Okay, great. Thanks everyone for joining. I'm David O'Connor, semiconductor analyst at Piper. Delighted to be joined by Tom Sepenzis, head of IR at Silicon Motion this morning. We've got 25 minutes, so anyone as well with any burning questions, please just raise your hand. Tom, thanks so much for coming.
Thanks for having me.
Yeah, you're welcome. Silicon Motion, so maybe just give us a quick overview on Silicon Motion, kind of the end markets you're selling into and what type of market share you have to kick things off.
Sure. Silicon Motion is the leading merchant supplier of NAND flash controllers in the world. We have grown to close to 40% of the PC market, ex- Apple and 20%-25% of the mobile market, ex- Apple. That historically, up until about 12 months ago, was 95% of our business was with mobile devices, including smartphones and PCs. Over the course of the last year, we've started a pretty massive transition to enterprise- class controllers and full solutions, which would mean a multi-chip module or an SSD, in the automotive market and also in the boot drive market, which boot drives are used across a wide variety of applications in AI infrastructure and enterprise infrastructure for data centers.
Awesome. What's the size of that market and how fast is it growing?
Well, it's changing almost daily depending on NAND price.
Good thing.
Also, as we learn more applications, we got into the boot drive market with our first customer, which is the world's largest GPU maker. Last year, we started shipping to them in the December quarter. They're having a transition from their Blackwell program to now the Rubin program. That is actually providing us with a lot more content opportunity, both from a unit perspective and from an ASP perspective, because they're increasing the capacity of the drives. But we've since won a large telecommunications provider as well, which we're going to start shipping to late this year or early next year. We're looking at the world's largest search engine, who's using our controllers. We're trying to convert them to a boot drive manufacturer, and we have probably about three or four other customers that we're in negotiations with now, companies that build servers for AI. It's a large market.
Okay.
It's growing almost daily as we find out, because we're getting a lot of interest in our DRAM-less controller for DPUs and TPUs and switches. Now customers are coming to us and asking us to see if we can make a controller that comes with DRAM, for CPU- intensive tasks.
Sure. They're super exciting. We'll get to them in a bit more detail in a minute.
Sure.
Just maybe level- set us on some of the incumbents exiting those markets that you mentioned previously, guys like Micron, SK hynix, SanDisk. You guys are taking share there. Can you maybe just talk about why they're exiting? Are these good businesses? Just your overall thoughts on that driver.
Yeah. Typically, the NAND makers, they've been in mobile, they've been in smartphones, they've been in PCs, boot drives, automotive. What they're finding is right now, everybody's got a limit on engineering resources. They're taking a lot of their resources, and they're moving them over to HBM and DRAM, which typically, when you're making a full solution, you're going to get a higher gross margin, and a higher ASP from HBM and DRAM than you would for NAND. Also, if you look at the NAND makers, they're getting very high margins just selling NAND now. When they make a full SSD, they're actually taking their gross margins down, which they don't want to explain to investors, right? They're less and less interested in providing a full solution and more interested in just selling the NAND.
We've seen a couple of major NAND makers exit mobile and exit PCs over the past 12 months. Micron, of course, announced it last December. We're seeing others follow suit. After Q1, SanDisk announced they're going to de-emphasize mobile. Most of these companies have started or are leaving the boot drive market as well. It's just not something that they would want to pursue because they can get higher margins from HBM and DRAM. They're going to participate either way because they're going to be selling NAND to companies like Silicon Motion, who will make those boot drives.
Yeah.
They're going to get a big piece of the sale anyway.
Okay. Awesome. Maybe switching to just the memory dynamics you just touched on earlier, the short supply, as we all know, translating to elevated pricing. What do you expect from the kind of memory from here, when you look into over the short to midterm, and into 2027?
Yeah. Are you talking about NAND prices?
Yeah.
Are you talking about HBM and DRAM as well?
No, just the NAND prices.
Yeah.
Or just your overall thoughts, what you're seeing.
Yeah. NAND demand is up likely 80% year-over-year this year. And supply is up 20% year-over-year. So we've seen a continuation of the very strong sequential increases in NAND price through the first half of the year. We believe that that's going to mitigate a little bit here in the second half, to the 5%-10% sequential increases in Q3 and Q4. But we don't really see an opportunity for NAND prices to come down meaningfully at least until 2028, 2029, when Samsung and Micron introduce two new fabs that they're building right now.
