Good afternoon, everybody. Hopefully you didn't eat too much of that great lunch there. We're joined here with the VP of Corporate Development of Superm icro, Michael Staiger. Thank you for coming, Michael. I think congratulations are in order in terms of anybody who's looked at the stock of late. Why don't we just jump right in? You guys have done some great job breaking out revenue in terms of these rack-scale Gen AI solutions. Now up over half of your revenue today. It's been growing at a rapid three-digit rate. Maybe just bring us right up to speed here and tell us what's driving that success.
Yeah. So I think the best way to look at what's going on in the end markets is that you have a transition from serial compute platforms to more of a parallel compute architecture. And so you're seeing a structural shift in demand from a standpoint of compute. So that's a main driver in the applications. And if you've kind of witnessed one of the larger application providers out there talking about penetration rate with an application of like 6% of the base of a very large organization that's adopting AI to drive revenue gains or productivity gains for its end users, you're at the forefront of a very large shift. And the equipment needs there are a little bit different. So we're aligned to those builds with our partners.
Talk about these particular customers that are going to be spearheading this growth of AI-related serial compute or Gen AI-related services like a Copilot, if you will, or some other services that you're just applying. Who are these customers that you're selling to in terms of the groups of hyperscalers, cloud service providers, and enterprise? Who's the core market for Superm icro?
Yeah. So that's kind of an interesting backdrop. So if you think about it, we're under indexed, in theory, to the hyperscalers. But there's an emerging class of cloud service providers that are standing up generative AI instances that are using GPU infrastructure. So those customers are getting more and more workloads. And it's ironic that the workloads are coming from the hyperscalers or from some of the large players where they are able to take advantage of the bare-metal generative AI platforms so that they can deliver these services to their customers. So the presumption is that the larger organizations that are using them are getting a benefit to them. So they're still early in their build cycle in rolling those applications out to the customers. And we're still probably early in the build cycle of supplying them and making sure that they're on board.
The same thing, if you just take a step back, there's the serial side of the compute architecture. And let's say it's x86-based. And so there's a lot to be done there in that element where you can densify, potentially liquid-cool those in the future for scale x86-type applications. So the fortunate thing is that Supermicro is indexed to our suppliers, Intel and NVIDIA, AMD, Arm. And as customers are rolling out applications that go on those platforms, we're able to develop the latest and greatest solutions that are power-optimized, power-optimized, excuse me, and dense and the highest performance capability, the best memory configurations. And we're able to put those all together rack-scale and deliver them to customers, fully tested, designed, validated, and plug-and-play and ready to go.
We have a lot of different opportunities across the market to serve any compute need, to build any accelerated compute platform. Customers are becoming more and more aware of that. The customer base that we like to serve is the folks that use these systems or use compute infrastructure to drive revenues. You can think of a lot of different companies that do that. We're serving those customers.
Yeah. I think you hinted at a couple of things there. One of them was the platform-agnostic and the diversity of design. Can you talk a little bit about competition and where does Supermicro kind of set itself apart, top couple of things that Supermicro sets itself apart in terms of the other OEMs and ODMs out there?
Yeah. So at the core of it, Charles has developed the building block architecture from the ground up, the motherboard design, power supplies, chassis, enclosures, and all the components to put a system together. And so anytime there's a design or a technology change, whether it's GPU, CPU, or a new architecture like Arm, we're able to put those types of systems together for the end customer. So it's really in the DNA for the R&D front to be at the forefront of computing to deliver to those customers. And so half of the OpEx is devoted to or over a larger portion is devoted to R&D. So we're innovating in product sets. And it just happens to be right now in the compute or if you want to call it, classify the enterprise IT market or the IT market in general, innovation is becoming a material force.
As we shift from serial or we add on a parallel compute architecture. So there's just a lot of different ways to do it. We have the DNA to be able to shift quickly. We haven't built a business model on one particular arena. We don't have a software and services infrastructure necessarily that's indexed to, say, x86. So we can flex pretty quickly in these areas. Downstream, as we gain share in these new market opportunities, we can look at adding those types of services, software, storage, switching downstream to the portfolio. But that's in the out years. Right now, the focus is on the compute platform and gaining share there.
