is like whatever typical pullback we keep getting every time Trump scares the market. I don't care about it at all. I just had the 10-year up on the screen. It looks like we're at a topping point. So, that's actually very bullish if that starts to come down, which it has every other time that we've reached this point. I don't know if it'll happen this time. This has been kind of a slower and steadier grind upwards um compared to the kind of the quick moves up that we've had previously. But, um I'm not concerned at all. I think uh macro looks good and I don't buy into Trump saying that he's not going to talk to Iran. That's all I don't even know how the market keeps getting caught up in that. It's crazy. And then everything around capbacks and AI just remains extremely bullish as I continue to post about and talk about every single day. >> 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 positions they're opening, it's just a dollar in Milkroad Pro at the link below. >> Beautiful, man. Well, I mean, we we I feel like we've tal we've seen this story of Trump scaring the markets many, many times. So, maybe we can just attribute the pullback to that unless you guys have other reasons. Kyle, I feel like we the main thing we want to talk about today, and this maybe relates to Melbourne losing power last week, uh, is is the question of power, right? And it feels like we've even on the show, we've we've probably held a relatively narrow view of the power trade and and haven't had the time to really talk more energy. >> Yeah. So, let me tee this up a little bit because it's something I've been looking a lot for the last week and a half. Um, power is a is a huge industry and there's a lot to understand. So, it's it's a very tough industry. Um, but I think uh and it's very clear that if we see where things are going, we're going to build a lot of compute over the next couple years. Uh Meta is looking to build like 15 gawatts by 2030. SpaceX 10 gawatts by next freaking year. Uh and you know all the Neoclouds, they're non-stop. All the hyperscalers is non-stop. And it's clear that the biggest bottleneck uh is going to actually be power in the United States. It's not so much in China. In China, they have plenty of power. They've been building solar and nuclear and everything. And so they've got um more power than you can absolutely imagine. Oh, there goes Melvin. There was a chart that came out. And I'm actually just going to pull it up right now real quick because what this chart showed, if I can find it here, is that starting next year, we are going to actually have a surplus of chips versus the available power to energize those chips. Meaning, you know, there's still it's still tough to to get as many chips as we need. We need way more chips with the amount of demand for AI. But the problem is we can't actually energize those chips um because we don't have enough power, at least in the US. So, this is a problem because if we make a bunch of chips and we can't power them, no one's going to want to buy them, uh, because you can't make any money if they're not energized. So, either the chip companies are going to go sell them to China because they actually do have enough power. Um, but I don't imagine that happens because as Trump's already came across and said, it's not actually allowed. So, there's one scenario. Um, or we really need to figure out how to create more power in the US. And the problem is is we don't have nuclear coming anytime soon. the regulatory and political side of power is is really going against the US at the moment. So even in Texas, the most friendly place for power and for data centers, they've paused all data centers getting energized or getting approvals for power. Um we don't know for how long, but for the next little while. And we just had Jensen Wong come out and say that the biggest bottleneck is now land power shell, which is basically getting land, putting a data center on it, right? even just an empty data center and connecting it to some sort of power whether it's the grid or it's on-site power and so this is the real bottleneck and actually the one thing that even if we have all the demand in the world for AI all the demand in the world for capex if we don't figure out power it could actually kill this whole thing um because otherwise we're going to build a bunch of data centers buy a bunch of GPUs and we can't do anything with them so we need to figure this out my assumption is the US figures this out right when you know this is the beauty of capitalism whenever there's a if there's a way to make money from it, then you know, businesses are going to figure that out. And Bloom Energy is the one example of a company that has helped to figure this out, right? Like they said, "Hey, if you guys can't connect to the grid, we will just bring you guys power onsite uh and sell you that." And they're, you know, their earnings are absolutely through the roof. But the problem is, Vincent, you can correct me if I'm wrong, but I was looking at Bloom Energy and is it like they only have like 1.5 gawatts of power that they've ever that they've supplied so far. Is that is that the number? because it is not a large amount and I don't know how much they can really scale to fix this problem. Is that correct? >> Yeah. Yeah, that's more or less correct. I mean end of this year they have the annual production capacity of 2 GB and they have the optionality to scale that up to 5 gawatt. So >> in how long? >> In so per gigawatt it takes between 6 to 9 months for them to scale production. So you could assume that end of next year we're somewhere close above 3 GW, >> right? So that was my problem with Bloom Energy is like they're they're already priced for absolute perfection for the next like 2 years, right? They're like 300 and something P. Um, and they can only scale so fast. If they could easily scale that business to 10 gawatts, 20 gawatts, like okay, this company's going to go through the roof. Still bullish. And they're actually down like 10% today, so I actually might buy after this call. Um, we'll see where this conversation goes here on energy. But even this, the one solution that we're like, okay, Bloom Energy, bring power on site, still can't scale as fast as the amount of data centers we're going to have. Like, it can't supply SpaceX in terms of trying to get these 10 10 gigawatts they want in the next year. So, the question is is how do we fix this problem? Because the grid's not going to fix it anytime soon. Tesla is kind of working on scaling up battery production, which is helping, especially in Texas. Um but again not going to solve the problem today or in the next few years. Um so that's not really a full solution though it does help. So how do we like what other companies you guys looking at to try to