Okay. So remain elevated through 2027. Best case.
Yeah. To be quite honest, the two new fabs are likely not going to change the demand supply--
Yeah.
--balance significantly.
Yeah.
But it may help keep NAND prices from going up as dramatically as we've seen over the last 12 months.
Okay. Previous cycles, you were constrained on the supply side. How is Silicon Motion managing the supply dynamics this cycle?
Yeah, we've been very fortunate. We've been working with the NAND makers for 20 years, so they've been customers of ours. We've been working with them to help them sell their products. So we have excellent relationships with the NAND makers. When we exited Q1 and we were ramping up this boot drive business with our first customer, we weren't sure if we were going to be able to get the supply. And we were working with a couple of NAND makers to try and get the supply from them. And by the end of Q1, we'd secured three of the six NAND makers who were willing to give us what we needed. And exiting Q2, that number became four of the NAND makers. So despite how difficult it is for everybody to get NAND right now, we've been very fortunate in being able to do that.
Actually, that's led to customers coming to us to try and help build their products with our controllers because we can actually get the NAND.
Okay. Those type of discussions with the NAND makers, do they stretch into 2027 or are they more short-term? How would you characterize them on average?
Yeah. The initial one that we did in Q1, that stretches through Q1 of 2027. Normally, these kind of arrangements are negotiated in the July, August time frame. We just completed that for next year.
Okay.
We're set.
You are pretty set for next year.
Yeah.
Okay, great to hear. Maybe switching over just to the boot drive business, that is the new opportunity for Silicon Motion. Just talk about your entry into that business as kind of the big difference versus the controller and just the opportunity in general.
Sure. Before boot drives, we had some, what we call our Ferri business, where we make a full solution. We buy the NAND, we will make a multi-chip package for automotive or a PCIe SSD for automotive. Boot drive was the first time we expanded that outside of automotive in a couple smaller markets into now the AI space with our first customer and the telecom customer upcoming. This is a full SSD. So we buy NAND, and we sell it as a complete product to our customer. It is for their DPU product, and also for multiple switches within the greater Rubin platform. Well, it was Blackwell platform for Q1 and Q2. As we go forward, it will be Rubin. So we will see.
Okay. Previously, boot drives, they were supplied by someone else or internally or--
Yeah.
--didn't exist?
Yeah. So actually, it's interesting, with our first customer, we sampled them a drive because we wanted to get into BlueField-4, and they were comparing it to what they were using at the time, which was from a NAND maker, and they realized that it was actually performing pretty much the same, which they were surprised by because the reason that they were going with that NAND maker was because the security firmware and software was excellent. It was better than anything else they could find. As we got into it, it turned out that that NAND maker was buying our controller from a distributor. We were already in there, in the product. We just didn't have the whole boot drive.
When that NAND maker decided not to move forward with BlueField-4, we became the natural successor and are now providing that customer with the full solution.
Okay. Just tying that in with the memory prices, continued price increases, is that good or bad for boot drive business?
Continued price increases, I think are kind of neutral. The customer has to pay whatever the price is, right? We are discussing potentially entering into long-term agreements to at least stop the prices from going higher, but we have not done so yet. I would say the reality is if prices came down quite a bit, it would help not only the boot drive business because our customer would, if for whatever reason, NAND prices dropped in half, people are concerned about this for us now that we have exposure to NAND prices. But the reality is our customer, if prices got cut in half tomorrow, they would likely double the capacity of the drive, thereby having no impact on the top line.
Okay.
That would actually re-accelerate our PC business, our mobile business, which are both units are going to be down in both those industries by about 10%-15% this year. So there's a pretty big pent-up demand--
Yeah.
--for PCs and mobile that's starting to get generated in the market and will likely extend through next year. It will also help our new enterprise SSD business, which we call MonTitan, because the availability of low-cost QLC NAND will help us become an alternative for hard drives in enterprise and AI data centers. We're doing that now. That's just starting to ramp, but I think it would accelerate if pricing was a little bit lower. So we're actually, contrary to what some people think, NAND prices going down would help us.