Where are we? So you talked about CSPs, maybe underindexing to hyperscalers currently. Where are we in the enterprise build-out of these solutions? I mean, that's a large unit TAM in terms of, say, something like the G2K. I know you have a couple of large customers now on the enterprise side. But where do you kind of feel when you're getting these bookings in, where are we in terms of the early stages of AI being pushed by the enterprise?
Well, it seems like on the enterprise side that everyone in boardrooms are looking at this and saying, "Hey, look, can we use this technology to drive revenue or get productivity gains?" And there's a lot of tasks out there that could benefit from an AI kind of infrastructure. So if you take a look at that, boardrooms are saying, "Hey, look, let's evaluate this. Let's move forward." And I wouldn't imagine it's an easy shift for the average enterprise. But there's a significant amount of large-scale enterprise customers that are interested in evaluating and testing and ultimately will be moving forward with rolling these platforms out. Some of them already have. So from that perspective, I think you would imagine that the first step would be to go to a generative AI provider, a CSP, so to speak, and test or validate.
And then ultimately, you would potentially bring that workload into your enterprise as you'd be able to plan it out on a move-forward basis. You'd know how to scale it out. So you'd have the opportunity to either use the CSP or do it in-house or do a hybrid model. We're all familiar with that. And if that's the case, we have an opportunity to serve the CSPs as they emerge and continue to expand and offer these to the enterprise, their services, and as well some of the larger-scale players. You think some of the verticals, the financial services, etc., would be a perfect candidate to take advantage of these applications. So it's really application growth that will drive the infrastructure. And so we're still in early days of the application growth there as an opportunity.
Supermicro has, you've disclosed some pretty chunky customers and in the past, too. It's not like you don't have other customers to fill some of those gaps. Walk us through the visibility that Supermicro has. You talk about increased backlogs of non-cancelable orders. Just for investors, I think they'll get a lot of questions about what Supermicro's kind of visibility is and how they kind of manage this backlog.
Yeah. This seems to be a pretty important topic.
What happens if you start growing 3% or 400%?
So where are we in the cycle?" That's a common investor question in the visibility. So the way we're looking at it, we're looking at it through the lens of the customer. And the customers typically are, like I said, again, using the platforms to generate revenues for their firms in some cases or save money in many cases. And so they look at their workloads. And they have kind of a vision or a plan. And as they roll these things out, it's a multi-quarter, if not potentially multi-year view as they expand, particularly along the service provider arena. And when I mentioned service provider, I want to segregate hyperscaler, the couple handful, to the more focused these applications. So they have a visibility and a plan. And then that maps into the product availability, whatever the platform they're going to use, and the timing of those platforms.
When you take a look at all these things, you mix it all together. We're talking to the platform providers on the compute side or the GPU side. We're talking to the customers. You put those things together, you kind of have an extended view as opposed to in the old days when a server was more of a commoditized element. They were just looking for capacity at the edge. Looking forward, it kind of builds the view of where we're going and where the customers are going. As long as there's an uptake of the applications, we'll have pretty solid visibility as we move forward.
Okay. And then one of your OEM competitors, if you want to, you might not allow me to call them that. But Dell had a solid report last week. And the callout here for Superm icro is that they talked about lead times improving on the GPU side. I wonder if maybe Superm icro had a comment on lead times in terms of the supply chain.
Yeah. So I kind of want to be a little bit cognizant of that.
Comment on Dell. Just.
No, no. But I think if you just think of it in context of what's going on in the end market and look at the market size and position and where the innovation is occurring, we're clearly innovating in these product categories and have been in the market for multiple, maybe two years ahead of the competition. That all said, I think current standards say that we're 6% or 7% of the total market, right? And so some of their competitors may have larger share. We're growing at a faster clip. And if you really take another step back and say, "Hey, wait a minute. This infrastructure build-out is secular change," then you've got to think of this market share statistic that investors like to lean on as a little bit, I want to say, misleading because the TAM is growing.
We've heard some of the chip suppliers on that platform front, GPU, CPU, talk about different TAM and an expanding TAM where I think investors are like, "Wait a minute. That's confusing. It's really big numbers. Where are you coming up with that?" And if you think about that context and we're 6%-7% of the market and we're growing at the rate we're growing, well, clearly, we're taking market share. And we have a view that we can get to a materially larger footprint from a market share perspective. So our focus isn't so much on where are we relative to our competitors as much as it is, can we bring this innovative product? Can we bring the latest and greatest to that customer? Can we integrate it for them? Can we get it scaled up for them in a really fast fashion?
