Penguin Solutions, Inc. (PENG) Earnings Call Transcript & Summary

Great. Good afternoon, everyone. And for the next session -- so sorry, I'm Samik Chatterjee. I cover the Hardware and Networking Companies at JPMorgan. For the next company we're hosting, it's Penguin Solutions. Thank you for coming to the conference. We have with us Mark Adams, who is the CEO of the company; and Nate Olmstead, who is the CFO. Thank you both for attending the conference. Thank you to the audience as well.

What I wanted to do is we've been asking all of our companies to sort of share their views of the macro first. To us, at least from where we sit, it seems like that's the biggest concern investors have at this point. And you talk to investors -- you talk to customers on a daily basis, how concerned are you about potentially much slower back half of the year or economic slowdown in the back half of the year? And what are the data points you're relying on to sort of inform you about where the macro is going?

Yes. I think if I were to categorize it into 2 buckets, there's those that have relatively high supply chain exposure inside of China and then there's those who don't. And then there's a dynamic of what is the supply chain cost impact -- and then it's not independent. It's almost tangential, but it's also then what is the customer impact. And I've talked to other CEOs that are struggling to quantify all this. The supply chain impact is something you can get your arms around, at least calculating. You can kind of understand that. The bigger challenge has to do with the customer demand impact. Of course, the markets reacted very favorably this week with the news out of progress with the talks with the Chinese. But I would say we haven't seen in our core business radical shift in demand yet. Now of course, when you have major bank CEOs saying the R word, you've got to wonder what the longer-term impact on demand will be. It certainly didn't help demand. But in an industry like the one we're in, AI is going to be pretty resilient because if you don't invest, you're out. And so I would say the supply chain impact is quantifiable and the customer demand impact is harder to quantify. I think one thing in the back of a lot of people's minds is if we can wake up one day and this can happen, what's -- what stability look like? How do we know when we're back to stability. Again, the capital markets are exciting this week, and that's all great for everybody. But what stops the next shoe to drop, so to speak. And I think there's a little bit more reserve if you were to talk to CEOs about how to think about the future. It's more like a little more cautious just because what just happened was out of the blue almost. So I guess anything you add?

No. I mean I think from our business, just on the supply chain side, most impactful for us on LED, which is about 15% to 20% of our revenue comes from LED, but only about 5% of our operating profit. And then within that, only about 25% of the LED business ships to the U.S. directly. We do have some additional business that would ship out of China from Chinese distributors to the U.S. So we can get our arms around kind of what the supply chain impact is. But as Mark said, it probably take a while to really understand if there's a demand impact on the memory or the compute side.

Okay. So maybe I'll follow up on that since you brought up the tariff side as well, what impact you're seeing. Do you see, firstly, in terms of your planning process, stability versus the unpredictability of the tariff? How is that impacting your planning process in terms of the rates that are put on these products that has continued to change in terms of the tariffs on China? And then in relation to changing the supply footprint over time, is that something you would sort of contemplate for the businesses that have exposure to China supply? Or is that more about, okay, this is the best cost we can get is in that geography and we'd rather sort of pay the tariff and still we would end up with a better landed cost than...

Yes. And to Nate's point, the biggest exposure by far for us is in this LED business. And yes, we are looking at alternatives. We actually were already working on this. For example, we moved some of our Level 2 manufacturing LED to our Malaysia facility. But it's like anyone contemplating this, this is not a flip of the switch and you're up and running. This is a longer-term process to get to a happy state, so to speak. So in the short term, you're kind of a taker. There are things people can do, and I can't signal this, but there are things like the country of origin declaration that companies can get, and we'd be a good candidate for that, given that a lot of our value add is outside of China. And there's also as you said, shifts in supply chain potential that allow you to do that, but that's a longer-term thing. So there are ways to think about this short and long term, but there's no quick, quick fix. And that's why, like I said earlier, the customer demand thing is really the unknown here because people have to react to all this. When it's China, I know some companies that have a very high presence in China supply chain. And the conversation gets -- it's 170%. It's a big pill to swallow and someone's got to pay it. So it does have an impact there. I think that's what the market reacted to today. Sunday was going into this week, the news Sunday was that there's a better outcome than where we're at today, which is good. But I think the big question mark is this, what is the sentiment really like? What is the market looking for? In AI, in our core business, we're generally positive on the perseverance of the demand, but it can't be positive, right?

