The NeoCloud Problem Nobody Talks About

you need to differentiate whether the market is puking or if just your your companies are not doing well. >> What's up everybody? It's LGD set here and welcome to Milk Road AI, the daily AI show that only gets bearish when it's convenient and we know the YouTube comments might kind of hate us for it. Today is July 28th, 2026. The market is nuking. It's very unhappy and conveniently, uh we'd we'd been planning the show for a while, but Martin has a presentation today for us about Neo Clouds to kind of counter Melvin's episode from a few weeks ago giving more of a bearish take. Not necessarily a totally bearish take, but basically showing the economics of Neo Clouds and how they actually have to finance all those gigawatts that they are promising that they can do and what happens once those chips uh kind of crap out in 5 years cuz I didn't know this, but chips only last a couple of years. So, it turns out that may be a reality for these Neo Clouds and that is basically what today's show is going to be about. We're going to look at that, their economics, and then, you know, it is a red day, so Martin is going to take us through how he manages days like today. What to do with your portfolio on days like today cuz I think that that is also important. If you want to get it access to Martin's portfolio, how he's managing this, you can pay a dollar for Milk Road Pro at the link below. And a reminder that our podcast today is free and it wouldn't be possible without our partners at Securitize, the regulated real estate tokenization, and Bitget Stocks 2.0 with real liquidity, real dividend. Keep an ear out later in the show for a message from them. What's up, Martin? Tell me what we are talking about today. >> Hello, sir. Well, I want to talk about CapEx because, you know, it it's um summer 2026 and all earnings call are about CapEx. And so, some companies are reporting huge CapEx even though they earnings are just, you know, uh really good. Let me actually share the chart. You can see the Google's numbers um for their Google Cloud revenue and operating income. So, this looks like really really good. I mean, it looks like it's accelerating and so you would expect that when Google announced their Q2 earnings and you would see those numbers, you would expect that Google will just shoot to the moon. But in fact, you know, Google dropped significantly. And the reason is because they increased their capex estimates for the future and it results also that in Q2 they free cash flow was negative and so the market is punishing them pretty heavily and it's underserved in my opinion. And the one point I want to make here is that market is punishing some players like, you know, um I just showed Google, also Tesla, but then there are some players that are still getting that AI label very hot sexy narrative and they are burning a lot of money as well, but market is not punishing them for burning a lot of money and also increasing capex year over year. So, that's what I want to talk about today. >> All right, so the market's being gentle with the the maybe the non-hyperscaler names who have been more, you know, uh hot picks this year. >> [laughter] >> Instead of Instead of boring old Google, boring old Google people are happy to happy to dump, but the more trendy picks people are are the market is happy to hold, but but you're basically saying that the bearish argument is the same for Google that it should be for the for the other companies. >> Yes. Yes. >> Okay. Okay, that makes a lot of sense. >> This is Corvid, which is the biggest now cloud today. And here is just how much money they are going to burn over the next few years. And you can see on the very chart at the top that um their EBITDA, how much money they are going to make is increasing, which is great. Like as as an investor, that's what you want to see. But then, if you deduct the CapEx thing, you know, because they need to like fi- finance all the GPUs that they are buying, then the story gets less interesting, I feel like. So, you can see here that this year they're going to they free cash flow is going to be negative um 28 billion. And like it's improving year over year, but they are not going to bring any cash home but until 2031. And at that point, like you can see here by 2030, they're going to burn over 100 um billion in cash. And so, this is very capital-intensive company, and market is not Well, they are down as well, like 40% since their highs, but market is not really punishing them for CapEx. They're punishing them for some other things, but you know, when they report their earnings, market doesn't really bother about those CapEx because they see that revenue is growing and they say like, "Okay, that's great. So, you know, let's just buy more Corveif, more Nebius." But my question is what is the difference here? Like why is market punishing Google or Tesla, but it's not punishing Corveif or Nebius? And like I need to be honest, I'm still holding Corveif and also Nebius. I'm just want to be aware what what I'm holding and what might happen at some point. And so, um I'm still building a evaluation model because I struggle to see what's the fair valuation for these new clouds and how they should be valued in in first place. And like later on, I'm going to show you um my mental model for how I'm thinking about the business economics that all these neural clouds are doing 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 Milk Road Pro are all over it. They spent the last couple of weeks making a lot of trades, getting out of some positions, and 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 Milk Road Pro at