← Run the Numbers
Run the Numbers
Cerebras IPO: S1 Breakdown - The Giant Chip, the OpenAI Deal, and the $24B Backlog
00:46:40
Loading summary
Transcript1 lines
- [00:00]
A
Welcome back to Run the Numbers. I love when there's an ipo. I get hyped up. Maybe I'm just overly caffeinated. I also just had four eggo waffles. So it's early in the morning and I was up late doing an S1 breakdown on Cerebras, who is going to IPO on the NASDAQ this week. And they're a chip maker. They think that bigger is bigger. Bigger is definitely bigger. I don't know if it's necessarily better, but what I'm going to do is take you through this upcoming IPO. I read every page of their S1. I'm going to explain to you what this company does, their target value, and some of the very interesting, I think I would call it circular finance going on, especially when it comes to OpenAI. Have you heard of them? All this and much, much more. Let's get into it. Is this thing on? Yesterday's price is not today's. All right, let's get this IPO party started. So Cerebras IPO breakdown. So for those at home following along in audio format, this person is dressed like a doctor in scrubs and it appears she's holding a chip the size of like a dinner plate. And I don't know if she's going to insert that in your chest because you need heart surgery. If your heart doesn't work, at least you'll have 44 gigs of on chip memory installed. So Cerebras builds the world's largest computer chip ever commercialized. It is 58 times the size of Nvidia's flagship and roughly the size of a dinner plate, which is also like the most American thing ever to say that bigger is bigger. Big as they are, the chips are the easy part of this tale. The financials are where it gets very, very funky. So Cerebras did 510 million in revenue in 2025. They grew 76% year over year and they are sitting on 24.6 billion in backlog. Okay, so those are contracts that they've signed with other customers. It's actually just one customer. I'm not gonna lie to you. Over 20 billion is one customer. That customer also loaned them a billion dollars and was issued 33 million shares for fractions of a penny. Who do you think it is? Who do you think it is? I had a lot of fun looking up sad Sam Altman picks. That is just a genre, a sub genre of the Internet. They're filing to IPO this week, today, probably Thursday At a range that implies a $33 billion market cap. So if you love core weave, you're going to potentially maybe like Cerebras. So strap in. There's a big chip sale going down, and I ain't talking Tostitos. How can that be profitable for Frito Lay? So what does this company do? Well, every chip you've ever interacted with is really small for a reason. Chip makers start with a silicon wafer, which is a round disc about the size of vinyl record. And they stamp hundreds of identical chips onto it and then slice the vinyl apart so each chunk can be packaged and sold individually. Your laptop and your phone each have one chip. And Nvidia's flagship B200 GPU has two that are mounted together. And so the reason chips are small is because defects happen. Silicon is finicky AF some part of every wafer has a flaw. And the bigger you make the chip, the higher the chance a flaw lands inside it and kills the whole thing. So it the industry settled into making chips smaller and stamping a lot of them per Wafer. So for 75 years, the largest commercial chip was about the size of a postage stamp. Cerebras kept the wafer intact. They invented a way for the chip to recognize defects in route work around them. They figured out how to cool a chip the size of that dinner plate without melting the motherboard. And they worked with Taiwan Semiconductor TSMC to develop a manufacturing process that had never existed before. No one had ever done this commercially. Today, it's pretty cool. They call it the wafer scale engine, or WSE, and it is 58 times the size of Nvidia's B200. As we mentioned. Here's why size matters for AI. They said in their S1, the enemy of speed is communication latency. And since communication is thousands of times faster on chip than a cross chips, the best way to reduce latency is to keep communication on chip chip. AI Workloads spend most of their time moving data between chips, not computing, which is something I learned when I was today's years old writing this S1 breakdown. So picture a GPU cluster as a giant room of accountants who can do math. But every time one finishes a calculation, they have to walk it down the hall to another accountant to continue the work. Now, the math isn't hard for them, but they walk really, really slow, and the walking takes up a ton of their energy. So another way to think about it is Nvidia sells you the accountants and the cables that connect them and Cerebras puts 900,000 of these accountants in the same room on a single piece of silicon. So the data never has to walk anywhere. And they get no lunch breaks. They claim this makes their system up to 15 times faster than Nvidia on inference at a fraction of the power, it's cool. But that pitch did not sell in 2020. So from their own S1, they said AI was nascent, it was raw and unproven. Training was time consuming. A black art. I love that. In the domain of a select few, GPUs were not yet the bottleneck, and our solutions struggled to find a home. They were founded in 2015, they were talking about this in 2020. So they did a handful of life sciences deals and they did like 25 million in 2022. So womp womp, Johnny, on the spot. Right place, right time. The inference became the workload that mattered the most. So the way modern AI works has changed in a way that finance people should honestly care about, even if they don't care about the technology. Old style AI. Old AI. The stuff that ran before late 2024 did its thinking upfront during training, and inference was just looking up