Loading summary
Transcript1 lines
- [00:00]
A
Welcome back to the rundown for another weekend deep dive. Today, we are talking about cerebras. This AI chip company just pulled off the biggest tech IPO Wall street has seen since 2019. And up until recently, most people hadn't heard of it. So in today's episode, we're going to break down what Cerebras actually does, why Wall street is so obsessed with it, the drama behind how they got to this ipo, the financials and business models, and whether this stock can actually live up to the hype. Or is this a warning sign that the A is getting way too frothy? We got a great one for you today. Let's dive in. Now, before we get into the IPO madness, let's start with what Cerebras actually does. Cerebras makes AI chips, but their chips are different than the chips that Nvidia and AMD make. Cerebras is best known for building what they say is the largest computer chip ever made. Their chips are the size of a dinner plate, about 100 times bigger than a typical chip with over 4 trillion transistors packed onto a single piece of silicon. Now, without getting too technical here, the advantage that Cerebras has is that since all of the computing is happening on one massive chip, instead of being split across hundreds of smaller ones, data doesn't have to travel as far, which makes the whole thing significantly faster compared to the competition. And, you know, speed matters a lot in AI right now, especially as more usage has turned to inference. A couple of years ago, the main usage of GPUs was for training AI models. And the best way to train AI models was to have as many powerful AI chips working together. That's how Nvidia came to dominate the space and become the most valuable company in the world. But now inference demand has overtaken training. See, inference is when an AI model answers your questions or when people run AI agents. Having a powerful chip is important, but speed also matters because nobody wants to use a slow AI. So Cerebras has positioned itself as the king of inference. Their CEO claims that they are 15 times faster than the nearest competitor. And as AI models get used by more and more people, the demand for fast inference will continue to explode. And that's how Cerebras is putting themselves in a position to take on the chip giants like Nvidia and amd. And that's got the attention of investors. So let's talk about Cerebras IPO this past week and the path they took to get There because it was a bumpy road. Cerebrus had their IPO this past week and they currently hold a title for the biggest IPO of 2026. The company sold 30 million shares at $185 a share, raising $5.5 billion in the process. That made it the largest US tech IPO since Uber back in 2019. And it's also the largest semiconductor IPO in US history. So this IPO is a massive deal, especially since just two years ago, Cerebras had backed out of going public. See back In September of 2024, CE filed for an IPO. But that process got messy. At the time, almost all of Cerebras revenue was coming from one customer. It was an Abu Dhabi based AI company called G42. That company accounted for 87% of Cerebras revenue at the time. And then it got even more complicated because G42 was also an investor in Cerebras. So that obviously raised some eyebrows, not just from investors, but also regulators. See, when you're a US company making cutting edge AI chips and having deep financial ties to a company in the UAE that also had connections to China, is going to get the attention of the US Government. Since it was a foreign investor, the Committee on Foreign Investments in the United States or CFIUS launched a national Security review. So this review by the US Government basically froze the IPO process and cerebras had to shelve the whole thing. But to Cerebras credit, they didn't give up. They kind of restructured the company. So G42 stake was turned into non voting shares which cleared the CFIUS review. And then In January of 2026, Cerebras landed a deal with OpenAI worth $20 billion. This was a multi year agreement for cerebras to provide 750 megawatts of computing power to OpenAI. OpenAI also gave Cerebras a $1 billion loan to help build infrastructure. And then OpenAI got warrants to buy about 10% of Cerebras stock. It was kind of a unique structure for a deal. We'll get to some of the red flags, but a little bit later in the episode. Now, on top of the OpenAI deal, Cerebras also landed a partnership with Amazon Web Services in March where AWS will deploy Cerebrus chips in their data centers. So look, within 18 months this company went from being almost entirely dependent on one customer in Abu Dhabi to having OpenAI and Amazon in the mix. And that's what really made this IPO possible. Now, again, there's still some red flags surrounding customer concentration, which we'll get to in a bit. But I got to say, Cerebra has tied this IPO perfectly because we are currently in the middle of the of a historic semiconductor rally. The stock semiconductor index is up more than 60% this year, and that's led to a lot of demand for Cerebras's ipo. Cerebras actually had to increase their listing price multiple times because of all the demand. Bloomberg reported that the IPO was 25 times oversubscribed, meaning that there were 25 times as many willing buyers as shares available. The company officially started trading on Thursday, May 14th at $185 a share. And the stock popped 68% on the first day of trading. G Cerebras a near 100 billion dollar valuation. So that brings us to the big question today. Are the financials good enough to justify evaluation this big? So let's dig into the numbers and the interesting business model behind this AI chip company. Okay, so we know what Cerebras does now. We know about their big ipo. Let's