the link below. >> Got it. Okay. So, then just for people that are listening on audio only, we're basically looking at a chart of Coreweave's like CapEx, which is enormous, and despite their EBITDA being growing a lot in the next like three or three or four years, the EBITDA won't be bigger than the than the CapEx or their interest and taxes for 5 years, right? And that's also, you know, that's been in the story this year as well, Martin, and that's going to I think with the >> market the market has had to balance that. There were these massive CapEx numbers, which will naturally flow into some of these companies like Coreweave, but even a company like Coreweave is also spending that money, right? It's like they're not that's not just profit for them. They also have to take that money and go and spend it, and as a result, that's kind of you're painting the picture that it's like and that's been the story about this AI trade, too, right? That's like, well, how long until all of these companies are actually in profit, not just revenue, but actually in profit? And clearly it's it's a long time. >> Yes. I think market really likes these days. Like obviously, $1 today is is worth much more than $1 in 5 to 10 years, right? Because like quite honestly, nobody knows how the world is going to look like in 5 years. Like AI is making such a big impact on everything we do. I I believe [clears throat] like everyone can experience it. And so, I think the market is really sort of rewarding the companies that are making money today, that generate free cash flow every year, and that are, you know, maybe not necessarily more disciplined, but that they want to capture more dollars today because this that might will will give them opportunity in the future when they see some better opportunities in the market. Because as I said, these days I feel like everyone is trying to find some AI strategy, what they should do, how they should approach it. How is it, you know, a risk to their business, what are their customers going to do, the the the behavior of of the customers is changing as well. Like I'm not Googling anymore, I'm not going to my online store, I'm just, you know, asking my agent to do some stuff for me. So, I think a lot of companies are right now going through a difficult period, and they need to figure this out. So, I think I really like the companies that are, you know, stashing dollars these days and just getting ready or just making sure that it sits on their balance sheet, and if they don't know what they what they should do with their money, okay, just do buybacks or give me dividends. That's fine. But, you know, these companies, while I understand the the logic or the business model behind it, I think there are still a lot of risks in the future that market is not really you know, considering or like accounting for it. And that's what I'm I'm going to talk about next. Because I know LG that you had Winson here, I think it was last week, and you were talking about Snow Cloud and how much backlog they have, right? So, you see that Corvee, for example, has 100, I think, 99 billion in the backlog. And you were like, what, 99 billion?" Yes, that's a lot of money, but it's only the revenue part. It's not It's not profit. And that's that's a big difference. I actually wanted to ask Melvin as well, like, how do you think that we should value these companies? Um what's the fair value, you know? Um yes, it's down 40% from its all-time highs, but what if that price was also way too high? And maybe now it's fair price, or maybe we are still way overpriced because the risks that I'm going to describe are real, and everyone needs to sort of create an opinion on that because otherwise you should not hold any of these near clouds at all. So, let me actually bring this chart. >> 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 Cathie Wood's ARK Invest, and chosen by the New York Stock Exchange, VanEck, 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. >> That's right. And one of our our our pinned post on our YouTube channel right now is is the Neo Clouds episode with Melvin where he's giving kind of the bull case for this and also explaining what they do, but um it's good to hear a counter balance. Maybe we'll set up a pod with the two of you in the future and we can kind of have a bit more of a debate, especially especially today Martin as as as the market is is maybe uh leaning more towards your argument, at least at least today. >> Well, but I think like everything is is red today, so it's it's not just, you know, >> Not just Neo Clouds. Yeah. Yeah. >> Okay. So, let me actually walk you through this table. It's called the economics of 1 MW of AI compute. That's pretty much what all these neo clouds sell. Like, you can go to CoreWeave or Nebius or Crusoe or some other neo clouds and you can buy um computation from them. But, it's not that easy. I mean, they need to uh get some money to buy GPUs and networking and servers and everything. And the industry average is that 1 MW costs something around um 30 million. Okay? So, when you see those data centers with gigawatts, those are just insane numbers that, you know, they are spending like tens of even hundreds of billions of money into those data centers. Yeah. But, you cannot just, you know, go to the bank and say like, "Hey, I need 30 million for each megawatt, so give me all that money." You always need to put some money on the table as well. And so, how these um contracts are usually funded, it's that um they can get 75% loan, but they need to fund the rest with their own equity or their own cash. So, you know, they might have some profits from the past, so they