the answer, which was both fast and cheap. Modern reasoning models like Claude, GPT5 and Gemini do most of their thinking during inference. That's why you always see it like talking through its logic and kind of almost buffering, planning the steps along the way and checking their own work and refining answers in real time before giving you that response. So it's the difference between a student who you could say memorize a textbook, and a student who actually works through the problem on the page live. Now, the second student is smarter, but the second student also takes a lot longer and uses a lot more compute on every single query for every single user. Inference is no longer a lookup, it's the bottleneck. And the bottleneck is exactly the thing that Cerebras built their chip to solve for four years before the bottleneck existed. That's why this IPO is happening now. Cerebra spent five years explaining wafer scale to a market that didn't need it at the time. And then in early 2025, the market arrived at their doorstep, right, right place, right time, and their revenue went nuts. It went from 78 million in 2023 to 290 in 2024 to 510 in the most recent fiscal year of 2025. And then OpenAI shows up, they sign a $20 billion deal in December, and then AWS signed a term sheet in March. So the business is booming. They said in the S1. We firmly believe that once you go fast you can never go back. I want to go fast. Damn, I'll bite. I'm pulling down 500 megabits a second via Xfinity, the world's most hated company in this biz op. So let's keep going some key stats for you. So quick scorecard before we get into the weird stuff. Stated this a second ago. Revenue 2025510 million. They grew 76% year over year. So it's up from 290 in 2024, up from 70.7 in 2023, and up from 24.6 in 2022. So they went 20x in three years. That's pretty awesome. And the quarterly trend is also accelerating. So Q1 of 25 was 99.5 million. Q4 of 25 was 1.71.4. So if you multiply Q4, annualize it as the cool kids do, you're at 686 million run rate going into IPO. Hey, thanks for listening. Be right back after a word from our sponsors. One more thing about Spendhound Teams using Spendhound reduce software spend by up to 30%. Here's how that actually happens. First you get the data. Real pricing benchmarks from over 1,000 companies across 10,000 SaaS and AI vendors. When a renewal comes up, you know what fair pricing is and what you should be pushing for. But knowing the number and getting the number are two different things. Spendhound also includes on demand procurement specialists who help with negotiation strategy and contract review and renewal emails. You stop overpaying for the tools you need, you cut redundant tools before they renew, and you negotiate from a position of strength now that you have both the market data and expert support behind you. Spend out is built by the team behind Yipit Data. It's rated number one on G2 in SaaS spend management and did we mention it's free forever for SMB? When we say free, we actually mean it. You can like keep all the savings. Plus it's only $10,000 per year for enterprise with $150,000 in savings guarantee. There's a reason we're growing more than 100% year over year. Do you want to stop overspending on SaaS? Then go to spendhound.com cj0risk guaranteed savings. That's spendhound.com cj Most finance teams aren't underperforming, they're actually under equipped. The problem isn't the people it's what they're being asked to do. Because there are two kinds of finance work. The kind that moves the business forward and the kind that just keeps it from falling flat on its face. And too many finance teams are buried in the second kind and they call it a good week when they just get through it. Expense reports that take longer than the trips they cover spend policies living in PDFs that no one reads, a three day approval process for a $40 receipt. That's a lot of burritos, but not worth it. Month end close, it stretches into weeks. These aren't edge cases, they're the default state of finance at most growing companies. And the cost isn't just the time, it's what your best people aren't doing. While they're buried in the wrong work, you become what you spend on. And right now, too many finance teams are spending on maintenance instead of momentum. And that is why I use Brex Agentic finance that captures receipts, automatically enforces policy before the spend happens, and closes your books in minutes rather than weeks. So your finance team stops spending on maintenance and starts spending on momentum. 35,000 companies like OpenAI, Coinbase, Anthropic and Doordash already run on Brex because high density talent deserves some high leverage systems. It's time to get Brex af Can't believe they let me say that. Learn more@brex.com metrics that is brex.com metrics. I've seen a lot of FP and A tools and a LAF is one of the few that gave me that aha moment. Within minutes I remember watching their founder shout out to Albert, connect my netsuite data and build me a full P and L live in minutes. Aleph is now trusted by hundreds of leading companies. I've had the CFOs from Turo 8, sleep, Zapier and more on the pod and every one of them is a huge advocate. I also just published my second annual CFO Tech Stack Report and Aleph has been on the podium both years including a number one finish in the 50 to $100 million segment. This year instead of being just another planning tool, they built a real enterprise grade data foundation for finance implemented at startup speed with AI native workflows woven into its DNA. All your systems, erp, CRM, hris, ats, product usage and more powering one clean governed data layer that finance can actually trust. With AI moving as fast as it is, they're pushing even further. Mcp, custom AI chatbots, AI powered variance analysis and the List keeps growing. Try