solve this problem or like how you guys thinking through this issue? >> Yeah, I think we need to contextualize the issue a bit more here. Uh because my numbers are even more aggressive. If you if you listen to the to the Moonshots podcast, they're saying that AI could need around 230 gawatt by 2030 and we're only building 100 gawatt of of of of US grid capacity. Right? So there's a massive gap. But then the the the more important data point is only 5 to 10% of AI data center costs are related to energy. The rest is for chips and everything else. So the core takeaway from this is OpenAI and Tropic and all the others who are building the data centers, they do not care about the price of energy. They're willing to pay a lot more than the normal energy price, right? They care about speed to energy. So they would pay whatever price, right, for energy to to to get it on site. And I think this is a core this is one of the core teases for Bloom Energy because they're able to deploy the Bloom the the fuel cell boxes within 60 to 90 days versus you on average you're waiting 7 years for your data center to connect with with the grid right and then one more one more uh data point here what Elon Musk with SpaceX did to to to bring on the data center that fast is they bought all the uh the guest turbines to to to to uh bring energy, but they're sold out for the next seven years. So, Musk even bought his own turbine company. I think in May it was it's called APR Energy. It's a private company that he bought just to make sure there is supply. So, what it will be or or what the solution will be is as you said it's capitalism. So what we're starting to see in the market is actually the there is this one company I forgot the name they were building the gas the the turbines for the for the airplanes and now they pivoted to produce with the same technology the turbines for the data centers to with gas produced energy right so there's a lot of companies um uh pivoting uh to to to kind of that behind the meter on-site energy market um nuclear will add capacity but that's 23 30 and and beyond. >> And then you also see the solar investments, right? I think uh Tesla and SpaceX together, I'm not sure, maybe Tesla just by itself are investing 10 billion uh into the production of solar uh plus batteries, but that's also down the road. Um so yeah, the market remains constrained and to me that's very bullish buying yeah, more or less all the behind the meter energy companies. Yeah, I got some interesting numbers from Morgan Stanley. Actually, this might be a tiny bit old, but I think the numbers that they put out is like super conservative. Uh because Morgan Stanley um predicted that, you know, data centers will need 68 gawatts of power between 2026 and 2028. That is way way off. But to just to give you some context, um data centers under construction is about like 15 gawatts right now. And as of right now, we all know how that has essentially become a bo bottleneck. Only like half of it is getting built. So it's not even 15 gawatts anymore. And then um they also said that capacity currently covers another 15 gawatt. Um and then 38 gawatt of alternative solution gets layered in. And just to give you some context on that 68 gawatt perspective, like one typical like nuclear reactor generates about 1 gawatt. So we need approximately 68 nuclear reactors worth to generate the all the power that we need. And guess how many the US has? Zero. Like while China has like 30 to 40 right now. So this is a massive massive shortfall. And actually what's interesting is Morgan Stally also made some projections on where they could add these you know um gigawatts from. They actually predict that turbines is probably the biggest near-term like we mentioned around 15 to 20 gigawatts. Um and then Bloom Energy obviously we talked in Vincent's favorite stock 5 to 8 gawatt and then you have like uh colllocating data centers like Bitcoin miners and all that they could add additional like 20 to 25 gawatt and even when you add that up and with the current estimates right this is a massive massive shortfall and I think like this is the biggest bottleneck like that's happening right now but like how do we make money from this right I still like just to give you some context a couple months ago I owned a company called Vistra and they're like one the second largest like nuclear uh reactor company in the US and uh I still don't I actually like exited that position because I still do not think the best way to play the power bottleneck is just buying producers like Vistra now because the because the the way you want to think about it is where does the actual value creation from each megawatt is going right. A utility or like independent power company like Vistra can obviously benefit from like rising like electricity demand but at the end of the day they're only they're only selling primary electrons. They do not capture full economics of the entire buildout that convert that gets converted into AI compute. And this is why you saw the chart going around in Twitter about like how much mega revenue per megawatt is being generated by companies like SpaceX and Iron and and u Nebus and stuff. SpaceX I think roughly generates around 40 to 50 billion per megawatt whereas like Nebas is around 10. Um iron coreweave is around 10 as well. What's interesting is Nebius actually is signing deals for $40 to $50 million short-term deals now because of how constrained we are. So I think that this is why if you look at companies like Nebius and Coreweave is getting massively like you know going on these huge runs because it's all about who can take the electrons and convert them into uh revenue. So that's the entire thesis. The the thing that I think is going to happen though and and probably already is is that whoever can figure out to supply the energy like Bloom Energy has, they're going to hold very strong margins for multiple years here until the energy like the grid and other parts of energy starts to catch up. So, they're going to have really strong margins here for a while. Um, and maybe the most powerful margins of all of it because again, don't really care if I can buy chips or not. If I can't energize them, then who gives a right? So, there's like a couple ways you can play it. Like Melvin just says, just buy the Neoclouds cuz they're the whole package. It's energy plus data center plus GPU. You could buy land power shell, which is what Chimath had talked about, and also Jensen just wrote about. Nvidia is now starting to um provide financing for this. So basically, they're going to backs stop um deals. They just did it with OpenAI and BS Energy, I think, was the company. I don't even know who they are, but basically they're going to allow and help finance a 20-year lease for OpenAI um to to