can put those money in or they might need to um issue more stocks and they're going to um sell them and use those funds to fund these CapEx. So, that's the 30 million per megawatt. But, if we start they are selling those compute and the revenue, the computation revenue that they are charging, you know, this is an example of CoreWeave made a deal with OpenAI and so they are charging 10.5 million per 1 MW. But then you have like CoreWeave, if you have those GPUs, but you need to put them somewhere. So, for example, in this case, I used a a deal with Galaxy because, you know, CoreWeave is bringing all these GPUs and their chips and servers into data center that's provided by Galaxy, but they need to pay for that. So, that's the rent. And then you have operating costs, which are like 80% of that is electricity bills. It obviously costs something to, you know, run all these data centers. And so, usually No Clouds has to pay for for these electricity bills, maintenance, and you know, all that stuff that's sort of operating data center. So, that's that's the cost there. And then there's interest because you you took the loan to get the money to fund your GPUs. That's great, but again, it's uh based on the contract that they made, it's per 1 MW, it's about 1.6 million. And then you because you also took the loan, and so you need to repay it. And because these contracts are mostly about 5 to 6 years, you need to pay them pretty quickly. So, it those are not loans that will last, you know, like 20 or 30 years. >> Yeah. >> And the reason is, do you know why it's only like 5 to 6 years those contracts last LG? It's because the lifetime of those GPUs is 5 to 6 years. >> It's 5 6 years, so they need to be replaced after 5 or 6 years. And you'll need you'll need a new loan >> [snorts] >> to buy more. >> thing. We will get into it. >> This is a good So, who's who's giving giving the loans? Who are they giving it like like the bank Who's giving the loans? Like are they they're borrowing from the bank? >> Nice bullish banks, man. I guess I guess So, so so So, all this goes to crap, all the all the banks are going to just going to end up owning all these data centers and land as they as they repossess the assets. >> [laughter] >> And then they'll just liquidate them. So, you'll be able to buy you'll be able to buy 200 acres in Texas on the cheap if if if AI turns out to be a bubble. >> [snorts] >> Yeah. Yeah, as they have to liquidate. >> Yeah, I'm just >> [laughter] >> I'm just making a joke, man. This is actually really interesting, Mark, because I I point. No, but I've never I've never thought about this this kind of balance sheet for these loans, right? Because again, we're we're we've always on the show and especially with the roll up with the guys, it's like we're looking at CapEx. We're like, "Wow, 7 and 1/2 trillion of of CapEx over the next 5 years." Like all these huge numbers uh and we've looked at like, "Well, the hyperscalers are borrowing that or like they're going to use that and it's going to go down to all the bottom line companies and then they're going to use it." And kind of this whole revenue workflow, but we've never looked at like, "Well, what is that loan, you know, which is $30 a megawatt?" So 30 30 billion for 1 gigawatt, basically. What does that actually look like on the balance sheet? And it's not just money that you get to borrow forever with no interest or anything. And it's not just money that you just buy you're just buy your GPUs and your racks and then that's it. There's so much more to it, right? Like you're kind of showing right now. Um and that that that timeline for repaying that loan and to make money is is way shorter than maybe a lot of people realize, right? >> Yeah. But we will get into it. You I think we are like we're good to go on this point. >> Well, I just [snorts] I think it's interesting. Yeah, good. >> And then there are taxes, obviously. So let's say the year one will bring you 1.4 million, okay? In for 1 megawatt. That's nice, but you know, you also need to remember that you had to put 7.5 million as a as a equity as a as a as a company. So if we deducted the first year and we are still operating at last 6 million per 1 megawatt. I'll make it short and after 5 years, you will finally turn profits. Because you know, like pretty much every year, cash is around 1. 4 to 2 million. But then, if we consider the the equity value that needs to be also put into the place, after 5 years, you might finally hit the break even and you might be in cash profit 1.3 million. And so, because all these contracts are usually 5 to 6 years, so after 6 years, you will get 3.5 million. >> Mhm. >> Now, [clears throat] you might say, "3.5 million and we put in, you know, 7.5 million. That's, you know, 45%. That's pretty good. But, you need to consider that we are talking about 6 years, LG. So, if you annualize that, we are talking about something like 6 to 7% per year." And that's probably something that I as an in as investor, like, "If I am, you know, diluting shareholders to fund these operations, but it only brings me 6 to 7% per year, like, hell, I'm not going to do it. I don't want to own that business. Like, no way." >> Mhm. >> But, that's where this story gets interesting because right now, you paid off all your debts, there's no interest, and so, from here, you only have GPUs. You can still, you know, um keep them in the data center, but the biggest cost are gone. But, the question is, and that's the most important one, what's the value of your chips in 6 years? >> Right. >> Do you Do you think they are like You probably will not be able to charge 100% because imagine the chips they are