저자: Kevin · 외국인
요약 보기
Cerebras(CBRS) IPO를 분석. 웨이퍼스케일 칩 기술은 Nvidia 대비 차별점이 있으나, 상장 직전 순이익은 일회성 평가이익(3.9억달러) 때문이며 실질 영업손실은 확대 중. 고객의 86%가 UAE 관련 특수관계자(G42, MBZUAI)와 OpenAI에 집중되어 있고, OpenAI 워런트·순환 계약 구조가 주가 부양 목적의 펌프앤덤프 가능성을 제기. 상장 초기 상승 후 실적 발표(8월 전후) 이후 변동성 확대 전망, 스윙 트레이딩 기회는 있다고 봄.
다듬은 글
Wow, we need to talk about this Cerebras IPO. So, this goes public on May 14th. It's going to go public on the Nasdaq Technologies Exchange. It is going public under ticker symbol CBRS, Charlie Bravo Romeo Sierra. It is said to be a pretty dang big IPO. It is 20x over subscribed as of May 10th, and the IPO is based on over 1-year long delay due to Citius reviews of the massive United Arab Emirates investors inside of this company and potentially this company being forced to divest some of its Chinese portfolio.
Now, we're going to break down some of the risks associated with this company, but let's first understand what the heck even is Cerebras. What is this company? Yet another tech IPO? Should you buy it? What's going to happen in the price action? Is this a good deal? What's the valuation like and everything?
Let's get into all of it. And the way we're going to do it is we're going to go straight into their S-1. So, the first thing that they do is this company promotes that they're basically the fastest artificial intelligence data center. And they do what they call a manufacturing process that helps them create the wafer scale engine.
The wafer scale engine is basically their new mouse trap. Everybody's trying to invent the new mouse trap. Google's got the tensor processing unit. Grok had the SRAM base processors, which Nvidia basically cord out. Nvidia is really big on GPUs. Other companies like Marvell are really big on ASICs, application-specific integrated circuits.
Companies like AMD are really big on the processors, so is Intel. Nvidia is now getting into processors, right? CPUs. I When I I should clarify processor there so we don't get confused. Uh well, the wafer scale engine is Cerebras's claim to fame. It is their reinvention of the mousetrap. And they want to sell as many of their chips as a chip designer manufactured by TSMC.
And they essentially want to be an AMD or an Nvidia play. Which, in fairness, they're going into a really rich market. I mean, understand this. Marvell Technology on ASICs, which we invest in, big fans, $156 billion market cap. AMD diesel, which we've talked about booming since early April, has hit a $731 billion market cap.
And look at that, skyrocketed early April. Valuation was ridiculously low over here. Boomed. Nvidia is breaking through its 227 line. Well, it's sitting right there right now. And in my opinion, that gets it up to 300. This is a $5.5 trillion company. Uh $320 is our price target for Nvidia. You can see all of our price targets over at mekevin.com when you join the membership.
We analyze stocks and fundamentals, technicals, and our alpha report every single day uh that the market is open. But, Cerebras is expected to go public around a $48 billion market cap, which basically puts it at less than 1/100 the size of Nvidia, less than 1/3 the size of Marvell, and less than 1/10 the size of AMD.
Probably closer to like 1/12 or 1/13. So, what do they do that's so special? Well, in order to help explain this, uh what I did is I brought uh some uh samples here, okay? And so, what we have here are four Cheez-Its. Now, I know this is going to seem ridiculous, but it's actually really important. When these Cheez-Its get made, they come together on sort of one giant sheet, and then they get cut up.
Let's just make this relatable for a moment. Let's say we're Nvidia, and we're trying to make 4090 or 5090 uh gaming chips, okay? We'll go with 5090s. So, we're going to go have one big wafer, and we're going to go put our little microchips and our processors and everything into all these, and we're going to slice this giant wafer into smaller components over here.
Chips. See, chips of a wafer. Get it? Chips, because now we have four pieces. Now, what we're going to do is we're going to look at each of these components and go, "Okay, this one has really high quality. We're going to call that a 4090 chip." Uh this one right here, really good quality, 4090. This one right here, slightly burned on the edge over here, you know, we're going to call that a 4080.
