Okay. Welcome to the Microchip fireside chat. My name is Mark Lipacis. I'm the senior semiconductor analyst from Evercore ISI. We're very honored and excited to have Steve Sanghi, the Chairman and CEO of Microchip, and Eric Bjornholt, who is the Senior Vice President and CFO. I think we'll just jump right into it. One of the biggest questions that we get is, how big is any one of our company's data center and AI exposure? You guys had a press release this week, which kind of spelled this out. You talked about two different segments here, and there's a Venn diagram of the data center and compute end market, which is 18% of your revenues, and then data center solutions business unit, which was $300 million in 2025, expected to grow 65% to $500 million in 2026.
Steve, maybe if you could help us understand, what is the difference between these two categories that you broke out for us?
Yeah. Thank you, Mark. I think many of the investors we have talked to were a bit confused, and they felt that the $303 million that we broke out was the total size of our data center exposure, and that is really not correct. Let me take a chance to clarify it. The end market that we break out every year is data center and compute end market, and that is 18% of our total business. When you do 18% of our calendar year 2025 sales, that comes out to be $787 million business. That is data center and compute. Now, out of that $787 million, our data center solutions business unit, which makes products 100% for the data center-only market, was the $303 million, which leaves the $484 million unaccounted for. Some of that is data center, and some of that is compute.
That is harder to break out because that has our catalog microcontrollers, our catalog analog products, mixed signal products, memory, security, timing, temperature sensors, fan control units, all that kind of stuff that goes in PCs, it goes in data centers, it goes into power supplies, and they could be for data center or PC. A portion of that $484 million is also in the data center and therefore the total data center exposure is at $303 million plus a significant portion of the $484.
Got you. The $303 million, that is products that you make specifically for the data center. Is it fair to say those are ASSP kinds of products or?
They are. They include PCI Express, they include storage controllers, storage accelerators. They include the storage memory products, PCI Express memory controllers, CXL memory controllers, all those parts.
Got it.
Exclusively made for the data center. You have huge amount of exposure to the data centers from catalog products.
Got you. Okay. Now you talk about a 65% growth for that business, the data center solutions business, the products that are specifically made for the data center business. You didn't talk about the growth of the broader category data center and compute end market. Is it fair to say that in aggregate, that 18% of revenues would be a higher growth business for you than what you would expect for your corporate average? Is that fair to say?
We didn't try to estimate the 787, because 787 is a mixture of it. A portion of those are microcontrollers that are going to laptops and loading docks and others, and their growth rate is nowhere close to the data center growth rate.
In that, there are some high-growth products for the data center, but there are also some low-growth products for PCs and docks and printers and all that, and we didn't try to estimate the overall growth rate of that. Year-over-year, our end market mix hasn't changed that much.
Fair enough. Okay. Now in the press release, I believe you also discussed price increases. We do channel checks every quarter, and this past quarter, we identified 16 separate analog companies that took their prices up. You were highlighted as one of the few companies that did not. The question is, why did you wait so long to do so?
It's a very good question. I think there are two kinds of products in the industry. There are commodity products where they're easily exchangeable from one supplier to the other, like the memory products. On those commodity products, often you can increase the price when the supply is tight, and then purchasing manager will beat you down when the supply is excess and the costs go down, and the ASP goes down. There are second kind of products which are proprietary products, which is what we make: microcontrollers, analogs, FPGA, data center products, and others. Historically, in our industry, prices have not gone up for the proprietary products. There is an unwritten promise to the customer that if you design with my proprietary products, you can safely think that I can deliver this price for the length of that design.
COVID was the first time in my memory, post-COVID, where the shortages were so acute that people took advantage and raised the prices on all the proprietary products also. There were also a high amount of input cost inflation at that time, so it became necessary to pass some of those price increases. That was really the first time. Historically, prices have always gone down, maybe low single digit. This is now the second time after COVID, where some of our competitors have raised prices just because they can. Now, we raised prices three times post-COVID, and we managed to damage the relationship with the customer. Some of them were upset. Pricing was not the only thing.
