저자: Kevin · 외국인
요약 보기
코어위브(CoreWeave) IPO는 자금 부족을 보전하기 위한 사실상의 구제금융 성격이며, 대규모 h100 데이터센터 투자는 과잉투자 우려가 있다. 자산 대비 수익률은 높지만 부채 의존도가 크고, 엔비디아 젠슨 황의 h100 가치 하락 발언과 알리바바 회장의 과잉 건설 우려를 근거로 신중론을 제시했다. IPO 직후 급등 가능성은 있으나 장기 리스크가 크다고 본다.
다듬은 글
fre IPO opportunities so they can get bailed out to the tune of $2 billion so they could pay their damn bills because they went so deep in debt the company cor weave may be IPO at the perfect time that means the stock will be available to trade very soon the only problem is will it be a good investment well a lot of that depends on what we think about buying out that core weave infrastructure and how much money it's going to return in the future cor weave is expected to trade under ticker C rwv this week and raise over $2.7 billion for the company which is really interesting because when you look at the finances for the company and you look at the history of the company you start understanding that this former crypto mining company called Atlantic crypto where they mined etherum for yield happen to strike literal gold during the AI reol solution think about it ethereum prices underperforming Bitcoin your gpus collecting dust because you're not making the yield you used to your break even costs compressing and then all of a sudden artificial intelligence comes around and what does that do for you oh let's just stop Mining ethereum and start doing Ai and Lease our gpus to other people who want to pay us to use our gpus and our data center data centers on a per hour basis this could be companies that exclusively use cor weave potentially startups or other larger companies that don't want to build out their own data centers or they could be companies like Microsoft that partner with cor for some of their artificial intelligence needs think about it this way let's say you have you're Microsoft and you have this really large machine at your home that you need a lot of power for I don't know let's say you need 100 units of power for that but then sometimes you need 150 units of power you don't want to install all of the Power needs to always Supply 150 so just do the 100 yourself and get the other 50 from cor weave that's an example of how cor weave could function in its business model so just because a company like Microsoft could use core weave for Access flow or potentially all of their work if they wanted to cor weave creates this flexibility for companies to basically use AI data setters when they need it and dial it back when they don't this is pretty straightforward cor weave is basically the infrastructure provider but they also provide the software around that data center infrastructure which is cool because it means you know these Executives were pretty smart in that when the AI boom hit they're like okay not only are we going to retool as an AI data center play and we're going to change the name of our company to cor weave but we're also going to develop our own in-house software to man manage these data centers which is frankly brilliant because I mean a it makes your lives easier but B it also gives you a USP a unique selling proposition to your customers where like oh yeah come use core weave for your AI infrastructure needs we got the software that you could just use because a it's easier for us but B it's also easier for you to manage you know what you're renting from us including all the pricing uh changes and and uh you know the turn off and uh turn on and on functions and the allocation functions of it anyway it's kind of brilliant so when you look at cor weave though what you really want to think of them as is a giant generator of AI services so remember that example I gave where like if Microsoft needs extra power think about cor weave is sort of the power plant except rather than providing power they provide AI data center access which obviously has been in massive demand over the 2023 and 2024 period because this has been sort of the AI boom cycle now there are some problems though with cor weave the first problem that I think is worth paying attention to is what industry experts are saying the chairman of Alibaba today wrote The Following quote I start to see the beginning of some kind of bubble I start to get worried when people are building data centers on spec there are a number of people coming up funds coming up to raise billions or Millions ions of dollars in capital to build AI data centers but what happens if they're out building these data centers and this Data Center capacity without an actual customer in mind and that's what Alibaba chairman is worried about he's worried that we might have actually overbuilt AI infrastructure with now a lot of h100 chips which are kind of the older generation right and we might not actually need that many data centers cor weav has got somewhere around 250 ,000 Nvidia chips Elon musk's got around 100,000 Nvidia chips at his Tennessee you know xai data center but the h100 chips are already the old version in fact Jensen Wong at the Nvidia GTC event literally said you won't be able to sell h100s to anyone and that well he kind of Bas he said it a little bit more bluntly than that he basically said you couldn't even give him away which is bad because it kind of suggests that the CEO of Nvidia is literally suggesting the value of h100 chips could be pretty dang low when their new chip sets come out now that's a downside now the upshot to this is hey you've got companies in China that are reporting like ant for example Jack Ma's an who co-founded Alibaba Jack ma uh and the Ant company they suggest hey you know we don't necessarily need the latest and greatest chips in fact we can do a lot of our work with older generation chips and then just use the more advanced chips for the final stages training llms or inference solutions that require heavier Duty chips uh this piece This research piece I actually broke it down uh let's see here I'll put that on screen so you can actually see it so it's right here so our experimental findings have demonstrated that 300b I mean okay you don't need to know can be effectively trained on Lower performance devices while achieving comparable performance to similarly dense models such as deep seeks version 2.5 blah blah blah just I'm going to translate this English here in the pre-training phase you can basically use cheaper chips which is potentially good although