talk about how Cerebras actually makes money and whether the numbers back up the hype. Cerebris makes money in two ways. The first is hardware. This is the traditional part of their business where they sell their CS3 systems, which are powered by their dinner plate sized chips, to companies all over the world. Their customers include everyone from Amazon to research institutions to governments and large enterprises that need computing power. And the hardware business is still the main source of revenue. In 2025, hardware sales brought in about $358 million, which is roughly 70% of their total revenue. But you know, it's their second business that has Wall street more excited. This is their cloud and services segment. So instead their physical AI chips. Cerebrus just puts these chips into their own data centers and lets customers pay to use the computing power through the cloud. This is the business model that the OpenAI deal is built around because, you know, OpenAI didn't want to buy Cerebras physical chips and have to manage them and put them in their own data centers. They want Cerebras to build the data centers and then just deliver the computing power as a service. Now, in 2025, the cloud segment brought in about $152 million. That's about 30% of total revenue. But it did grow 94% year over year. Compare that to the 69% growth for hardware so it's a faster growing segment. And this is where the company thinks the growth will come from down the line, especially as the OpenAI deal ramps up and the US partnership kicks in, cloud revenue should become a much bigger piece of the pie. And that's what has investors excited, because Wall street loves reoccurring cloud revenue. It's stickier, it's more predictable, and and typically comes with higher margins over time. So Cerebras is essentially in the middle of transforming from a hardware company into a cloud infrastructure company. Now, when you zoom out and look at the overall finances of this company, I mean, the growth over the last few years has been Incredible. Cerebras did $24 million in revenue in 2022. That jumped to $79 million in 2023, then $290 million in 2024, and then $510 million in 2025. So in other words, they 20 just three years, which is crazy. And the 2025 growth rate was 76% year over year, which is pretty solid. But here's where things get a little bit complicated. On paper, Cerebras reported a profit in 2025. The company reported a GAAP net income of $238 million last year, which would be a massive swing from the $482 million loss they had in 2024. So that looks like an amazing turnaround on the surface. But then when you start digging into the numbers, you realize that the profit was mostly driven by a non cash accounting adjustment thing related to the restructuring of the G42 investment. So when you strip out all that accounting nonsense, Cerebras actually had a net loss of about $76 million. And what I like to look at is the operating income or loss in this case, because it tells you how much money the core business makes from day to day operations. Cerebris reported an operating loss of $146 million in 2025. Now, a big reason for that was because the company is spending heavily on R and D right now. Also, Cerebras gross margins are around 40%, which you know, is decent for a hardware company, but it's well below Nvidia's margins, which are closer to 75%. So, yeah, big picture here is that Cerebras is not yet profitable and they're still burning cash to build out the technology and infrastructure. So now let's really dig into some of the other business risk and serious questions facing the company. Now that we've gotten to know about what Cerebras does and how they make money. Let's talk about a few of the big risks that investors need to know about. I think the biggest one by far is customer concentration. It's funny. Cerebras had to delay their 2024 IPO because of this concern. It's still a big problem. 86% of Cerebras's revenue last year came from just two customers. And both of them were based in the United Arab Emirates. Now, I've already mentioned G42. They're still one of Cerebras biggest customers, accounting for 24% of their revenues last year. The other big customer is a research institute in Abu Dhabi called the Mohammed Bin Zayed University of artificial intelligence. That AI research institute accounted for 62% of Cerebras revenues. So when Cerebras says they've diversified away from G42, I guess that's technically true. But what's really happened is that revenue shifted from one Abu Dhabi entity to another Abu Dhabi entity. Now, I'm not saying anything fishy is going on, but it's just worth paying attention to. You know, it's pretty concerning that Cerebras is relying on just one country for a majority of their revenue. That carries a ton of geopolitical risk because the US Government could potentially tighten export restrictions on AI chips. Again, you know, I think the risk of that are low at this point, but it's still a risk. Now look, Cerebras did sign Those deals with OpenAI and AWS earlier this year. So that should diversify the revenue. Now, according to the latest filings, cerebras had about $24.6 billion in contracted backlog. Most of that is coming from OpenAI. But what I was shocked to learn is that only 15% of that backlog will show up as recognized revenue over the next two years. In fact, the AWS deal won't go into full production until next year. So the revenue diversification that investors are betting on is still mostly a promise and not a reality yet. Now, the other risk here is circular financing, especially with the OpenAI deal. Now remember, as I mentioned earlier, OpenAI gave Cerebras a $1 billion loan as part of their agreement. Cerebras is using that money to build data centers packed with their