SiTime Corporation (SITM) Earnings Call Transcript & Summary

Okay. Good afternoon, everybody. Welcome to the Goldman Sachs Communacopia Technology Conference. My name is Jim Schneider. I'm the semiconductor analyst here at Goldman Sachs. It's my pleasure to welcome SiTime CEO, Rajesh Vashist; and CFO, Beth Howe, to the stage today. Welcome to you both.

Thank you.

Thank you.

Before we get started, the company has asked me to read -- remind you that today's discussion includes forward-looking statements that involve risks, uncertainties and assumptions, which are further described in SiTime's SEC filings, including SiTime's most recent Form 10-Q, and actual results could vary materially and adversely from those anticipated or implied. SiTime assumes no obligation and does not intend to update any such forward-looking statements. For more information, please visit SiTime's Investor Relations website at investor.sitime.com. That was quick. Okay. So let's start off on big picture strategy question. Rajesh, Beth, maybe start off with kind of a brief kind of like view on the company, but specifically explain the role of precision timing, specifically MEMS-based timing in advanced applications and what's been driving the shift away from quartz-based timing?

Yes. So to take a little bit of credit, about 20 years ago, we saw and believed that there was an increase in throughput, decrease in latency, decrease in face noise, tougher environment and we felt that timing had to be a solution presented to customers. So that was the bet we made, and that bet has proven true when we went public in 2019, but it's become even more true in the last few years as the pace of innovation along those axes has increased so dramatically. So precision timing is essentially the heartbeat. It's the rising -- it's literally a square wave, which shows up so that a CPU, GPU, TPU, WiFi, wireless, modem or whatever other ASIC knows when to be on and when to not be on. And given the throughputs, given the rates, it becomes really important, really critical. Yes. So our most expensive chip, not in military application, sells for about $50. Our lowest priced chip sells for around $0.35, $0.40.

Great. Okay. So now the CED segment that includes data center represents the majority of your revenue now, I think, 65% -- 64% last quarter. Can you run through us kind of the increasing importance of precision timing, specifically in data centers? What's the key driver for SiTime in terms of content and your outlook for that segment?

Right. In general, we grow in about 4 different ways. In general, we grow because the end customer units grow, easy. We grow because ASP grows because functionality grows. We grow because the number of chips per unit of system grows. We also grow, of course, when we get new opportunities, new applications. And so the great thing about the data center AI business is that they're firing on all cylinders and they're firing on all cylinders at a very high rate of innovation and speed. And so we are in the GPU, the TPU, the CPU. We are along with the switches. We're sitting next to the NIC cards. We are in the smart cables. We are in the pluggable optical modules. And the new generations of these products are moving fast. And with our acquisition of the clocking business as well as our own road map, the revenue opportunities are growing dramatically in that segment. When we went public, we were probably in that space. There was no data center AI space, but there was a data center business, which was around $800 million in SAM. Today, that would be probably close to $3 billion in SAM, opportunity. So it's an amazing opportunity, but half our business is not in this space, as you indicated, outside that.

Yes. You recently completed the acquisition of the Renesas IC clock business. Can you maybe articulate the strategy behind why you decided to make that acquisition and how that kind of bolsters or augments your position in the market?

Yes. So we always knew we wanted a clocking business because it's a complement to the oscillator business we have. The oscillator business is generating the frequency. The clocking business is disseminating or spreading the frequency. And in that business, we know that the most best brand name, the best technology, the best team was in Renesas, which used to be IDT, which used to be ICS. So it's a 25- to 30-year history. So that was the best business to buy, and we were fortunate that we were able to get that. So it was the golden property, if you will.

Yes. Now longer term, how do you think about competition in the overall MEMS timing market? Because this is a category you mentioned you created in many ways, but it's no longer small. I mean this is a target now. So how realistic is potential for new competitors to kind of enter the market in a much more meaningful way?

So while we have to give a lot of credit to our MEMS technology. Our MEMS technology is really just a way of enabling precision timing. At the highest level, we are thinking of timing, precision timing as a system solution, solving system problems for about 500 different applications. So MEMS unlocks that, but we also have advances in digital design, analog design, packaging, testing, building and infrastructure. And I see our 500, 600 engineers involved in all aspects of that. And that makes us the biggest team. That makes us the most focused team. And the best part is we're focused on the most innovative parts, the most differentiated parts of this business. So we're not interested in a me-too part of this business. We want only the best parts, which has allowed us to go from 45% gross margins at IPO to 67% last quarter.

Yes. Good place to kind of move to the end markets, I think. So I want to touch on a few of those, starting with CED and specifically data centers, as I touched on earlier. You discussed the data center end market and the critical role that precision timing plays. On the 800 gig to 1.6T transition, can you qualitatively or quantitatively kind of describe the incremental opportunity that drives for SiTime?