it does make some people question companies like the dominance of companies like Nvidia because you know hey you could just go to AMD or you could go to other chip creators such as chip manufacturers or designers in China that uh compete with AMD or Nvidia especially more likely after this trade war that uh is undergoing you know occurring under Trump uh and Trump has directly argued that there will be specific tariffs on chips now if that happens or not remains to be seen But the argument here is maybe there's still demand for older chips right so I'm kind of giving you a pro and a con here on one hand we might be overbuilding data centers on the other hand that's okay we could still make use of the older chips but just remember what you're getting with core weave are a lot of h100 chips and again the CEO of Nvidia says you couldn't even give away Hoppers uh now that's obviously a big claim so I should point it out to you in the actual video it's at 1 hour and 19 minutes in in the GTC presentation and just listen to Jensen saying it himself because cor weave has a lot of these and it's kind of important that you hear this because I kind of think they're iping well I'll give you my opinion in just a moment and what their balance sheet tells us let's just listen to this for a moment hey just a quick update house Haack has doubled its fund raise from the last update that we gave so if you're looking to diversify into real estate check out house act.com it's amazing we you know the last fundraising of dat we gave we've already doubled that and I think that's because people are looking from for some more diversification from stocks but anyway uh updates over at the house Haack homes YouTube channel or just go to house.com you can see the properties and more about that but it's diversification into real estate if you haven't heard of it before uh and uh there's a non- accredited round available for it 5% yield upside uh included because you know you get the stock if it goes up and there's also downside prot protection so check it out over at h.com you know I've said before somebody actually asked you know why would I say that but I said before that when Blackwell starts shipping in volume you couldn't give Hoppers away and this is what I mean and this makes sense if anybody if you're still looking to buy a hopper don't be afraid I'm it's okay but I'm the chief re Revenue Destroyer my sales guys are going oh no oh don't say that I like Jensen okay I I think he's actually a very sincere person and I like Nvidia a lot and don't get me wrong I I don't have bets on either of these plays like you know for NVIDIA or against cor weave I don't care I'm just trying to provide information on YouTube because I think when I provide value on YouTube you might consider subscribing to the channel and coming back for more value and that's my job I sit here and try to figure out how can I provide you value on a daily basis and I think when you you know when you're like me and you research this stuff I watched that Jensen interview and I didn't say to myself oh okay you know I'm going to I'm going to make a whole video on this I didn't even know I would ever reference it again it was only right now when I'm going through the core weave documents for their IPO their S1 I'm like oh wait a minute what what did Jensen say again that's right so now you kind of make these links and these are the kind of like aha moments that I tried to provide to you now I'm not saying that they don't have black Wells of course cor weave has black Wells but most of their infrastructure buildout occurred in 2023 and 2024 you could literally see it right here which most of that actually excuse me I would venture to say almost all of what you see on this balance sheet is Hopper infrastructure because Blackwell only started shipping at the end of 2024 we don't even know if that shows up in plant property and Equipment yet or if it's just you know showing up as restricted cash because the product hasn't actually shown up yet and haven't paid for it yet right but they have orders for it I don't know but the point is look at this explosion in the balance sheet and I want to show you a risk because look at this plant property and Equipment exploded from uh I mean nothing in like 2022 to 3.4 billion in 2023 and then 11.9 in 2024 these folks were doing whatever they could to in Europe in the United States anywhere they could sell you an AI data center at low latencies they were doing it they went all in and honestly they did it at the perfect time they wrote they caught the wave and they wrote it perfectly I'm jealous of these bastards they did so well with their timeing honestly good for them like they they played it perfectly wow we're sitting on all these gpus for ethereum let's get into AI data center Leasing and make some software around brilliant boom billionaires overnight okay I don't I think that is so smart but as an investor you you should be asking yourself what am I buying or what am I doing to bail this company out yeah don't get me started on the bailout that's coming in just a moment so this company 3.4x their plant property and Equipment assets from 3.4 billion to 11.9 in one year they basically plow money into Data Centers right now this company has cash and cash equivalents of $1.7 billion but they have bills to pay of 3.6 they are $2 billion short they need a bailout they literally don't have the cash to survive the next 12 months unless they get bailed out by retail investors being promoted this stock through Robin Hood preo opportunities so they can get bailed out to the tune of two billion dollars so they could pay their damn bills because they went so deep in debt to build data centers cuz there was no limit to the amount of demand that people thought there would be for data centers the only problem here is what happens when all of a sudden that demand starts drying up and people aren't asking for your AI data centers that much anymore because either they don't need as much h100 compute power because it's the older version or because they just straight up don't need as much h100 computer power anymore because maybe it was overbuilt just like the Alibaba CEO says not my opinion I'm just putting the connections together see I'm taking from Nvidia I'm taking from Baba and I'm combining it with sort of experience and