Okay. That's a large percentage of the BOM, the bill of materials cost in boot drive?
Yes.
It is. Okay. The largest GPU maker, the opportunity there in boot drives, what does that ramp- up look like for you guys? We had initial kind of Blackwell, then Rubin layers on top. How should investors think about that, kind of the cadence basically, as you go into 2027?
Yeah, for the GPU maker?
Yeah.
Yeah. They're switching from Blackwell to Rubin now. In the Blackwell and BlueField-3, we have the dedicated BlueField-3 DPU, so we have boot drives in there. BlueField-3 was an option. You could buy it. You didn't have to. The attach rate was nowhere near even 50%. It was closer to probably 20% or 30%. In Rubin, BlueField-4 is integrated. You have to buy it, so it's 100% attached. The units are going up significantly, just from that. We also now have content or boot drives in some of their switches, like NVLink and Ethernet switches. From a unit perspective, we're going to see a market increase.
Yeah.
It's not going to happen all in Q3, right?
Yeah, sure.
It's going to take a moment for Rubin to ramp, probably 6-12 months.
Yeah.
This is a very large opportunity for our company.
Okay. Both unit and content side of things as well--
That's right.
--as you spoke about. You also mentioned on boot drives, the largest search engine.
Yeah.
AI, also as a potential boot drive customer. What does that adoption look like?
Well, we have to win them, so we haven't won them yet. Although amusingly, we found when we were doing the exploration with our first customer, that they also use our controller because of security, with another NAND maker. We're just trying to now get them to convert over to our full boot drive solution. In terms of timing and when the ramp is, since we haven't quite won that customer yet, I can't really talk to it.
Yeah, sure. No, understood. But the discussion is broadly similar to the discussion you had with the GPU maker.
That's right.
18 months, two years ago.
Yeah.
Okay. Maybe, you mentioned MonTitan on the enterprise. Maybe just mention what that means for Silicon Motion, what that opportunity look like on the enterprise side.
Yeah. This was actually our first product that we were really trying to get into the enterprise boot drives. Really kind of, they didn't come out of nowhere. We worked on it, but it was a much more rapid adoption. MonTitan is something we've been building for five, six, seven years now. We really wanted to understand what enterprise customers wanted, so we worked with them for two to three years to develop at least the feature set. Then we spent the next two or three years developing the product. The firmware, which is out now, we have seven tier one customers. Our original two, one in the U.S. and one in Asia, then we won five CSPs that we announced at the end of Q1, all of which will be ramping here in the second half of the year.
Well, two started in June, but the other five will be ramping here in the second half of the year. This is a very large market. It's actually bifurcated since we started. Originally, we were just looking to be using QLC NAND. We wanted to be an option for storage. So instead of using a hard drive, you would use a NAND flash drive, which is going to be faster, it's going to use less energy. But unfortunately, NAND prices have gone up so much in the last year, we're no longer really price- competitive. We hope that when NAND prices start to level off, we'll get to see that market really jump. I mean, people are going to use us now because there is no alternative. You can't buy a hard drive.
Yeah.
They're sold out for two years.
Yeah.
So, we will see that business grow. But the other side of it, what MonTitan can do is it can control what's called a KV cache drive or a compute drive. NVIDIA and their Rubin platform want a 16 TB NAND flash drive for KV cache. MonTitan has got a number of design wins with its customers that we're starting to ramp with here in the second half of the year. This is a multibillion-dollar market that will have more competition, ironically, than anything else, probably, at least in the beginning, because the NAND makers do still want to play in this space.
Okay.
But like we have done historically with the NAND makers, they won't have product for everybody, and they certainly won't have engineering teams for everybody. So one of the major limitations in enterprise and AI data storage is most customers want that storage solution customized. And so you need to have the engineering team to go in and do the customization. So SanDisk or Micron or Samsung, they might be able to get customer A, B, and C with their own internal teams and their own internal solution, but they can't get customer D, E, and F. So what they'll do in that scenario is they'll, instead of using their own internal solution, they'll use MonTitan, and therefore they can leverage our engineering team to go and do the customization work.