it with your own data at get a left.com run that is G E T A L E P h.com/run. Tell them CJ sent you. Let's talk gross margin 39%. This is actually down from 42% the year before. Hardware margin is 43%. Cloud and other services margin is OOH only 30. Cloud is the newer part they want to grow and cloud is the part with worse margins which we will indeed discuss later. Net income FY 2025 it was positive 237.8 million. Don't get excited my friends. The 238 million includes 336.3 million in one time gain from extinguishing a forward contract liability which for all the SEC accountants at home that is a non cash accounting event tied to preferred stock arrangements. So if you strip it out, they actually lost 75.7 million on a non GAAP basis. And if you dare, if you want to do this, some people will criticize you. You can strip out stock based comp and you get a clearer view of operating performance which is roughly break even on ops and burning a lot, a lot a lot a lot of cash on capex and working capital. So they're not really profitable. Operating cash flow negative 10 million. So for comparison 2024 operating cash flow was positive 452 million. What is going on here? We got a whipsaw folks. The reason for the swing here is not that 2024 was a great year, in 2025 sucked. It's that 2024 included 640.3 million in customer prepayments from a customer and investor G42 which sounds like an airplane, which lands in operating cash flow. And in 2025 saw those prepayments worked down by 286 million as cerebras delivered against them. So net net customer prepayments are funding the business, not profits. Remaining performance obligation RPO. This is a very large number 24.6 billion. So just to pause here for a sec, this both justifies their valuation also comes with a shit ton of asterisks. It's 48 times last year's revenue. My goodness. And most of it is OpenAI. They expect to recognize 15% over the next 24 months through the end of 2027, 43% in the 24 months after that and the remaining amount sometime in 2029. So fun comparison here. Core we've IPO'd in March of 2025 with 1.92 billion in revenue and 15.1 billion in RPO. Cerebras is going public with roughly a quarter of Core Weave's revenue and a 63% bigger backlog. What's going on there? Customer count and concentration. The S1 does not give a precise number, only that the top 10 customers grew their spend with Cerebrus by approximately 80% within 12 months of their initial purchase expansion. Two of the companies in the top 10 are MBZUAI, probably the worst name ever, and G42, that airplane. Both based in Abu Dhabi. Never been there. Here it's hot and related to each other. In 2025, MBZUAI was 62% of revenue. In 2024, G42 was 85% of revenue. That's what you would call concentrated. The third major name is OpenAI, who signed in December of 2025 and they account for $0 of 2025 revenue. But most of the 24.6 billion backlog employees. What it do 708 people as of December 31, 2025. About half are international with offices in Canada, India and the UAE where those customers are. Revenue per employee is 720,000, which is pretty healthy for a hardware company at this scale. Target valuation 33 billion at the 155 per share midpoint that they filed of the 150 to 160 price range with 215 million shares post offering. And that's comprised of Class B and Class A which we'll get into. The fully diluted number is actually higher once you factor in the OpenAI warrant. It's a large warrant, AWS warrant, another warrant, an RSU pool and options. The Series H we're getting into the Alphabet in January of 2026 priced at $89, which means the new IPO buyers are actually paying a 74% step up four months later to call out the Core Weave in the room. The comparison is too clear. It's important to note that Core Weave works with Nvidia souping up their chips. Cerebras has TSMC manufacture their chips and then supes them up. So same same but different. But the rest of the story is kind of eerie. Both companies built infrastructure for AI training, then got supercharged when inference became the bottleneck. Both have customer concentration concerns. Microsoft was actually 67% of Core Weave's revenue in 2025 when they IPO'd. MBZUAI is 62% of Cerebras'. Both have a giant OpenAI backlog. 22 billion for Core Weave. Most of Cerebras 24.6 billion went public with no real cloud business yet. Now let's look back. Core weave priced at $40 in March of 25, down from an initial $55 target if you recall, which was honestly kind of a dark cloud over the IPO at the time. They had to go down in price and the stock opened flat, then ran up by 300% in June and then gave half of it back. It's currently trading at roughly 19x current sales after 5.1 billion of 2025 revenue in bonkers growth. They're doing well, but also the market does not know what to do with this category. And with Cerebras, that uncertainty is growing. How do they make money? I'm so glad you asked. Cerebrus has two revenue lines. The first is hardware. That was 358 million last year. That was 70% of their revenue if you look at the mix. And they sell that wafer scale engine inside a fully integrated system called the CS3. And they sell racks of those CS3s wired together into what they call an AI supercomputer. Also it sounds like a midsize Audi with a twin turbo. Customers buy these and put them in their own data centers. This is the original business model and it's where most of their revenue still comes from. They have this funny picture of like someone in a data center and it's like this hipster in a lumberjack costume with a hood on and a mask and it's like, is that masked lumberjack hipster robbing our data center? Well, the good news here, hardware grew 69% year over year from 212 million to 358. And the S1 attributes most of that growth to that one customer. MBZUAI how many times am I going to have to