Great. Well, good morning, everybody. My name is John Vinh. I cover semis here at KeyBanc Capital Markets. We are delighted to have the onsemi team with us today. We have Hassane El-Khoury, CEO, and Thad Trent, CFO. Welcome, guys.
Thanks, John.
Thank you.
Hey, maybe what I would like to start off is maybe talk about the earnings call. You talked about prioritizing data center demand over autos and industrials, and I am wondering if you could just unpack that comment a little bit, because I did not think that short-term capacity was fungible, right? Because as you think about parts that you supply in a data center is very much different than the parts you supply to autos and industrials. Maybe we can start there.
Sure. The comment I made, obviously, we also framed it into a temporary adjustment that we have made, not a strategic redirect. We expect that to normalize, call it, over the next couple of quarters when the impact of utilization that we took up will start coming out from material. To answer your question more directly, at a product level, it is not fungible. When I would call a product is when the product is fully assembled, because then it is a footprint, it has to go on a certain board, a certain physical form. From a fab perspective, not just from a fab, all the way to technology, it is fungible. When we talk about 1,200 volt silicon carbide, it is the same 1,200 volt that we supply, for example, in automotive or in 800 volt systems in AI data center.
It is fully fungible up until it gets to almost die, which is two weeks away from finished goods, if you think about a two-week cycle time for a finished good. All the way up until that point where it gets its, call it, marketing part number, or MPN, it is fungible. That is the decision we made. When you get orders in short time and you have to make priority calls, as long as those priority calls are on a two-week. If it is packaged and it is kind of ready to go, you cannot just change the mailing address. That is the flexibility. Compare that two-week change in end product to a little over 30 weeks of lead time, we are able to make those decisions while we start those new wafers for the auto and industrial orders that we got.
That takes about, call it, three-six months to come out. That is why I said it will normalize in the next couple of quarters, just a matter of time. You saw utilization went up, so we already took utilization in response to that demand, but the output is yet to come.
Great. Other than increasing utilizations, is there anything else from a supply perspective that you are doing to respond to this uptick in demand that you are seeing?
No, we have the capacity. We talked about we took utilization up to 83%. Fully utilized, you can think about it in the 92%-93%. We are not fully utilized. We did a step function based on the demand that we saw, but that is not fully utilized, meaning we are not capped out on capacity. What we are managing now is the slope of the orders that came in. If you get a lot of orders within 13, let us say everything is at 30 week lead time, if you get a lot of short lead time orders, you are just not going to service them, just a matter of fact from a lead time perspective. It is not like we cannot do the capacity and we need more CapEx.
Our CapEx remains unchanged. CapEx remains mid-single digit. We will take up utilization as we see demand coming in, and even when we get to that 92%-93% utilized, we can flex to the outside. It is not like CapEx is going to have another CapEx cycle built in here.
Great.
We feel pretty comfortable with the growth trajectory that we have for the company overall and the flexibility we have. Remember, as utilization goes up, the margin gets the tailwind from the utilization, and that is our short-term margin expansion, and you start to see it in Q3, even throughout this year.
Right. Maybe on that front, Thad, can you just remind us just what the puts and takes on gross margins over the next 6 - 12 months is? Is it primarily going to be driven by utilizations, or are there puts and takes that we should be thinking about here?
Yeah, utilization's going to be the primary driver. If you think about where we've gone with utilization, exiting last year, we were at 68%. Last quarter, we were at 83%. So you can see how quickly we've taken that utilization up. Every point of utilization is 25- 30 basis points of gross margin improvement just two quarters later, right? So you saw our margins step up in Q2. We guided a significant step up in Q3. I expect more because of the utilization tailwind. I expect more in Q4. Based on the visibility, we're getting much better visibility than we were a year ago from our customers, so we got good visibility into 2027. In some cases, we have customers even starting to layer in backlog into 2028. So it gives us that confidence that utilization will continue to improve.
Now, we will get over that hump, as Hassane said, kind of that cycle time, that fab cycle time, but I expect further gross margin improvement into next year as well. The one thing that we've had as a headwind is the input cost to hit the P&L where the pricing hasn't yet. We did our first round of price increase April 1st. We're doing a second round now. That'll take several quarters for it to fully layer in, but that's a nice tailwind kind of in the short term. Longer term, let me back up to Q2. Q2 had 650 basis points of underutilization charges. If you get from that 83 to 92, 93, there's 650 basis points right there if you just do the math that I was giving you.
