Okay, we're gonna go ahead and get started. My name is Harsh Kumar. I cover semis, a lot of other exciting companies like NVIDIA, AMD, Intel. Today, we have a special company. We have Supermicro. We have Mike from Supermicro. He's the VP of Corporate Development. What's interesting here is that we feel that Supermicro is a very good way to play the growth that some of the companies that I mentioned are seeing. So with that, let me just kind of dive into it. When we talk to our companies, these giants that I referred to, NVIDIA, AMD, Intel, they all love you.
You know, to just perfectly honest, there's companies that are much larger than you, that have been around longer than you, whatever, but they have a special affinity for what you guys do and how you deliver the product. So I wanted to dig into that. What is it that makes Supermicro click with these computing giants?
Yeah. Thanks for having us. You know, on that front, it's really technology-driven and core competencies of speed, performance, and cost. And the real magic here is that when the company was founded 30 years ago, we started off as a motherboard company, and basically optimizing the design of these chipsets and the layout on a board. Over time, customers asked us to do more, and more, and more. And with the complexities in the systems now, and the thread, you know, the core counts, et cetera, and the integration with memory storage, I/O, networking, they really need someone to partner with them-
Yeah
... to put these things into these systems, to get them into to customers' hands. So we're really an enablement partner, and we're very close. You know, obviously, we're-- I shouldn't say obviously, we're headquartered in Silicon Valley. We're right next to these folks, and our engineers work hand-in-hand on each product chipset. And so when the future roadmap comes out, we have a pretty good view of what that would look like and how we can optimize that. And as I mentioned before, on the motherboard level, we had originally optimized those for air airflow, for air cooling, and energy conservation.
Conservation.
On that front, literally, the x86 platforms we have a 10% less power draw on average. In, I'd say, the old days, a couple of years back, you know, enterprises didn't really care about that. Now, customers are using infrastructure to scale, to drive revenues. They care about the power, they care about how the systems are laid out, they care about the application optimization now more than they ever have, and it's factoring into our growth.
Great. So you, you do a couple of things. I mean, you do the servers, but you do, you do some pieces of storage. You, you're also getting into other aspects of your technology, where, where computing might be required. Could you talk about what some of the major areas, initiative, product lines, however you wanna classify them, are? And then maybe... I mean, I can guess which one's the fastest growing, but maybe you can, officially clarify for us the fastest growing areas.
Yeah. So let me start with the fastest-growing areas, I suppose, which would be, as everyone wants to hear about, you know, AI, artificial intelligence. So if you take a step back on this, and you look at the markets, the x86 market was fairly obviously grew quite, quite large. We did quite well in the x86 environment, optimizing applications for customers that cared. Those tend to be CSPs, large enterprises, and large enterprises, that use that equipment to drive revenue, so it's a critical part of their infrastructure. And so now, as we're moving forward, the new workloads are emerging in the AI world, and it's a parallel compute, GPU-based type workloads.
Yeah.
They're different than the x86 workloads, the systems of record. So customers are coming to us, obviously, the generative element, the large language models, everyone has a different version or a different thought process, and there's just a lot of different iterations of on that front. So there's a lot of innovation going on in the markets, whereas and the investment community is may not be as aware of that, the growth and the use cases that are emerging for AI-based applications. So that's been a phenomenal backdrop and driver, and those systems are complex, and when people deploy them, they tend to deploy them in large unit numbers.
And so we talk about rack scale manufacturing capabilities and the ability to deliver to these customers, and we're able to supply them with these footprints. That's been a strong part of our business as of late.
Okay. How about... So you brought up Gen AI space. I'm assuming the activity has been pretty hot for you guys, and, and volume has been pretty good for you guys. But maybe you can talk about what you guys are seeing, if there are any numbers you could share with us, or maybe some perspective on how fast this is growing relative to x86, some aspect of, you know, some, some kind of factor of growth-
Yeah
... if you will, from other technologies.
So we're roughly 50/50, in mixed on an x86 GPU. On a move-forward basis, that growth on the GPU side has been rather explosive. I think we've you know talked about you know some extent, year-over-year, might even be triple digit. And the outlook is pretty solid on that front, as obviously, as we all know, that there's a supply challenge-
Yeah
... on that front, and customers are still building out these footprints and refining these footprints.
Yeah.
They're not yet complete, and so the commentary comes around about Gen AI versus inferencing, and maybe we can get into that later. But on that front, the Gen AI platforms haven't been fully built. So demand's been really robust there and a driving factor. On the x86 side, growth rates are a little bit slower. Obviously, there's been some new cycles, new products from our partners, AMD and Intel, that we're addressing, and I think they've guided to back half of the year as improvements. So we benefit from those builds because those are the, you know, first to market with those products... for customers that care, and there's plenty of them that do.