We also had a very unpopular PSP program, which forced the customer to take the products when they no longer wanted the product, then they got left with a year, two years' worth of inventory, which they have been bleeding for the last three years, finally, it's coming close to correction. We came to an era of significantly negatively affected, damaged customer relationships. When I returned back to this job in November of 2024, for the past 18 months, we have visited thousands of customers around the world and are repairing our relationship, which I consider was a very successful project. Our relationships are good. We're winning business. We're gaining market share.
Therefore, we have taken all the input costs that have gone up so far in the first half of the year, and they're mostly internal cost, people cost, gold, copper, some raw materials, and others, and we have absorbed them without giving a price increase to the customer. In fact, despite that, our gross margin has gone up 950 basis points in the last year. We were pretty pleased with that, and we are in the striking range of a long-term target of 65, and while we were repairing the relationships, we took the high road, and while others were raising prices, our customers were praising us how we were doing. Now why the switch? During this first half, while we had some internal cost increases, we absorbed them. We did not get external cost increases from our foundries and assembly test partners.
We were able to hold them at bay. Finally, we can't hold them at bay. Now we have price increases with effective dates and percentage wafer cost or assembly test costs is going up in the second half of the year. Therefore, we felt prudent that we can no longer do that, and we needed to take those costs and pass it to our customers. What we're doing is we're taking the cost we have absorbed in the first half, plus the new cost increases that are coming in the second half. We have totaled them up and we'll margin it up and distribute it among our customers. Considering we have absorbed the first half cost ourselves, it will be slightly margin accretive. The goal is really to be margin neutral.
The fact that it's a very complex equation across thousands of customers and 100,000 SKUs, we do not know how to hit the bullseye. We're going to hit on the right of it and not allow it to go on the left of it. Therefore, it could be slightly margin accretive.
Got you. I appreciate that you don't want to give the specific variance, even though I know, I'm going to guess you have the variance, but how were you able to absorb the cost? What did you do to absorb those costs? Your gross margins went up despite higher input costs.
We had very large inventory reserve charges. We had very large underutilization charges. I believe, three or four quarters ago, I seem to remember the number, we had $122 million of inventory reserve plus underutilization charges in one quarter. I believe that was the June quarter last year. As our business started ramping and as we cut production in the fab, our inventory charges came down by a huge amount. These internal cost increases was a small fraction of it. Our gross margin was rising while we absorbed those small amount of costs.
Got you. What is the impact to the customer relationship? As you held off raising prices and to me, it was remarkable to see 16 companies that we identified, and I'm sure there's more, that they raised their prices. Does that help you win some share back?
It did. It did. We have thousands of customers coming back to us. We've gotten praise from our customers, how we treated them. We told them we were not increasing the prices, and they were very happy with it. One, it helped us improve the relationship, which we already were doing it through other means. When the competitors started raising prices, it accelerated our customer relationship improvement, and we're winning new designs. We're winning every push because customers think others are raising prices and we're not. The question obviously comes now as we also raise the price, how would that go? Number one, our price increase would be small. Number two, it will be very justified.
We are passing on our cost increases. Customer will understand. Competitors have raised prices, some of them more than once. We held them off in our first half. We're going to raise some in the second half. Most of these will not be effective till probably in September. Mostly effect will be in the fourth quarter. I think customer will really appreciate. It will not damage our relationship.
That's interesting for me because our checks indicated that those 16 companies were effectively raising them instantaneously.
Yeah.
It sounds like you're still like a six-month window where you're not raising them relative to the competition out there.
On proprietary products, you can essentially raise the price and put the gun to the customer's head and force it.
Doesn't mean you have to do it.
Right.
You can do it, but it doesn't mean you have to do it.
We have decided not to do it.
That, in part at least, is driven by your desire to rebuild the goodwill with your customers.
Yeah.
If we look at the data center solutions business, you mentioned the three buckets, like storage controllers, PCI Express, CXL, and switches and retimers. Rough numbers, are these equally sized businesses, roughly? Is there a skew on one group versus the other?
They are equal size, wide brackets.
They're roughly equal. One third in the storage controller area, one-third in the memory controller area, where we have the solid state disk drive controller, as well as hard disk drive controllers, and some CXL, and then the PCI Express. The retimer revenue is zero.
