Super Micro Computer, Inc. (SMCI) Earnings Call Transcript & Summary

It's 2024 TMT Conference. My name is Asiya Merchant. I cover the tech hardware and tech supply chain companies here at Citi Research. Very pleased to have Mike Staiger here from Super Micro. We also have Krishna Shankar, who is VP of Investor Relations -- of Finance and Investor Relations. Mike, here, is VP of Corporate Development. Very pleased to have them here with us. This session is for Citi clients only. And I'm first going to turn it over to Mike for some opening remarks before we jump into Q&A.

Hope, you can hear me. Thanks for having us, Asiya. First thing, I hate to go to the safe harbor, but investors should refer to our cautionary safe harbor statement regarding risk factors and forward-looking statements that's on our website. Also recall, last week, we disclosed that we needed additional time to file our 10-K, and we said in our filing that the Board has assigned a committee to diligently conduct a review. We also said, based on what we know, we don't expect any material changes to our fourth quarter and full year '24 results. So I appreciate everyone's understanding that there's nothing more that I can say about the topic at this time, and we remain focused on customers and executing on our ambitious business plan. With that, we can talk about the business.

Okay. I'll be leading most of the Q&A. And towards the end, we can have it open for investor Q&A. [Operator Instructions] So Mike, thanks again. First question, we're pretty much asking all the companies here. It's just about end demand. You kind of alluded to that right now in your opening remarks as well about the ambitious demand outlook that you guys have. Maybe just given you guys recently reported, how would you characterize the demand environment and how Super Micro is positioned relative to that demand?

I think from that perspective, the numbers speak for themselves, 110% year-over-year growth kind of gives you an idea that what we're providing to our customers is paramount and high demand. So it's mostly AI-related products. We said a majority of our demand has been in that category. So I think it's been pretty robust.

Okay. And then just given your market positioning, clearly, AI is the one that's driving the bulk of your demand, the majority of your demand outlook above the industry what do you -- how do you think about your market share positioning relative to your peers?

So I think that was pretty obvious relative to our peers, we're clearly leading. We're well out in the forefront. We're doing it with innovative products, new products and new technology. So we think our market position is quite healthy.

Okay. And just the conversations that you guys are having with your end customers. What are they focused on? Maybe just relative to where you -- when you report it, how confident are you in those conversations?

So I think in general, the customers that we're dealing with, the majority of the customers are worried about cost -- total cost per watt per compute cycle. So from that perspective the conversations are revolving around how can we get more out of the products? How can we get more density? How can we get more power efficient, thermal efficient. So the conversation is just really is about liquid cooling and data center architecture. So those conversations are right in our wheelhouse from an innovation perspective. If you think about what we've been doing with the Hopper or the DGX reference architecture that NVIDIA first roll out with the Hopper. We've been able to double the density, liquid cooled so that customers can basically get the most performance out of it by throttling those -- the performance on those GPUs. And so from a technical perspective, we've been able to do quite an excellent job, and that's one of the reasons why we're gaining share.

And again, as it relates to end demand, I think there's just a lot of concern. Is this sustainable or not sustainable? Just based on your conversations with your customers, what gives you the confidence that this outlook that you guys provided for your next fiscal year remains well on track?