And we think over time, as they're adopting the language models in the enterprise, at least the high end, that they'll ultimately wind up upgrading the systems of record for efficiency gains within the x86 complex as they start to develop AI-type applications for their end users.
So, we get this question a lot from investors, like, almost at least a couple of times a week. Is there cannibalization going on, you know, between x86 and GPU systems? And you're kind of right in the middle of it.
Yeah.
You get to see a lot of stuff, so I'm gonna throw that question to you.
Yeah
and see what your thoughts are.
Yeah, it's nuanced, it's complicated. The market's really, really large, so it's, you know, really hard to pin it down. But you would have to imagine that at least at the hyperscaler level, where they were using x86 for everything, and that's obviously not the case anymore. So from a hyperscaler level, I think they're a little bit more focused on GPU or AI-related ways to use that to accomplish their goals. We're under-indexed to the hyperscalers. We're having more conversations with them as we move forward because they have different application needs that could benefit from the different builds that we do, that can be optimized.
But from the x86 complex, if you wanna call it that, or the average enterprise, as I mentioned before, I think over time, as they look to upgrade or as they're looking at AI applications, it's probably taking their eye off a little bit on the maintenance side, so to speak. And as they get a little bit more firm on what their plans would be, I do believe that they will upgrade that complex, the x86 complex, to make room for AI. And so I think really what I'm trying to get to is we have a bifurcation, so to speak, on like what these markets will ultimately look like. So you have your systems of records, record.
Mm-hmm.
And then you have anything that's parallel processing, you know, audio, models, you know, language models, excuse me, video-based models, all kinds of different use cases and different verticals that require different platforms to get an optimal experience. So one size won't fit all in the AI world as we move forward, so standardization, we don't see happening anytime soon. And since we're half engineers, we're pretty indexed on developing new platforms for these customers. So very a lot of activity there, but at the same time, the x86 complex, there's still plenty of innovation and a large footprint to capture on that front.
Okay. And if I had to, like, push you a little bit and say, you know, maybe what kind, as a factor of GPU demand, are you seeing GPU server demand are you seeing versus x86? I mean, you mentioned triple digits for the GPU-based systems. Is it a 3x right now or a 4x right now? Is there a number you feel comfortable pivoting to?
I don't know if there's a number we feel comfortable committing to other than, we do have... You know, it's early days, and you've seen some large companies starting to roll out services, right? They haven't They haven't really fully hit the market.
Yeah.
They haven't had any full adoption rate. So when people talk about Gen AI, you know, wait a minute, we think that footprint is, you know, almost built and nobody will come. They're coming, and the applications aren't fully fleshed out, and as those become more available, they'll, they'll drive demand on this, in this category alone, and that will underpin the growth rates. And, as, as you probably well know, we've got it to a roughly a 40% growth rate into, our fiscal 2024 year, which we think is a pretty, a pretty, pretty phenomenal-
Pretty healthy number, yeah.
Pretty healthy, healthy number. We do indicate that if you know, supply was a little bit more available, at least in the short run, that we would be able to easily, you know, satisfy the end market.
So-
It's really dynamic.
Yeah
... I'm trying to say.
So you hit upon something interesting. We hear that supply is pretty tight, particularly for GPUs, and you're effectively saying that it is, and that you see it too?
Oh, yeah. Yeah, of course, we've mentioned that-
Okay
... on record.
You brought up Gen AI training versus inferencing earlier.
Yeah.
Is all the activity and all the action that you're seeing today, is that training based, or are you starting to have people knock on your door for inferencing-type servers? And what, what do those servers look like? Are they CPU? Are they GPU? Are they a mix of something else?
Well, so on that front, one, the demand for the Gen AI, Gen AI is still strong. The builds there-
Okay
... are not even nearly complete. So that could take-
This is training?
Yeah, it's GenAI.
Okay.
Yes.
Okay.
And so on that front, ultimately, the inferencing element will come. We have SKUs for it.
Okay.
We have some early POCs with customers, so we're anticipating that to move forward. You've seen some movement from some of our partners with new cards, if you wanna call it that, the L40S, and a discussion point and an adoption rate that will, you know, ultimately, you know, occur into fiscal 2024, or excuse me, into calendar 2024. So it's on the come, but most of the activity right now is in the training arena, and obviously, we wanna leverage the training with the inferencing downstream.
Okay.
So it's an emerging market, so to speak, and it's expanding rapidly as well.