get power and land for for for that 20 years, which previously was kind of a harder thing to do. Galaxy is another one we've talked about a ton on this show and and our analysts, some of our analysts like, which is again just getting land, building a data center, and having the ability to connect it to power. Part of getting power is also just having the relationships with the politicians and the energy companies. So Galaxy's done a good job of kind of like working its way in with with the the Texas grid aircot and so they can get kind of preferential access. So even though a bunch of data centers have been put on hold right now because aircott's not allowing anyone else to get approved, they're still approving Galaxy because they've already been approved previously. So they know they're they're a legit customer. So there is that way or there is there are going to be companies that help solve this like you just said Vincent that company that just pivoted from uh you know they were building engines for planes and now they're using that technology to to provide power to data centers. This is very similar to two years ago when all the Bitcoin miners decided that they were going to pivot from being a Bitcoin miner into uh being a NeoCloud essentially right um because they knew that they had better economics over on the AI side. That's the thing that you can start to potentially find in the market today is what companies are providing power elsewhere and will they pivot into AI and provide power here because they're going to have way better margins. I don't know those companies today, but um trying to figure that out because I think that's that's still a big opportunity. >> I think that the pricing argument here is really important especially for Bloom Energy as you said Kyle because you you you asked how they can scale right and how they can increase revenue. Yes, they have a certain volume in terms of gigawatts, but pricing is the core thing here because remember I said that only 5 to 10% of total data center cost is energy. So it doesn't really matter when a Bloom Energy increases its prices to the total cost of a data center operator. So they have a lot of flexibility in terms of increasing prices and therefore increasing margins, right? Then if you look at the Bloom Energy Q2 earnings, you see a massive growth in operating margins because they have their pricing power. Um, so that's that's point number one. Point number two here is is another way on how to play this and this is kind of a different angle and is connecting back to what I was saying in the intro around the 800 volt DC architecture of Nvidia. What this basically means is that we are um uh introducing 800 volt DC inside the power racks. Um so we have a higher energy density inside the Rex where the chips sits. And this implies that we need many more power semiconductors. Basically you can think of them as the chips that manage the electricity that step the voltage up and down depending on on on the voltage needed for the GPUs to work. um and there companies like Infinian um and others are positioned well. So this is is an angle and and and then again Nvidia just came out I think it was last week saying that we're starting with 800 volt DC in uh second half of 2026 and then massively scaling it into 2027. This is a an angle that I think the market has not fully figured out yet and is because >> the efficiency trade you mean? Yeah, the efficiency trade but then 800 volt DC just requiring or enabling for an infinion for instance to to to increase its volage here. They can sell just many more of those power semiconductors into the data centers versus they can now because Nvidia is switching to that higher uh voltage of energy that is flowing into those GPUs. Um so that's an interesting trade from my perspective that I'm positioned for as well. How long do you guys expect this issue in power uh like this bottleneck in power to last for? Like at some point we're going to figure it out and we're going to scale power whether it's nuclear but that's 10 years or something else. Um or we get really efficient at using power and we just need less of it. But I I'm from my perspective is we're going to have so many more data centers being built on Earth for the next little while here that it's going to be a problem for years. But then eventually we'll have data centers that go up in space and then okay, it's no longer a power issue here. um or we find solar or whatever that can fix it. But it's what a multi-year story. How far out? Because this is what's important for a bloom side of things. Like on-site power is very important until the grid can catch up. If it can catch up and if it does catch up, then on-site power is only there for when there's like issues um or they need a little bit of extra, but it's not a thing that's going to scale out to like it is today where it's so needed. >> The core issue essentially is the grid, right? As long as you're waiting 7 years and more for a data center to connect with the grid, there will always be demand for behind the meter power solutions like Bloom. Now you can look at companies like Quant Services who are providing all the things uh or or an ABB or Eon or Schneider Electric that are all providing the switch gear, the transformers that are needed to build out the grid. Now they sold out for years, right? They're racing to increase capacity themselves. The stocks are up massively because they revised guidance up and up and up. But it takes years for them until the grid like 2028 2029 the earliest. But but then the gap in terms of how much energy we need for compute is only narrowing but it's not fully getting solved. So to me those will be trades that I hold at least until the end of this decade because there there is no obvious solution and how this bottleneck can be solved. >> Okay. So let me flip the switch. Melvin and Vincent is is or should we not be looking for where to invest in energy? Instead should we be worried that this is going to screw over the entire infrastructure buildout trade? >> Yeah 100%. I think that the the trump card in all this is like Elon Musk uh because I feel like he or essentially he only has the power right now with everything that he's doing with solar batteries to really scale this up. I I think he's probably the biggest bottleneck remover. And if you really want to know, this is a tip for subscribers. If you really want to know if something's becoming a bottleneck, just look at what Nvidia is investing in. And we do see that time and time again with TMMC, we see that with memory, we see that with like packaging and right now Jensen things like we said is power. So this is this is a multi-year multi-year like shortage and I honestly don't know when this is going to this going to be, you