say that buying a lot of on Prem systems. The less good news, hardware revenue is lumpy af it depends on big customers cutting big pos and then waiting for delivery. The S1 puts it very bluntly. Each individual sale tends to be large as a proportion of our overall sales, which has impacted our ability to accurately forecast revenue and manage cash flows. So a miss on one customer is a miss on the quarter. Hey, thanks for listening. We'll be right back after a word from our sponsors. So here's a pattern I keep running into when I talk to finance leaders at fast growing companies. You've outgrown the spreadsheets, you've probably outgrown your billing tools built in Revrec. But you're not quite at the point where you can throw a 20 person team at the problem either. That's exactly the danger zone right? Rev owns right? Rev is revenue recognition done right. It handles the messy stuff like high volume subscriptions, usage based contracts and mid contract upgrades. The things that break your ERP and the billing platform bolt ons. Here's the thing though. Your sales team isn't slowing down for you. They're closing ramp deals, usage commitments and mid quarter upgrades. And the longer you wait to fix the engine the further behind you fall. So stop scrambling at month end and stitching together allocations across 3, 4, 5 spreadsheets to just have the numbers read. Well that's it. That's the whole pitch. CFO's telling me it's like a glow up for the revenue books. If that sounds like where you are right now, right rev is worth the look. Head to rightrev.com cj that's right rev.com cj check em out. I got news for you. The ERP category is finally getting disrupted. And if you haven't heard of RYLIT yet, please pay attention. It's the AI native ERP built specifically to replace netsuite and it's already won over hundreds of finance teams. Their mission is to make these zero day close a reality. And they're actually doing it. We're talking teams closing the books at 1:35pm on the first day of the month. Companies like Windsurf, Mercore and hundreds of others run their entire finance stack on really revenue recognition, close management, multi entity, native stripe and Salesforce integrations. Woo everything a scaling company needs. They've got 5.0 star 50 wow 50 stars on G2. They're better backed by A16Z and Sequoia. Heard of them and CPA led implementations that get you live in 45 days. That is simply unheard of in the ERP space. If your books aren't running as fast as your business, check out row it. Book a demo at rillet.com cj that is R-I L L E T.com cj that's me. Scalia Tech Company is thrilling. It's also really really messy. Just ask anyone who's done it or anyone who's tried. Better yet, ask ey. They've seen startups at their best and in their most fragile moments. EY knows you don't start a company to burn cycles on regulatory hoops, discounted cash flows, or the fine print of sec form S1. Although you probably do know I love myself a good S1. If you've listened to or read my stuff long enough. You can't ignore these things. That's how risk compounds kind of like negative interest. What you can do is work with EY from day one. They'll help you get it right early and often so you can stay in builder mode and keep the trains running on time. EY shape the future with confidence. Learn more at ey.com techstartups that is ey.com techstartups let's talk about the next revenue line, which is very much still emerging. Is there cloud and other services? I love when they just say other services. What could be in there? Are you also doing Uber on weekends? 152 million or 30% of revenue they launched the Cerebras inference cloud in August of 2024. Customers can rent compute by the token by the month or for dedicated long term capacity and they can buy it directly from Cerebras or through partner marketplaces like AWS, Microsoft, IBM, Vercel, OpenRouter, and Hugging Face. The good news Cloud grew 94% year over year from 78 million to 152 million. Subscription revenue inside that line grew even faster. Nice. The less Good News Cloud Gross margin is 30%. That's a low number. Hardware gross margin is 43%. One of these things does not look like the other. The business they want to grow has worse margins than the business they already have. So sit with that for a sec, because it's actually the inverse of what the IPO narrative typically craves. The standard SaaS pivot story is that you start as a lumpy hardware or services business and migrate to recurring software revenue with high gross margins, sticky retention, and a clean multiple. Cerebras is doing the opposite. The hardware sale is a one time invoice with reasonable margins. The cloud product is the recurring revenue line, and it's costing them more to deliver than the hardware they're trying to migrate customers away from. Let's look at the quarterly trend here. Not to get too pedantic on you, but the cloud gross margin in 2025 went from 68% to 26% to 16% to 21%. What are we doing here folks? The Q2 crater is what happens when you stand up a bunch of new cloud capacity faster than you can fill it. So the S1 attributes the gross margin decline due to higher data center costs related to our cloud inference capacity services. Two things to know about where this goes. First, Cerebras is locked into the cloud direction by their largest customer. The OpenAI deal is structured as cloud capacity, not hardware sales. Most of the 24.6 billion backlog converts the cloud revenue, which means the gross margin mix is going to get more cloud heavy, not less, as the backlog burns down. So the same pattern applies to that AWS term sheet that I mentioned. The cloud margin will probably improve once the capacities they're building gets filled. The 16% Q3 number reflects unfilled data center capacity just sitting there earning nothing while still costing money