Yes. So maybe we'll get to AI in a bit, but just last follow-up on that tariff front. Have you tried to pass through the tariff already in terms of price increases? And how has that been received? What have you seen in terms of elasticity of demand?

I don't think anyone said thank you when we did it. So it was not received super well. But I would say people are shortening their ordering windows because of the dynamic, like it can change in a day or a week or a month. And so people are a little more conservative on their inventory positions, obviously. And they're only buying kind of hand-to-mouth, so to speak, to make sure that they can meet their biggest customer needs. And yes, we're not looking to eat those costs. And so we have to pass that on to our customers at some level. And of course, we want to be good partners and all that stuff. But financially, it would be tough for us to take the burden. So we look at this on a case-by-case basis. But in general, we're pretty cautious with how we manage that.

Okay. So then moving over to advanced computing or sort of the AI opportunity that you talked about. Maybe sort of talk to about the -- when you look at sort of the broader customer base that you're catering to, which is sort of enterprises, firstly, where do you think AI enterprises are in relation to their AI overall sort of adoption curve? And what are the data points you're looking for in terms of to call an inflection in that enterprise adoption?

Sure. I think just to step back for a second, you're calling out enterprise adoption, and that would be different in our nomenclature than hyperscaler adoption. And the hyperscalers have been at this for 3-plus years and accelerating their investments into data center build-outs. And of course, we have one of the large -- it's not our core market, but we have a large hyperscaler partnership that we do have. And so when I compare that to the question of where the enterprises, the enterprises are in the early innings of this deployment cycle. Why is that? One of which is they're doing different things in terms of deploying AI and the value creation for their business and the measurement of the investment. That's all playing out at a different pace than the hyperscalers. Secondly, these enterprises have started out over the last couple of years with more of a proof-of-concept mindset. So if you're building a large language model, if you're Meta or Amazon or Google or Microsoft and you're building out, you just need -- I mean, if you think about where the slope of the curve -- the big curve on parameters around AI and the slope of that acceleration, that's just compute power you need more and more and more, and that's why they're out buying up data centers as fast as they can. As an enterprise, it's a little bit different value proposition. And so this proof of concept is, hey, we have a theory, we want to build it. Does it really work for us? Is it going to do as we think? And sometimes it's a large language model, but in the future, a lot of that return will be in inferencing. And so this is really the early innings. We've already seen, though, and we commented this on our last earnings call, we've seen a pickup in enterprise deployment as the kind of next leg of the stool for AI and -- but it's super early. And I think that's encouraging for our business because primarily, we're most attractive to the enterprise customers, Fortune 200, 300 companies as well as there's a new class of cloud service providers. There's a term that's out in the industry now called neocloud. And think of a broad range of about 200 or so people who have access to data centers and they have the right amount of power to be able to run significant scale AI at a time where there's a dearth in terms of supply, there's not good AI data center availability globally, definitely not in the U.S. And that's not something you can change over time. Like you can't just wake up and say, we're going to flip the switch and have more power tomorrow. This is kind of a 2- or 3-year challenge. And so what Microsoft and the big -- and Amazon and the likes, the big hyperscalers are out just buying as much data center space as they can. And there's these 200 or so kind of smaller cloud service providers that technically came from somewhere else. There are crypto companies that had access to data center and power, and they're trying to convert away from a crypto business that's not a great return into a different model. That's a great customer for us, and they're all out trying to do this, and we're building relationships and some of our success more recently has come from them. They're well capitalized. And they have the assets. In a lot of cases, they have assets that people don't have. So between the enterprise early inning deployment model for real production systems shifting from proof of concept and these neoclouds, this is where most of our focus is. And the logo, the new customer acquisition and the build-out of these opportunities has been something we've been very pleased with over the course of the year so far.

Okay. Great. Maybe you mentioned the neoclouds and the enterprises, but can you just give us more color in terms of what an enterprise use case looks like? Because neocloud, I think we most understand.