So while people think it's going to be kind of the Wild West with everybody competing with everybody, it's more likely that we will be in partnership with the NAND flash makers, as we always have been.
Okay. Very good. So, big agentic AI inference play you see for the MonTitan business.
Yes, absolutely.
Awesome. Maybe just explain to us again that dynamic as the industry moves from TLC to QLC. How does that benefit you guys?
That really would benefit, at this point, mainly MonTitan.
Yeah.
Because, if you want to challenge or be an alternative to a hard drive, what you need to be able to do is store a lot of data cheaply. QLC is 4 bits per cell, NAND versus SLC, which is 1 bit per cell, MLC is 2 bits per cell, TLC is 3 bits per cell. So when you're using 4 bits per cell, you can pack more data into the same space. The other benefit or, I guess, the QLC is really perfectly suited to host large language models. Because one thing that people don't like about QLC is, even though it's starting to get solved at this point, that you can introduce more write errors because you're squeezing so much information, 4 bits versus 1 bit.
But when you are hosting a large language model, anytime you ask a question or I ask a question, it's just going in and reading our question, it's reading the data, then it's outputting an answer. There's no writing involved. So actually, what you're doing, if you use NAND instead of an alternative storage solution, you're going to get a faster response from the NAND, than say, a hard drive.
Yeah.
It is really perfectly suited for hosting large language models.
Okay. You talk about HDD sold out for the next two years. At the same time, the price of flash is very elevated.
Yeah.
Does that benefit you? Are you having more discussions with customers, or are you still in the middle, you are caught on both sides here? HDD sold out should drive more customers your way, but at the same time, SSD prices are elevated, so maybe they are not adopting. How does it play out basically, in terms of what customers are doing there?
Well--
How does that benefit you guys?
I mean, realistically with MonTitan, we're seeing so much interest in our product from the KV cache and compute area, that we don't have to focus as much time on the storage, pure storage or, that's a bad way of putting things, but pure storage. The storage alternative using QLC. But people are still coming to us, even though the price is so high because, right now the alternative is a tape drive.
Sure. There's no--
No one wants to use that--
Yeah.
--in an AI data center, right? Or any enterprise data center other than as backup.
Okay. So yeah. The business really transforming for you guys with--
Yeah.
--boot drive and Ferri. Now I think was it 30% of the mix versus low single digit. Maybe just talk about what that does for margin. So different products, new products coming in, different scales of ramp-up. How does that impact the margin profile?
Yeah. The Boot Drive business and the automotive business are closer to the 40%-45% gross margin range. But they have lower operating margin impact because they do not require the same level of investment as MonTitan does, or some of our newer mobile or PC tape- outs for the controllers. So from an operating leverage perspective, they are both very good, even at 40%- 45%. But MonTitan is probably closer to 60% gross margin, so that would be our highest, and our mobile and PC are in that 48%- 50% historical range that we guide to.
Okay. Awesome. Maybe just, two minutes we have left, maybe just talk about the MaxLinear arbitration in Singapore. Where do we stand with that? What is the next signpost that investors should look out for?
Well, we completed the closing arguments in March, beginning of April time frame, and they told us we'd get a decision within three to six months, so hopefully they stick to that.
Okay. Maybe lastly, just on the cash side of things, maybe if you could just talk about the capital returns. Again, multiple products ramping, different inventory levels. How should we be thinking the cash and just capital returns, I guess, overall?
Well, we just completed a bond offering for $1.15 billion. Initially we were looking to raise $800 million. Most of that is going to be for NAND. So, that's to support the ramp of our first two customers and boot drives and hopefully we get one or two more. So, this gives us cushion to make sure that we can get the NAND that we need to support multiple customers and boot drives, and automotive, where we're seeing significant strength in automotive, which was less than 5% of our revenue a year ago and now is closer to 15% of revenue. So, we're actually seeing significant growth in automotive as well as the boot drive business, for full Ferri solutions. So most of that cash right now is going to be earmarked for buying NAND.
Okay. And, maybe, since time is almost up, maybe any closing remarks that you'd like to leave with investors on Silicon Motion?
No, I'd just like to thank everybody that's come today and that's online, and look forward to speaking with you all at another time. Thank you.
Awesome. Well, Tom Sepenzis--