You've got that as a tailwind. You've got the FabRight initiatives, so we've said that's about 200 basis points of gross margin improvement. We announced last quarter the divestiture of two manufacturing sites. That's 50 of the 200 basis points. Some of that will start to layer in in 2027, most of it in 2028. We have favorable mix, so we'll talk about the products, the differentiated products. Over a multiyear period, there's another 200 basis points of improvement there. Then if you go back to the fab divestitures that we did a few years ago, we built that bridge inventory that we're burning through. You can see we actually burned through quite a bit of that last quarter.
As we start manufacturing that insight, that's another 200 basis points. If you start doing that math, you get over 50% pretty quickly. Utilization is the biggest short-term driver, though.
Great. Thank you. You talked about increasing backlog. I assume with your lead times are extending, backlog is getting better. Maybe, Hassane, can you just talk about what you're seeing from a restocking perspective, specifically from your automotive customers at this point?
Yeah, it's not very consistent. We've had, obviously, last earning or Q1 earnings call, I made the comment of we need to see the backlog in automotive, where we haven't seen a replenishment cycle. We haven't really seen an uptick in demand per se. SAARs has been very flat, sometimes slightly down. Overall, we haven't seen the replenishment cycle, which is a reason for concern. A couple of things. One is when demand does pick up in automotive with no replenishment, the constraint that we're seeing today will get even worse. That's point number one. We haven't seen that yet. Point number two is OEMs are no longer waiting for the Tier 1s. Typically, the replenishment cycle will happen through the Tier 1s placing orders ahead of an OEM demand cycle.
OEMs are concerned about the fact that the Tier 1s have drawn down on their inventory because even their balance sheet can't support a small inventory build. They're buying directly from us. Not enough to solve the potential constraint with a demand uptick, but at least there's a realization that things are not healthy from an inventory perspective in the supply chain in automotive, that OEMs are starting to take a little bit control of the situation. That tells you the Tier 1 business model needs to change. If we need to go down a path of you have to have inventory when you have AI data center competing with the same capacity, the fungible capacity, as I mentioned earlier, that business model is going to change.
I think some of the OEMs are realizing that, and they're working directly with us on direct orders. We ship to the OEM. When the OEM has the demand, they decide which Tier 1 gets it. They're doing the allocation.
Right. How broad based is that in terms of OEMs?
Not very broad based. Obviously, the, I am going to call the disruptors, a lot of the Chinese OEMs, North America EVs, that is their business model. They started. We deal directly with every single one of them. That model is kind of the new model. The traditional OEMs, I would say it is not enough, not enough to make it a trend yet. But look, I think when constraints or that challenging environment will happen with the demand, I think there will be OEMs that will get through it very nicely and OEMs that will not, and that is going to be like: Hold on, what was different? Well, they bought direct, and they held the inventory, and that is going to become kind of a lessons learned.
I communicate as much as I can to make the lessons learned be pre-error versus post-error, but sometimes companies want to make their own mistakes to learn, but that is not mine to judge.
From a cycles perspective, I am just wondering, is there anything unique and different about this cycle versus past cycles? Or as you alluded to, is this just going to be history repeats itself?
I would say I wish it was history repeating itself. At least you know what the heck is going on. It is cyclical, but I do not think any cycle in the last, call it three, four, five years have been anything. You have seen the up, you have seen the down. We have all seen many ups and downs. That has a very different feel to it. Not because every cycle is different, but the technology and the breadth of technology is just very unique. You do not have typically a lot of secular growth, very aggressive secular growth that are all converging on a technology. Let me give you kind of where. And it is actually a good opportunity for us, and not just today, but over a longer period of time. It is the convergence around power. If you think about it, everything translates into gigawatts of power installed.
Whether you talk about AI data center, everybody measures the overall capacity in gigawatts. If you think about automotive and the kilowatt per car translates into a gigawatt deployment in automotive. Everything is measured. The unit of measure of the secular, the same thing in industrial, SSDs, ESS, it is all measured in gigawatt deployment. The fact that every secular growth market in every end market that matters is being measured in a unit of power is unique. It puts us in a much unique and a favorable market, because that is what we do. We do power, and we do power conversion, translating that gigawatt overall into whatever end use cases. You have seen us do it in automotive very successfully and get a dominant share, and you have seen us doing it in AI data center today, and we will continue to do that.
Industrial, I think, we talked about being number one in energy storage already. That is going to continue. We talked about a 40% growth, even in this environment. We feel very good about the cycle and what it means to a technology basis, and the fact that it lands in the technology domain that we are really good and strong at.