So from a demand perspective, I mean it's kind of interesting, it's kind of missing from the general public. The larger -- the CSPs, the GenAI focused CSPs, they are getting a lot of workloads, not only from the hyperscalers themselves for various different reasons. But from the enterprise. And if you think about an enterprise setup and they usually have -- they have testing dev on their X86 platform. So they're all built out on the X86 platforms. And they usually reserve 10%, maybe 15% for -- to test applications, et cetera, in that footprint. An interesting thing here is that few, if any of the large enterprises actually have a GenAI footprint built out. And all of them are now testing large language models, industry vertical models or anything that's like specific to their particular business to drive revenue or to save on costs. But from that perspective, our customers, the enterprises using the CSPs and so the demand cycles at the CSP level are pretty accelerated. And so as they develop their applications within their footprint and they're going to roll them out to their user base. They'll either have to build out a GenAI internally for their own regulatory reasons or they'll continue the CSPs. So under the covers, the demand from an AI perspective is really broad-based, and that's a really good underpinning of our business because one, we could serve the CSPs. There's more of them coming. There's new ones forming to deliver industry-specific models as well. And then you have the enterprises that -- from a regulatory perspective, and I'm thinking financial services might be one of those will ultimately build out. And so that's a pretty big backdrop from a demand perspective. And even if they just continue to stay in the CSP envelope, that will get bigger. And what will likely happen in that space is that you will continue to need low cost per watt per compute and from that perspective that's where we shine, where we're able to take whatever our partners are able to deliver from the compute element, we can package that in a total system, total rack, total data center elements and provide that to the customers so that they can drive their cost of service down to the end customer. Now if things continue to travel in the distance in a way that we see right now, the next phase will be on inferencing side -- inferencing point. And the enterprises and end users will likely need devices from that perspective that probably look a little bit more like a general-purpose cellular, so to speak. And from that perspective, there will be a mass unit requirement. So we'll potentially have a customer that has an AI footprint on the generative side, and we'll have customers that will need the units to do the inferencing. So they'll kind of go hand in hand from a demand perspective. So that's -- I think that's how the market will likely unfold. There's just a lot of noise right now with respect to is it going to be sustainable and the benefits of the technology are pretty clear, and we think that there's a lot of runway ahead.

Okay. How do you think about drivers between pricing and volume as you -- as we go through these chip transitions from your major chip suppliers. Do you view that as a pricing opportunity or as a volume opportunity?

It's kind of an interesting question because you would have to think that the market that we're serving stay static, right? And so if it's just one particular platform that serves the entire market to get into that like the old commodity off the shelf, COTS X86 kind of platform. And what we're seeing now is a large amount of variance of different form factors, different setups to address different workloads, and you can just take a look at some of our partner's product road maps. You can take a look at what the hyperscalers are doing, and there's a massive forking away from just the standard one unit serves all. So the pricing and volume element there becomes a little bit different or it becomes more spread out. And when we hit a little bit earlier was the inferencing, we'll probably be more of a volume kind of a game potentially, but we'll see. But so like, I don't think you can look at it from very static pricing and volume perspective. The products that we're delivering continue to change. I mean, very, very quickly and our ability to keep ahead of the market by being able to put our partners' products on the floor, in volume, at scale so that customers can run their applications is a pretty unparalleled position right now in the marketplace.

Yes. The first-to-market is generally what Super Micro is known for. That's fair enough. With talking about which -- I mean, I think you guys talked about it a little bit on your last earnings call is about the Blackwell delays and how we should think about that as it relates to your business and the revenue ramp ahead with the Blackwell?

Let me just say one thing about first-to-market, right? Like I think the characterization is that we're able to put technology into the marketplace sooner than anyone else. And the ability for the rest of the market to actually keep up is becoming more and more of a challenge than we've ever seen before. So it's an underestimated thing. So referring to the Blackwell element, it kind of goes back to our DNA with what we've done with Hopper. So I think it's getting common from an industry perspective that if you want to call it a delay, I'm not really sure because it's a moving target on when availability and shipments. And I think the -- it's very widely known in the industry that select few customers would be getting those in volume in the initial phases. Our customer behavior has been to build out and put as much compute on the floor, cool it, liquid cooling, so they can get the most performance out of what they have available today, which is the H100, H200. And so our forecast kind of reflect that. Our thought processes reflect that. And over time, all of our customers that are concentrated on these types of footprints are going to be ultimately migrating to whatever Blackwell variant is available. And the key there is that we are understanding that there are several. Interesting thing is it's not even on the market that I know of. And we're already talking about what it's going to be and how it's going to be, and we haven't got to that point. But whatever it is, we will be able to take that platform and enhance it for our customers so that they continue to come to us for the low cost per watt per compute, the most efficient systems, et cetera, the most performant and the densest. So we'll be well in that discussion point.

Okay. Well, how would you characterize the current supply environment then for these GPUs?

Supply affects a lot of different arenas from our perspective. We kind of mentioned on our call that we had some supply, the challenges with the liquid cooling element because we were scaling up some pretty large builds using basically a new supply chain, a new manufacturing set up, new manufacturing lines. So supply is ramping on the liquid cooling side to support our forecast. Most people think of supply in the forms of what GPUs are available. And from that perspective, I think it's pretty well known. It's looser than it was a while back, but I don't have the exact numbers for you today. And we've been able to supply our customers and sustain growth rates that are well above the industry average, well above our peers and competitors. I think supply is a challenge and it changes all the time depending on what we're going to be delivering to the customer. So it's an ever-moving target.