Yes, sure. The enterprise model, the markets that are top of mind for us, the verticals are financial. When you think of financial and a good example would be high-frequency trading or hedge fund, where seconds of information matter in their business. Energy in terms of exploration is another good market for us. Education still is very favorable for us. Federal integration and defense is a good category for us. Those are kind of the 4. We do have opportunities in terms of some new opportunities with -- through partnerships and sovereign cloud opportunities that we're starting to build out and compete for. But those first 4 verticals are there. One that I think is longer in the tooth for us is health care. I'm very bullish on health care, but they are early in their investment stage for infrastructure build-outs. And so if I look at all that, that's kind of the attraction of the enterprise. And in each one of them -- I mean, the big challenge right now is we're at Meta, which is our biggest customer, they can quantify the impact of the AI infrastructure investments they've made with us in terms of ad revenue generated from AI, they can quantify it. The biggest challenge in enterprise broadly, not necessarily with us, but just in the market is enterprises, why they're slower, they're having a tougher time getting their arms around an ROI on the massive investments in capital. And so what's nice about this for us is that we're starting to see that it's actually happening. And that's good. Like I said, in the applications I've made reference to, AI can be a valuable piece of the commercial benefit of the company as opposed to just buying some GPUs and hacking around. People are really on this high-frequency trading algorithm or they're on general exploration. You can imagine in consumer, for example, major retail and e-tail trying to understand customer behavior better. These are real-life applications that are evolving nicely where I think if you went back a year or 2 ago, there was interest in curiosity, but not enough of a support for the investment model.

Okay. Okay. Good. So maybe moving back to neoclouds. It's an emerging opportunity, definitely more relevant when it comes to the AI landscape than what it does in a classical cloud landscape. But have you tried to quantify what that opportunity could look like with the neoclouds for Penguin? And then how does that relationship typically differ when you are working with or engaging with the neocloud relative to Meta as a hyperscale customer? Where do you see sort of differences in that relationship?

The attraction for us in neoclouds is that generally, the model they're coming from, they don't have the capabilities to go out and stand up AI infrastructure. The crypto model is a lot of hardware centricity and just raw compute power to handle something that's very well known. They just don't have the depth or the capabilities. And so in our relationships, we're a significant extension of the neocloud models that we choose to be in. I add that last piece because if you look at neoclouds, there's a lot of money out there that's been invested in these companies. Our biggest challenge was selecting the winners and making sure we're investing in partnerships that lasts. Because I think there's a lot of people who've got round one investing in neoclouds that won't get further investing because they don't have the commercial viability to back it up. And so we're very selective because our model is much different than a Supermicro and HP or Dell. If you look at our gross margin structure, it highlights our value add in terms of services and software. And so we look at our customer relationships like engagements, not -- we're not just selling something and moving on. We only stick in relationships that we're adding value through services and support and advisory services and the like. And so why that's important is we don't want customers just to buy one generation of product. Once we get in, we're very sticky. And as I said, like if you look at the margins in the kind of traditional OEM model, the AI infrastructure margins are in the low teens, 12%, 13%. If you look at our advanced computing margins and our corporate margins way, way higher than that, somewhere in the neighborhood of 2.5 to 3x that. And that's because our model is different. It's one on engagement, not volume and transactions. And so we're going to select the winners that we see in neocloud and build partnerships that way, just like when we made the selection on the partnership with Dell and what we could do for them and help them get to market in AI and more recently last week with CDW coming in as a partner for a company who's got tremendous reach in terms of corporate Fortune 200 enterprise, but doesn't necessarily have the capabilities of deploying these large data centers. We're trying to find those values. Even I would mention, some of you might be familiar, we closed an investing round with SK in December. And that's -- we did that for the commercial viability of what we're trying to do together. It wasn't like we were crying for money per se. Our balance sheet is phenomenal. We have 0 net debt. But the SK partnership put us in a position where we can leverage their technology and our capabilities to provide them the solutions they need for themselves and for their customers. And so we look at these little pieces about how we get to market in terms of long-term sustainable value, not one-off on customers. And I think where you see a lot of the GPU traders, so to speak, who are just throwing GPUs in the boxes. That model is not sustainable. That will go up and down with the cycle. I think with us, when you look at our business model today, we reported last fiscal year $230 million of services, high-margin services. Well, that's kind of a run rate business in AI that we have in addition to new customer systems and all that stuff over time. But the model for us is based on a solution and value add over time, not just a transaction upfront.