logic into their S1 and then giving you perspective now you could say hey ma'am look they're going to make plenty of money they're going to grow Revenue blah blah blah great we hope that they do but understand when you look at this balance sheet this company does not have enough money to get through the next 12 months and the only way they're going to survive is by either dumping stock onto retail share holders buying it at IPO and it could do really well or raising more debt to pay for their existing debt again I'm not here to say the Stock's going to go up or down or whatever I'm here to tell you what's going on the finances are such that they need money they need about 2 billion which is really interesting because if you look at the estimates for what they plan to raise they're about $2.7 billion it's right here so it's kind of interesting now this is going to be great so let's get past this okay so now let's say they IPO they raised the $2.7 billion if the stock crashes cor weave doesn't care because now they got their bills paid for the next year and they've got their big fat IPO paychecks and bonuses good for them no shade at that at all the question is is this a company that I want to buy okay well to understand that we have to understand a little bit about their revenue and how sort of their revenue works and what projections could reasonably be made off off of this so uh we have that on screen right here so most of this loss that you see right here is a loss of fair value related to bonds I actually don't think this loss is that bad and I actually think in 2024 they did a a good job making money uh they actually made money when we look at uh uh you know taking out depreciation and these fair value adjustments because most of this technology and infrastructure here about $800 million of that in 2024 is actually depreciation so they don't line that out which is surprising to me because if you do that you actually kind of get more bullish on these numbers uh they also have a lot of interest expense but then again they did whatever they could to beg borrow and plead to build up these data centers because that's that's the ride they were waving there's a lot of demand I mean anybody would have done the same thing so again I don't blame them I'm just saying I think they're kind of trying to IPO a peak here which is smart also I'm telling you these people are geniuses we don't have enough money we're at Peak AI deployment we're probably at Peak AI use in 2024 at least for Hopper chips which is the infrastructure that they have anyway and things are getting uncertain in the economy quick IPO before it's too late and tariffs crush us into a recession and oh my gosh AI demand slows down and people realize we've overbuilt data centers you know and obviously then the stock drops off a cliff but that doesn't have to happen right away you know the stock could be up 50% on IPO day because usually what they'll do is they'll pre- price it at like $50 and then it'll trade for like $75 or $80 and it'll be like oh my gosh it's up 50% on IP day they do that by just limiting how many shares are available because they want it to seem like it's up a lot on IPO day so more people get foam wo and buy it yeah it's just like this is Wall Street Wall Street is literally let me put it this way somebody at the Security and Exchange Commission once told me this line you ready for this Securities are sold not not bought think about what that means for a moment Securities are sold not bought well what it means is you have to sell you got to sell the sizzle that's what an IPO is why do you think they put a banner on Times Square or the NASDAQ is or the party that they do over in uh you know on on on Wall Street if you're iping at the New York Stock Exchange why do you think they parade these because it's part of the marketing which is normal none of this is abnormal what matters more is that we understand this Revenue section right here of course the revenue looks like it exploded but you have to also consider that they brought the difference between 11.9 and what 3.9 worth of assets online so 11.9 minus 3.9 they brought about $8 billion of servers online imagine you bought $8 billion of rental real estate do you think your rental re Revenue your Top Line would go up well of course it would now the question is did it go up by a greater multitude yes or multiple rather yeah it actually did see if we go back to the balance sheet they oh they had about 3.4 rather than 3.9 so that means they added about 8.5 billion uh they 3.4x their plant property and Equipment but their revenues actually exceeded that their revenues popped up to the tune of 1950 by 228 uh oops 19 15 ided 228 their revenues actually popped up by about eight times so that's pretty good an 8X increase uh in their revenue in exchange for uh what did we say here 11.9 divid by 3.4 about a 3.5x in an infrastructure buildout or you know an increase in infrastructure buildout not bad it is possible though and this is what some folks are speculating that a lot of this is because this infrastructure really didn't start producing Revenue until in 2024 so in other words there was some delay in getting the 2023 revenues online from the plant property equipment making that growth seem a little bit more extreme that's a little hard to say but so far if they're bringing in $1.9 billion and their costs of Revenue are around 493 million less depreciation and not a lot of sales I mean these people sell themselves which is honestly really good like this is amazing very very very little sales effort here they're actually doing pretty dang well because again add back in depreciation to their operating income they're bringing in somewhere around a billion bucks which if you look at that that's about a 99.4% operating return just on their plant property and Equipment their servers now the question is do we want to invest in this company that has a 99.4% return on their assets you know before interest right we got to consider interest as well so if you take off the interest of $360 million that return is going to go down a little bit but just ignoring interest for a moment uh we're going to look at operating income 1.9 million that's that number right there look at operating expenses minus 1.59 equals we got 31 right now we're going to minus out interest uh minus.