Haven't even put the first part in production.
Got you. Can you drill down on the PCI Express Gen 6 product that you have? I believe that you had mentioned your market share on the Gen 4 product was pretty good. Gen 5, you missed a product cycle, and so quite a bit disappeared. Now you're coming back with Gen 6, and you seem to be optimistic there. Can you help us quantify, what did you lose, and what could you come back, and where's the confidence come from?
This business was acquired through Microsemi. Microsemi had acquired PMC-Sierra, and when we acquired Microsemi in 2018, we got that business. PMC-Sierra/Microsemi had done pretty good on Gen 2, Gen 3, Gen 4. Actually, Gen 4 happened on my clock.
After 2018, between 2018 and 2021. In all those generations, we had a good share of the market, was very good business, high profitability. During the COVID and post-COVID years, was the era of Gen 5, and Microchip was about two years late to production on Gen 5. Basically, we tried to design. I wasn't the CEO at that time. I'd kind of stepped down, so it was a different management team. We tried to design SerDes inside, and every two weeks it looked like we were two weeks away. It kept on for too long, and nobody really pulled that trigger in saying, "Enough is enough. We need to get the product out to the market, and let's go license a SerDes from somebody else." Which is what I did on Gen 6, and that's what we're doing for Gen 7.
When we were two years late to the market, we basically lost the market. People designed it with whoever had the part. Eventually when we came out with Gen 5 two years late, we won some designs in the last year and a half, but they're all in the second source position because the pole position had gone to somebody else. What I have generally said is we lost several hundred million dollars of business per year on that Gen 5 alone.
We haven't quantified exactly, but it was several hundred million dollar. It was a very large loss, and that way we can never allow it to happen again. When we came on Gen 6, now we have one of the best parts in the industry, the only 3 nm part in the industry on PCI Express Gen 6. We now have eight design wins. One of them is a $100 million per year design win that we have talked about. Every customer that we had in Gen 2, Gen 3, Gen 4, and including all the customers that we lost in Gen 5, they're all evaluating our chip.
Because it's a better chip in the market. Now, the question is, how many of those we will get back when we hurt them on Gen 5? Because many of them were waiting on Gen 5. We made commitments that we didn't meet, we withdrew the part.
From the market because we didn't have it. We got some amending to do, and so therefore we are conservative in the amount of penetration we can make on Gen 6.
We think we should do better than anybody's saying, but we're not quite willing to say that yet.
Well, if you have one design win at $100 million annually, it seems like you're climbing some of that back.
Yeah.
Climbing some of that back.
It seems like in the data center market, winning a $500 million design is not too much.
That would be a lot for me. I would pale.
I'm looking for my billion-dollar design win.
Right. Can you help us understand, to what extent is this a product that ships into data centers that are enterprise-oriented versus hyperscalers? I think there's a view that it is not a hyperscaler product per se. What can you tell us about.
It is a hyperscaler product. Every hyperscaler, every server manufacturer, and lots of semiconductor manufacturers that build boards by bundling their CPU, GPUs together with PCI Express, they're all looking at the part, and they're all either customers or potential customers of PCI Express. One point I wanted to add is, over the last two couple of years, two or three years, while it was the era of training the large language models, the GPU to CPU ratio was like 10: 1.
As that is shifting towards more inference, the GPU to CPU ratio is dramatically changing in the favor of CPUs. Most people are saying it's at least 1: 1. I've also read people saying it's one GPU to two CPUs. Therefore, when in the data center, the percentage of GPUs to CPU moves in the favor of CPUs, it's a bonanza for PCI Express because PCI Express connect GPU to CPU to memory, CPU to any peripheral, GPU to any peripheral. When there are lots of GPUs, the GPU to GPU is through NVLink. GPU to anything else or CPU to anything else is often through PCI Express. Therefore, any estimates that we have seen so far on the size of the PCI Express market, and I don't have any numbers to quote, would be found underestimated.
Got you.
Therefore, it's a big opportunity for us.
