Thanks for joining us. My name is Simon Leopold, Raymond James Semiconductor and Data Infrastructure Analyst. Pleased that you're able to join us here in New York or folks on the webcast. This afternoon we've got for our next fireside chat, Michael Staiger, who is the SVP and Head of Corporate Development for Supermicro. Michael, I think you've got a Safe Harbor.
Yeah.
We'll dive right into it.
Thanks for having us. Let me just make the following legal disclaimer. Please note that today's discussion may contain forward-looking statements and actual results could differ materially due to risks and uncertainties discussed in our SEC filings, which can be found on our website. All right, now that that's out of the way.
Great.
Thanks a lot.
Thanks for having us. So we were joking before about how we've heard it so many times that we still have to read it. So I imagine most folks who came are somewhat familiar, but I'd like to see how do you prefer to introduce Supermicro to potentially a new investor?
All right, that's an interesting approach. I would say from a new investor perspective, if you're thinking about what we're trying to do, what we're trying to accomplish, and where we came from, most importantly, where we've always been focused on the dynamics of performance, power, and price, so to speak, in an application-optimized system, we've noticed, obviously, that the AI generation was going to require different platforming, so to speak. So if you think about the landscape and you think about NVIDIA, you think about AMD, you think about ASICs, you think about Arm, and you think about the applications, and you think about where hyperscalers play, and they have all kinds of applications that go beyond just using CPUs now, right? So there's all kinds of infrastructure being stood up. We're optimizing for those applications across the ecosystem.
As we move forward and we're thinking multi-years ahead, two, three years ahead, end users will require different formats in different segments of the data center for different applications, whether it's models, whether it's inferencing, whether it's edge applications, whether it's IoT. We're optimizing the platforms for that. Since we don't have legacy architecture to support, and since we are 50% engineers, we literally are at the forefront of designing and developing systems to bring to customers on a global basis.
So I want to talk a little bit about your take on the AI demand environment. And I think part of the reason I think this question's not as obvious as maybe it first appears is press reports in the last couple of months about the idea that there's an AI bubble and it's not sustainable. I'm in print, so people know I'm an AI bull, so I'm not going to hide that. My biases are not to be hidden. But when you sort of encounter any kind of challenges, how would you describe your take on not just the current environment, but the durability?
Yeah, so our take's a little bit different than, let's just say, Wall Street's take, right? Our take is derived from our customer interaction and the customer demand and the ecosystem and what we see. And so from that perspective, it's very robust. I'll just go back a little bit and just talk about where we recently guided our full year numbers up from $33 billion for fiscal 2026 to $36 billion. Obviously, there's some demand in there.
Your fiscal year end?
Fiscal year is June.
June.
End of June, and so when we're thinking about the demand environment, we also discussed and disclosed that we had $13 billion in GB200 orders alone in December that we discussed that we said that would be probably a multi-quarter deployment, and so if you think about what our customers do, NeoClouds, enterprises, and sovereigns, they're coming to us from all different directions, and from what we're seeing, the demand just keeps getting better and broader, and then our ecosystem partners, as you can see, they're all expanding capacity, and so there's a lot of smart people across the industries trying to address what essentially will be new applications, and I don't know if anyone here has taken a Waymo or there's an AI element to that, of course, right, so it's just that we're at the forefront of a pretty big application-driven market.
So it looks pretty good, and we're going to be building to that across all the different stacks.
And maybe double-click a little bit in terms of who those customers are for Supermicro, because so much of the, I think, headlines are dominated by these hyperscalers. And then we talk about terms like Neo and sovereign and enterprise. Help us understand what that mix is and who those are. What are they doing?
So the way we think about customers and the optimal customers for us is anyone that wants to leverage data to advance their business. So I think it's fairly well known that we've had customers like Tesla. We've had customers like Twitter, which is now xAI. I'm sure there's plenty in that stack. So those types of customers and the NeoClouds, the CoreWeave of the world, the Lambdas of the world, the other NeoClouds that are coming into the fold to serve this. And the interesting, like the Applied Digital, et cetera. The interesting thing is that the NeoClouds are getting workloads from the hyperscalers. And so the NeoCloud environment's starting to expand quite dramatically.