own chips. And then OpenAI is going to pay Cerebr rent compute from those data centers. And remember, OpenAI holds warrants to buy up to 10% of Cerebras. And those warrants become more valuable as Cerebras Stock goes up partly because of the revenue that OpenAI is providing. So yeah, it's another one of those complicated circular financing webs that have become pretty common in the AI space these days. The risk here for investors is that what happens if OpenAI stumbles. You know, we talked a couple weeks ago on this show how OpenAI has been missing internal growth targets and that their CFO is worried about paying for all their computing contracts. So if OpenAI ever has to scale back spending, Cerebras could lose its most important growth engine and their stock could tank because OpenAI's revenue was the whole reason for the valuation in the first place. So that's definitely a concern that investors have to watch out for. And I think the third risk is competition. You know, I mentioned earlier how Cerebras chips are different than Nvidia because they're faster and designed for inference. But Nvidia isn't sitting still here. They're also designing chips to that specialize in inference. In fact, late last year, Nvidia paid $20 billion to acquire assets from a startup called Grok, not Elon Musk's Gro Chatbot. This is a different Grok. And these Grog chips actually resemble what Cerebras does. They're designed for the same fast inference workloads that Cerebras is targeting. And in fact, Nvidia already announced plans for GROK based products. And then, not to mention Google is building their own inference chip and then Amazon has their Trainium processors as well. So Cerebras is going up against multiple multitrillion dol that have massive pocketbooks. That's a pretty tough spot to be in for a company that only has around 800 employees. And finally, I want to cap this off by talking about the valuation at the time of this recording, Cerebras is sitting at around $280 a share. That would put the company's price to sales ratio at 130x. So what that means is that you would need to multiply the total revenue the company made in the last 12 months by 130 to get the company's current market cap. For some context here, Nvidia trades at about 25 times SAL. If you assume that Cerebras grows revenues by 150% this year, which would be double their growth rate from 2025, the stock would still be trading at about 54 times forward sales. That would still make it the most expensive semiconductor stock right now. So as one analyst puts it, the market is already pricing in years of flawless growth for this company. And you got to keep in mind the chip business has historically been cyclical, so sustained growth over multiple years almost never happens. So does that mean that this stock is doomed? Well, not necessarily. Especially in the short term. So what's my take here? Well, look, I'm not going to act like I'm a chip expert and fully understand the nuances and advantages of having a chip the size of a dinner plate. All right? I do think the shift from training to inference is a big tailwind for Cerebras. I think at some point people are going to expect their AI tools to work faster and faster. So Cerebrus is in a good spot there. Also, having partnerships with well known names like OpenAI and Amazon is also great. But personally, as an investor, it's hard for me to get excited and jump in at these valuations. There's just too many hurdles right now. You know, the company trades at 130 times sales, they're not yet profitable. They get 86% of their revenue from two customers in a foreign country. That's just too high of a bar for me. Not to mention, I also read this great piece from tech analyst Ben Thompson that made the case that the biggest AI market in the future won't be humans won't waiting for fast answers from AI. Instead, it'll be AI agents running tasks on their own in the background. And for those workloads, speed doesn't matter nearly as much. You know, an AI agent processing a thousand tasks overnight doesn't care if it takes 2 seconds or 10 seconds per task. So the speed advantage that Cerebras has built up right now might not be as important down the line as the whole industry moves towards agentic AI. Now, all that being said, we are in the middle of a massive semiconductor bull run right now. And that hype alone can keep the stock price moving higher for a while. And look, maybe it's possible that Cerebras proves the haters wrong and puts up the big growth numbers to justify the valuation. A lot of people thought that core weave was overvalued when they went public last March at a $23 billion valuation. And the stock has gone up 160% since the IPO. So we'll have to see if Cerebras can do the same or if the market gravity will eventually pull it back down. Well, all right, guys, that's it for today's weekend deep dive. Hope you guys enjoyed that one. You know, this is definitely going to be a stock that we keep an eye, especially as the AI IPO wave heats up with SpaceX, OpenAI, and Anthropic all potentially going public later this year. So make sure you subscribe for the podcast so you don't miss our coverage of when those IPOs finally drop. Also, let me know in the comments on what you think about Cerebras. Are you buying the Cerebras hype at these prices or do you think the valuation is way too stretched? Drop your thoughts in the comments on Spotify and YouTube, and while you're at it, consider giving us a five star rating as well. Wherever you listen to your podcast, know all that engagement really does help us out, and it helps other people find the show. Thank you guys so much for listening, watching and commenting. Shout out to Mike for all the work behind the scenes and we'll see you guys back here tomorrow.