Oh, this one's uh a 5080, whatever. You get where I'm going with this. This has a whole chunk missing over here. We're going to throw that one away and redo it. So, in other words, the 4080 really just has some defects in how it was manufactured. The 4070 really just has some defects in how it was manufactured.
The 4090 is considered more perfect. Same for 5090 70, right? And then if it's too imperfect, they just throw it away and recycle it. Okay. That is the Nvidia strategy for uh making chips. And this is the traditional way that you would make a chip. Totally normal. Now, the Cerebras strategy is very different.
The Cerebras strategy says, "You know what? We're just not going to chop up the chip at all. In fact, we're going to leave it all together as this giant wafer. And inside of this wafer we'll put all of the components we need. We'll put our memory or whatever. And the benefit of this is we aren't going to have to wire all of this stuff together.
See, with all of these little chips, you end up putting this into a server rack, all these little guys, you kind of have to tie these all together. Nvidia uses InfiniBand, which is a form of Ethernet communication technology. Marvell sells fiber optic communication tech, which is great for like data center to data center communications or even, you know, building out cell towers or whatever.
Cerebras' theme is why cut the chip at all? Well, people counter this and go, "Well, well, what if there's a defect?" And they say, "Oh, we just use software and we have enough redundancy inside of this that if there's a little bit of a defect, who cares? We'll just use our software to not use that portion of chat the chip or just use redundancy elsewhere on the chip to where that doesn't matter.
That defect is essentially insignificant and it doesn't matter. That's the Cerebras strategy. So, when you see comparisons between the size of a of an Nvidia chip and a Cerebras essentially wafer scale engine, you're going to be blown away because, you know, if you look at a B200, a Blackwell 200 versus a Cerebras chip, the WSE, the the wafer scale engine, is 29 times the size.
And the chip has 44 GB of SRAM on the actual chip. It uses the 5 nm production, which is basically a sibling of the 4 nm. It's not like a new generation. That's not that big of a deal, not that big of a shift. But the Nvidia chip has about 126 MB of SRAM on the chip and then another 172 GB of usually high-bandwidth memory off of the chip.
All of that crap has to be connected via cabling, which is expensive and introduces latency. This is why Cerebras says, "We are the faster mouse trap." We are the faster mouse trap because you don't need all this cabling. It's all on the same damn chip. And we use SRAM, which, remember, Nvidia bought Groq, which gives a lot of credence to SRAM.
Okay. Now, a Cerebras chip is also about 57 times the size of an H100, H100 being the older compared to like a B200, right? So, this is actually really interesting. And a lot of people say that Cerebras is the only architecture that's able to beat Nvidia on a regular basis. Cool. That's great and all.
But, what do we know about the details of this actual company? Like, who actually provides revenues to this company? And does this new wafer strategy actually work? That is the big question. Let's take another sip of coffee, and then we'll get right into the tab that I bookmarked called voting power.
If you haven't yet, make sure to subscribe to the channel. Appreciate you being here. Ooh. Ah. Why would somebody not edit out them sipping coffee in the middle of a video? I don't know. Okay. So, let's go to voting power. So, we jump over to voting power over here. And if you look at this particular segment right here, what you're going to find is there are three classes of stock.
There's class A stock, class B stock, and then I think it's class N stock uh somewhere around here. But, there's there's one other form of stock that's basically convertible uh into Yeah, there it is. Class N stock is um convertible into class A stock. So, class B stock is really interesting because class B stock is basically where 99.2% of the voting power is going to be held because these shares have extra voting rights.
And the vast majority of that is going to be held by insiders. After the IPO, they expect that officers, directors, and current stockholders that already hold more than 5% of the outstanding capital stock along with their affiliates will control 50.9% of the voting power of the company. This is not a surprise.
Mark Zuckerberg has voting power over at Meta. A lot of people say founder-generated companies perform better when the founders have control of the company. This is one of the reasons why Elon is so frustrated. He's down to like 15% control stake over at Tesla, and why I think he wants to get it absorbed over into SpaceX where he's got like 70% control.
So, that's voting power. So, it's very consolidated, but that's not actually the biggest weird thing. See, if you look over here at the hardware section, you can see their hardware sales are up 69%. Their cloud sales are up 39%. And their total growth averaged here is up 75%, which is all really fantastic.