I will say, our own checks indicate, when we talked to about a couple dozen hyperscalers, that CPU to GPU ratio goes to 1: 1 or 2: 1, and that's kind of where the average is landing, between 1: 1 and 2: 1. We also have learned, this week that, with agentic, that the agents are calling more database workloads, and that's causing more demand for your standard CPUs also.
It's not just an agentic CPU idea, it's agents cause demand for others, the standard CPUs also, which is quite interesting.
Those CPUs will require PCI Express to access memory, storage, RAID cards, peripherals, anything else.
Got you. Okay. I want to shift to the aerospace and defense market, which I believe is one of your fastest-growing businesses. Was there anything else on the data center business you think is important for investors to keep in mind?
Anything else on there?
I don't think so.
That we covered. We covered it. Okay. Got you. All right. This was, I believe, it made up 18% of your revenues in fiscal 2026. Can you talk about the segments here? What's driving it?
There are three segments in aerospace and defense. There is aviation, there is military hardware, offensive and defensive, there is space. There are things happening on all three. For a while, the aviation production was way down when the MAX 750 planes were not really being manufactured. Now Boeing has the largest backlog ever, and they're building planes like crazy. We're in other planes, too, French planes and others. Parts are loaded with Microchip's products, all these planes. That part of the business is doing very well. When you look at the number two, which is the military hardware and software, first, it is the largest among the three. We used up about half of the arsenal we had built in 20 years in two months of war.
They're trying to rebuild it, and Trump has asked all the primes to ramp their production 4x. To ramp the production 4x, many of them have to build new buildings and new factories. This is not a short-term kicker, but it's a very long-term sustainable growth from the military hardware buildup.
Those primes are coming to you.
Those primes are coming to you. I've taken direct phone calls from CEOs of primes, saying, "What is your capability to ramp up?" I'm asking, "Give me the order." They just say, "Well, are you ready?" We're ready to do what? We make thousands of products. You have to give me specific product. How many do you want? Is it missile? Is it plane? So the thing is that we are in every single military offensive and defensive hardware. We're in every missile, every plane, every battle tank, every rifle, every radar installation, every interceptor, every drone. It's very broad. There are hundreds and hundreds of products that go into that structure. They really need to work either by a forecast or by order on specific parts.
We're getting those orders, but majority of them we haven't gotten yet because it's still being worked, and they're really getting their own act together because our product is not the only product needed. It starts with ammunition and metal and housings and all that, casings.
You need a complete kit.
Yeah, need a complete kit.
That furnace.
They got challenges all over the place in being able to ramp it. That part of the business looks very good. Third piece of that is space. America has new fascination with space that I haven't seen since the Apollo years. We want to go back to moon and land a man on the moon in 2028. The recent lunar mission where we went around the lunar orbit, it was loaded with Microchip product. We have multimillion dollars of product in that mission. We'll have in every mission, I think there are two or three missions before we land a man on the Moon. Then we want to go to Mars, and we want to colonize Mars. I don't know why, but as long as I'm making money, I'm okay with it.
You have the fourth piece of that, which is the new space. These are low Earth orbit, the LEO.
Where SpaceX and others are trying to build a constellation of these satellites. That market is a little different because they're not trying to send them up for the last 50 years like NASA is trying to do. If they last five years, then come back in and burn up in atmosphere, they just launch another thousand of them. They just do it in volume. They do it cheaply. It's a different business model. Many times they're using lower-grade parts. They're certainly not using deep space parts. Sometimes they're using the radiation-tolerant parts. Other times they're simply using the automotive-grade parts and using them in parallel with high amount of redundancy. If one fails, the other one will work, and things like that. We are participating in that too, and we don't know how large that market becomes.
When I think of this business, because you have to build in the redundancy or the concept of rad hard packaging, I think of these are higher gross margin businesses. Is that fair?
The space part of it, which is radiation tolerant as well as radiation hardened, are very high gross margins. Some of the highest. They're just incredibly high gross margin because they're very hard to make and takes a long time to make them. Then you have to build the parts and then take a sample and burn them in and test them with radiations for a period of time, and then they pass, then you can ship the lot. It's like a nine months to one year production cycle. The parts that are going into LEO, especially if they're not radiation hardened, if they're automotive-grade parts, then they're just off the shelf. They can be bought from distribution. That model is different.