know, resolved. So >> Kyle, you're making the exact point my the the biggest position of my portfolio builds on. I think I and I always thought this right. My very first Milkro Pro report was built on this argument that energy is the ultimate bottleneck of everything here. We do not need memory if we cannot energize the chips, right? And yeah, the to me the favorite expression of this trade today is still bloom energy. Uh yeah, but there are many other ways. >> So why are you buying memories? How come you bought memory? Because memory uh companies are still able to sell more memory moving forward and increase the margins because power will yes it will be a core bottleneck and it will be always there and the companies will benefit massively and it will slow this build out down of for sure but it will not stop this buildout. Um and that's why I'm buying the dip here also today. I've never actually I'm still trying to figure this out. Actually, for like the last two years, if you actually look over on Twitter and everywhere, we knew this was coming. We knew power was going to be the bottleneck. Yet, if you look at the companies like Vistra and all these energy companies, they did not move up. I mean, they had a bit of a run up like in 2025, but 2026, they're flat. All the energy companies, they're essentially flat. So, is >> same not all. >> Yeah. So is the market knows that there is a there is other alternatives alternatives coming like Bloom Energy and stuff or is there or are we missing something here? >> Well no the the big reason is that they used to make so those kind of businesses the way they actually work like a Vistra is they don't sell the power directly to you know a data center they sell it to air. aircot is like a marketplace for power and then they choose who the buyer is, right? And so what happens is the way that those companies make most of their money is whenever you hit like peak energy times where like everyone, you know, needs to turn on their AC at the same time and the cost goes way up. That's when they make their money, like the high margin moments. And the thing is is that Tesla flooded the market in Texas with batteries over the last 2 three years and they never have those peak moments anymore. And so they've lost a ton of revenue. They still have a ton of demand, but they don't get those high high margins anymore. This is why electricity costs have actually come down uh in place like Texas because batteries have smoothened out the the the energy like the grid. Um and so they make a lot less on their margins in those peak moments. And so that's why their companies their revenues continue to grow, but their multiples are coming down because they're not getting those big spikes anymore uh because of batteries. Um and so that's happening kind of all over the US. So that's a big part of why those companies haven't actually moved. Um, it's helped the grid and helped consumers, but it's not necessarily helping their business. >> It's not true that the energy companies didn't move. I mean, Quant Service is up 82% year to date. Bloom is up over 100%. >> Yeah, he's talking about ones that attach to the grid, not not Blooms. Those are different, >> but also not true. I mean, look at Eaton share price. Let me just quickly uh check this year to date is up at least 31% right after a big draw down. ABB also up massively. Uh so 46% year to date, right? What you need to understand is those are massive companies, right? Couple of billions of of of of market cap here. And I think the mark what's important to understand here the market and the investors in the AI infiltrate are really focused on on compute on the computer right the neoclouds etc. but then also memory of course and this and behind the meter energy you could argue and there most of the capital was flowing to I think as we're part of this midcycle slow some of that capital will also shift into that broader trade and will they perform as well as the memory companies did over the last couple of months probably not but they will perform really well >> the the other thing to note of why memory is still bullish in my opinion is so this chart um is is showing accuracy per per jewel So you can think of like intelligence per jewel, right, from Frontier models. And you can see that we're getting more efficient uh over the years. Okay, the big some of it is just because the models are getting more efficient, some of it is because the hardware is getting more efficient. The big jump is um is Nvidia Blackwells, right? Um which is moving it more efficient, but also um memory inside of GPUs makes this a lot more efficient. So if you remove memory, which Nvidia talked about potentially doing, they can do it and it works, but it's a lot less efficient. And so if they want it to be more efficient, the GPUs to be better, they actually need more memory inside of them. Um, which is why I think, you know, we're going to push that efficiency trade, which we talked about, Infinion is one of them. More memory is another one, more HBM. So, um, I think, you know, if energy is a problem, we're just going to need more memory for the GPUs. Now, we still got to be able to energize the GPUs, otherwise it's all for nothing. But, um, that's another big part of it. >> You know, one thing we've talked about a lot on this show is that crypto is quickly becoming a huge part of the global payments infrastructure, and nowhere is that more obvious than in Asia. But if you're actually running a remittance company or a payment business, you know that the hard part isn't moving the stable coins. It's dealing with local banking partners, compliance, liquidity, and all of the operational headaches that come with sending money into places like India and Southeast Asia. That's why today's partner is Saber. They give payment companies stable coinpowered infrastructure to collect and make payouts across Asia without having to build all of that complexity themselves. They've already processed more than $3 billion in transactions across 40 different countries. So, this isn't just a concept. These guys are actually doing this for real. If you're building payment infrastructure or expanding into Asia, make sure you check out saber.money. >> Speaking of memory, >> you got some of that memory m there's been a lot on memory lately. So, let's let's hear it. >> Yeah. So, actually there was a couple interesting news that came out this morning. One was from uh Bank of America and they basically addressed what you essentially said um um Kyle about you know there was reports that you know Nvidia may cut like uh Ruben Ultra from one terabyte HPM down to like 192 or to 220 288 HPM or gigabytes and um and everyone was losing their mind. This is immediately bad for freaking memory and whatnot. But the reason the reason why