Okay. And on the DLC-specific, I think you just talked about it? The supply and DLC still pretty tight or it's improved?

I think it's improving. I don't know specifically if there's anything in particular, if we do a product change into the next variant, what that would need. But everything changes from a standpoint of the new product that we're delivering across any of our partners. And so aligning our suppliers to match our forecast, our customer forecast is somewhat of an art -- underappreciated art. And we've been doing a phenomenal job, if you just look at the track record of the growth of the company and it kind of supports that.

Yes. Maybe if you can dive a little bit into your competitor -- your competitive advantage and you talk about the modular approach. Some of the other competitors, peers also talk about something similar. So maybe if you can just -- how you guys think about Super Micro's moat as it relates to -- you talked about speed to market, you talk about this modular building. How you guys differentiate yourself from some of your peers on that one?

Well, I think it's clear the differentiation is pretty pronounced at this point in time. I don't think we're the only ones that are doing the double density, if you want to call it, or plus on the Hopper itself. We're the first to roll out a large volume of liquid cooled racks. So from that perspective, our time to market, our ability to take our partners' products and make them better is a pretty significant advantage. We are -- control our own engineering. We control our own manufacturing. I'm sure we have partners that help, but we design everything that we do and we integrate that from our perspective to put the best products on the floor of the customer. So there's a lot of art to this. It's a really difficult challenge anyone here to build the liquid-cooled rack at the densities that we're doing. I think they find it pretty difficult to integrate all of those to get the right elements in place to satisfy customers. So I think the competitive advantage is our ability to flex and use our modular design to extend our advantage. We're also moving in a direction where we'll be able to deliver the entire data center stack so to speak, our data center building block. And there will be more color and more comments as we develop those products for our customers downstream.

Okay. So by that you mean storage and ...

I think there's -- we're talking about going from the setup of the data center to the -- from the tower liquid cooling to the compute platform that we do already supply storage, to switches, software and service and wrap that complete package around for customers.

Okay. Maybe on DLC, just remind investors where you are from a capacity standpoint? And as it relates to, like you mentioned, these Blackwell variants that might be there, there's some chatter that maybe the NVL36 might be preferred, which tends to be more air cooled versus liquid cooled. How does that kind of factor into your outlook for DLC?

Well, I think there's...

Direct liquid cooling.

So from that perspective, one, those products aren't necessarily in the mass market. So there are some customers that may not be set up for liquid. We can help some of those customers for sure, depending on who, how, where, where they are. So the air definitely is an option and the liquid, like a more discerning customers that have very -- a big concern over power consumption and they want to direct the power away from a traditional air cooled data center to a direct to chip liquid cooling environment. There's plenty of customers that want that. There's some discussion that that's the forward motion for almost all systems in the future which is right around the corner and so we've spent a considerable amount of time and effort and investment in developing our liquid cooling technology and there'll will be many, many different variants as we move forward and we'll be able to put anything to on the floor, liquid cooled, we have X86 liquid-cooled offerings as we speak. So over time, when you want to compress the X86 footprint and get denser and run hotter, we'll be able to liquid cool those alongside of the GenAI platform. So we're not just thinking about AI system, so to speak, we're thinking about the total data center. So we'll be able to offer a customer their serial compute, payroll compute, basically X86 platforms or AI platforms and whether it's generative or inferencing, it won't matter. We'll be able to wrap all those around and integrate those for end customers.

How do you measure where you are relative to your competitors on the liquid cooling and direct liquid cooling format? Is there like some industry standard that you compare yourself to what your peer offerings are. And therefore, you say, okay, you're the leader when it comes to...