What kind of visibility do you guys have with the GPU-based servers right now? We've heard... You know, I've seen stories where they're sold out for a year or a year plus. Is that something that you see at this point in time also?
On that front, yeah. Yeah, there's lots of different, you know, points on that front, but yeah, it depends on what the customer's doing with respect to allocation. So, we believe NVIDIA is allocating to the customers, our customers, you know, everyone's kind of in line waiting for the supply to continue the build-out. So yeah, there's a waiting period, and 'cause for our guidance, we will, we see that-
Yeah
-easing, you know, quarter-over-quarter into the rest of the year, fiscal year.
All right. What plans do you guys have? You've got some, some footprint here in the U.S., some footprint outside the U.S. Can you talk about your footprint for manufacturing and, and what are you in a position to be able to supply the demand that, that you are seeing, that's visible to you, and anything in the future?
Yeah. So, on that front, we're a global manufacturer, Silicon Valley, Taiwan, Netherlands. We've disclosed that we're building out a facility in Malaysia, and we have plans somewhere in the Americas. So from a capacity perspective, when we take a look at what our customers are saying, and now our customers are in a little bit of a different position they were, let's say, two or three years ago, where customers now, as I mentioned before, are driving revenue with their infrastructure. So what they want is a partner that can supply them on a global basis and can supply them not just next quarter, but two, three, four years out, and carry them through each one of the iterations of the CPU upgrade or memory upgrade or whatever happens in that system.
It's very complicated, you know, componentry that we're talking about. It needs to be integrated from software level up, firmware, et cetera. We do all those elements for our customers. On that front, when we were talking to them and we see their build plans, it kind of informs us of what kind of capacity that we will need downstream to reach the targets and the goals that we have. As you... I think most people know, but maybe 5% or 6% of the current market, as we're discussing it, our plans are to obviously gain a lot of share. With the innovations and the new products and the new opportunity sets, they've never been richer for us as a company.
Where is your biggest factory? Is that the U.S. factory or the Asian factory?
Well, I think over time, potentially, we would be it, indeterminate. Right now-
Okay
... the United States is the largest on the capacity front.
Okay, U.S. is the largest.
But, Taiwan, and then Malaysia, but I think if you add the Malaysia into the mix, it effectively doubles our current footprint-
Got it
... once it's once it's finished.
You've got gross margins that, you know, we think about. What we hear about your company is that you're very good at customization. You're very good at working with people and their requests and more flexible than a lot of other bigger guys. How does that work with your gross margin when you're that flexible and that accommodating, or sort of almost borderline customization? Would you not expect to see some benefit on your margin structure with that?
Well, I think the most important thing or takeaway on the margin side is that we're in a growth phase, period. Like, we're a growth company, and we're focused on customer footprint. We're focused on... Obviously, you know, we want to have a great healthy gross margin, and I think we've proven that in the past couple of quarters, in our outlook as well. But I think as we move forward, as we get a larger footprint, I think we could think about, let's say, two years down the road, about really optimizing for margin and, you know, really taking advantage of that. But right now, the focus is to gain share.
So, demand is strong enough that you're just basically, you know, meet the demand at this point in time and fix things later on?
To some extent, but it's really like to win the customer, okay? So when we go into a customer, and they have an application that they want to... We want to win that footprint. We're going to make sure that what we're building at that moment in time fits our current margin profile. But over time, you know, we feel we could probably, you know, go back and work on that to improve that, you know, whether it's, you know, cost savings, whether it's manufacturing it offshore, whatever it is, to satisfy our margin goals in the long term.
In a typical setup, your customer comes to you, they have some desire of a certain plan. They probably have some blueprints of what they want to see on the board. Do you have much input into that? Do you... Okay, do you have, you know, ideas, "Hey, we cut this, cut that-
Yeah
- or improve?" Could you just give us an example of something like that?
You know, we have one, well, not more than one, but-
Okay
... customers or larger customers that are trying to optimize their software to run their business. So, we collaborate on-site with their engineers, work with them, and yeah, for sure, we'll make suggestions with respect to how the process, you know, how things are set up-
Okay
The alignment, the firmware adjustments to match up with their software stack. So yeah, when we get a customer that cares that deeply, that's a customer that we want and we like, and we'll work with them because they're going to be using more and more equipment because they're going to be scaling up whatever service that they're providing for us. So for sure, that's part of the equation and one of the reasons why we have a strong and large, very technically competent customer base.
Great. I'm gonna poll the room to see if anybody's got a question. If not, we can, we can call it a day here for this meeting. Is there anybody that has a, has a question here for, for Mike? I guess we're good. Thanks, Mike. Thanks for your time.
All right.
Good to see you.