selling today, and it's going to be, you know, in 6 years it's going to be 2032. >> Mhm. >> The [clears throat] price will needs to be lower and >> Mhm. >> it's it's [clears throat] called residual value and >> Mhm. >> I don't really know what that value is going to be. Here, in this model, I used assumption that I will be able to rent those GPUs at 60% of the initial price. Okay? For the next, let's say, 9 years because the the contract with Galaxy to rent this data center is for 15 years. So, I modeled it for 15 years. >> Mhm. >> So, even if you agree that you are going to be able to charge 60% of the initial price, you will end up bringing in um 19 million per 1 megahash and if we annualize it, that's 23% and that's that's already much better, right? >> Mhm. >> S&P [clears throat] grows 10% and so, for a lot of people, that's that's the hurdle. And so, 23% that sounds reasonable. >> Mhm. >> But now, there is that one key assumption that everyone needs to understand, which is what if my GPUs are worth zero in 6 years? >> Yeah. >> I'm going to have just, you know, a bunch of chips that nobody really wants because they are not performant enough and I made, [snorts] you know, barely my principal on all that business. So, what is going on? >> 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. Bitget 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 .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/bitget to get started. >> I can't believe I can't believe you're saying all this on the day that that like on the week that the market is nuking. People are People are People are going to be in the comments on YouTube and be like, well, of course now you're bearish that the market's down. They're going to say like, where was this Where was this 2 weeks ago when Melvin was giving his pitch on the old clouds? Uh you know, this kind of reality. And you know what's funny? You know what's a good analogy for this, Martin, maybe is that it's like, you know, by the if you if you have to buy a new fancy iPhone and you you repay it, you know, you pay it into your your cell contract or you you, you know, you're borrowing or whatever to pay it off and you're paying like 20 bucks a month and it takes you 3 years to pay it off. After 3 years, there's a new iPhone that you need to buy anyways, right? That it's like by the time you have your iPhone paid off, you have to get the new model because you have the old one. You have your still You still have iPhone 13 and 17 has come out or something, right? And this is kind of what that feels like. That it's like, well, by the time they actually are cash flow positive and they're actually making money on these, the chips will be obsolete, so they'll have to start over again. >> Yes. But actually, I'm using old iPhones and I'm fine with it. So, your analogy [laughter] doesn't apply to me, but yeah, you got a point, yeah. >> That's it. But for people who want new iPhones, it's annoying uh to have to do that every time and there's new features and all that. So, okay, so this is interesting. So, so why Wait, why are you bullish then? If these if all these neo cloud companies are just going to have to restart every couple years and by the time they pay off their things that there's there's they have to get new chips. Like, what What What makes you bullish? >> Actually, I'm getting more and more bearish and the reason is that >> [laughter] >> Now >> Yeah, go on. >> more GPUs suppliers than like ever before and everyone is trying to penetrate that market, right? I'm not saying that like anyone is going to replace Nvidia or something like that. But the more supply is coming in, I think it will just create more pressure even for Nvidia and all these chip makers. And so, I think the the the difference in performance from the chips that will be, you know, new in 2032 versus the performance of chips that are coming out today is going to be like like very different. And so, I'm really getting worried that even that 60% here will not last and stand. And so, if [clears throat] you put something like 30%, then this whole trade is over. And it doesn't make much sense because then the the the return for investors is, you know, below 10% and it's like, "Okay, this is not what I was hoping for. It doesn't make sense." And in that case, you should probably park your money somewhere else and not in in CoreWeave. >> So, is this is this is this something that you think the market has realized yet? >> I think not yet. I think again, I think the like all the investors in CoreWeave and Nvidia, I don't think that they are like, you know, doing this math. Or it they just see the backlog, you know, you see 100 billions. So, you are saying to yourself, "Wow, this company like you are bullish on the right thing." I am bullish Compute, too. So, like don't get me wrong. I think this like the demand is just going to accelerate. So, like I'm very confident about that. But, you need to go and look at the companies that you invested and try to understand what's their business best about and what are the risks to the business. And so, I hope I laid this out clearly that I think if you don't have an opinion about what the GPUs are are to be worth after the first initial contract expire. Do they going to renew your contracts? And if so, at what price is? Then um you know, that's that's the bear case. >> Now that we know you're bearish on these, you still have them in your Milkroad Pro portfolio. So, what what is your plan? And again, this is you know, and maybe well and you answer that Martin, you can also walk us through um kind of your psychology on days like today, right? Because it happens It just so happens