and lease power and depreciation. So as OpenAI ramps and AWS comes online, the utilization will go up and the margin hopefully recovers somewhat. But the bull case is that cloud gross margin eventually approaches hardware gross margin. That's kind of backwards. And in the high 30s or 40s. The bear case is that running a data center against hyperscalers who do it for a living means structurally lower margins forever. So terrible now, but maybe not terrible forever. I don't know. Either way, the company that's ipoing is not a software company. It's a hardware company with a margin dilutive cloud bolt on. Let's Talk customer concentration. Two customers account for 86% of Cerebras 2025 revenue. Those customers are. I have to do it again. Damn it. MBZ UAI and G42, both based in Abu Dhabi. They are also related parties to each other as defined by Accounting Standards Codification850. Meaning there's some stuff going on there. Oh yeah, there's some stuff inside that stuff. G42 is an ABU Dhabi based technology holding company with subsidiaries spanning energy, finance, cloud security and healthcare. MBZUAI is the Mohammed Bin Zayed University of Artificial Intelligence. I also went to University of Phoenix, also in Abu Dhabi. Also operating in the orbit of the Same sovereign backed AI strategy that. Let's take it from the S1. They say we have established strategic relationships with groups 42 holding limited in the Mohammed bin Zayed University of Artificial Intelligence. G42 and Mbzuai have acted as our customers, vendors, partners and or research collaborators on multiple initiatives in model training, inference and AI compute infrastructure. So customers, vendors, partners and research collaborators. Maybe even roommates. And they were roommates. Oh my God, they were roommates. Here's how the concentration has moved. So in 2024, G42 is 85% of revenue. MBZ UAI was not material yet in 2025 MBZ UAI was 62% of revenue. G42 is 24. But if you combine those two, do the little addition, it's 86%. The accounts receivable is even more concentrated than the revenue. At year end 2025, MBZ UAI was 78% of accounts receivable. At year end 2024, G42 is 91%. Working capital concentration is a risk. I hope we get paid. So, good thing they're investors, right? Oh, and then there's how money moves between them. So G42 prepaid cerebras 640 million 2024 to fund the build out of the hardware they wanted Cerebras to deliver. That prepayment is what Cerebras is 2024. Operating cash flow made it look healthy, made it look like it was real. Without it, operating cash flow was deeply in the red. G42 also holds warrants. Boy, do they. In December of 2025, Cerebrus issued G42 a warrant to buy 1,857,516 shares of class and common stock at $0.01 per share. The warrant was fully vested and exercised in January of 2026. In April of 2026, Cerebras issued G42 another warrant. This time for 1,655,975 shares, also at $0.01 per share. That's a good price. Also fully vested in exercise the same month. What could be lower than $0.01 per share? We'll see. G42 is. Let's just recap. A customer that prepaid hundreds of millions of dollars for product was 85% of revenue one year, 24% the next. Was issued two warrants worth at the 155 IPO midpoint. Approximately 554 million in equity for an aggregate exercise price of $35,000. Tell him what he's won, Johnny. Doing the math, that's roughly half a billion in equity for the cost of a slightly used Nissan Altima. There is also AWS. In March of 2026, Cerebra signed a binding term sheet with AWS to be the first hyperscaler to deploy Cerebra systems to its own data centers. The term sheet is binding on pricing, exclusivity, minimum capacity and protections in favor of aws. But the definitive agreement hasn't been negotiated yet. So AWS also got a Warrant up to 2,696,678 shares of Class N, as in Nancy. Common stock at 100 bucks per share vesting tied to product purchases beyond the initial lease. And the AWS warrant is structurally different than G42S. The $100 exercise price means AWS pays real money 270 million if fully exercised and only vests if they actually buy a lot of compute. That's a more normal commercial agreement. The S1 names a dependency layer cake and its risk factors. They name the four customers who essentially the whole company. They say a reduction in demand from or a material adverse development in our relationship with any of our significant customers including OpenAI, G42MB, Zuai and AWS or our failure to meet our obligations under the MRA with OpenAI would harm our business financial condition, results of operation and prospects. So Cerebras also has supplier concentration on the back end. They use one foundry, TSMC to manufacture every wafer. And the S1 acknowledges they have no formalized long term supply commitment with TSMC who also fabricates wafers for Cerebras competitors including Nvidia, who's many many many times larger and buys many, many many more wafers from tsmc. And on top of that, the data centers hosting the cloud business are leased, not owned. So the picture is TSMC manufactures the chips, OpenAI funds the working capital and accounts for most of the backlog. G42 and MBZ UAI account for most of the current revenue and have prepaid hundreds of millions of dollars in advance of future deliveries. AWS will host the cloud business at scale in their own data centers with exclusivity provisions and then lease data centers host the rest. So looking up the cap table is like attending a Targaryen wedding. The customers are also the owners. The owners are also the lenders. The lenders are also your sister. Let's talk about the OpenAI warrant because I do think it deserves its own section. This is like this is Funky Town. Take me to Funky town. In December of 2025, Cerebra signing Master Relationship Agreement, which sounds like a uranium production deal