I like the way you're thinking about it, but I don't think we think about it that way. I think what we think about as we look at a system or [indiscernible] does and figures out how can you get the most out of it and the format, what it would look like, whether it's bladed or racked or however and where you get the most performance out of that system and then be able to deliver it so that the customer can use it and plug it in, turn it on and not to worry about architecture of it, not have to worry about the design of it, right? So it's become way too complicated at the data center level for the average operator to figure out how to get the most benefit out of the systems that they're running and it's efficient to get some time to market. They get some time online to serve the customer. So if you think about it, if the service provider is not running like the highest performance systems, they're potentially going to get their lunch eaten by their other competitors. So if we can help accelerate that and accelerate them, then that becomes a standard, and then we'll build to that standard in volume. And when the platform shifts or changes, we will take that and we will look at that and develop another platform. So the good news is there's so many different versions out there right now that it benefits our business model. And so that we can really do wonderful things for the customer and they can have a pretty solid experience.

Okay. Great. One of the concerns generally as it relates to OEMs in this market is AI, maybe a lot of the primary growth of AI hyperscalers is being supported by ODMs in Taiwan. When you compare yourself to proxy ODMs, where is Super Micro's competitive advantage specifically? And as you talk about gaining share in this AI TAM, how do you think Super Micro is positioned in terms of gaining share there?

One, we don't compare ourselves to OEMs, right? We don't do what they're doing. Like you said, what we're doing is we're designing and developing systems, the most performative systems for our customers, whereas an OEM is essentially getting a recipe from a couple of select few that are designing their own systems. And so all the R&D that goes into those systems is born -- those costs are born by, let's say, the hyperscalers, the Microsofts and Googles, et cetera. And so what we're doing is designing our systems and not taking a recipe from someone else and just -- and building those for those customers. So I think the comparison there is clear that that's not what we're doing. And I think it's showing up in our customer results. There's a significant amount of customers that are scale customers that require a certain set of parameters are designed. If we do that work for them, and that's kind of leads us into the differentiation between us and them as well as the OEMs because it's -- the OEM model is a little bit different than what we're doing as well.

Okay. Speaking of customer concentration, can you maybe help us understand what your current customer composition is, what does customer concentration look like today? And as you're talking about inferencing, how do you see that customer concentration -- customer composition are evolving?

So customer concentration, I think in fiscal '24, we only had one customer over 10%, which is pretty phenomenal relative to the growth rates of the company. So we have a pretty broad base of customers. So we have a few larger customers, let's just put it that way. They are doing some pretty amazing things with their technology that hopefully change the world. And those customers are demanding the most performative, the densest, compute platforms, the liquid cooling, the lowest total cost per watt per compute cycle. And so we serve those customers quite well as well as a broad base of enterprise customers. So the composition would be the emerging CSPs, which I think have not just emerged but are quite well known on the stage now, and there's more of them coming with large enterprise customers. We have OEM customers. Everyone knows, it's familiar that we have Nutanix, which is one of our larger partners and their systems. Their software is running on our systems and quite a few of the Fortune 500. So pretty pervasive. And if you think about the CSPs, if the large enterprises are using their models and they're running in the CSPs, they're running on Super Micro equipment, right? And so the user base or the community base, they're very technical, the technical elements, they know that it's Super Micro, that's powering some of these large language models. So from that perspective, it's been driving new logos for us. There's also been quite a bit of interest from sovereign nation type accounts looking to update or utilize GenAI across their platforms. And we've made a few public announcements about that in the past that kind of give you an idea of flavor where we are with those types of customers. So it's pretty broad, and we'd like to get it broader. We said on our last call that we were looking to double down on the enterprise. There's a lot of aged equipment out there. A lot of enterprise customers are looking to do new things. And if we can densify the X86 platform, it'll help them get ready for the AI if they're going to do GenAI in-house. And the inferencing units, we have plenty of SKUs, and we'll be addressing that in those end markets as those applications come up. And so there's quite a broad base of opportunity sets for us to continue to fuel our growth.

What is your current market share with these enterprises? Are you able to share that? Or...

From an IDC perspective. I think we've rolled into the #2 at 9-ish percent. And I don't think that accurately reflects it because the IDC numbers don't categorize the OEM revenues that we have. So I think it would be -- I don't know if I can say its larger, but I think if you factor that in, it would be a better number, from 2% or 3% share a while back. So we've clearly -- let's just say, from an OEM perspective, coming to the #2 position. So that's -- and our goal, as we've said this publicly many, many times, our goal is to be #1. So we're pushing ahead. And if we can use our innovation in this new era, which is very -- is changing rapidly. I think that we have a really good shot of owning a pretty big chunk of the compute. It's also important to understand there's a huge market. It's very fragmented. No one has a majority share. And so I don't think that's happened in the past. I mean, since the days of IBM, maybe if you want to categorize it that way. So there's a lot of opportunity. So this whole notion of like share, shifts and movements and et cetera, we steadily are increasing our position. And if you take a look at that cadence, I think that should give people a pretty good level of confidence where the direction we're heading.