Got you. Some companies talk about, "We'll give a TAM per satellite." Is that something that you have shared in the past?
No, I don't know what that is. Yeah.
Okay. Got you. Okay. I know that you are a student of the cycle. I want to get your perspective on how do we think about where we are right now. Do you think your customers downstream from you, not just one layer deep, but multiple layer deeps, where do you think they are in the inventory restocking cycle? Are they below normal? Are they normal? Are they above? Are they trying to restock? Where are we in the cycle?
We had a lot of inventory post-COVID because of the PSP program. We didn't allow customers to push out or cancel the parts they no longer needed. As a result, we had really up to one year to two years of inventory. A lot of customers and distribution inventory was more than twice of really what would be normal. We have spent the last two, two and a half years now cleaning up that inventory. The distribution inventory is now normal. Actually, I could say it's slightly below normal, but it's not perfect. Some distributors are below normal, some others still have a little high inventory, so it's kind of out of mix also. The customers is a very long-tail customers, 110,000 customers.
Hard to know who has what, seeing that thousands of customers who were no longer buying the part because they were using inventory and now they're buying the part and customer count is increasing by the thousand, is a very good indication that thousands of customers' inventory is coming down. They're reengaging to buy the part. I think that part of the cycle where the inventory correction happens is largely done, could have another quarter or so left, and then the customers are now buying to their consumption. Now, at the same time as they're buying to their consumption, volume has gone up significantly. We're up 34% versus the same quarter a year ago. In that situation, and with the AI growth, AI is crowding out some of the assembly test capacity at some subcontractor, some of the processes also.
As those shortages start to materialize and work in the system and lead times start to go up, I think that will start the real upcycle of restocking, which in our business hasn't begun yet. Maybe it's begun in AI and all that, automotive customers are not rebuilding inventory. Industrial customers are not rebuilding inventory. Appliance customers are not rebuilding inventory. Would they build in the second half? Maybe, they're not building yet because our internal inventory is still coming down. Customer inventory has just corrected. Some of them haven't even corrected yet. It's early part to use the word restocking. I don't think it has begun yet. It will begin later. Now what is happening on the top of the inventory-driven cycle is the innovation-driven cycle, and innovation-driven growth is layering on the top of that.
You're seeing the innovation-driven growth in data centers, we talked about. You're seeing that innovation-driven growth in aerospace and defense, we talked about. We're seeing the innovation-driven growth in automotive. We haven't talked about automotive, and we're seeing it in industrial. Industrial, as the on-shoring is happening and factories are moving from China to here, they're not duplicating the Chinese labor-intensive factories in U.S. They're automating them. There are moving arms. There are robots. There is conveyors. Factories are really being retooled for lower number of people working in the factory and high amount of automation, which means they're loaded with our products, with sensors and controllers and PCI Express or USB and Ethernet connectivity and all over. There are this stuff happening in the industrial sector. There are things happening in medical.
Advanced medical equipment is loaded with semiconductors, open-face MRI systems where you don't have the claustrophobic feeling that they push you in for 40 minutes inside that tube. All these are innovation-driven growth things that we're seeing in the industrial market. The automotive market, what's happening in the automotive market is the unit growth is low single digit, but the content growth is much higher. We got 81 chips in a Hyundai Genesis. We have 61 chips in Mercedes S-Class cars and on and on and on. The number of parts are multiplying, and what's happening in the cars is today there are six or seven distinct protocols for connectivity. There is a Controller Area Network which is called CAN bus. There's a LIN bus. There's an entertainment bus that's called MOST. There's USB connectivity in every car. There's Ethernet. There's RS-232. There's RS-485.
When the car wasn't a connected car, then these protocols can reside in various modules themselves because they're not talking to anybody. If you want to be able to press a button in the cockpit, like I can do in my Tesla and saying, "What is my tire pressure?" Three seconds later it shows a tire pressure of all four tires on my screen. You're connecting to every part of the car. When you're connecting to the car and you have multiple protocols, then you need bridges to convert signal from one protocol to the other so you can centrally access it on whatever your screen is connected to. That is expensive and then it's a complex design. What automotive manufacturers are trying to do is converge all those protocols to a single Ethernet-based 10BASE-T1S protocol that runs 10 MHz per second.