with OpenAI. OpenAI committed to purchase 750 megawatts of coin cloud compute capacity over multiple years with options to expand to 2 gigawatts. The deal is valued at more than 20 billion and represents most of the 24.6 billion backlog. The IPO pricing is built upon alongside the MRA. Cerebras issued OpenAI warrants to purchase 33,455,026 shares of Class N as in NED common shares at an exercise price of oh my goodness, we can go lower than $0.01 is 0.00001 per share. At the $155 IPO midpoint, that warrant is worth approximately 5.18 billion in equity. The exercise cost to OpenAI to take possession of all 33 million shares is $334.45. So like about what it costs to take my three kids to a character breakfast at the Contemporary hotel in Disney World. OpenAI also advanced Cerebras a $1 billion working capital loan. Like we mentioned, it accrues at 6% interest that is lower than my mortgage rate, matures in December of 2032 and is repayable in cash, compute capacity or hardware. Dealer's choice. It is secured and it comes with strings. If the MRA is terminated for any reason other than OpenAI's material uncured breach, OpenAI can direct the bank to freeze the cash and demand immediate repayment of principal plus interest. They will send Doug the Bounty Hunter the warrant vests in three tranches. A first slug of 691 million in equity already vested in January of 26 when Cerebras accepted the $1 billion loan. A second slug of 864 million vessels. Cerebras's market cap clears 40 billion on a 30 day rolling average. Meaning the CEO is very much incentivized to hit a valuation milestone that automatically transfers almost a billion dollars of equity to his largest customer. The remaining 3.6 billion vests as cerebras delivers compute capacity with the full amount only. Vesting of OpenAI exercises every option and scales the deal to 2 gigawatts. So Cerebras largest customer is also the customer most economically incentivized to keep buying more success, be expensive and dilutive as fuck. Let's talk about the financials here. R and D was 243 million. It grew 5.54percent year over year and represents 48% of revenue. So almost half of revenue is going to engineers. So for Comparison, Nvidia spent 12% of revenue on R&D in 2025, which is also a very terrible comparison because they're very very big reminder. Cerebras is a process node company designing wafer scale chips against a competitor with 50x their cash flow. So that's a tough comp set. They have to spend at this level very much so to have a shot. Sales and marketing 70.6 million 237% year over year growth, 14% of revenue. So the build is more for cloud. That's my take here, not for hardware because hardware sells through a few reps closing these massive deals. They have like two customers, remember just kidding their staffing. A SaaS go to market on top of a hardware go to market that is very much nascent. GNA was 31 million. Actually down 31% year on year. Down What? GNA went down in an IPO year when revenue was up 76% and headcount was growing. The S1 attributes it to lower legal expenses and litigation settlement costs. Okay, so there was some stuff in that stuff. Operating loss 146 million versus 101 million in 2024. OPEX out ran gross profit. And this is a part of the financial story that gets buried under the GAAP net income headline which is technically positive but philosophically misleading. Other income and expense was plus 391 million and that includes the 363 million afford contract gains. So that line is what makes the 2025 GAAP profitable and has like nothing to do with the underlying business. Let's talk about the balance sheet. What it do balance sheet 72 million of cash on hand at year end 2025. The S1 also discloses that OpenAI Advance, a $1 billion working capital loan. We covered that. Which it's just important to know because that wasn't in that cash number. The 702 million that now sits on top of that cash but can be clawed back if the agreement terminates the wrong way. So much of the cash they have access to today is tied up in OpenAI cash burn. Operating cash flow was negative 10 million investing was negative 668 including 383 million of capex for cloud capacity. Remember it was that unused cloud capacity is still ramping. Electricity's on honey, turn off the lights. Financing was 1 billion. Almost all preferred stock. So they raised their series G and H Recently cerebras funded 2025 by raising a billion dollars of equity and taking on a billion of open AI and loans which Cerebras funded 2025 by raising another billion of equity. About 400 million was burnt on that cloud build out and then they have a revolver. So $250 million facility from Morgan Stanley signed in April of 26 and that can upsize to 850 million post IPO. Morgan Stanley is also the lead left underwriter. So your book runner is also your revolver lender. But that's kind of normal in IPO season but worth knowing. Let's talk about RPO. That is not a peptide. Next 24 months through 2027 15% 3.7 billion months 25 to 48 43% 10.6 billion after 2029 42% 10.3 billion. That's 1.9 billion a year of recognized revenue coming up over the next 24 months. I love that growth coming from the backlog alone and it grows to 5.3 billion a year in years 3 to 4. So if it plays out 26 and 27 combined revenue from the backlog is more than 3x what they did in 25. That is what's getting people excited, even if it's only coming from a couple players. That is very much the bull case. Asterisk time it assumes OpenAI takes delivery on schedule, it assumes MBZUAI keeps buying, it assumes no no no no mra disputes with OpenAI and it assumes the cloud capacity gets built empowered on time. Tell those construction people to pick up the picks and shovels and get back to work because if any of that slips, revenue recognition shifts out out out to the right potential red flags 