Great. Just going to ask if anybody here in the audience has any questions, please raise your hand. Okay. I will say more that I'll be talking about then -- margins. Top of mind for everybody. Is AI, how detrimental is it to corporate average margins? I know you guys talked about it at your last earnings call. So maybe you can just remind investors what happened there and how you guys are thinking about as your direct liquid cooling ramps. How you guys are thinking about margins there?

So I think we were pretty vocal about what our view was on our last call. With respect to margins, there was a pretty considerable investment in the liquid-cooled technology jumping out ahead. There was a lot of investment in the supply chain, a lot of investment in the capacity element and getting the lines. They're probably a lot less efficient. We also talked about strategic customer wins and strategic customer pricing. And it was all about customer satisfaction, and we wanted to get a few lighthouse accounts under our belt. And our view is that if we can please those customers as we iron out the cost inefficiencies and the new technology because typically a margin hit in new process, so to speak. As we roll forward and we stand alone, the opportunity to recapture some of that margin, recapture potentially some of the pricing elements and get back to -- we mentioned getting back to our corporate average range of 14% to 17%. So I think we've been fairly clear about that despite some questions from the community.

And just as investors kind of think about your ramp there for direct liquid cooling, and just if you can expand a little bit on that, like what is that scale that you're referring to that direct liquid cool needs to be as a percentage of your business before perhaps you get back to corporate average here?

I don't think we have a lot to share with that number. What I can say is that we've been investing heavily in the capacity. And one of the keys to this is the AI product set, GenAI, in particular, is a scale build, this is a scale use case. And when the customers all in, the systems tend to be quite large. And we have a few customers that have seen large appetites coming from a lot of different directions, a lot of enterprise into the CSP, a couple of interesting applications from innovative -- an innovative individual. It's just trying to change the world with AI, let's put it that way with multiple different companies. And so we're trying to -- we've been aligning ourselves to about 5,000 racks of production by the end of fiscal '25. We said 3,000 of them would be liquid cooled. I think that's up from 1,500 or so. And we think that there will be a clear need for those. We wouldn't be moving in this direction if we didn't take the hint from our partners who are all heading in that direction of needing liquid cooling for their GPUs or CPUs, and we wouldn't be heading in that direction if we didn't hear it from our end customers who are -- I don't want to say demanding, but are very vocal about the direction they're going in. So I think we've aligned ourselves with a customer for customer success. And we're working with our partners to enable them. So if you think about what we've done with from an NVIDIA perspective, instead of having 1 unit to the end customer, like 2. So it's a pretty favorable proposition that we are able to leverage their platform and be able to sell more of their product into the customer base. So I think that's a pretty powerful thing as we move forward.

As it relates to penetration of liquid cooling as we start to go down more and more of this path, what are some obstacles for data centers to get more? Is it space, obviously, it seems like there's a lot of architecture that has to be redone within data centers to enable direct liquid cooling. And what we hear is there's not -- people aren't there yet. So what else needs to happen?

So I don't have the specific numbers of like who's ready, who's not from a standpoint of what they can do. I know that we can help them, get ready for the ones that are in need, depending on the facility. I know from our perspective, we put chillers and towers in -- for our own particular use case in data centers. It's fairly quickly. Of course, will be a little bit more motivated than others to do so. But I don't think it's a material. I think the plan is to do it and new data center bills, I believe, are contemplating or aren't going in the ground with outage. So that gives us a pretty good pathway in the future for filling those data centers. So I think the industry is flexing pretty hard in that direction. And right now, as we speak, our growth rates support a pretty ambitious ramp from our perspective. And we still have a lot more share to go. So I can't really speak to the individual like which company in your financial services has the capacity to adopt liquid cooling. And there's other different form factors outside of the directed chip, but we believe the directed chip is the most -- the best commercial available today and over time, maybe we move to immersion, we've done it in the past. So there's all kinds of different flavors. And once again, it's a differentiator for us. It's a differentiator for customers. And if we can get their cost, compute cycle down, that will drive more customer satisfaction from our perspective and the people will look to us for more and more systems to help them.