It's fast enough for the automobile. It's not your office Ethernet wire which is an eight-strand wire and can carry 10 gigabit. The wire we're talking about is single-pair, twisted pair, two-strand wire, and it can carry 10 Mb per second for the length of the car. That is good enough. All those protocols are converging to this single standard and we're leading the standard, winning virtually every design. These are 2028 production year starts, some in 2027, more in 2028 and 2029. That's the innovation-driven growth happening in the automotive. I think that's how I will describe the cycle where there is inventory cycle-driven growth which is a normal semiconductor and then there is just incredible innovation in these four segments which seems like it's below normal innovation activity. It has really accelerated. With the help of AI tools, customers' design cycles are shrinking.
Our design cycles to deliver these innovative products are shrinking. Time to write software is shrinking. It's very productive delivery of this innovation that should drive revenue and be a very profitable revenue. Exciting time.
We have a cyclical dynamic here from my standpoint, our own. We debated this yesterday, right? Looks to us like you are still shipping below your trend line and if your customers are shipping to demand but they haven't started restocking, I would say that's consistent with our analysis. You have to bring your inventories back to a normal safety stock to get there. Then there's this secular dynamic and so it seems like you have innovation cycles on four dimensions. If I think about the past, I would say, oh, you have an innovation cycle in handsets and then, or before that in PCs.
It seems like you have four and I'm wondering if you think about the industry which has grown about 5% or 6% over time, if you have AI and automotive innovation and industrial innovation, is there a chance, do you think about this opportunity for that long-term growth trend line for your industry or for Microchip to go higher than that historical 5%-6%?
I mean, absolutely. We completed a five-year strategic long-range plan in January putting all this understanding into it and I asked our 20 business units to give me a long-range plan that you can sign in blood means it's conservative. They all did that, and two quarters have gone by. We're well ahead of what they put together, which means that they were conservative. Even the conservative numbers they came up with for the next five years, its growth was pleasurable. It was good. It was much higher than.
I guess you're not going to share that.
No, we're not going to share that. It's much higher than the numbers you talked about.
If that is conservative, that's even more music to my ears.
We have a little bit of time left. I don't want to let Eric off of the hook.
Okay.
I want to ask him one, and I got one last one for you. Eric, assuming we are in the place in the cycle that it seems like we're on the same page on, what are you going to do with the cash, the free cash flow that you generate?
Obviously, we're going to continue to pay the dividend at the levels that it's at, but we are still really focused on paying down debt and getting our leverage down. We got into trouble in last cycle when the EBITDA fell so far, and with that, we had to actually come to the market with an equity instrument last March during this mandatory convertible preferred. We don't want to find ourselves in that situation again, so going to maintain the dividend where it's at, not do stock buyback, generate cash, pay down debt, and get the leverage down. This quarter, we will end with net leverage below three. It's not where we want to be, but much improved from where we were, and I think we should make rapid progress through the rest of the year.
Got you. Okay. Fair enough. Steve, last question. You, as much as any CEO, talk to investors, your shareholders. What do you think is the biggest disconnect or misperception about Microchip versus how you understand the fundamentals of the company and the opportunity?
I think, somewhere close to 60%-75% of the time gets taken by the data center talk, and they just seem to not be able to get off that topic. In most meetings, automotive doesn't come up, industrial doesn't come up. Aerospace and defense, when they get off data center, then start talking about SpaceX and others. Investors are very focused in these couple of segments and sometimes they miss the beauty of Microchip's broad customer base, broad end market exposure into definitely four of these markets that are doing really great. Clients market is kind of so-so. I think that's sometimes they kind of miss out and any remaining time gets spent on trying to push us on why the margins should go higher than 65%.
I've heard that on the earnings calls.
Yeah.
All right. Well, fair enough. Well, listen, we're at time. Steve and Eric, thank you for joining. Thanks for sharing the insights about Microchip and the industry. Really enjoyed the conversation.
Thank you, Mark.
Thanks, Mark.
Thank you.