Their costs are only up 62%, which implies that the company does have pricing power, which is great. Now, they do have $390 million you know, personally, I think it's a little bit ridiculous because it's like, okay. I mean, if it weren't for that $390 million, you guys wouldn't have a net income of $237 million.
So, it sort of suggests that some of your other income is boosting your your actual net, and you're using this other income as a way to quickly jump in and go public with a positive EPS, but then what will actually likely end up happening is they'll end up showing some form of gap loss in the future and people go, "Wait, I thought you had profitable EPS." Oh, well, that was because of all the other income.
Oopsies. So, that's not great. Now, they actually do break down their other income and expenses. Well, you can see it here. Other income and expenses, they should provide a note here on it. Other income and expenses comprise of the following. No, they don't actually make it very simple here, which is Okay, here we go.
Change in fair value was recognized in other income. So, a lot of this had to do with the change of warrant liabilities. Uh, which is super complicated and basically gives them a boost in income from a change in Yeah, there it is. Remeasurement or extinguishment of forward contract liability and warrant liability.
So, basically because of the valuation of some warrants changed, they got this massive boost to their income. Now, I want you to see where that is because I find it personally problematic. If you go over here to revenue, you could see that they have a loss from operations that has been growing. Even though their pricing power is growing, they're spending so much more on research and development that their loss has actually grown.
See, 2025 on the left, 2024 on the right, you could see their losses are growing. But, the only reason they're showing a positive EPS right now is because they took this $390 million remeasurement gain, which is totally like one time, right? So, people are going to invest into the this business going, "Oh, it's profitable." No, it's not.
No, I'm not here to bag on it. I actually I'm really excited about their chip and what they're doing, but I'm looking at this going, "Wow, they're choosing a very convenient time to go public where they happen to have profitable EPS and they're going to imply to most people that they're profitable, but they're actually not." Whatever.
Now, if you look at their balance sheet, balance sheet is actually fantastic. I've got really nothing to say here. They have very, very little debt. They have bills of less than a quarter of a billion dollars. They got some leases, but they have over a $1.1 billion of cash and investments and they're going to raise even more money on this IPO.
So, you know, who really cares basically? Uh their their balance sheet is fine. On top of that, if you look over here, yes, they have a negative cash flow of $390 million and and yes, they raised that $1.1 billion from raising stock by selling stock. They're going to raise a bunch of money from this IPO.
So, I'm not worried about their balance sheet. I'm not worried about their cash flow. They're going to have plenty of capital to reinvest into the business. I love the phrase reinvest as you know. Plenty of money to reinvest. The biggest question here is how is the stock going to perform when markets wake up and realize that their EPS is going to go from positive, you know, a buck 38 a share diluted, which is right here, uh and this starts turning negative again.
It's also worth noting that one of the ways that this is going to turn negative has to do with how they're going to end up clarifying their OpenAI investment into Cerebras. See, OpenAI gave Cerebras a $1 billion working capital loan at 6%. OpenAI ended up getting 33.4 million penny warrants at basically no expense.
OpenAI is going to get filthy rich off of this IPO. And Open AI, as we know, desperately needs the money. So, I want you to think about this for a moment and start seeing how this puzzle goose, if you will, whatever you want to call it, is getting put together because first of all, we have this negative impact coming to EPS from that other expense.
Second of all, when these warrants vest in tranches between now and 2028, they will be recognized in what's known as a contra revenue manner, which basically means their GAAP revenue is going to decline by the value of those warrants. It's crazy. A lot of complicated crap going on here. And if you don't fully understand what's going on with that revaluing measurement or the contra revenue that they have to spend basically on warrants for Open AI, then just don't be surprised that this is very volatile over the next year or two.
Now, it'll probably pump at the open because, you know, this stock market is very momentum based. But I want you to understand this for a moment. So, let's take out the good old handy-dandy whiteboard. You know what? Let's hold it like this. So, we get the handy-dandy whiteboard. And what we find is that Open AI and then we'll we'll just do Cerebras over here, okay?