Okay. Any other questions here from the audience? Okay. Working capital, free cash flow conversion. If you can talk a little bit about that, there's been some concerns around that?

Yes. So as you can imagine with the product cycle is with the supply issues, I would say issues, but the supplies, the chain, et cetera, trying to align those into the turn cycle, the order cycle, the 90-day cycle, it's -- we're working to try to lower those turns which kind of lessen the burden on the capital front. So from that perspective, the working capital, we're very, very, very interested in being -- increasing our efficiency there. So I think that's how we're looking at it.

Okay. All right. And maybe if I can just double-click again on liquid cooling because I still do get this question. Aside from, obviously, the GPU side of things, what other things have to happen in order to maybe like increase the penetration and kind of go more your way of more liquid cooling systems versus going back to maybe air cooling or is it a GPU thing? Is it...

No. I don't necessarily...

Is it a component?

I don't think it's necessarily that. I think from a standpoint -- so we've been on a few customer calls and discussions where they're planning out their future data center footprint. And there's many other way. I don't want to talk about air, We're preparing for liquid. So I think it's kind of a matter of time. So it depends. I mean obviously, if you have an existing data center that's all set up for an air kind of environment to convert that might be a little bit more of an effort. I'm not saying it's necessarily all greenfield. There is an ability to -- I want to say miniaturized, but like there's different ways to accomplish all within rack. And if we can bring the rest of the componentry in chillers and towers, et cetera, to a specific site, we can do that. And there's some folks that are using sites that were never contemplated to be data centers so that they can put the plumbing in and get them liquid cooled. So I think there's a lot of effort to go that way. And the difference here is, if you're going air, you really -- you're kind of limited by the heat envelope of that system. And you can't really get the most performance out of the GPU at least in the current versions. And the new next gen, I'm sure those parameters will still exist. And physics kind of wins here from a standpoint of liquid being a cooling element versus air and the amount of energy to cool air relative to the amount of energy and heat dissipation from a liquid perspective. So I think it's a pretty -- it's pretty obvious, and we've aligned ourselves to meet in the particular product categories, the liquid cooling environment. So I think we're all in on helping our customers, bringing innovation and getting that to them as soon as they need. And right now, the demand for compute has been -- it's always been strong. It just -- it feels like it's getting even stronger.

Okay. Any other questions here? We have one.

I guess one question that I had is a lot of the at least ODMs in Taiwan are talking about the Blackwell series being slightly lower margin, even though it's significantly higher ASP. And I guess -- I know you guys aren't exactly well on ODMs and your value add is much higher just given the design. But can you talk about just do you -- to the extent you guys like have firm orders or have conversations around that and having able to sort of map out what you think your ASPs are versus your costs are, where you think you see margin for just the Blackwell series going? And I guess on top of that, how much more sort of design and I guess, thought has gone into the NVL36s and the 72s from NVIDIA versus, I guess, like the DGX Hopper series.

I think it's a great question. There's a lot of unknowns in that at that element with respect to one, the product set themselves and what we're actually doing with it from a design element. I can assure you that we are looking at all angles to make that particular product, [indiscernible] better. I mean, it's not even necessarily out. And so our goal is to innovate with that and to bring it to our customers and at the margin levels that we like to operate in. So it's a TBD with respect to what that all is going to look like when it starts showing up in volumes. And as you well know, the ODMs are busy for whatever format that they're getting from their partners. We're going to be busy making ours the best and unique. So I think it's the best I can do from a standpoint of answering that question.

All right. Maybe as we wrap it up here, Mike, what do you think is underappreciated by the investment community about Super Micro?

Well, I think the pace of innovation is widely underestimated. I think the ability to put that together for customers, the capabilities, the engineering skills are really underestimated. And I think I'm going to butcher this but I think Charlie Munger paraphrased someone who said that. Never underestimate someone that can -- that overestimates themselves. And I know we have pretty big ambitions. So I think we'll leave it at that.

Okay. Appreciate it. Thank you. Good luck with the rest of your meetings.