That is great. Speaking of secular growth, I think you highlighted that your AI data center business is going to more than double this year. I think that you had raised that a little bit from previously, just doubling. Can you just help us understand how you are participating in the AI data center power tree, right? You have got high voltage, Stage 1, Stage 2 power. Help us understand where you are benefiting along that power tree.
Yeah. Just for clarity, the way we define our AI data center power revenue is revenue we generate within the walls of the AI data center. So, I just briefly mentioned the energy storage systems. We put that in industrial. Although it is growing because of the deployment in AI data center, we keep that separate. Our AI data center, within the walls, is what is more than doubling this year. Where it is, historically, we have come at it from the high voltage. That is kind of our pedigree. The doubling is happening from last year's $250 million to this year to over $500 million, is actually across the power tree. Although we started at it from the high voltage, where we are today, we have gained share across the whole power tree.
The strength that we saw in the second quarter was actually across all of our content, not one piece did better than the other. That is actually a very good sign of a broader deployment. If you kind of consolidate the last few earnings call that we talked about, you have heard me talk about Flex Power, Delta, Lite-On, Great Wall, AWS, NVIDIA, and so on. Our approach is the proliferation and the breadth of not just platforms, but breadth of customer and geographies in order to maintain that very broad exposure, so you do not see the fluctuation overall. It is the same playbook we have had in automotive.
If you recall, when we started automotive, it was a concentrated position, and over the years I have always said, we have to have a geographical distribution and a customer deployment distribution. We have achieved that. We are doing that from the get-go in the AI data center business.
Right. I think on your most recent earnings deck, you had talked about your SAM and data center increasing from $15,000 per rack today to $115,000 per rack by 2030. Where is most of that incremental growth in the rack coming from for you?
Sure. If I break it down, what I would do is I break it down in high voltage and then medium and low voltage, just to separate that 800-volt transition, because that is what is driving the 115. If I break it down, the $15,000 today is about 30% high voltage. So a third of it is high voltage. That puts it about $5,000 of content in high voltage, $10,000 of medium and low. Primarily, of course, by the GPU, with the SPS and so on, last stage. When you fast-forward and you talk about the $115,000, it actually splits 50/50. So the high voltage is now $55,000, or 50% of that, while the medium and low is about the 55, give or take.
From a growth perspective and a dollar content, the high voltage is growing 10x. The medium and low is growing 5x. Still very healthy, both, but there's a bigger skew of content going into high voltage, which again, it's where we came from, as a company, and we'll continue to win share and grow in that side of it, while we continue to introduce new products, whether it's trench-based or other, on the medium and low voltage.
Great. I think you also recently announced some key wins with NVIDIA on the MGX platforms, and also with AWS on power supply and battery backup systems. How meaningful are these wins?
Look, I think every win in the AI data center is meaningful, as far as exposure to the end market. That goes back to the breadth of the approach that we have. We want to be as broad as possible because, at the end of the day, whatever the market does and however fast the deployment, people ask, "Well, how fast is 800-volt deployment going to be?" Well, it doesn't matter what it is, we still have a very good share in the existing platform. It will be incremental and higher in the new 800 volt, but it's not like we have to wait for the 800 to participate in it. Our approach is ensure continuity and ensure we always have the highest breadth of customer and platforms, more importantly, because you have now more not just training, you have the inference platforms coming out.
We talked about it last night, the agent platforms coming out. You have to have that breadth because you can't pick a side. It's all growing, and if you want to do power conversion, you have to be able to convert power regardless of the end use case.
Great. I think you've highlighted vertical GaN previously. Maybe just talk about where we stand in terms of development and ramp of vertical GaN. I think historically, from a technical perspective, vertical GaN has proven to be a very good technology. But the issues historically have been just being able to scale that into volume production. Maybe talk about where you guys are at there.
Yeah. Where we're at with GaN, we've sampled our high voltage vertical GaN. You think about it, monolithic 1,200 volts, so it's non-stacked die. Single die, 1,200 volts, high power. We sampled in AI data centers, and we sampled in automotive. Quals are progressing as expected, and the fab is-- We build in our own fab. All of the big hurdles that I would say is, can you manufacture it? Is it yielding? Does it qual? Is it performing? All of these are check the box. Right now, just like any technology, we're going through the technologies, but the products is out already.
Great. Maybe switching to M&A, I wanted to ask you about the Synaptics acquisition. You obviously have had a chance to talk to quite a few investors since the acquisition was announced. Maybe just come at it a different way, based on your conversations with investors, what do you think is the most misunderstood perception about the acquisition at this point?