like Nvidia is testing lower memory is because because it needs more memory. They can't find enough memory from Micron and all the skinex and Samsung in the world. And that's why, you know, they they're redesigning their entire, you know, um um chip. Uh but what's interesting is this is actually a this is actually a bullish sign long term because if you actually like if you for workloads like robots and like um other models um you know having GPU like having memory close to like um next to your GPU is extremely important because they they need access to memory right away when you get into robotics you know that you got to be able to remember all these contacts and whatnot. So this is extremely extremely bullish for memory and actually UBS came out today and they also put a price target on Micron as well today. They're actually expecting Microns to hold their gross margins above 90% um exit by 2028. So I think uh Micron right now currently has like 82%. So, they expect that to go higher even. And um and they also like, you know, we talked about this time and time again, long-term agreements. You know, they're signing all these long-term agreements. And actually, one of the notes that came out a couple days ago, um Micron is actually expected or want to sign 50% of long-term contracts with um all these hyperscalers and whatnot, which means they're going to hold these margins and all all the they're going to get rerated pretty soon. and they haven't started buybacks. That doesn't start till December. The free cash flow of Micron is about to be uh actually I think by next year they will be in the top 10 companies for free cash flow over hundred billion dollars a year. Uh which Micron's not today, but there's only I think 10 companies that have that or whatever. And Micron's going to hit that next year. And they're going to be able to use a lot of that to buy back shares, which they've never done. They're not able to until December 9th. So I think that's going to come. SKH's probably going to do the same thing. they don't have a a block on it. So they already could. They've talked about it. So I think buybacks is the next big leg up and the market will front run that so we don't have to wait till December. I think it'll happen before. >> One more point on this whole memory story. I think from a geopolitical perspective there is one important angle because the US administration Howard Lutmik the the commerce secretary I think he is came out or or actually approached Apple telling them to not source Chinese memory because of security issues because of independence reasons right etc. The important part is what they're doing with that is they I think unintentionally create kind of a a premium for the memory of of of US-based companies like Micron, right? Um and as we're moving into this world where memory is really this strategic important infrastructure part of this whole buildout then the multiples of those company should should rerate to the upside plus the earnings per share that will grow plus uh all the other developments that that go around memory makes me really bullish and to your question Kyle that's why I bought the dip on my because the outlook is really nice. Uh, I got a question. This actually just made me think about something. So, we were just talking about power. Let's say that the US can't get enough power to power their data centers. Okay. So, China could, but I think as we know, they're not going to sell their chips and, you know, and and data centers over into China. We're not going to put them in China. We see that cuz, you know, literally everything Trump does is like, "Hey, don't sell memory there. Nvidia, you can't sell in China, blah, blah." Like, I know that's not going to happen, but that doesn't mean they won't go to other countries. What if they went to other countries and built data centers that are energy rich and have a lot of resources? Would they think about doing that or are they are are they already doing that? I don't I don't I have zero clue. I'm just wondering if they can't get the energy in the US, would they go elsewhere? >> I don't fully get it. You You're saying they're building data centers in other countries because they have energy and the US has not. >> Correct. Would they do this? Yes. I'm just spitballing. And maybe they make some sort of agreement with I don't know Argentina or India. India, Canada actually like semi analysis I think posted something saying that India is supposed to have 10 gaw gigawatts by certain year I forgot what the date was but I can absolutely see a world where this happens cuz this is what we do with all our supply chain issues right wherever there is like cheap things we go and conquer that side of the world and like you know >> but this is happening I mean here Melvin you tell me but which of the neoclouds did build data centers in Wales, I think in Finland or Norway. >> Yeah, Nibbius. >> Nibbius, right? So, and and and the UAE is also building data centers, right? Because they have abundant energy, right? So, this is not an if. This is this is happening for sure. >> H interesting. I got to do more research on that because it feels like there's more opportunities there. Melvin's like buys. My my angles on this is still that sovereign and AI, right? South Korea is pushing that way, Japan is pushing into that way, Europe is starting to do it is really bullish for the and and and and this also ties back to the four models that came out this week, right? The more sovereign AI we see, so meaning the more con countries build their own AI stack, their own data centers, the more different models there are coming out, the more of AI, energy, memory, uh, all the the kind of stuff we need, right? So to me, this is all really bullish for the AI trade from a really ba really basic perspective, right? This is this is how I view that. Did you guys see the projections from Jeff? Um I think it was earlier this week or last week. Uh they basically projects that anthropics uh ARR run rate hit $74 billion by July 2026. Just to give you some context, 1 billion, they were at 1 billion in January of 2025, nine billion in December, um, of 2025, 30 billion by April, 47 billion in May, and right now they're roughly sitting at 74 billion. >> Well, we got confirmation this morning or yesterday. They're at 65, not 75, 65, but but um, it is still insane numbers. So, like if you look, they hit 11.5 billion in Q2. Um, and so that was 14x year-over-year growth from uh the quarter previous to it or sorry from from last year. But the Entropic is expected to end the year at 100 to 120 billion AR. But what's also cool is OpenAI reached 40 billion. Um, so it's not just one company anymore. It is it is two, which I think is huge. And then I really think Grock and Curser are going to become a