As long as we can do that, our growth rate should be in a very healthy rate. And over time, we should have a meaningful share. So instead of getting into the elbowing with whose versus what, there's plenty of share here for us to be a clear leader. And there'd be room for one or two others to do well. So I think we'll keep it at a gentleman's view of how this is going to evolve. But the underlying thing of this is we are the innovator in this space. And we will help our partners get their products to market so that the customers can actually get the benefits that they're looking for. And we can get rid of some of the menial tasks like taking notes in conferences where they could just be immediately transcribed. And we can search it.
So stuff like that, you don't even think about, right? So there's revenue opportunities galore. And I think in a couple of weeks, we might hear even more of what the future may look like from one of the leading innovators out there that we're closely partnered with.
The conference coming up, GTC.
Yeah. Being gentlemanly-like and not calling anybody out, right?
We'll see what comes from there and its impact on Supermicro. I appreciate the tease. The question was on lead times. So are they improving?
Yeah. So on that front, we come back to not necessarily calling out anyone specifically. But there's clearly been an improvement with respect to the availability of some of the components that we need to put the systems together for the customers. And the messaging from the industry has been that they will gradually continue to increase as we move forward. And I imagine as we move forward and there's new versions coming out, there'll be pockets of there's some more availability here. So there'll be a lot of noise in the end market. But overall, the trajectory and the direction is to satisfy the demand in the end market. And the demand for these types of applications could be quite large and expansive.
The 6%-7% share, just to double-click on that, that's a general number you're citing, right, in terms of that includes general-purpose servers? Because that wouldn't be fair to qualify that as your share in AI servers.
Yeah. So on the general side, if you want to classify it as that, when we talk to the street in our earnings reports, we talk about what the majority is and over 50%, etc. So we don't want to be known as just an AI-focused company. We're focused on innovation, right? So let's not think that there isn't any opportunity to innovate on the x86 platforms because that is an important massive infrastructure out there. And so there's plenty to do. And like I said, once again, from a standpoint of solving customer problems with a total solution, that's what we're focused on. And that requires innovation and an ability to bring some of the leading-edge technology into something that may be a refresh cycle. And some would argue right now, we're at a cyclical low on IT traditional spending, right?
And we're at a secular emergence of parallel compute or GPU compute, right? You put those two together. And what happens? And so I'll backfill that one step. So if you're going to do AI in the enterprise, theoretically, you're going to want to populate your language models or your training models with the data that you have collected over the years. And so there's plenty of firms out there that have cold storage that they push off the side, different tierings of storage. They'll want to bring all that data in and train it. So if you want to bring all that data in and train it, there's got to be some sort of cognizance of what your x86 platform is. And so you're all of a sudden bringing that data that's active in the GenAI model or the inferencing model. So it's active there, right?
So you have more data kind of circulating in the entirety of it. So there's a catalyst here for the entire industry to think about retooling and re-architecting. And once again, let's go circle back. What do we do? R&D, innovation, time to market with these newer-type leading-edge products.
The re-architecting of data centers and install bases would definitely be expanding your TAM. Are there any questions from the audience, Carol? We have plenty more. Supermicro has been also, again, a lot of good disclosure from the company. One of them is on the amount of capacity that you've been building for these rack-scale, plug-and-play solutions that are typically targeted at Gen AI and GPUs. But I understand next to the x86 opportunity, that you've articulated well. Whereabouts are we in terms of I think the latest is 4,000?
Yeah. 4,000 racks a month.
Yeah. Per month. So about.
1,000 on the liquid-cooled side.
Where are we in terms of capacity utilization there? And from my math, we're not close to even being over half utilized in that. And you guys are talking about building more. So walk us through the indications of demand that you're intimating to the market in terms of building out more capacity.
I think we've been pretty clear advertising that we have capacity bills. We have Malaysia coming on. We've talked about some of Americas locations that would come to be announced, so to speak. But where we're trying to position ourselves for more demand downstream, we've always been scaling up capacity ahead of the theoretical demand. And now, we're just in a much better position. So if you think about the three things that really are pretty big drivers for us, it's R&D capabilities, our sourcing capabilities, managing the supply chain, if you want to call it that, and the manufacturing now at the scale level. So these customers and these applications are scale applications. So they need our end customers will be like, "Hey, do you have the capacity to build?" And they know that there's other customers out there.