Yeah, I think at this point, there's way less confusion than it was when we announced the deal, because one, people have had time to not just digest the deal, but also do work, and I'll explain a little bit what that work entailed, and also our communication and the last earnings call was the proof of one of the concerns that existed. Let me kind of give you at a high level. When we announced the deal, there were a few categories of investors, whether it's our investors or the Synaptics investors. But it fell into investors that are familiar with both companies, us and Synaptics today, were favorable from the get-go. I say the anecdote, I got emails before we even got on the call to talk about the deal. That was favorable from an awareness perspective.
Now, what surprised me, and in hindsight, of course, it could've been differently communicated, but what's surprising is how many people were not familiar with the new Synaptics. A lot of people were anchored on Synaptics, going, "Oh, what are you doing with the five, six years ago Synaptics?" None of this is true today. If anybody did the work on Synaptics, based on the transformation they've gone through over the last five years, it's a very different company. Now that this work has been done and continues to be done, the sentiment is much more favorable on Synaptics as a company.
The strategic intent that we talked about, there was a lot of concern at the beginning, which is kind of the bear case of, "Oh, my God, there's something wrong with the onsemi core business." We just delivered a much better quarter than, whether it's AI data center and so on, and we actually incrementally added to our position across the markets. That basically highlighted, and of course, the margin expansion that we had this quarter and talked about it, margin expansion throughout the year, all of these are signs of a very robust core business. It wasn't about, "Oh, my, is the Synaptics going to overcome some weaknesses in the core?" Or, "Is one or the other? Are we going to reduce focus on AI data center or automotive because" The answer is, we're coming at it from a position of strength.
We have proven the strength of our core business, not just in the reported quarter, but in our outlook. We have proven that it's the right asset. There's a lot of different conversations with some of my peers, even microcontroller peers, about thinking about, "Hey, we need an AI first compute platform." That tells you that we actually have the right strategy to do it. A lot of our peers have compute, so that gives us that competitive edge because we're not getting a microcontroller franchise, we're getting an AI first franchise. All of these make the strategic deal strategic.
Oh, by the way, between the synergies and really the health of the business and the margin of the Synaptics business, it comes with free cash flow that's actually going to fund the growth of the AI first compute, the Astra platform, rather than diverting R&D dollars from the core, which we're going to continue to invest in our core at the same level because we see the growth potential. Net net, it is still a favorable deal. It is still a financially favorable and strategic deal, and it does not change my view of the announcement or the deal. Since the day we've done it till now, maybe the communication, in hindsight, that would've been better learning there. From the strategic intent of the deal, it's going to deliver the value that we are here to deliver to investors.
Yeah, what I would add is it's accretive to our gross margin target today. We believe we can take our scale and leverage that platform. We can take our distribution network, our thousands of customers, and we can take that product broad very quickly, which is something they've struggled with. The other thing, over time, is we can think about our manufacturing footprint and say, "How do we improve those margins by bringing some of those new products in-house?" As they develop the next level of products. Accretive, cash flow positive, and scaling is really what we'll do with that business.
One thing I do want to highlight, when we announced the deal, we talked about $200 million synergies for the deal to be accretive in 18 months. I want to highlight that this is not the extent of the synergies. We put a trigger of 18 months. Everything that Thad talked about the margin expansion, the manufacturing, bringing manufacturing, which is all incremental to their margin and incremental to the benefit of the combined company, all that is synergies that happen outside of 18 months. You can think about the $200 million is kind of at that 18-month line, but it doesn't stop at that. All of the distribution, one design cycle, you're outside the 18 months, but that's all revenue synergies that kind of complement our growth and theirs.
That's where the one plus one is more than two, not just for the $200 million synergy. That just gets us to an accretive deal in 18 months. It's going to continue with the scale that we bring and the technology that they bring.
Good. Do you have a sense of right now how much of their products you're going to be able to take in-house and get those COGS benefits?
It won't be anything non-Astra. Astra is an advanced node. We don't do that internal. Everything else is a candidate, I would say. I say a candidate because if you have a business that's going to be replaced by a new generation next cycle from Synaptics, we may introduce the next one in and keep the older one. All the technology nodes they make, we can do internal. The question is, from a cost and CapEx and so on, it has to be favorable overall. That's part of the integration planning we're doing now. From an overlap perspective of what we can, a high percentage of the non-Astra business can go in.
Great. Looks like we're out of time. Thank you, guys.
Awesome. Thank you.