serious contender in the Frontier model race here. And so I think look, they're much smaller. I think they're around maybe like 6 to 8 billion um AR right now. So it's it's much smaller. I got a feeling they're going to grow very fast and they're going to eat some market share here. It doesn't mean it's going to kill open athropic, but they're all going to grow together. But I I got a feeling by like the end of this year, Grock could be closer to like 20 billion, which would be insane. Um, and then I don't know if you heard the all-in podcast on Friday, Gavin Baker and whoever were saying, David Sax, I think that they think Anthropic will end next year at 250 billion to maybe that was their bare case, I think, and their bull case is 400 billion. >> Exactly. This is insane, man. >> ROI question mark. >> Yeah. Well, the ROI is everywhere. But the again, if power is a bottleneck, what do we do? Um, we're going to have to build data centers outside of the US, I guess, >> which is probably going to cause World War II. >> Kyle, I haven't seen your SpaceX video, and give me your one minute TLDDR or why I should watch it because SpaceX keeps coming back to my mind these days around short-term, they're so interesting from from Elon Web Services. Long-term, they're interesting from from compute, from up above, right? Um, Starlink, etc. So, uh, yeah, >> you should watch it because I'm your boss. No, I'm just kidding. I'm just kidding. Uh, so what what SpaceX is building is optionality. Uh, so I think um compute is is the biggest moat uh in where we're going. Um, I think there's no one better positioned to build out more compute than SpaceX. They're about to do the fastest buildout of compute in history. So the fastest pre So SpaceX is trying to build get to 10 gawatts by next year. If they do that, that's 8 gigawatts in one year because they should end this year at 2 gigawatts. Um, this is almost double the fastest builder ever. AWS did this or Amazon did this years ago. They built it, I think it was 3.4 or 4.3 gawatt. Uh, and it took them I think 18 months or or something like that. Um, SpaceX is going to double that in a year, which is crazy. Now, here's the reason why I'm so bullish on this one. If they really and the reason why I think they do it is they have the best team to figure this out. They have the best engineers in the world. They built reusable rockets. They have Tesla to get batteries and energy. Like if there's anyone that can do this, it's them. They built Colossus in 90 days, which is insane. So where I think this why I think this matters is they are going to have optionality in this 10 gawatt. So they can sell it for 30 to to to 50 billion a year. If they sold all of it, which I don't think they'll do, but if they did, that's like a $500 billion uh in revenue by the end of next year if they did that. Okay. Now, I don't think they will. Elon even said 10% probably goes to Grock Cursor. But the key thing is is a lot of their deals so far anyway are short-term deals that they can get out of if they want 90 days. So, the optionality they have is if um Grock Cursor actually starts to crush it and do really well and compete with OpenAI and Enthropic, they can get out of those deals and flip it and put that into training inference for Grock at any point. So if Grockbot takes off and they become the agent, you know, harness um and the developer harness with cursor, they can give the compute over to that instead. And that's an even better multiple than the multiple you get if you if you resell your compute, right? And so what they're going to do, I think, is if they and it all depends on if they can build this out. And no other company has the ability to build 8 gigawatts in a year. No one. Not even close. Not even the hyperscalers. Like they're not even close to this. I believe they can do it. And if they do that, they're at $500 billion in revenue by the end of next year. Whether that revenue comes from resell and compute, which has its own multiple, or it comes from there being the next frontier model, which has an even better multiple on it. Um, and that doesn't count the scale, I think, that they can get Starlink um from the Starship uh that's coming. So, so, so you have this like let's say $500 billion coming from both uh either compute or the frontier model and then you have this like connectivity which I think is one of the biggest moes that they're building up right now uh in Starlink which really really ramps up next year once they start launching the V3 uh satellites and V3 satellites enables them to have basically as good as download and upload speed as fiber does but you can get it anywhere in the world. So right now like Starlink is great and it's generating uh I think it's going to hit like 20 25 billion by the end of this year and that is from like more niche areas where like you can't get fiber. So like they're selling in developing countries, they're selling to like airlines, they'll sell to cruise ships, but it's harder to get someone in like Toronto or I mean I was going to say Chicago, but I don't know you have tornadoes every day so maybe you need Starlink but like you know in in developed countries and developed cities like they're just going with fiber because they don't need Starlink. But what I think is going to happen is if they can get this thing good enough and they can go direct to device, they're going to start to compete with AT&T and T-Mobile, etc. Um, and so this becomes an also very massive business for them that really starts to ramp next year. Um, and so like Starling could easily become a 50 billion plus AR company as well. And that doesn't even count the rocket ship business that they have that's incoming too. So like they just they have three big moes which is rocket launches, connectivity, and compute. And all those feed off of each other as compute moves into space. So I just think like if you were to do I did like a fair valuation on their business by end of 2026 just saying they hit 100 billion and the fair valuation of that company actually sits at around $130 um per share if they hit the numbers that they're expecting to hit by the end of 2026. But really you typically would value a company out farther than what are we four months away from the end of the year. And so if you start to look at them really hitting those numbers on those gigawatts, I actually think the company's undervalued. And this doesn't account for the fact that something like SpaceX usually gets an Elon premium, which I don't actually think is because of Elon. It's a moat premium, which is what Tesla gets on their robo taxis and their humanoids, their ability to scale, manufacture, and have their own um vertical uh manufacturing capabilities. SpaceX should have a premium because of their moat around connectivity