So can you stand up, let's say, 1,000 racks for us? Another customer comes in. It's like, "If we don't have that capacity, then we're going to miss out on the opportunity." So at the same time, all the three things I just mentioned are really a competitive moat for us because I argue who else is putting together this kind of infrastructure to support the end markets? And as the end markets continue to evolve and the need for new GPU, new CPUs come up, and we're indexed to being able to market, and we have the ability to scale them up and get them to the customers to test and validate and drop them in, we're talking L11, L12 kind of validations. We're going to gain share because of our capabilities. And I think it's widely known that we have those capabilities now.
More and more customers will need our services and need our products. They'll be able to build a much larger ecosystem than people would expect.
This might be related. We talked about CSPs, hyperscalers, and enterprises. I think a third bucket or fourth bucket, rather, that folks have been talking about in this industry. I should have asked earlier about sovereign AI. Is that something that you're building for in terms of these capacity additions when you're not even at half capacity utilization of the targets that you have related to sovereign AI?
So this is an interesting topic. It don't have anything necessarily to specifically say. But if you do take the mantra that large language models across the entirety of the different country bases - and some countries are more forward-looking on that - that we have the ability to serve them because there will be scale applications. And we would be indexed to helping any of those out. And I'm sure we're having conversations as we speak with some, or if not many, of those to drive business. And that would be a nice thing to have, so to speak. But we have the capabilities of serving those customers today. So we welcome them.
The only reason I asked, Michael, is because it would seem like those particular opportunities would want to make sure before they start going down the pike of designing, which is part of your expertise, that you have the capacity to deliver.
I think we will have the capacity to handle that. So those customers I don't know if you mentioned that they were designing themselves. But we could design something for them pretty easily.
Pretty easily.
Yeah, yeah, yeah. And be able to deliver quickly because I think no one wants to be left out. And there's leverage to be gained at the country level. And there's many state-owned entities that would probably leverage off of that, think the utilities, telco kind of situation. So there's a lot of opportunity there that's not necessarily factored into people's models.
Any audience questions? No? Liquid cooling. So these things create a lot of heat, suck a lot of power. So where are we in liquid cooling? Let me start right up front. Do you have liquid- cooled installed today? Do you have customers that are utilizing your service today? And what kind of ARPU uptake? I won't ask about penetration rate because we're very early here. But do you have something in the market today? And what kind of ARPU lift or what kind of pricing lift is this going to bring to Super Micro?
That's interesting. There's a lot to talk about on that front. But the one thing that I want to mention is that if you really do a deep dive and you look at what we were doing in the x86 world, we develop our own power supplies, right? So I think the stats are a 10% better power draw, period, the lower power draw per unit than anyone else in the market, 20% on the blade side. And the enterprise didn't care, right? The enterprise IT manager's like, "Well, that's not my budget," right? Well, so the core DNA is like, "Hey, power," right? Power density, performance, I mean, all these elements. And so now, if you look forward and you see the AI complex is power-hungry, right? It's performance-hungry, right? It's square foot. There's capacity. So we're indexed to that. This is in our DNA.
So we've already delivered liquid- cooled solutions more than the HPC side of different flavors. And it's small as a % of revenues as we move forward. But if you think about what's going to happen ultimately with the power envelope exceeding the 1,000 kW per unit, customers are actually asking us, "What can you do for us? What can you design in this situation so that we can turn off the AC and get rid of the air cooling and the 40% power that we're sending into these units? And we can put liquid in these things. And we can use the electricity that's available to us to power more systems because these are scale applications." So there's a lot of focus on that. And from a standpoint of what does it do to the model?
Once you get to the highest level, it's another piece of the pie we're solving for the customer. So we go back to the solution, right, innovation. It's another thing we wrap into it. It's a whole complete thing that we're going to deliver to that customer. So we would like to believe that and would hope to think that there would be some enhancement from a model perspective, an ASP perspective. It is costly to do. And there's a TCO benefit to the customer. So I can't really give guidance on what's going to happen to the model in 2025 as a result of liquid- cooling. But what I can say is that our capability of doing that for our customers as they ask for it is going to keep more customers in the seat. And I'm sure that we'll be able to monetize that to delight our shareholders.