And so I think people are, and then you've seen a few, well, not a few, but plenty of announcements about infrastructure and RPO and big transactions that the infrastructure may not yet have been built. And so the forward look there is there's going to be a lot of iron that needs to be engineered, manufactured, delivered, integrated, and brought up to support the ambitions of our customers.
Now, I guess we can accept the fact that there's going to be some lumpiness to the business because they're big projects. One of the transitions that seems to have been disruptive is going from NVIDIA's Hopper to Blackwell. Where are you through that transition, and what lessons maybe did the company learn in terms of execution so that when we go through the next generation, how do you avoid that? What sort of needs to happen?
From our perspective, to understand that we're scaling capacity, we're scaling go-to-market, we're scaling engineering capabilities. When we're looking on a forward basis and you think about what we talked about with respect to December, essentially you're standing up a massive brand new product, right? The cadence of any new product launch, any new hardware product launch tends to have a little bit of a headwind. We had a couple of factors in that element. This is the largest, most complicated system ever been delivered to mankind, and we're delivering it at scale. The market needs to understand or appreciate that.
And so the lessons there from that standpoint is the scale of the size of the opportunity we see going forward and putting all those efforts into establishing that footprint, establishing the dominant position there, and establishing probably the most reliable build from a system standpoint. Because when our customers are deploying our systems, the reliability of the systems and the engineering of the systems are very, very paramount to what they're doing. And so if we can deliver a lower cost per watt per compute for them, it directly translates back into their business models, especially in the NeoCloud and the service provider market. So we're doing an excellent job of doing that, and we will carry that forward in future generations.
Now, there's another sort of technological transition that maybe you can help explain this to an audience that might not appreciate the terminology, but you're shifting to more rack-scale systems. We talk about the idea of nodes to rack-scales. First, explain to people what that means. And then my real question is, is that a risk? Is this a phase where it's another challenge that could have stumbling blocks?
So I would say that it's not a risk for us, being the engineering powerhouse that we are. We started off motherboards, right? And then partial chassis and then servers, which you would consider nodes. And then we've evolved into what you would consider rack-scale. So at each step of the way, each turn of technology has become more and more complex, and there's more integration work that's required. So we've actually taken that a step further, and we're developing what we call our Data Center Building Block Solutions, which is all the ancillary equipment, power, cooling, storage, and other elements to stand up an entire data center.
Because when our customers come to us and they say, "Hey, we need to, we want to get this compute platform together by virtue of delivering racks in the past," where we've done a phenomenal job, and we talk about the capacity that we're able to put together for our customers, we see abilities to productize around that. And that's supportive of a better margin from our perspective, but a better experience for the customer. And we tend to retain our customers. And once again, we mentioned one earlier, it was started off a long time ago and has graduated with us. As things got more complicated, we're able to do more complicated things for them at every turn. And that's what's scaled the company from $2 billion a year to we're on our way to $36 billion.
And when you talk about retention or stickiness, what's the rationale? Is it essentially you're doing some customization, or is there something unique or something differentiated? What creates that?
So we try to do, again, lower the cost per compute per watt for a customer, whether that is accomplished depending on what the customer's requirements are to deliver that application to them in a way that's best useful for them. So not every customer is the same, not every requirement's the same, but there's a lot of engineering and skill set that goes into what we're delivering to the customers for those that want to take advantage of that. We can do what we will call more reference architecture capacity kind of builds that are a little bit lower margin, but that's kind of where we're focused. And it's the degree of difficulty that keeps advancing with the ecosystem itself. NVIDIA's platform's way more powerful than it was two years ago. AMD's platform is way more powerful as well than it was a few years ago.
So it's more complicated. Tolerances change, tightness with respect to the build itself and putting in cabling. There's the cooling element that we're introducing into the systems that we're perfecting, optimization of power, et cetera. There's a whole complication of the firmware, the integration, the cabling, all those types of things to get the customer to run at scale. It's a very complicated setup.
The company's been growing its capacity fairly rapidly. Could you help folks understand maybe a little bit of sort of where you've come from, what you've added, where we are today, and then where we're going? How much more capacity do we need to add? How are we doing today?
I'm struggling to figure out where our prior capacity was, but I can remember that we were talking about 2,000 racks a month, and it's 2,500 a month. And now by the end of fiscal 2026, June, 3,000 racks a month on the liquid cooling side, total 6,000 racks. So we've come a long way into that kind of production environment. And if you were to think that the liquid-cooled racks or the higher engineered solution with the higher ASP, the market's suggesting somewhere around $3 million-$4 million per rack, and it's scale production, you can see that the revenue support there is quite large. And we've always built capacity in line with what we see long-term demand is. So we're very particular about making sure that they're stepped in process, and there's no outsized CapEx requirements, and scale it as we move forward.