10 yard penalty. They told the SEC they don't have a real good accounting function in their own S1. The material weaknesses that we have identified related to one inadequate or missing resources who possess an appropriate level of expertise to timely review account reconciliations and identify, SELECT and apply U.S. generally Accepted Accounting Principles GAAP pertaining to several financial statement areas including rev rec, Inventory management and costing, Data Center Assets Accounting and Equity administration. My goodness, the list goes on. And the failure to maintain adequate IT general controls including ineffective segregation of duties. You don't want to print that. Why'd you have to print that? Listen, I've spent enough time in audit meetings to know what ineffective segregation of duties means. In practice. It basically means that somebody is writing the journal entries and reviewing the journal entries. It may also mean you have one shared QuickBooks login, which is the kind of thing that gets you on the front page of the Wall Street Journal in a bad way. And remember, this is the company about to go public at a $33 billion implied market cap with 24.6 billion in RPO under a complex rev rec agreement with OpenAI telling you and writing that their controls over rev rack kind of suck. Next Red flag one. Foundry to make chips no long term supply commitment so every Cerebras wafer is manufactured by TSMC with no formal long term supply agreement. How could that even be true? TSMC also manufactures wafers for Nvidia, AMD and most of their competitors, all of whom buy a lot more wafers than Cerebras does. If TSMC reduces allocation, raises prices or prioritize one of those competitors, Cerebras says no. Second source. It's not like you can call up Samsung Foundry and just be like yo, can we get some wafers on Friday because remember they have this new technology that they came up with, wafers. Scale adds a wrinkle here because they buy an entire wafer per unit on a specialized manufacturing process that was built in partnership with tsmc, the ones they don't have a long term contract with which is not easily portable to another foundry. So bigger is a bigger risk. Number three, the UAE concentration is very much a geopolitical bet as well as a customer bet. Remember, 86% of their 2025 revenue came from two Abu Dhabi entities. Yeah, a lot of political stuff going on over there and. And G42 has actually already drawn scrutiny from the US government over historical ties to China, which is why Microsoft put 1.5 billion into them in 2024, basically buying them a US passport. The Commerce Department continues to update its export framework for advanced semiconductors in jurisdictions where G42 and Mbzuai operate. And the framework, it's getting tighter, not looser, over the next 18 months. So I don't know what's going to happen there, but hypothetically wouldn't be good if you couldn't import and export out of the uae. Next one. Number four, the cloud business is unproven and they're betting the company on it. So Cerebras launched the Inference Cloud in August of 2024. As of the S1, it is fewer than 18 months old. I've been doing this podcast for two and a half years longer than that. OpenAI, MRA, the AWS term sheet and more. You know that big backlog are all structured as cloud capacity, not hardware. So just a reminder, the big deals are big cloud deals and the cloud business isn't really big yet. Let's talk about the cap table and also some context on the recent Series H. So the founders and the board still hold pretty meaningful equity. But Cerebras has been a heavily funded private company for almost a decade and institutional names dominate the cap table. So the top holders post IPO before any over allotment. Fidelity comes in at 11% benchmark, just under 10% foundation capital, about 8 eclipse 7% alpha wave 6.5 sounds like a Call of Duty level. Andrew Feldman the co founder 54 Sean Lai the co founder and CTO 2.9% so Feldman and Lai have actually been a duo since C Micro in 2007, which AMD acquired in 2012. They both did time at AMD post acquisition and then co founded Cerebras in 2016. So the operating partnership between these two 18 years of continuity. Feldman is a founder CEO running his second company. He is not a hired gun. So founder CEO the combined founder ownership is 8.3% which if you were to benchmark that is on the lower end for an IPO stage company. Dylan Fields of Figma had more himself. Cerebras founders got diluted harder than most you could say because of the business needing a ton of capital before the inference moment showed up in 2022 most recent round that it's going to be benchmarked against. In January of 26 Cerebras closed a $1 billion series each at 89 bucks per share with Alpha Wave, Benchmark and fidelity all participating. Four months later they're pricing that IPO to midpoint of 155 which is a 74% step up from the series H2 IPO. So for context the series G price at 36 bucks in September of 25 the company was reportedly thinking of actually going public back then I remember I almost wrote this good thing I didn't at that point did not figure out the market situation going on there. So it's inside of 12 months 36 bucks a share to 89 bucks to 155 that is 4.3x in a year. Almost all of it riding on the OpenAI deal that was signed in December of 25. So other thing to flag is that fidelity put in 700 million at the that is oh my goodness, 700 million at the Series G round and another 100 million at Series H. That is a big old fat IPO position. I wonder if they cash out founder voting control. That's an important thing to talk about huh? Cast your votes. Cerebras has three share classes, Class A one vote per share. The public IPO shares Class B 20 votes per share held by insiders and founders and class N as in Nancy non voting held by OpenAI and G42 