There's lots of puts and takes. There's a lot of discussion points about it. It's something that we're 100% indexed to. We're going to take advantage of it as it develops. We'll see commercialization ultimately in the form of a couple of larger clusters that'll be delivered using this. Obviously, the ecosystem has to be sort of in sync. It's sort of in sync now but completely in sync downstream. I'd love to give you a much clearer answer than that. It's something we'll be talking about many quarters down the road.
Well, I mean, the power draw of 10% or 20% better than your competitors is a lot more important when you're dealing with AI as opposed to back in the day. On the liquid- cooling, who are your competitors there? And do you have the same type of moat that you have in other areas in terms of design or power?
So I don't think we think of direct competitors in liquid- cooling. I think when we're engaging a customer, they're like, "Look, we have an application. We need the best horsepower. We need the best solution here." And so the liquid- cooling becomes an element of what we're doing for the customer. The conversation doesn't start as, "Can you liquid- cool this?" It starts at, "Hey, what's the platform? Is it whose compute platform we're going to use? How are we going to put this together? What's the networking envelope look like?" All those factors. And so we put all those factors together for the customer. So the liquid- cooling is another element. It's like, "Oh, wait a minute. You can do that for me too. And you can solve these oh, and I can get more. Okay. This is perfect." Boom.
And then the engagement with the customer is completely different than you have a part on the shelf, "I need how much is it?" It's a complete element. And there's a loyalty on the customer end. And the retention rate of our customers that are dealing with us that we're providing these platforms to is solid. So we have a really good working relationship with the customers. Surely, price is a part of the component, right? And it's a competitive market. But we're going to use liquid- cooling to advantage ourselves on a long-term basis.
Okay. Shifting to the model if there's no other technical or market questions. Gross margins have come in a little bit. Just walk us through the puts and takes and the levers that Super Micro looks at in terms of expanding off the 15% gross margin level currently or maybe even going lower.
Yeah. David, who couldn't be here today, the CFO, was pretty clear about the advantage we have relative to the margins in the early part of the lifecycle of a product. So as we move forward, the lifecycle of a product, there's going to be a little bit more competition at the margin. And so that has an impact on price. But what we know as we move forward, there will be first-to-market in many, many new variants that are coming downstream. So as we take market share in the current parameter of gross margins in the model, the expectation is that we'll be first-to-market with some newer products. And that will be helpful from a margin perspective. So more to say on that in future quarters, I suppose. But we're operating comfortably in this 14%-17% range. We expect to be in that range.
We strive to be better in that if we can. So let's keep it there.
Have you disclosed if liquid- cooling would be gross margin accretive?
No. No significant update from that perspective.
Okay. All right. Recently, maybe it was my final question, you've just completed a very successful secondary equity raise as well as a recent convert. Maybe you want to give the investors an update in terms of usage of this cash and your overall working capital needs.
From that perspective, we thank you for your help in both of those raises, by the way. Well, not necessarily with KeyBanc's help. Let's put it that way. It was to fund continued growth and working capital. The velocity of the business has been such that we wanted to make sure that we were well-prepared for the new product launches that are coming and some of the customer builds that we anticipate to win into 2024. I think these are really well-received transactions. It puts us in a better competitive position and a stronger balance sheet. I think it was expected by the street for us to do those things.
Let me squeeze one more in if no one else had. How about the general-purpose server market, the general state of give us insight into the general state of IT spending. You have meaningful share in x86 general server market. What are you seeing there?
So that goes back to the earlier comment that if you expect the end users to repatriate data to fulfill the needs into the Gen AI element, that the broad enterprise market, like the IT managers, if they're even as they evaluate Gen AI or any of those applications, they're going to have to really take a good hard look at what's on the floor today. And I think some of our partners have mentioned four- or five-year-old infrastructure sitting out on the floor. It's probably not going to be a good idea to let that sit another year or two. So to the extent that some of the newer core counts, denser x86 platforms, are starting to hit the market and masse, and they've been tested and validated, we would expect that there would be some refreshes. And once again, I'll go back to our market share, right?
If we're 6%-7% of the market, whatever the number is, it doesn't matter to us if there, we will find customers that will need to upgrade. So we'll continue to focus on helping those customers out and have a broad, diversified mix relative to the market. And if there's any innovation on that front, real dense racks and scale customers like, "Hey, I want to move to this," or even if they want to inference on an x86, we'll be there to help them solve those problems.
Great. I'm going to thank you for your time, Michael.
No, awesome. Thanks for having us.