Okay. I know you talked about services and the opportunity there. But maybe break that down for us a bit in terms of what are the nature of services that customers opt for? What are the different tier levels that they can like how much of that is installation versus actually post-installation service or even extending to sort of managing managed service...

That's great. So if you break apart our services, one way we look at that is there's about 10% to 15% of our services are point in time. I think it was like 30 or 40 out of 240 roughly, those are rough numbers that are point in time, okay? And then there's ongoing renewed annually services that have -- can be installation services, upgrade services, fault diagnostic services, software management services all the way out to spare parts, inventory management and the likes. And so there's a whole host of services we provide. And then there are cases where we have major customers who actually run the data center. There are -- our employees in their data center running the infrastructure. And so there's a whole scope of things we do. The point-in-time stuff is upfront onetime. That could be design related. So people forget that when you design a data center, you've got all these complex technologies that have to kind of come together at the right time. So that would be power, cooling, compute, networking, memory, storage and the software layers to manage all that. Designing that before you even get to installation, what's a cold aisle, what's a warm aisle, where does the power come in? How do you put the storage interface and the networking design of all that? It's a lot more complex than the average consumer knows or the average person knows. And so our design feature upfront gets us to the point where we're a trusted adviser [ early ], and then we go on from design to build it out and test for them and then we deploy and manage.

Okay. Got it. No. So you talked about Dell and maybe highlight sort of how do you -- how does the relationship work because in some cases, you're competing with Dell, I would imagine, for certain customers, but at the same time, you're partnering with them in certain sort of engagements. So how does that relationship work? And who do you see as more -- maybe a more direct competitor to the breadth that you offer?

Sure. It's actually really interesting because our commitment to differentiation and a solutions mindset, we don't typically run up against Dell or Supermicro or HP head-to-head. If a company is going to go buy a server or a cluster or what have you, we're out. That's all they want to do. We're the wrong company. If they want help in managing AI infrastructure deployment and the success of putting all this money to work and getting time to money sooner and more reliable and more scalable, that's our game. That's what we do well. And we're not going to sacrifice that nor am I going to try to get artificial revenue growth just to get revenue growth. And so you've got to be sitting there like the people who follow us and like, hey, how is it that your margins are 2.5 to 3x your competitors? Well, they're not with our competitors. And by the way, they're great business models. I'm not saying it's wrong or right. I'm saying ours is different. And from our perspective, we're going to stick to our knitting of what we do, and that comes from 25 years of being in high-performance compute, which when you look at the characteristics of high-performance compute, large clusters, power, cooling, networking, software, management of a data center. Those are all the same characteristics that actually led to AI. I wish I could tell you 25 years ago, I saw this coming. I didn't, but it's actually better to be lucky than good. We've got a great capability and knowing how things will go in a data center. By the way, things will go wrong. How do you treat the things that go wrong? How do you remedy them? How do you get people back online? Some of our biggest successes are when we've encountered problems and people like, h***, you guys knew how to do this. In our software today, we have predictive software that can tell them which GPUs are likely to go down. By the way, GPU failure rates are astronomic, much higher than CPUs. And NVIDIA benefits from that because they just sell more GPUs. And what I mean by that is people then take a reserve approach. And so like our business model is different. So when you say about Dell, if we're up against Dell on the same proposal, it's likely a hardware proposal, we're not going to win because I don't want to win at 12% margin. That's just not our game. And so we're really more -- again, think about it like a different model. Think about it more like a consulting model. We want to be in there. And then Dell will bring us in to customers. And by the way, we'll even -- if they don't want to buy -- if they want it -- if it's a big company and they have a big IT relationship with Dell, we have no problem letting Dell sell the hardware or working with them on that because what we really want is what we're good at, which is the services and software and the design piece of it. So it's about discipline. And do we give up GPU low-end revenue to be disciplined? Yes, because I think long term for this company, that's the right model for us to be focused on. And again, it shows up in 14.5 -- sorry, in 4.5 years, 18 quarters, gross margin has gone from 19% to 20%, 31%. Balance sheet is great, and we've had -- we made money every quarter.

Looking at it by brand, just Stratus, Penguin, OriginAI, just talk about sort of where the different positioning for the different brands are? What are the typical sort of customer targets for these individual brands?