spa uh space launch and and compute. And so if you add like an Elon premium which is usually like 40 75% this company's actually extremely undervalued to where it should be by the end of this year. So that's kind of the the the I don't know what was that 30 seconds 60 seconds >> or like five minutes. That's all right. Um, so I just think I think SpaceX is a way bigger company than people are realizing and it's the speed at which they're going to get there is is really crazy. Um, now there's risk like maybe they don't build that 10 gigawatts and maybe they can't figure it out. That's the risk. But if they get that 10 gigawatts, this company's a whole beast, a whole different beast. I think the underestimated part is really Grock bot here because the way I think about this this new product which I think last week came out right is it's the same as open claw but kind of productized and ready for mainstream because you just install the app and you create the bot and it runs on the cloud. You you don't need Mac minis etc. you do not need to handle the permissions all that kind of stuff right so it's a more mature version of open claw >> and if they are able to follow in the kind of entropic entropic footsteps in terms of growing revenue yes it will be not that scale definitely not and it maybe will also not grow as fast as entropic is able to but similar then this could also grow into a massive business as well >> yeah I think Gavin Baker actually tweeted something interesting out the other day he said that um it took him like about 15 seconds in Grockbot to build like a AI podcast summarizer, you know, and so which is which is insane >> which is that said it took him for like so long to do previously. >> Yeah. This is why I think the the the closed source models that they're in open are not getting crushed by open source models is because it's about the UI and the UX, right? which doesn't last for forever, but it's the it's the fact that like if you want to get an enterprise to get their thousand employees to use even just chat, but let's go agents too. They're not doing that on Kim K3, right? Or on on on on anything else that's open source because anthropic and open have built the layer to be able to manage usage and connect into your like companies uh CRM, etc. Like we have that in both of my companies um and create agents that just work and everyone can use inside of your Slack. That's so hard to do if you don't have that UX. And this is what Grockbot just did is they're taking like I'm an AI power user and I used OpenClaw on a Mac Mini. It was very hard for me to get a Hermes agent set up and I ended up just giving up on it. I just use Cloud anyway because it's just annoying. I just didn't want to spend the time to do it and I can't. Now I can just use Grockbot to do all that. Like it's I I think it's going to grow really fast. And that's the part that the market started to price in. So two Fridays ago, semi analysis put out the idea of SpaceX hitting 10 gawatt and SpaceX re repriced from like 110 or whatever it's low hit back up to like 140. And I don't think and I think that's only because they can sell compute. What the market's not understanding is this is going to be a frontier model like an anthropic or open AI and that has a whole different multiple on it. Like Enthropic is about to IPO at two trillion. >> Yeah. >> SpaceX isn't even worth two trillion and it not only could be to that size. It's not today but it could be. Plus, it has all the other businesses to go along with it, right? Um, so I just think, um, SpaceX is going to quickly become >> Are you checking the side of Farset now? I saw your beef on X. >> Beef. >> Yeah, that was just not beef. He just was like, >> yeah, he's just a Super Bowl and just like he was like, "If Enthropic's worth two trillion, then Grock should be worth two trillion." I was like, "That's the dumbest analysis I've ever heard in my entire life." Like, what are you talking about? He was like, "Yeah, you're right. I I I got a I'm going to take LG's job real quick and I got a fun question for you guys. Okay. What do you think the highest price target on SpaceX is um as of right now? Any guesses? >> Oh man, isn't it like 800 bucks or something stupid like that? >> Is it? >> Yeah, it's like 800. I didn't expect you. >> What is that like 10 trillion market cap? >> Yeah. >> Yeah, it's insane. And what's the date on that, Melvin? What's the projection on that? >> Uh no, I don't have the data. I just remember it was >> 800 by when though cuz that's not that's not that could be like 20 years. >> I think it's like 12 month or 24month target. So by like 2 years some crazy insane number >> look think about it two years we're sitting in August 2026. So this is August 2028. We're talking about them hitting a $500 billion potential revenue run rate by uh the end of next year by 3/4 of the way through 2020. Um but so I guess August 2028 that's a whole different story right like what they could be close to a trillion by at that point. >> So what's your price target for uh you have any price targets for end of this year for next year? >> End of this year no I don't the end of this year no not at all like um I don't I don't really have a price target for SpaceX to be completely honest. I think it's too hard because of the unlocks makes it difficult to see where it goes in the next you know 6 months but I just buying on dips. Who is it in the chat that said they don't do price targets? Was it you, Kyle? >> Yeah. >> Did you say that? You said and Melvin said he only does price targets, right? >> What's your price target? >> What's mine? 200. >> I think 200 by the end of this year, which is I think >> 200. >> Yeah. >> What's that another 30% 30 40% from here >> pretty much. >> Yeah. It's 144. >> It's all going to depend on the only thing that matters for Well, actually, there's two things. First is the amount of gigawatts they can put online. So if that tracks SpaceX is is ripping, right? If they're actually getting close to 10 gigawatts, SpaceX is going way higher. And then the second thing will will be what are the revenues turn out for Grock? And if that actually starts to ramp at a similar speed to open athropic, SpaceX needs to go much higher. I talked with Melvin about this in in the chat this morning. Vincent, I want to get your take. Let me pull up in this chart real quick. So speaking of optionality of of compute, okay, meaning being able to Oh, I haven't posted this yet. Actually, it's coming out soon, so I got to find uh but I do know the chart here. Um the So, we talked about the optionality of like building compute and then using it internally or selling it externally. Another company that I think might have a really good chance to do this and Melvin's talked about this a few times is