Presumably, we shouldn't build models assuming 100% utilization all the time. You've obviously given a full year revenue forecast, but when we think about kind of the long-term business model, is there sort of an optimized or rule of thumb of what's the appropriate utilization to think of?
I don't think we've ever really given a specific kind of number because the boundaries kind of shift, but half utilization plus would be kind of a nice zone to be in. And so we also take a look at the market opportunity itself, and we've noticed that some of our platform partners are talking about a $1 trillion opportunity or in some cases a $3 trillion opportunity. And currently, we have about 10% share of these platforms in aggregate, I think higher in AI-related infrastructure. And if we were 10% share of these higher numbers that our partners are talking about, well, that's a $100 billion revenue opportunity that is literally right around the corner for us.
If you match that to what our capacity numbers are right now today, you're saying, well, we're in line to serve that, and our customers want us to be able to serve that. So I think we're in a very good position to serve the market, to serve them at the leading edge, and broaden out the base. Like I mentioned before, we have multiple different platform partners. We're not just supplying one particular product. We're architecting designs across the board for our customers to take advantage of it. So we'll be a platform provider of accelerated compute on a long-term basis, and we think that I wouldn't say stand alone, but we look very well positioned for future growth.
Now, the other topic that the investors keep harping on, profits. So I guess it's great that you can sell a lot of products, but the investors also would like you to be profitable. Gross margins have naturally been under pressure in, I think, the whole space. I don't think it's unique to Supermicro, but I feel like it's always sort of tomorrow, tomorrow, tomorrow. What do you see as sort of the timeline and the outlook for improving margins?
One, you pinned it. The dynamics in the market right now are tremendous. I don't think we've ever seen a period where we've had so much change so quickly. From a margin perspective, we're guiding one quarter ahead on a margin front. If you think about what we're doing right now this quarter and with the parameters that we provided, this is the largest, most complicated revenue product cadence that we've ever been on. I don't know if the market really understands or appreciates that. As we grind into that and solidify that as we move forward, we had mentioned that we will leverage this product set as we move forward. The second piece of that would be we're also expanding the offering to the customers.
From a margin perspective, you're thinking customer mix is important, product mix is important, and then manufacturing efficiencies are important. And we can't control all three. One of them is a little quicker to control than the others. The customer mix is starting. We think the customer's mix will broaden out. We see the product mix from our perspective broaden out as well as we move forward. And all these things will come into play, and we've always wanted to operate in a much higher profit zone. And that's the goal of the company, and that's the goal of to get there.
Now, the other sort of aspect of, I guess, cost that's really been a hot topic is memory. In that anytime you build an AI platform, you have to buy memory chips, High Bandwidth Memory, DRAM in particular. How have you managed through that? Just as an aside, I think we've all been surprised watching some of the earnings reports. It hasn't been as bad as feared, but it's our nature to worry about what comes next. So with that.
Yeah. So in general, if we're thinking about what we're delivering right now, currently, and likely in the future, there's a heavy skew towards AI-related systems. And so the DRAM element there is less of an issue for us. We tend to contract with our customers on a, I wouldn't say real-time basis, but we're not forward buying mass amounts of memory. We're not making any bets on our supply chain. So like the period during COVID, we have very flexible capabilities from a supply chain perspective. So we'll manage through it like we always have and try to minimize any impacts as we move forward. So not as exposed as the PC, and it has an impact on the general server business, but again, our customers are really leaning in on the AI side. So not as exposed as some others.
Do you have ability to contractually pass these costs off?
Ultimately, if customers are looking to build out and the price of memory is higher and that's what it is, they're going to be in a position where we'll have to be passing that cost onto them. Yeah.
So you've got some flexibility.
We have some flexibility. Yeah.
So the other thing I sort of think about how the company's been evolving is I always looked at it as somewhere between the white-box vendors and sort of your traditional OEMs, and you've run lean. You generally are a pretty efficient company. But as you're sort of moving up market, moving more sort of branded type behavior, do you see the business model, the operating model changing, either becoming more R&D heavy or more marketing heavy, or are you sort of the culture's the culture and live with it?