Overall, the optical module business in a couple of years is probably $0.5 billion for us. So it's big, very large. We expected -- we joined this business in 400, where we were about 40% of the market share. In 1.6, we probably double that, close to double that. The ASPs have gone from around $1 to 40%, 50% growth. And this market has become more and more important. Eventually, the 3.2, which is already in design win will probably get to be even more in market share as it gets to higher frequencies, lower jitter, greater resilience to temperature. So I think that's a perfect market for what we are doing.

Got it. Now talk a little bit about your customer mix in the data center. How should investors sort of think about penetration levels alongside different numbers of the chip providers in AI servers among the different hyperscale customers and sort of talk through that, if you can?

Sure. Maybe I'll talk a little bit about. So one of the beauties of SiTime is the diversity of our applications, the diversity of our customer base. So if I think about CED and the data center, we have a broad range of customers, the optical and transceivers across a wide range of providers, but also the hyperscalers, the neo clouds, the sovereigns. So really, we think about all of those as our customers. And in addition, selling into the other semiconductor providers as well as the OEMs that are providing different kinds of solutions, SoCs, as well as systems to those hyperscalers and other AI and data center customers.

Yes. And then -- can you say anything about sort of like position in GPUs versus ASICs versus other kind of chips?

Yes. So we have design wins across all of those. If I think about GPUs and CPUs, you have timing and clocking on the trays there, NIC cards, the timing associated with those. Switches, you have multiple chips with typically maybe $20, $25 worth of content in a typical switch. So our ASPs range from $1 to $2 at the low end to $4 to $7 up to $10 for certain higher-end oscillators. The clock ASPs run in the typically $4, $6 kind of range. And so different solutions will have different combinations of buffers, clocks, XOs as well as TCXOs and OCXOs depending on the configuration. And then as you build those up into racks, you can see multiple hundreds of dollars worth of content in racks and sometimes up to over $1,000 of timing content in certain rack configurations.

Okay. And then lastly, related to CED, you previously talked about beginning to move from sort of discrete components to chiplets, advanced substrates, modules, things that get you to higher performance and compute density. So how large of an incremental opportunity do these kind of system solutions represent? And when could we see the first ones launched?

Yes. That can be rather large. We think it could be anything from $1 billion to $1.5 billion more of SAM. The reason for all of this, of course, is because timing needs to get closer, timing needs to get more clean. The signal needs to be less attenuated. And just like in power, you have VPD vertical delivery, we think there's a concept called VTD, which is getting closer and closer to the chip either below embedded in the substrate or above. And I think that this could start as early as '28, maybe '29. But I think this is a long-term trend that continues in the market. And I think only SiTime has the technology to make this happen in the way that we are making it happening.

And would that be completely incremental? Or is that more sort of like partially cannibalistic relative to your...

I think it's mostly incremental simply because that becomes something that gets used at the very high end. But one of the things that Beth was also saying was that -- there are new opportunities, whether it's in sovereign data centers or enterprise data centers. In our case, we're not very well penetrated, for example, in Chinese data centers. There are opportunities in those areas as well. So I think there's a lot of incremental business for us. And that is why when you think of timing TAM, we see it doubling almost between now $12 billion, $11 billion in -- by the end of the decade, which is faster than we would have expected. We would normally expect it to do that in 7, 8, 10 years, but we think it's going to happen in the next 4, 5 years.

So $25 billion by 2030.

Give or take, it could be a little bit less, maybe $22 billion or so, but significantly higher.

Okay. Yes.

Because you see the semiconductor business itself is going from $700 billion, as you know, to pick your number, $1.6 trillion to $2 trillion.

Yes.

And if that's true, then timing needs to scale along with it.

Understand. Okay. So I want to turn to the mobile IoT and consumer market. You've been a beneficiary of your largest customer transition to internal modems. We got some new products today. I think clearly, you're seeing the benefit of that product transition in your Q2 with 80% sequential growth in that segment. So if you think within that market, how are you thinking about the revenue trajectory through the back half of this year as your customer continues to ramp that internal modem in those devices? And then is it still reasonable to expect traditional seasonality in that business in the first quarter of '27?

Starting with the last one, yes, we do expect a little bit of a slowdown, as you rightly point out, in the consumer market in Q1. But at the same time, growth continues in Q3 and Q4. And then the size of the business next year, even though the seasonality down in Q1, is much bigger because it's the first full year of, in my opinion, 100% internal modem business next year. It could be a little bit less than that, but I think it's 100% internal modem next year. So I think it's a very good business, very solid business. We, in addition, continue to get design wins in other areas. We also think that the number of opportunities will grow under the new management because I think there's -- we're expecting for more hardware innovation and centric around the product innovation piece coming up. So yes, we expect lots of new good products.

Yes. And then -- but to clarify, once you get to that kind of elevated full run rate of fully penetrated...

It can grow after that.

Yes, you grow units then.

Yes.