Open AI ended up providing a loan, a working capital loan of $1 billion. So, $1 billion loan to Cerebras. Cerebras ends up giving what we say this was a crap load of warrants. These are 33.4 million penny warrants of class N shares, which are convertible to common stock. So, just to understand what that would look like, if this ends up being a $175 stock, 30 3 million of those would work out to three zeros, three zeros, about $5.8 billion, okay?
So, $5.8 billion worth of stock options, because a warrant is just a stock option with the company itself. And a stock option is usually a contract not with the company itself. Of course, that gets complicated by stocks. Anyway, stock options like compensation, it's ridiculous. Just don't worry about that.
And forget I said all that. $5.8 billion in stock of Cerebras stock going to OpenAI. But OpenAI wasn't able to go public because they had too much of an investment from the UAE. And so, they ended up having to clean that up. One of the ways they ended up cleaning that up was by getting a partnership with OpenAI, which is really interesting because now $20 billion of the remaining performance obligations for Cerebras come from OpenAI.
Now, that's really interesting. So, we've got a billion dollar working capital loan going to Cerebras. We got $5.8 billion in stock options basically going to OpenAI. And OpenAI to help pump this IPO is going to say, "You know what? We need $20 billion of compute, and we're going to go ahead and give you guys that contract." Of course, there's something really interesting to remember about all of this is Mr.
Brockman, who's getting sued by Elon Musk. He's got, uh, you know, he's he's the president of the board over at OpenAI. Okay? Mr. Brockman has a massive massive position in Cerebras. Ah, interesting. So, the very president of OpenAI who's getting these options and is creating this circular computing contract to essentially help pump the Cerebras stock even though their revenue is likely going to go negative to account for those OpenAI warrants and the fact that their other income is going to disappear over time, this to me quietly looks like a little bit of a pump and dump.
Let's pump up Cerebras. Let's make OpenAI rich. If this goes up to $300 a share, this could be $10 billion of free money, basically, that's going to OpenAI. Free because they have plenty of tax write-offs. So, it's free money that goes to OpenAI. So, Greg Brockman gets even richer at OpenAI. And OpenAI gets richer because Greg Brockman also has personal exposure to the stock, right?
So, OpenAI gets richer and he gets richer. And it's all predicated heavily on revenue concentration from OpenAI and the UAE, which is really interesting because Greg Brockman is responsible for helping establish some of that revenue. Yeah. This is an interesting one, boys and girls, but it doesn't actually end there.
See, one of the reasons Sifyus was all over this company was because originally the investors, so not this $20 billion OpenAI deal yet. The $20 billion OpenAI deal is really just starting, okay? That started April of 2026. They also have a letter of intent with Amazon, but a letter of intent doesn't mean much, okay?
It's an idea. It's a way to pitch an IPO. But look at this. Related party income from 2025. In 2025, customer A represented 62% of this company's income, and customer B represented 24%. That means 86% of this company's income is coming from related parties. And if you take out that other income we saw, they're not profitable.
They lose money hand over fist. 86% of this, by the way, is coming from the UAE. Mohamed bin Zayed University of AI, 62%, and G42, 24%. G42 is a major backer, but also a major customer. So the very investors that stand to profit off of this company, like the UAE and these UAE entities, and Greg Brockman, and OpenAI, also happen to be the biggest customers of the company.
So in other words, that was a very long way of trying to say, you see these? Oh, whoop. You see these? These are Nvidia GPUs. They generate like $50 a quarter of revenue. They sell a lot of chips. These we don't actually know if it's commercially viable yet, because the only people right now really buying it are people who stand to win off the pump and dump.
And since we expect EPS to go negative, I wouldn't be surprised if this is actually a little appetizer of what we might see with SpaceX. You get a delicious pump, you get a dump, and then you got to evaluate, is their product actually commercially viable? The first public earnings will probably be in August and maybe we could actually see some real revenues coming in from Amazon or some other entities.
But right now there are a whole lot of red flags for me. Now, is there an opportunity to swing trade it? Hell freaking yeah, because that pump is probably going to be glorious. No guarantees. Anyway, if you want to see exactly what my strategies are, come join us over at meetkevin.com get analysis like this on a regular basis and folks, .
Comments 0