Yes. The more sophisticated, more technology-driven implementations, they tend to want to use Penguin Solutions because we're doing something unique in the design of the system itself, not just to mention the service and software, but we're doing something upfront, whether it be a different processor architecture or a different memory configuration or a different unique storage environment. I'll give you a good example of another one like this. We have a large oil and gas company. And as many of you know, they're not in great shape with ESG metrics and initiatives given their other business activities. And so they're looking for ways to be more friendly on the environment level. They came to us and said, hey, can you use our recycled oil in immersion cooling? It's never been done before. So our team designed it and brought it into our factory in Fremont. We tested it. And now we literally have immersion cooling where they're dumping their computers in the tanks of recycled oil and running their data centers. Now that's not something that a Dell or an HP or Supermicro is going to go do that. We knew what we were getting involved in, we designed it, and that was -- so we'll take that business because that's a good margin opportunity, and it's also differentiated. OriginAI is a model that says if someone has the interest in more commoditized hardware and what have you, we can still layer our services and software on top of that. It's just we don't want to get in that model. So when it's more custom and performance or reliability driven, whether it's on our memory, for example, we have a memory architecture or a tool called Zephyr, which is a 0 failure rate for memory. We can test memory models before you put them in the computer to weed out the bad memory die. And so all these things we do are a little unique. And to the common person, it sounds cute, but the data center people who live in this every day, it's really significant.

Okay. Moving to another part, which we've heard good things about is your ClusterWare software. Really good feedback from customers, but maybe you can talk about what the underlying value add to a customer is from the ClusterWare software -- tends to be underappreciated in terms of the complexity.

Yes. The 2 layers that we're focused on strategically are cluster management, and I'll tell you a little bit what that is in orchestration. Cluster management really has to do with the platform that could manage all of these compute resources. And why I say that's important is some of our data centers are 24,000, 36,000, 60,000 data center environments to be able to monitor every one of those computers because every one of those GPUs is effectively a computer inside of a cluster and multiclusters. And then you've got networking links, you've got -- as I mentioned earlier, about the memory. To have the tool sets capability, cluster management allows you to diagnose all of that, schedule and use the tool capabilities and working with like a Kubernetes type application to allow for proper scheduling for workload optimization, provisioning the system resources where they're needed because in multiprocessor is not always even. If you think about simple things like how do I get -- if I'm scaling in one case at Meta, we went from 6,000 to 16,000 GPUs. How do I get software installed on all of those? We can do that from one single point. So there's diagnostics, there's provisioning, there's scheduling. There's new features we've added. We recently launched a feature called multi-tenancy, which is good for neocloud because this allows you to virtually set up a number of different clients in one platform or one data center install. So -- those are the type of features that, again, unless you work in a data center, you wouldn't have an appreciation. But to be able to manage all your data center from one platform software architecture and you have your user interface effectively to give you health statistics on what's going on. Those things are vital to making sure you have the most uptime. But at the end of the day, GPU reliability, as I mentioned, is not great. Everything you could do to preserve uptime and reliability is money in your pockets, time to money. And in a lot of cases, that's how people look at the investment here is if GPU reliability is 50%, if you can improve that by 25%, the money in your pocket from that investment is significant. And that's how they're evaluated internally as well.

I mean relative to just competition on that front, like what do you usually run into, particularly in terms of hyperscalers? Is it more in-house solutions that?

That's right. And by the way, that's why I said earlier, the hyperscalers are not really our core business because they live in data centers. That's their core business, and they've had their own strategy around using open source tools and they've got the breadth of resources in-house. It so happens that we actually still have the hyperscaler as our biggest customer. But our bigger opportunities are when there's third-party enterprise companies who don't have those resources. If you think about, as I mentioned, a neocloud company or a high-frequency trading company or hedge fund, these are companies that don't have massive infrastructure, and they need our help in our advisory service kind of technology-agnostic consulting model to really to implement because they're not going to go out and spend precious capital on building that internally in the short term.

Okay. Got it. You referenced this in a bit earlier, the partnership you have or the investment from -- with SK. Just talk about the sort of what opportunity does that create? When do we start to see more material revenues out of that partnership?