I don't want to be as succinct as saying the culture's the culture, but I would say is that we're focused on the customer, okay, and we're focused on the dynamics from the end customer's perspective, and so that changed radically, and there's always something new and better to do and to offer, and there's always a bevy of customers that want more. I like to frame it in the fact that 10% of the market's like the most high performance you can get. 10% of the market might be the lowest cost, the most efficient you get, so somewhere in between, we're covering both of those ends, and we may move more towards the middle, but it really is going to be driven by what customers want, and what we're seeing is platform expansion, right, and opportunity expansion and product expansion for us to address that in DCBBS.
So as we move forward and you think about what the company will be, we'll be supplying AI-accelerated platforms across data centers across the globe. That's where we're positioned right now.
So I'm going to ask a question and admit upfront, it's a little bit unfair. And so I'll preface it that way.
I love unfair.
Acknowledge it. In that Supermicro has sold to enterprises in the past. That's the unfair part is to pretend you never did. But when I think about what the sort of dominant branded IT companies spend on sales and marketing, it's insane. It's a lot more than you spend. So it feels like a big challenge. How does Supermicro become more and more competitive without the same kind of sales and marketing budget for the broader enterprise market, or is that just not on the table for you?
You know, I think the view of what the market will be is changing, right? So it will favor what we are doing. To have the concentration of selling equipment to every mom and pop shop, so to speak, in that branded environment isn't really where we think the market may be going. In fact, in some respects, it could be going a little bit more to service provider and broadening out of the service provider, and then follows on with an enterprise element. And so what's been helpful for us as we serve some of the enterprise customers that are technology forward and the service providers that are technology forward, there's sort of an interplay there. And they're coming to us. DevOps people know that we are building the best gear, the best equipment.
And so the branding element is a factor in the current environment, but it will gradually fade as we do more for our customers. And they come to us and go, "Hey, Supermicro can do this for me." They did it for CoreWeave. They did it for Tesla. They did it for just that's very powerful from our perspective. And it's been a driver of business for us. And you can see it in the growth.
Yeah. It's interesting in that at the time when we did the initiation of coverage, you were the sort of leader among the branded products and sort of chipping away at the white box. So I think I want to ask you about how you think about the competitive landscape and frame it in two dimensions because you sort of are competing against the white box on one side and the branded on the other. How do you think about that? Is it fight on two fronts?
I mean, sort of in a way, right? So the two fronts that you're talking about, I think in general, we view our ability to bring a better product, whether it's for we'll design a better product. So what will potentially could happen is that some of the customers, the hyperscalers are being served generally by some of the white box folks en masse. If they need something more and better and beyond their own design, and we can deliver that, then that opportunity set opens up for us. And again, if you look at the behavior of the hyperscalers and the numerous amounts of new applications that they're trying to develop and the chipsets they're trying to develop, et cetera, we should be able to serve we could serve them if they wanted to.
But the rest of the market, at the same time, needs the skill sets that we have to bring them up. So two fronts. We're very focused on, like I said, the customer and delivering the most performative systems for those folks at the lowest price and the best experience. So I think that, in balance, the market's big enough to support what we're doing in a major way. So if the white box model is, I don't still think it's going to be as static as people might be underwriting.
And then I guess we're thinking about other tools you can employ to grow into some new customers. And things I wanted to check on is you don't have a financing arm. Some companies do. Are there ways you can tackle that without maybe burdening your balance sheet? Or are there sort of you pursuing partnerships for expanding your market, working with other, yeah, I think you did a deal recently with Nokia as an example.
Yeah. So there's multiple different ways, channel programs and partners. And again, it's an example of folks coming to us and looking for our expertise to help them get to market, which is basically what we're doing for our platform providers. But from a standpoint of offering financing, I don't necessarily think that that's a limiting factor. Almost any of the customers can obtain. There's plenty of sources for financing. So I don't think there's anything specifically on the table from a financing perspective. And I don't think it's holding us back at all. Obviously, it's in the numbers that hasn't held us back.
What about partners, OEM partners, go-to-market partners? How important is that?
I think they're important, and we will support if there's opportunity sets with any number of these folks to bring platforms jointly developed to the market to enable them and to enable the industry. So I think we're more than open to that. And I'm sure we're working with many on that front.