36 uh and then what I'm going to do is I'm going to add back in about $800 million of depreciation you could get that number in the actual uh financials so I add that back in uh and we're doing pretty well we're making somewhere around $800 billion versus about 1 to 1.1 is without that interest 8 billion or sorry $800 million off by zero here $800 million divided by about 11.9 in assets uh is a return of about 6.7% net of Interest 6.7% return on their server infrastructure pretty damn good these are rough numbers here right but the point is if we're at about a 6.7% return on infrastructure after interest or 99.4% excluding interest because they could always refinance that debt in the future if rates come down uh then the question is is that return going to stay steady or can we grow it and that's where the big question marks are what is the growth going forward now there are a lot of different thoughts on this my B thing that I'd like to ask is and this is what I would want to know as an investor is do we see pricing decays one of the problems that I've seen so far is cor we actually changed their pricing model so you can't see it as well on the way back machine from a per GPU to a per instance basis so it kind of makes it harder to find out hey you know what has their pricing looked like what kind of pricing power do they have now you do have a piece out from the information which provides a little bit of color which we'll look at in just a moment uh but that that makes me curious so I would love some insights on like hey how are we doing here which to me probably means we'll have to wait a few quarters and let some of that IPO hype go away so we can really start seeing what are they saying in their earnings calls in terms of pricing power to is is revenue act like what is revenue actually growing at that's what I'd like to see and how much are they continuing to invest in new infrastructure because right now they spent about 2.4 let me see here uh let's see ah yeah okay there we go so net cash provided by operating activities was about 2.7 in 2024 but their purchases of plant property equipment around 8.7 so the free cash flow negative to the tune of about five bill because they're throwing it all back into servers the question is what happens when that server you know expansion this right here slows down are we going to see a big slow down in Revenue growth obviously it's not going to look as meteoric as this Revenue growth that we've seen over here we already know that that's why I'm looking at sort of a return on assets because to me I'm if I'm investing in this company I'm buying h100 chips the old stuff and I want to see that continue to print money but that does create risks because people might be going into this thinking you know this is the next biggest and greatest AI play that's like an open AI or something is really just you're buying a company that's just going to lease out the chips I'm not saying that's a bad thing just something to keep in mind now the information has some interesting projections because they argue that JP Morgan and Goldman Sachs suggest that cor weave's revenue is going to explode to $4.6 billion this year which is great that's about a 2.4x increase now markets already know this so this could already be priced in but that's going to come by increasing their cash Burn by nearly 3x which probably means their infrastructure Investments are going to rise to close to $20 billion maybe1 18 but $20 billion divided by what they spent last year about 8.4 would be about a 2.4x in infastructure so in other words they're still plowing money into more uh uh you know data centers which is good because it's a promising sign of demand or we're overinvestigation ala says but this Revenue growth a lot of people might go oh my gosh they they grew from you know $15 million of ethereum Revenue to 228 to 1.9 now to 4.6 they're making so much money most of that is debt financed infrastructure again you go from you know uh uh $3 billion of rental properties to 11 billion to potentially next year $30 billion of data centers of course your revenue is going to go up but beyond that we don't have much guide so I'm really curious to see what the per unit Revenue increases are for this company this is expected to be a really big IPO and a lot of people are going to make a lot of money on this mostly the Insiders already and so there's going to be a lot of hype in marketing expect this to be one of the most hyped up IPOs of the year and I think they're doing it potentially still relatively close to the top of the market at a brilliant time for them because if we are at sort of the peak of how much they could earn renting out these data centers and we're at the peak valuation for an h100 data center that Nvidia CEO J suggest could end up being you know worthless then they're playing their cards really well here congratulations to the executive staff for being really really smart if thean company nah man AI data centers take on as much debt as possible and before you crushed that Peak IPO would dump it all on retail there you have it thanks so much for watching and see you in the next one subscribe for more
Comments 0