that we're doing kind of a Neo Cloud bearish episode on a week where like the market is very unhappy. Uh so, even even tell us like what is your plan for Nebius and Corvus which you own in your Milkroad Pro portfolio. And also, what is your you know, what how are you managing mentally on a day like today where things are not not great? >> Yeah, I'll start at high level. I think this is like more market sentiment thing. It's not anything particular to Corvus or Nebius. So, like you know, you need to differentiate whether the market is puking or if just your your companies are not doing well. So, that's that's the one part. The second part is that yes, I still hold both, but right now I'm working on some valuation models and I am chatting with couple other analysts about like what's the right approach, how we should value these companies. Because if you look at Nvidia or Google, their business valuations are like you know, pretty standardized and that's pretty easy. But for companies like these who are, you know, not even profitable, you you can see the story why at some point they should have some serious value, but to get there, it's not a free way. It's not a free ride. So, I'm trying to, you know um consult couple analysts to find the right approach here, how we should value these companies, because I want to know if I am owning Corvive and Nebius, but you know, I am actually uh owning them at very expensive valuations. That's not my style. I want to make sure that all my companies that I hold have limited downside, but very very big upside. And here, I mean, the upside imagine that they're going to be able to rent those GPUs after the first initial contract is over at the same prices. So, that, you know, that's the bull story. But if what if not? And my problem is that I don't have a strong opinion about what's the value, what's the residual value of these GPUs is going to be in 2032. So, I want to create a scenario where I can play with the numbers and see once I figure out which valuation method I want to use for companies like that, is it is it right now priced fairly, or is it still too expensive, or is it actually cheap? So, that's what I'm trying to do next, and yeah, I I I hope that by um by tomorrow or maybe uh Thursday, I'll share my model with, you know, pro community to >> Oh, great. >> And then, based on that output, I'm going to probably do some action. >> Great. Okay. Great. Well, I guess that's that's a good incentive for anybody who hasn't gone in pro yet. Like if if you want to see what that is, and and Martin is promising it this week. So, um you go in in there and check it out. You can pay just a dollar to check to check out pro for 7 days. I think the most important part is that if you want to go in at the end of the week, pay a dollar, just see Martin's model, and then you think it's stupid, and you leave. You can. >> [laughter] >> But I think people like it. You always have really great models. Uh Portfolio S is is a really great tool, it's something you're kind of building right now as well, um that you use to to model this stuff. And and I always appreciate uh both in your reports and in in these podcasts that we started doing together now, like the the the way you approach these. And I think also, Martin, today it's it's despite it coming on a red day, it is refreshing to to hear the bear case. I think it's very important to consider that because I think for months it's it's been up only, you know, when we hit that euphoria last month that it was like, well, you know, everything all these companies will be trillion-dollar companies soon after Micron did it, and that's not necessarily true, right? Or it or it won't happen immediately as well. Well, they'll have to figure out uh other ways to get kind of get there. >> I agree. Very good point, LG. I think a lot of people are just, okay, this is AI trade, and everyone's talking about it. I just want, you know, I want on that board as well. But then, are you jumping in? Is the valuation okay? What are the assumptions for it the business to really work out? Like, >> [clears throat] >> those are the things that people usually don't do. I mean, it's very easy to just jump on the hottest thing that you see on, you know, Twitter or YouTube or elsewhere, but like, yeah, I would I would recommend everyone to do the homework, and before you buy something, you you do this due diligence, and at least like prepare your head for what can goes wrong, what what if it goes right, but, you know, if you are not doing that exercise, and now you see, you know, um Nimbus is down couple percent, like, are you worried? And if so, why? And like, that's that's something that helps me to manage my portfolio. So, for things like here, I still don't have my model because again, uh we are just keep having those conversations about how we should value these companies, but for some other companies, I have my models in, and it just lets me, okay, I I have Galaxy. I shared my model many times on X and with my with our community as well. So, it's 9% down today. I don't really care. I'm fine. I'm adding more because I have that model, so it's just like, you know, I'm not panicking at all. I'm actually adding, so yeah, if there is one thing that I should recommend everyone, that's that's this one. >> I appreciate it. That's a great way to go into it. Thank you, Martin. Another great episode and good to see you and we'll be back next week. We're going to be doing these every every week with Martin, so make sure you tune in and subscribe. Thanks, [music] man. 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's 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.