via their warrants. Class B will represent 99.2% of voting power. Immediately after the offering. The new public shareholders own 14% of the economic upside and 0.8% of the voting power. LOL. Let's talk valuation at the $155 midpoint, cerebras prices at roughly 33 billion on 510 million of 2025 revenue and 2.5 to 3 billion of expected 2026 revenue. I'm using the RPO backlog there plus the base business so that's an EV to forward revenue multiple of 10 to 13x. So let's let's stack rank that with public peers so Nvidia has a $3 trillion market cap, has 130 billion in revenue, will probably do 200 billion in revenue next year. And it's trading at 18 to 22x. 2026 revenue. AMD 280 billion. Market cap, 26 billion in revenue, 32 billion probably next year. Trading at 6 to 7x core weave. Remember them? Market cap 62. Wow. 5.1 billion in revenue. They're ripping 12 to 13 billion in revenue, probably next year. 7, 8 forward revenue. Keep that in mind. 7, 8 Nebus is another one, which I didn't know was a company until I did this. 41 billion market cap, 530 million in revenue. That sounds crazy. Oh, that's why they're going to do 7 and 9 billion in ARR next year. So 5 to 6 revenue, big backlog. And Cerebras at the midpoint. Like I said, 10 to 13x. So they're pricing at a premium to every peer except Nvidia. Aggressive. So bold strategy, Cotton. Let's see if it pays off for the semiconductor peers in the neo cloud peers. Core Weave and Nebius both strateg trade at half the multiple Cerebras is asking for. So only Nvidia gets a higher multiple. And Nvidia prints free cash flow at industrial scale, which Cerebras does not. Let's be real though. What's actually defensible at 6 to 7 or AMD multiples? That would put them at 16 to 21 billion. The argument there is, hey, you're a fabulous semiconductor comp. The chip is good, but the customer concentration is real. Cloud's still improving at 7 to 8x core weave, that's 22 to 25 billion in market cap. Core Core Weave has more revenue, more customers and more operating history. So if Cerebras reads as a core weave with worse concentration, the multiple compresses below that at 10x. That would be 25 to 30 billion. Okay, so credit for owning the silicon. No benefit of the doubt. For the cloud margin at 13x in midpoint, that's 33 billion. Like we discussed in the market, is buying the wafer scale tech durable margin advantage? I don't know. Nvidia. Ooh, imagine if they got 20x. That'll be 50 billion. OpenAI narrative wins the day. The midpoint is the high end of what I think is defensible. With the current information and the Nvidia style, multiples with the OpenAI deal narrative will be selling. Whether it sticks depends on what the market reads. Some miscellaneous stuff of Note. Hate that I have to say this. The CEO has a 2007 guilty plea for circumventing accounting control. So Feldman was a defendant in SEC versus Pereira in 2008 related to his time as VP of Corporate Marketing Corporate Development at Riverstone Networks from 2001 to 2002. That's a while ago. The SEC alleged he was aware of and aided in sales transactions that were improperly accounted for. He settled with the SEC without admitting wrongdoing, agreeing to a permanent injunction against future securities violations and paid 289 grand plus interest in a parallel DOJ action. He pled guilty to one count of circumventing accounting controls of an issuer. He was sentenced to three years probation and a $5,000 fine. So these are 18 year old events from a different company. And Feldman wasn't barred from serving as an officer or director. America, land of second chances. He's gone on to run C micro, sold to AMD and now Cerebras. And the S1 discloses all this? They're not trying to hide it. Three classes of stock the same except for the parts that matter. The S1 says that the three share classes is so funny. Are identical except with respect to voting and conversion rights. Okay, so the important stuff. Class A like we said, one vote. Class B, 20 votes. Class N, no votes. Oh, Class N is what OpenAI and G42 got. Founders keep control. I think the public should just be aware that they're going to get diluted by Class B's voting power and Class N's economic share when it converts while paying the full price for the very privileged to get diluted. 40 billion market cap equals 10D. So the Tranche 2 OpenAI warrant 864 million in equity to OpenAI vests when Cerebras's market cap clears 40 billion on a 30 day rolling average. And at the $155 midpoint, if they run up, they could be a couple months away from that. Yeah, the company is one good month of trading away from automatically transferring almost a billion dollars in equity to its largest customer. And lastly, if you look at the COVID where it lists all the banks, some guy named Craig Hallam is on the COVID The lead left underwriter is Morgan Stanley. The bracket is Citi, Barclays, ubs, Mizuho and td. Then further down the COVID Needham, Wedbush, Rose, Bland, Academy, Credit Agricole, mufg, First Citizens and Craig Hallam Capital Groups llc. I smoked pot with Craig Hallam. It was Craig Hallam and Sloan Kettering. And they were blazing that shit up every day. All right folks, none of this is investment advice. Please do your own homework. Wishing you an IPO with ample float and no material or county weaknesses. Peace Run the Numbers is a mostly media production. Yelling an intro by Fat Joe Artwork by Meg d'. Alessandro show is executive produced by Ben Hillman. Nothing said on this podcast is intended to be business or investment advice. It's the sole opinion of me, a guy who feeds his dog way too much ice cream and has a history of net operating losses. Lol. If you like this podcast, hit subscribe and give us five stars. It will take like two seconds and our algorithm overlords love it. Drink water, call your mom and have a great day. Peace.