Meta. So, this is Rothschild and Co Redburn. They came out and they said they think Meta will reach close to 15 gawatts by the end of 2030 and they will launch a cloud business. So doing similar to what SpaceX is doing. And they will be able to use that compute on their ad business and their social media business first. Whether that lasts for a year, two years, whatever, they'll be able to make their return on that right away. And then they'll go sell it afterwards, that compute, that access compute afterwards to other companies with a new cloud business. Um, if that's the case, Meta is significantly undervalued. Curious if you believe this is going to happen, Vincent, because I know you have some takes on hyperscalers. And then Melvin, we'll let you wrap up after. >> Yeah. So what I really like about Meta is that they came out last week with their MetaMuse Glimmer model. It's an openweight Aentic model that you can run locally on your PC with just one GPU. This is basically showing you how much a Genti can become cheaper. But what's the biggest story is if they're able to deploy those agents inside their legacy businesses and increase kind of conversion, increase ads, start to understand even more about their consumers, the strategic value of this is pretty interesting. Now that being said, there's still meta glasses. there is still them competing against Elon, competing against the Neoclouds which have proven that they're faster, that they're probably more efficient in terms of using the GPUs. And those are the arguments why I probably would not buy it here be because I like the others more. But if I see them adopting those AI agents in their legacy businesses and the tools we're all using Instagram Facebook WhatsApp etc., then I will become more interested. >> Melvin, what's your take? >> Yeah, just to give you some context, I think Meta is, you know, we said SpaceX is um building like 10 gawatts by end of the year, but this is a long term, but Meta is um building 20 21 gawatts roughly by end of 2028 and roughly about they only need about 13.2 of that um for internal purposes. So they can very well do what SpaceX is going to do and sell out the excess compute um towards like other um other you know what SpaceX is doing. So that's one side of the business. I think they're going to massively massively like um capitalize from um adding I I believe they bought recently um I forgot what the name of the company was, but they want to essentially add that into all the meta and like um into their models as well. Um do you guys remember the name of it by any chance? it was uh some Chinese company, some Singaporean company uh for like $23 billion. So I think they can capitalize massively on that and the meta glasses I think that's where I think the future unlock is. I think all the models and everything will get put into that and that would be like a worldwide adoption. I feel like cuz I even thought about getting meta glasses like which are worth like three 400 bucks even though it has like less capabilities now but in the next two three years I think they're going to massively revamp and create new ways to capitalize on this as well. So I think there is a lot of short-term um you know uh price action with this stock largely because of the lawsuit that's going on right now. Um, so I don't know how that lawsuit and how that's going to resolve and how much they're going to have to pay out for that, but this would be a good time to like add in shares here for, you know, for 2028 or for 2030. So I I'm I'm I remain bullish on that. Actually, >> what I would love to see on that chart is the excess AI cloud. So this this bright green part much larger, right? You were saying earlier Kyle that SpaceX is doing that SpaceX is doing what 90% selling capacity versus 10% using it internally or let's say it's anti 8020 right here it's much more internally right we also see Google shifting towards that end right with with deis uh staying but actually leaving because they're shifting focus to what towards the cloud business and the market is actually saying we're valuing this more because the the the revenue ramp from this is much more clear versus you guys building new tools and using it internally and the difference here >> Meta has more of a business to use their compute on whereas SpaceX doesn't yet right so SpaceX is 90% 10% because it doesn't have as big of a business yet whereas Meta has you know these how many platforms with like billions of users >> business owner support and all other things. >> Yeah, but you don't need 5 GW and then eventually 6 G and more for that to run. What they're using glasses. >> Exactly. This is what they're using the gigawatts for and glasses is zero revenue today and you do not have any uh transparency onto how much they will bring in. >> My other fear is I so to build you know any gigawatt in a year is a tough thing to do and I think SpaceX is going to be able to do it. Meta is not a hardware company. They are a software company, right? A social media company. Whereas like SpaceX is literally a hardware engineer company. So to go and build data centers is easy for them. >> I don't think it's going to be as easy for Meta unfortunately. And so I don't feel as strong or bullish in their ability to do it. However, I do love the optionality in the business if they can succeed even at some level. Uh Meta is priced at freaking nothing right now. So um I do think it's a great time. Um, and is probably something I might buy after this call. Where are they? Anyway, you can wrap us up now. Uh, sorry about that LG. I'm sure people will appreciate the extra the extra chatter. Um, I did want to ask you guys what what you're looking at buying this week, but I think Kyle just told us his, and I'll let you guys answer that in the chat for Vincent and Melvin, or at least answer that for pro members. So, if you guys do want to see maybe what they're going to scoop up, maybe they'll buy some of each other's portfolios because not all these guys own all the same stuff. You can check out to buy a bunch of So, if you're not a Milk Road programmer, go right now because literally I think all three of us about to go buy >> I mean, you probably have done it by the time this gets edited and put out. But it is an immediate thing cuz there is a dip happening today. So, you guys can get in on it. Uh, and follow these guys along. Is this a dollar for Milk Road Pro? So, check that out at the link below. Guys, thank you for the marathon episode. Uh, lots to learn from this one. So, appreciate your time. Want to stay ahead of the biggest technological shift in history? Subscribe now to get insights straight from the sharpest minds in technical 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. Thanks for tuning in. See you in the next one.