Yes. As I've talked about at the time of the announcement of this investment and obviously, on the closing, where we look at SK and I've known them for a long time, given my background in memory. SK has traditionally had an OEM model. They're a fascinating technology company that no one knows about. It's like a $250 billion conglomerate in Korea. And unlike Samsung, SK has not really entered the U.S. market strategically. Now they're starting to do more of that. But I would say they're not a well-known customer engagement brand. They're a technology company. And so when we met with them, their interest was how can they get their technology to the end customer and AI infrastructure. And they saw us as an enablement to them growing in the AI space. More in short-term needs, even their OEM business, their OEM business, they don't have a great solutions group that's touching the U.S. enterprises directly. They -- most of their business goes to consumer-related phones and compute and what have you, where a company, and I'm very familiar with Micron does a better job of touching the end customer. Well, they're using us in our model at Smart Modular to do that successfully. And so those 2 areas led me to not -- we didn't technically need the investment for financial means only, although that's super helpful. What we really like about is the commercial viability. In the real, real short term, we're already starting to see benefits on the memory side because there's opportunities that we can help extend this SK portfolio of products to other customer types. And it's very accretive to them and great for us. There also is an initiative at SK, and they've already invested this way, that they're starting to build out AI infrastructure, and we're confident that we're going to be a good partner for them, and that means -- and the way to measure that will be how we are able to talk about successful wins with them. And I've got to be careful, but I would say we're really encouraged more broadly with that relationship. It's a great relationship so far. And I've seen ones that don't work and ones that work. This is a really good relationship so far, and we're really excited about it.

Got it. Maybe for the last couple of questions. So Nate, maybe on the margin front, just talk about the strategy of how you are thinking about managing the volatility in the margins outside of advanced computing. Advanced computing obviously, is doing well on margins, but we see a lot of volatility in the margins outside of that. So any thoughts in terms of how you can manage that better through a cycle?

I think on the memory side, it has a relatively fixed OpEx structure. And so we do see -- we've seen margin expansion in the last few quarters actually, sequentially in memory. As the volumes have picked up, you start to see that flow through pretty well. So the gross margins are in the low 20s, but operating margins got above 10% last quarter in memory. So it gives you an idea that on a relatively fixed OpEx structure, pretty lean, well-managed, mature business, as we drive higher volumes, we'll see that flow through at pretty good rates. On the LED side, it's a little bit -- it's got a little bit thicker cost structure, I would say, in OpEx. It's got sales and marketing. It's a little bit heavier there. So again, big impact in LED would be customer mix, product mix and then the overall volume levels. I think in terms of a strategy for managing those margins, I think about when there's volatility or variability on the top line, you want to try to variabilize your cost structure as much as you can. And so in the areas of that cost structure where we have variability, we manage those very closely. We've got weekly spend reviews and keep a very close eye on things to try to align those areas of cost with the top line outlooks that we have.

Got it. And just lastly, capital allocation strategy. I mean you -- what are your thoughts currently on where can you add in relation to M&A versus sort of maybe using it more for overall capital returns to shareholders?

I'll just start like philosophically, I'll hand it to Nate. Philosophically, we're in growth mode. We move to an annual guidance because of some of the lumpiness in our business and the way the business transact. It's hard to know from the 13 weeks of the quarter if we're going to get customer validation or the first week of the next quarter or what have you. So we set a target this year back in September or August of last year that we're going to grow 15% in our fiscal year. In our last earnings call, we raised the midpoint of that guidance to 17%. Now mind you, the LED business is a single-digit grower, so to speak, and best case scenario, so you can kind of get a sense that our business is growing, and we want to grow this from here. So we will continue to look at opportunities from an M&A perspective. Having said that, we've been very disciplined. Now there's a balance sheet implications to other alternatives. I'll let Nate -- maybe just to talk to...

Yes. I mean, definitely a growth mindset, and that's our #1 priority. And then I think beyond that, it would be sort of managing the debt levels and driving down gross leverage. We got down to 3.8x this last quarter. I'd like to drive that down further, maybe get it down to 3 or below and then also opportunistically buying back shares when we see opportunities in the market.

Great. I'll wrap it up there, but thank you for coming to the conference. Thank you to the audience as well. Thank you all for coming.

Thank you very much.