Are there aspects of the value chain that maybe you'd want to either become somewhat more vertically integrated? Obviously, you're not going to become a GPU vendor, but technologies like cooling or the racks. It seems as if your sort of core competency is the engineering and the manufacturing capacity. But so many companies sort of go down this path of incorporating elements to improve margins. Is that a logical part of strategy?
So I think from a standpoint of the founder, from Charles' perspective, that doing the best for the customer works and sometimes going down and owning something that you may not need to own because it may be out of favor two years, three years. So we'd be very, very careful in what we would consider a durable asset that would help us out. That said, as the companies move from, let's say, the server level, right, to the rack to the data center level, there could be some things that could fit. But we'll have to cross that bridge when we come to it.
I want to talk a little bit about sort of supply chain resilience. Please correct me if I get this wrong. You've got U.S., Taiwan, and Malaysia as your primary locations.
We have primary, and we have the Netherlands.
Netherlands.
Yeah.
And so if you think about your supply chain, what do you think are the risks? How do you see your ability to manage it? And what are the sort of greatest sensitivities? What would be the choke points? What's your weak link here?
Obviously, I'm not going to be saying that. I don't think there is a weak link. But what I would say is that, one, we're expanding in all locations. We're expanding capacity in all the locations, including in California. We're looking at some new locations. We've been vocal about that. That could be mid-U.S. or whatever. So we're always evaluating what we can do for our customers, where the customers may need that. You've seen a couple of things that we've done. We've established a federal subsidiary or federal entity to address that business directly. We've established a presence in the Middle East. So we're global, and we're expanding that. So if one site is advantaged from a delivery perspective or a production cost perspective, we can flex our capabilities there. So that's kind of how we've been operating, and it's been working pretty well.
So, I guess maybe my constraint or sort of sensitivity question was, I feel like in the past, it was just the allocation of GPUs was what we could point at. More recently, it sounds like it's high-bandwidth memory is very tight. So what are the tightest parts of the supply chain? Maybe that's a different way to think about it.
Yeah. So I mean, obviously, memory is the thing that people are pointing to. I don't think there's anything that's necessarily constraining us in that construct. So our customers tend to get the allocation of the GPUs. So they're aligned with us as their provider. And we've done a fantastic job of delivering to our customers, right? So that's why they keep coming back and ordering from us, and that's where we're getting the allocation. That's the predominant point. So we'll just continue to, whether it was the supply crisis and COVID or whatever, there's all kinds of different challenges that we've managed through every one of them and continue to keep on the trajectory that we're at from a growth perspective.
So if we sort of step back and think about kind of the vision of what Supermicro might look like five years from now, and let's do this on the assumption that the CEOs of AMD and Broadcom and NVIDIA are all correct. Let's just accept that. In that world, where do you see Supermicro in five years?
So if that world exists, and I think it may, it'll be a much larger company. $100 billion is more than on the table. Market share will be probably beyond where we are now because there will be customers that will be flexing in size and scale that needs our capacity that we'll be able to deliver to that and the engineering support that goes around it, and it'll be a very good outcome for the company long term.
And so.
A branded name more than likely at that point in time because we will be supplying all the different platforms.
So if you were sort of encouraging an analyst who's monitoring your company, what metric or metrics to watch? Is it market share? Is it revenue growth? What should they be focused in?
I think right now we're focused on revenue growth and market share, and those are very important because we can leverage a lot of different elements with the size and capacity that we will have, and we can modulate how we want to behave in the end market, so I think those are the primary goals I would keep you focused on.
I want to check with the audience if we've got any questions out there before we. A question I always like to close with is what do you think is either the most misunderstood or least appreciated aspect of the Supermicro story?
I think one of the least appreciated aspects is this product launch and scale is unbelievable, and it's fantastic. That's just people are looking at the margin as opposed to what we're doing, right? That's one thing. So the second piece that I think that people are underestimating is the applications that are going to be utilized in the future, like the bubble talk, right? So they're just starting to get underway. The more applications come to the market, the more the use of this data becomes useful in various different platforms. It'll look different. I think analysts tend to look at the server business of the traditional how things worked in the old days versus what they will look like tomorrow. We're working on what they will look like tomorrow, and people are completely missing that.
Great. Michael, thank you very much for joining us today. Folks, thanks for joining us. Wrapping up with Supermicro.